National Academies Press: OpenBook

Inner-City Poverty in the United States (1990)

Chapter: 4 The Social Consequences of Growing Up in a Poor Neighborhood

« Previous: 3 How Poverty Neighborhoods Are Changing
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 111
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 112
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 113
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 114
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 115
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 116
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 117
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 118
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 119
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 120
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 121
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 122
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 123
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 124
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 125
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 126
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 127
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 128
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 129
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 130
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 131
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 132
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 133
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 134
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 135
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 136
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 137
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 138
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 139
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 140
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 141
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 142
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 143
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 144
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 145
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 146
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 147
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 148
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 149
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 150
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 151
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 152
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 153
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 154
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 155
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 156
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 157
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 158
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 159
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 160
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 161
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 162
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 163
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 164
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 165
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 166
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 167
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 168
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 169
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 170
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 171
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 172
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 173
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 174
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 175
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 176
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 177
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 178
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 179
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 180
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 181
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 182
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 183
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 184
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 185
Suggested Citation:"4 The Social Consequences of Growing Up in a Poor Neighborhood." National Research Council. 1990. Inner-City Poverty in the United States. Washington, DC: The National Academies Press. doi: 10.17226/1539.
×
Page 186

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

4 The Social Consequences of Growing Up in a Poor Neighborhood CHRISTOPHER JENCKS AND SUSAN E. MAYER INTRODUCTION Children from affluent schools know more, stay in school longer, and end up with better jobs than children from schools that enroll mostly poor children. Children who live in affluent neighborhoods also get into less trouble with the law and have fewer illegitimate children than children who live in poor neighborhoods. Similar patterns are found when we compare white neighborhoods to black neighborhoods. These patterns have convinced many social scientists, policy analysts, and ordinary citizens that a neighborhood or school's social composition really influences children's life chances. But this need not be the case. The differences we observe could simply reflect the fact that children from affluent families do better than children from poor families no matter where they live. Similarly, white children may fare better than black children regardless of their neighborhood's racial mix. In order to determine how much a neighborhood or school's mean socioeconomic status (SES) affects a child's life chances, we need to compare children from similar families who grew up in different kinds of neighborhoods. This study examines what social scientists have learned from studies of this kind. We give considerable attention to the policy implications of the studies we discuss. Many observers (notably ~ Wilson, 1987) believe that when poor children have predominantly poor neighbors, their chances of escaping from poverty decline. If this is so, a strong case can be made for govern- mental efforts to reduce the geographic isolation of poor children. Yet such evidence as we have suggests that the poor or at least poor blacks are becoming more geographically isolated rather than less so (Jargowsky and Bane, in this volume; Massey and Eggers, 1990; Weicher, in this volume). 111

112 INNER-CITY POVERTY IN THE UNITED STATES At present, the main goal of federal subsidies for low-income housing is to build as many low-income units as possible for as little money as possible. The best way to achieve this goal is usually to build in low- income neighborhoods. As a result, federal subsidies are quite likely to increase economic segregation. If the federal government wanted to reduce economic segregation, it would either have to help poor families move to better neighborhoods or encourage more affluent families to remain in poor neighborhoods (perhaps through mortgage subsidies). In assessing these policy alternatives, it is important to ask how they will affect both rich and poor children. We cannot answer that question, because social scientists have not yet accumulated the information we would need to answer it. The best we can do is summarize the available evidence and offer some guidelines for interpreting it. We focus on quantitative studies that try to separate neighborhood or school effects from family effects through statistical analysis of survey data. We ignore qualitative studies, not because we think them incapable of answering the question that concerns us but because we found no qualitative research that tried to answer this question. The ethnographic studies we reviewed never tried to compare children from similar families who lived in different neighborhoods. Nor did they follow families as they moved from one neighborhood to another, describing how the moves affected the children. As a result, they cannot help us disentangle neighborhood or school effects from family effects. Our definition of a "neighborhood" is very broad. We include elemen- tary school attendance areas, which usually coincide fairly closely with what people mean by a neighborhood (hence the term "neighborhood schools. But we also include high school attendance areas, which are usually larger than what most people mean by a neighborhood. We include research on the effect of living in one kind of census tract rather than another, even though census tracts are much smaller than elementary school attendance areas. And we also include research that uses postal zip codes to define neighborhoods, even though zip code areas are likely to be somewhat larger than a traditional neighborhood. Although our definition of a neighborhood is broad, it is always ge- ographic rather than social. We have not tried to review the effects of nongeographic communities of various kinds, such as friendship networks. Nor have we tried to review the work of social psychologists on the way "so- cial context" affects behavior. Readers should not interpret these omissions as an implicit judgment that nongeographic communities or social contexts are less important than geographic communities. The available evidence suggests the contrary. When placed in a room with a group of stooges who claim that the longer of two lines is the shorter, for example, most experimental subjects will reject the evidence of their senses and agree with

oROwING uP IN A POOR NEIGHgoRHooD 113 the stooges (Asch, 1951~. This demonstrates that individuals seldom defy the unanimous opinion of others, at least in the short run. The relevance of this fact to the study of geographic communities is minimal, however, because geographic communities are never completely homogeneous. The experiment just described shows that homogeneity is crucial, for when even one stooge concedes that the longer line is indeed longer, most subjects give the correct answer. When the opinions of others vary, in other words, individuals do more than just count noses and espouse the news of the majority. The principal conclusion we draw from work like Asch's is that the way social context influences individual behavior varies with the problem the individual confronts, his or her experience, and the mix of opinions and role models available in a given social context. This variability makes it almost impossible to generalize from laboratory experiments to neighborhood or school settings. There are currently three schools of thought about how the social composition of a neighborhood or school affects young people's behavior. Most Americans assume that advantaged neighbors or classmates encourage "good" behavior. A few assume that advantaged neighbors or classmates encourage "bad" behavior. And some assume that advantaged neighbors or classmates have no effect one way or the other. Each of these three schools of thought is compatible with a varietr of theories about the mechanisms by which neighborhoods and schools influence individuals. We take up the three theories in turn. The Advantages of Advantaged Neighbors Most Americans assume that children who grow up in a "good" neigh- borhood are more likely than those who grow up in a "bad" neighborhood to work hard in school, stay out of trouble, go to college, and get a good job when they become adults. Social scientists have suggested three mech- anisms that could produce this result: peer influences, indigenous adult influences, and outside adult influences. Those who emphasize peer influ- ences usually construct what we call epidemic models of how neighborhoods affect individuals. Those who emphasize the role of indigenous adults con- struct what we call collective socialization models. Those who emphasize the role of outside adults usually construct what we call institutional models. Epidemic models focus on the way in which peers influence one another's behavior, and they assume that "like begets like." If children grow up in a community where a lot of their neighbors steal cars, for example, the children will be more likely to steal cars themselves. Conversely, if children grow up in neighborhoods where all their neighbors finish high school, the children will feel obliged to finish school themselves. Because

114 INNER-CITY POVERTY IN THE UNITED STATES "bad" behavior is more common in poor neighborhoods, epidemic models predict that, if we compare children from similar families, those reared in poor neighborhoods will behave worse than those reared in affluent neighborhoods.) Many writers assume that bad behavior is contagious, but few examine the implications of this idea in detail. Many seem to assume, for example, that each school or neighborhood has a single dominant set of norms, to which every child, or at least every teenager, tries to conform. The dominant norm about any given form of behavior derives, in turn, from observing what others do. If "most" teenage girls in the neighborhood wear short skirts, then "every" teenager wants a short skirt. Similarly, if most teenage girls have babies before they marry, every teenage girl wants one. If this simple notion were correct, however, all neighborhoods would end up internally homogeneous. Either every girl would have a baby before marrying, for example, or none would. 1b be convincing, epidemic models must allow for individual differ- ences in susceptibility to neighborhood or school influences. Epidemic models of antisocial or self-destructive behavior usually impute differential susceptibility to differences in upbringing, but the model works in the same way if we impute individual differences lo heredity or to chance. The critical feature of the model is that among individuals of any given sus- ceptibilit~r, the likelihood of antisocial or self-destructive behavior increases with exposure to others who engage in similar behavior. If children from low-SES families are more susceptible to such influences, increases in the proportion of low-SES families in a neighborhood will lead to exponential increases in bad behavior. Whereas epidemic models focus on the way in which peers influence one another, collective socialization models focus on the way the adults in a neighborhood influence young people who are not their children. Those who believe in this model (e.g., W. Wilson, 1987) see affluent adults as role models whose existence proves that success is possible if you work hard and there and throughout we use the adjectives "affluent," "advantaged," and "high-SES" as syn- onyms. Thus, when we refer to "affluent" neighborhoods, we mean neighborhoods that have a variety of social and economic advantages besides high family income. We also use the terms "affluent," "advantaged," and "high-SES" in a relative rather than an absolute sense. When we speak of "affluent" neighborhoods, for example, we often mean all neighborhoods that are more affluent than "poor neighborhoods," not neighborhoods that are more affluent than the national average. Likewise, when we speak of "high-SES" students, we often mean all students whose so- cioeconomic status is higher than that of "low-SES" students. As a result, "high-SES" students may merely be students whose parents hold steady jobs and earn average incomes, not students whose parents are high-level executives or professionals.

GROWING UP IN A POOR NEIGHBORHOOD 115 behave prudently. They also see affluent adults as potential "enforcers," who keep children from running wild on the streets, call the police when trouble occurs, and generally help maintain public order. Institutional models also focus on the way adults affect children, but they focus primarily on adults from outside the community who work in the schools, the police force, and other neighborhood institutions. Almost everyone assumes, for example, that elementary schools in affluent neigh- borhoods get better teachers than those in poor neighborhoods and that this affects how much students learn. Many people also assume that the police treat delinquents differently in rich and poor neighborhoods and that this affects a teenager's chances of acquiring a criminal record. If such assumptions are correct, a neighborhood's mean SES could affect children's life chances even if neighbors per se were irrelevant. From an empirical viewpoint it is often difficult to choose among these three models. All three predict that students will learn more when their schoolmates come from affluent families, for example. The institutional model attributes this to the fact that affluent schools have better teachers and a more demanding curriculum. The contagion model attributes it to the fact that affluent students serge as role models for the less affluent. The social control model attributes it to the fact that affluent parents force their children's schools to set high standards. When we look at real schools, the three models are hard to distinguish. Because this issue is difficult to resolve empirically, social scientists often try to resolve it ideologically. Conservatives tend to espouse contagion or social control models that focus on the way the poor affect one another's attitudes, values, or behavior. Liberals prefer institutional models because they shift responsibility for what happens in a poor neighborhood to middle- class outsiders. The work we review throws little light on this controversy. Almost all of it relies on a "black box" model of neighborhood and school effects that makes no assumptions about how social composition influences individual behavior. Models of this kind try to answer the question, How much would an individual's behavior change if he or she moved from a low-SES to a high-SES neighborhood or school? They do not purport to explain why moving has an effect. As a matter of literary convenience, we sometimes attribute hypothet- ical changes in individual behavior to neighbors or schoolmates rather than neighborhood institutions or school practices. Readers should treat this as verbal shorthand, not as an empirical judgment that the contagion or social control model is superior to the institutional model. What we describe as an effect of having affluent neighbors may be an erect of the neighborhood institutions that the affluent create for themselves and their neighbors.

116 INNER-C~ POURS IN THE UNFED STATES The Disadvantages of Advantaged Neighbors Epidemic models, collective socialization models, and institutional models all assume that growing up in an affluent neighborhood encourages children to do what adults want them to do: learn a lot in school, stay out of trouble, and get good jobs when they grow up. Models that emphasize concepts like relative deprivation, cultural convict, and competition for scarce resources imply, in contrast, that affluent neighbors often influence children's behavior in ways that most adults regard as undesirable. Relative deprivation models assume that people judge their success or failure by comparing themselves with others around them. If people want to know how well they are doing economically, for example, they compare their standard of living with that of their friends and neighbors. It follows that if their income remains constant, they feel poorer when they have rich neighbors than when they have poor neighbors. Likewise, a college dropout feels less culturally competent if his or her neighbors all have Ph.D.'s than if they are all high school dropouts. The same logic also applies to children. Children judge their economic position by comparing their standard of living with that of their schoolmates and neighbors. They judge their academic success by comparing their school performance with that of their classmates. Other things equal, low-SES children do worse in school than high-SES children. Low-SES children will therefore form a more favorable opinion of their abilities if they attend a low-SES school than if they attend a high-SES school. (The same is, of course, also true for a high-SES child.) Some children who do not compete successfully respond by trying harder; others drop out of the competition. The relative frequency of these two responses depends on a wide range of factors, which are not well understood. But if most young people eventually respond to poor academic performance by refusing to do any more work, moving them from a low-SES school to a high-SES school will not only lower their relative performance but also reduce their academic effort. As a result, moving a child from a low-SES to a high-SES school may also increase the child's chances of quitting school, becoming a teenage mother, or committing violent crimes. The theory of relative deprivation is a theory about individual psy- chology that purports to explain when people judge themselves successful and unsuccessful. It interprets deviant behavior as a by-product of these individual judgments. Theories that emphasize cultural conflict are similar in their underlying structure, but they focus on the way groups create a common culture. These theories suggest that when large numbers of indi- viduals are unable to do what society as a whole expects them to do (finish school, get a respectable job, create and support a family), they will try to create a common culture to deal with their common failure. This culture

GROWING UP IN A POOR NEIGHBORHOOD 117 will accept as "normal" and in some cases even praiseworthy what the rest of society regards as aberrant and reprehensible. If the creation of a deviant subculture is a collective reaction to relative failure, such a subculture is more likely to arise in settings where success is very unequally distributed. Deviant subcultures will therefore be stronger in neighborhoods or schools where the poor rub shoulders with the rich than in places where the poor only rub shoulders with one another. Competition for scarce resources can also make affluent neighbors a liability. We noted above that schoolchildren compete for grades and that the competition is tougher in high-SES schools. But the same logic applies when teenagers compete for jobs. In both cases a big frog in a small pond is probably better off than a small frog in a big pond. The Irrelevance of Advantaged Neighbors Strong individualists-especially economists-often assume that neigh- bors have no direct effect on an individual's behavior. They believe that people base their decisions on their own circumstances and long-term in- terests, not on their neighbors' ideas about what is sensible, desirable, or acceptable. Most anthropologists and sociologists, as well as many psychol- ogists, reject this view, arguing that individual decisions consist largely of choosing among a menu of culturally defined alternatives and that an indi- vidual's menu depends in part on the alternatives his friends and neighbors are considering. This "sociological" view need not deny that most people are rational utility maximizers. It merely denies that they are imaginative utility maximizers. Even if individuals restrict themselves to choosing among familiar alter- natives, however, a neighborhood's social composition may not have much effect on individual behavior. Most people prefer friends like themselves. So long as neighborhoods and schools are moderately heterogeneous, most young people can indulge this preference. Even in the poorest neighbor- hoods, a teenager can find friends who stay out of trouble, finish high school, go on to college, and get good jobs. And even the most affluent neighborhood has some teenagers who hate schoolwork, reject adult stan- dards of behavior, and get into the same sorts of trouble as teenagers in poor neighborhoods. Prospective troublemakers can therefore find cocon- spirators in a rich neighborhood, even though they are scarcer than they would be in the ghetto. There are, of course, some cases in which a neighborhood's social composition has a big effect on friendship patterns. Rosenbaum et al. (1986) found, for example, that poor black families who had been lured to white Chicago suburbs by Section 8 housing certificates reported that

118 INNER-CITY POVERTY IN THE UNITED STATES their children had more white friends after moving than before.2 This is an extreme case, however. In the absence of strong financial incentives such as those that lured these poor black families to the Chicago suburbs, families seldom move to neighborhoods where their children have trouble finding friends like themselves. Nonetheless, a neighborhood or school's social composition surely has some effect on a youngster's choice of friends, even when the neighborhood or school is somewhat heterogeneous. These contextual influences on friendship patterns must, in turn, have some effect on the alternatives that young people consider open to them. These effects may well be weak. Indeed, they may be too weak to deserve serious attention. But they are unlikely to be zero. There is, however, a plausible scenario in which the social composition of a school or neighborhood will not appear to affect individual behavior. Suppose that both the epidemic model and the relative deprivation model are partially correct. In such a world high-SES neighbors might have two offsetting effects, one positive and the other negative. If these effects were of roughly equal magnitude, a neighborhood or school's mean SES would not appear to matter at all. As we shall see, this is roughly what we found when we tried to disentangle the effects of a high school's mean SES on its graduates' chances of attending college. The remainder of this chapter proceeds as follows. In the next section we discuss six methodological issues that will recur over and over when we try to interpret the results of studies that assess the long-term effects of neighborhoods or schools on children's life chances. We then review the evidence about how a neighborhood or school's social composition affects children's eventual educational attainment, cognitive skills, crime rates, sexual behavior, and labor market success. In the closing section we summarize our findings and discuss their implications for those who do research and those who finance it. PROBLEMS IN MEASURING NEIGHBOREIOOD EFFECTS ON CHILDREN Anyone who wants to make policy inferences from the currently avail- able studies of neighborhood or school effects confronts two difficulties. First, it is hard to be sure whether the causal inferences that social scien- tists make from survey data are valid. Second, even if those inferences are 2 pursuant to a finding that the Chicago Housing Authority (CHA) had deliberately segregated its public housing projects during the 1950s and 1960s, the Gautreaux decision ordered the CHA to provide some black public housing residents and applicants with Section 8 housing certificates that could only be used in white areas. Rosenbaum et al. (1986) studied families with children who had volunteered to move in order to get those subsidies.

GROWING UP IN A POOR NEIGHBORHOOD 119 valid, they are seldom sufficiently detailed or precise to predict the effects of specific public policies. Because these general difficulties recur over and over in the studies we review, we discuss them here rather than rehearse them throughout the chapter. We begin with the problems of making causal inferences from survey data. Controlling Exogenous Influences Perhaps the most fundamental problem confronting anyone who wants to estimate neighborhoods' effects on children is distinguishing between neighborhood effects and family effects. Family characteristics exert a major influence on children's life chances no matter where a child lives. A family's characteristics also influence where it lives. This means that children who grow up in rich neighborhoods would differ to some extent from children who grow up in poor neighborhoods even if neighborhoods had no effect whatever. From a scientific viewpoint, the best way to estimate neighborhood effects would be to conduct controlled experiments in which we assigned families randomly to different neighborhoods, persuaded each family to remain in its assigned neighborhood for a protracted period, and then measured each neighborhood's effects on the children involved. Fortu- nately, social scientists cannot conduct experiments of this kind. In their absence, social scientists rely on surveys that collect information on both family and neighborhood characteristics. They then compare children from apparently similar families who live in different neighborhoods. This kind of statistical analysis poses several problems, however. First, we must decide which parental characteristics are exogenous and which are endogenous. (Family characteristics are exogenous if they do not depend on where the family lives. They are endogenous if they change when families move from one neighborhood to another.) There is no simple formula for deciding whether a family characteristic is exogenous. Many people believe, for example, that neighborhoods affect their residents' job opportunities. If this is true, conventional measures of parental SES, such as father's occupation and family income, are partly endogenous. Some part of what we attribute to parental SES may therefore be traceable to the neighborhood in which a family lives. But while neighborhoods may have some effect on adults' job opportunities, no one claims that they explain a large fraction of the total variance in adults' occupational status or income. (We review this literature in Chapter 5 of this volume.) It follows that estimates of a neighborhood's effect on children will be far less biased if parental SES is controlled than if it is not. Similar arguments apply to family composition. As we show below, the neighborhood in which a teenage girl lives affects her chances of having

120 INNER-CITY POVERTY IN THE UNITED STATES a child out of wedlock. Neighborhoods may also influence marriage and divorce rates. This means that both the number of children and the number of adults in a family depend in part on where the family lives. But no one has argued that neighborhoods have anything like as much influence on family composition as family composition has on where people live. Thus, if we want to estimate a neighborhood's impact on children, we will get less biased results if we compare children from families of similar size and structure than if we treat family composition as endogenous. We have restricted this review to studies that control at least one measure of parental SES when estimating neighborhood or school effects on children. But the studies we review seldom include all the standard indicators of parental SES (mother's and father's education, mother's and father's occupation, and family income) or family composition. Omitting or mismeasuring these family characteristics tends to inflate neighborhoods' estimated effects on children, because a neighborhood's mean SES is a partial proxy for unmeasured variation in individual SES. At present, we have no idea which specific family characteristics we must control in order to get relatively unbiased estimates of neighborhood effects. Such information is crucial for assessing the likely degree of bias in studies that include only one or two measures of parental SES, as most studies do. Longitudinal Versus Cross-sectional Models A second possible way to estimate neighborhood effects would be to study families that moved voluntarily from one neighborhood to another. Studying families that move allows us to control all the stable family char- acteristics, measured and unmeasured, that influence both where families live and their children's life chances. If we found that moving to a better neighborhood lowered poor black teenagers' arrest rates relative to those of their older siblings, for example, we would have more confidence that this was a true neighborhood effect than if we merely found that poor black teenagers who lived in good neighborhoods committed fewer cranes than those who lived in bad neighborhoods. Longitudinal data on the characteristics of the neighborhoods through which families have moved were just becoming available for the first time when we finished this review, so none of the studies we discuss uses such data.3 Even when such data become available, they will have important 3Rosenbaum et al. (1986) tried to assess the effects of moving from segregated inner-city Chicago neighborhoods to white suburbs, but they relied on retrospective parental reports to describe childrents experiences before they moved, and they did not examine any of the outcomes that concern us in this review.

GROWING UP IN A POOR NEIGHBORHOOD 121 limitations.4 Families usually move because their circumstances have changed. No survey can identify all the changes in a family's circum- stances that lead to a move. As a result, if children's behavior changes after they move, we can never be sure whether these changes reflect the influence of the new neighborhood or the influence of the factors that led to the move. If a father takes to drink, loses his job, and is unable to pay the rent, for example, the family may move to a cheaper neighborhood and the children may start misbehaving. Unless we know about the drinking, we may erroneously impute the change in the children's behavior to the change in neighborhood. The studies we review ignore the issue of change; they measure neigh- borhood characteristics at a single moment in time and implicitly assume that these neighborhood characteristics have remained stable throughout the respondent's childhood. If neighborhood effects accumulate slowly as we might expect in the case of school achievement, for example-measuring neighborhood characteristics at a single point in time can lead to serious measurement errors. Just as failure to measure a family's past income may innate neighborhoods' apparent effects (because current neighborhood is a proxy for past income), so too failure to measure where children have lived in the past may inflate the apparent importance of individual characteris- tics (because individual characteristics are proxies for prior neighborhood characteristics). Even cross-sectional surveys could tell us more than they now do about the effects of changing neighborhoods if they asked respondents how long they had lived in their current neighborhood and whether their current neighbors were richer or poorer than their previous ones. If we had this kind of information, we could determine whether the strength of a neighborhood's apparent effect depended either on how long the respondent had lived there or on having lived in similar neighborhoods before. If neither length of residence nor prior neighborhood characteristics proved important, we would have to abandon many popular theories about how neighborhoods affect children. Nonlinear Effects of Socioeconomic Mix We turn now to a series of problems that arise when we try to predict the likely ejects of government policy from the kinds of causal models that social scientists usually estimate. Unlike the problems discussed in the two 4The Institute for Social Research at the University of Michigan is currently adding neighbor- hood data to the Panel Study of Income Dynamics. These data should be available in 1990.

122 INNER-CITY POVERTY IN THE UNITED STATES previous subsections, which are widely recognized but hard to solve without better data, the problems we discuss in the next four subsections are for the most part easy to solve once we recognize them. Most of the studies we review assume that a neighborhood's mean SES has linear effects. None of the studies that make this assumption tries to test its validity, however. Social scientists have used linear approximations because their aim has been to determine whether the mean SES of a neighborhood or school has any effect whatever on a particular outcome. Linear approximations are usually satisfactory for this purpose. But if we want to predict the likely effect of housing policies that alter the degree of residential segregation, linear approximations will not tell us what we need to know. Suppose, for example, that poor neighbors encourage antisocial be- havior and that the effect of each additional poor neighbor is the same. In such a world, distributing the poor more evenly across a metropolitan area will redistribute the cost of having poor neighbors from the poor to the more affluent, but it will not reduce the cost to society as a whole. 1b see why, imagine a town with two neighborhoods of equal size, East Side and West Side. The poverty rate is 30 percent in East Side and 10 percent in West Side, and the annual burglary rate is 15 percent in East Side and 5 percent in West Side. Now imagine that the town provides housing vouchers to the poor in East Side so they can move to West Side and that it subsidizes developers who want to build expensive housing in East Side so as to attract affluent residents from West Side. Five years later the poverty rate is 20 percent in both neighborhoods. If burglary rates are a linear function of poverty rates, the burglary rate should fall from 15 to 10 percent in East Side and should rise from 5 to 10 percent in West Side. The citywide burglary rate will therefore remain unchanged. Needless to say, if the residents of West Side anticipate this outcome, they will oppose housing vouchers. Since the affluent have more political influence than the poor, and since in this case they have the status quo on their side, they will probably prevail. Dispersing the poor will only command majority support if most people believe that it will improve poor neighborhoods a lot more than it will harm more affluent ones. The effect of poor neighbors must, in other words, be strongly nonlinear. Only one study (Crane, forthcoming) devotes much attention to the shape of the relationship between individual outcomes and the mean SES of a neighborhood or school. Several other studies do, however, present data that bear on this question. Most of these studies divide neighbor- hoods into three or more socioeconomic levels and report mean outcomes for individuals in neighborhoods at each level. But since none reports the characteristics of neighborhoods at different levels, we cannot use $

GROOVING UP IN A POOR ~RHWD 123 their findings to estimate the likely effect of redistributing people between neighborhoods.5 Recent discussions of the underclass and concentrated poverty have focused on the effects of living in the worst 5 or 10 percent of all neigh- borhoods. Many believe that such neighborhoods have become breeding grounds for crime, drug abuse, teenage pregnancy, and welfare dependency. Many also believe that we could reduce the incidence of such problems by moving the poor to better neighborhoods. If we want to know whether very bad neighborhoods have substantial effects on children, we must look at what happens to children who actually grow up in such neighborhoods. Extrapolating from differences among children who grow up in neighborhoods that are a little better or a little worse than average may be quite misleading. The work of Hogan and Kitagawa (1985) on teenage pregnancy in Chicago underlines the potential dangers of such extrapolation. Hogan and Kitagawa studied black teenagers living in the city of Chicago. They divided the black neighborhoods in which the girls lived into three groups (best quarter, middle half, and worst quarter), based on their mean SES. Girls who lived in the worst quarter of black neighborhoods were substantially more likely to become pregnant than girls living in better black neighborhoods, even when their family characteristics were the same. The effect of living in the best quarter rather than the middle half of black neighborhoods was trivial. Very few Chicago whites live in neighborhoods as bad as the worst quarter of black neighborhoods. (Four-fifths of the families in these neigh- borhoods lived in public housing.) Thus, if we were to survey a repre- sentative sample of all Chicago-area residents, fewer than 1 in 10 would probably live in a neighborhood bad enough to have a detectable effect on the teenage pregnancy rate. If this were the case, using a linear model to estimate neighborhood effects would show that mean SES had very little effect on teenage childbearing. That conclusion would be correct for most neighborhoods, but it would be seriously misleading if we were mainly concerned with the effects of the very worst ghetto neighborhoods. Interactions Between Neighborhood SES and Individual SES Most of the studies we review assume that neighborhoods and schools have the same effect on everyone, regardless of their family background. Yet those who advocate governmental programs for reducing residential 5 In order to estimate the effect of moving people between neighborhoods, we need to know the shape of the relationship between neighborhoods' mean SES and the outcome of interest. We cannot determine the shape of this relationship from grouped data unless we know the mean SES of neighborhoods in each group.

124 INNER-CITY POVERTY IN THE UNITED STATES segregation usually argue that such programs would help the poor more than they would hurt the more affluent.6 There are at least two plausible rationales for this assumption. First, there is some evidence that the social networks of poor families are more geographically restricted than those of affluent families (Bott, 1957~. The social composition of a poor child's neighborhood may therefore have more effect on the child's choice of friends than the social composition of a rich child's neighborhood does. Second, poor parents seldom have the skills they need to join the middle class, so poor children must learn such skills at school if they are to learn them at all. Middle-class children, in contrast, can learn the required skills at home if they do not learn them at school. As a result, good schools may be more important for the poor than for the more affluent.7 The simplest and most reliable way to test such theories is to conduct separate analyses of high-SES and low-SES respondents. Older studies that used cross-tabulations often did this. More recent studies that use multivariate methods rarely do it (but see Corcoran et al., 198~, Crane, forthcoming). Several studies present separate analyses of neighborhood effects on blacks and whites. Because most areas with high concentrations of poverty are either black or Hispanic, dispersing the poor more evenly across a metropolitan area means dispersing blacks and Hispanics more evenly. Knowing whether such changes would help blacks or Hispanics more than they would hurt their new white neighbors is therefore quite important. But even if the average black or Hispanic gained more from residential desegregation than the average white lost, poor blacks or Hispanics might not gain more than the average white lost. Suppose schools in affluent white areas ignore the special problems of the poor or track them into "slow" classes. Middle-income blacks might then gain a lot from moving to such neighborhoods while poor blacks gained very little. Recognizing this, some middle-income blacks would probably move to more affluent neighborhoods when the opportunity arose, but most poor blacks would spend their limited resources in other ways. The result would be a decline in racial segregation and an increase in economic segregation. This is precisely what appears to have happened between 1970 and 1980 (Massey and Denton, 1987; Massey and Eggers, 1990~. 6 If the effects of individual SES and neighborhood SES are not additive, neighborhood erects will ordinarily appear to be nonlinear. But neighborhood effects can be nonlinear even if the edects of individual SES and neighborhood SES are completely additive. 7If low-SES students are more sensitive to school quality than high-SES students, the between- school variance in, say, test performance should be greater for low-SES students. If the within- school variance is the same for both groups, the total variance should also be greater for low~ES students. We found no studies that compared the variability of outcomes for high-SES and low- SES children.

GROWING UP IN A POOR NEIGHBORHOOD 125 None of the studies we review presents separate estimates of neigh- borhoods' effects on poor blacks or Hispanics, who would presumably be the primary beneficiaries of housing policies aimed at reducing residential segregation. Filling this gap should be a top priority in future work. Choosing Appropriate Measures of Neighborhood Composition Very few studies offer any strong theoretical or empirical rationale for focusing on one measure of a neighborhood's mean SES rather than another. The studies we review typically include one or more of the following measures: mean or median family income, the mean education of one or both parents, some measure of occupational mix, the percentage of families with female heads, and the percentage on welfare. Many investigators combine several of these measures into a single composite. We describe all such measures as indicators of a neighborhood's "mean SES," regardless of how they were constructed. Despite this common terminology, however, what we call "mean SES" does not always measure the same thing or rank neighborhoods in the same way. Different measures of mean SES are likely to be quite highly correlated with one another, but even when investigators constructed several measures of mean SES they did not report their intercorrelation. Nor have investigators who experimented with several different measures reported the results of these experiments in enough detail for us to generalize about how the choice of one measure rather than another might affect other studies' findings. From the viewpoint of a policy analyst, composite measures of mean SES have several drawbacks. First, they make it impossible to determine which particular neighborhood characteristics are important and which are not. It makes a big difference, for example, whether poor black teenagers' pregnancy rates fall when their parents live in affluent black neighborhoods or only when they live in predominantly white neighborhoods. A second drawback of composite SES measures is that investigators seldom report the weights they gave different neighborhood characteristics, so we cannot use their results to predict how any specific combination of neighborhood characteristics would alter children's life chances. Nor can we be sure whether the characteristics that went into the composite were weighted in such a way as to capture their full effect on the outcome of interest. If racial mix is critical, for example, and other neighborhood characteristics are of marginal importance, an index that includes racial mix as one of four equally weighted components may seriously underestimate neighborhoods' actual importance. A third problem is that composites seldom tell us the effects of the most politically relevant neighborhood characteristics. ~ test the hypothesis that poor neighbors affect children's opportunities or behavior, for example, we

126 INNER-CITY POVERTY IN THE UNITED STATES need to estimate the effect of a neighborhood's poverty rate with nothing else controlled. We found only one study that did this. Likewise, if we want to predict the effect of racial desegregation, we need to estimate the effect of racial mix with nothing else controlled. A few studies present this information, but most do not. Estimating Neighborhoods' Overall Effect This review focuses on estimating the effects on children of neighbor- hoods' and schools' social composition. For policy purposes this is part of a larger question, namely, whether any feature of a neighborhood or school affects children's life chances. 1b answer this larger question, we need two kinds of information. First, we need to know how much the outcome we are studying varies from one neighborhood or school to the next. Second, we need to know how much of this variation is attributable to exogenous factors like family background. If we want to know how much influence a high school has on its students' chances of graduating, for example, our first question should be how much the graduation rate varies from school to school. Our second question should be how much the graduation rate varies once we adjust it to eliminate the effects of differences among the students entering different high schools. Studies of schools occasionally do this. Studies of neighborhoods almost never do. Once we know how much impact each neighborhood or school has on the outcome that interests us, we can ask whether its impact is attributable to its mean SES, its racial composition, or some other measured or unmea- sured characteristic. School studies suggest that socioeconomic mix plays a relatively modest role in determining a school's overall impact on its students. Jencks and Brown (1975a) show, for example, that white students learned quite different amounts in different high schools in the early 1960s. Yet a school's socioeconomic mix had no consistent effect on how much its students learned. Other factors, which Jencks and Brown were unable to identify, were far more important. Analyses of neighborhoods' and schools' total effects are easy to do, but they are quite rare. As a result, we seldom know whether unmeasured neighborhood characteristics have important effects. In the absence of such data, readers should not interpret negative findings about the effects Alto estimate neighborhoods' explanatory power with the erects of family background controlled, we need an analysis of covariance, a regression equation that includes a separate dummyvariable for each school, or some statistically equivalent method. Hauser (1971) and Alwin (1976) discuss the statistical rationale for such methods. 13ryk and Raudenbush (1987, 1988) discuss hierarchical linear models for estimating neighborhood or school effects.

GROWING UP IN A POOR NEIGHBORHOOD 127 of mean SES or racial composition as evidence that neighborhoods or schools "don't matter." EDUCATIONAL ATTAINMENT We use the term "educational attainment" as shorthand for the number of years of school an individual completes. Educational attainment is the cumulative product of a series of decisions about whether to complete high school, enter college, complete college, and so on. These decisions are made at different ages and are subject to different social and economic influences. The factors that shape decisions about whether to finish high school, for example, may not be the same as those that shape decisions about whether to enter college (Mare, 1980), so we discuss them separately whenever we can. High School SES and College Attendance The earliest quantitative studies of "neighborhood" effects were con- ducted in the late 1950s and early 1960s and dealt with the effects of a high school's mean SES on graduating seniors' college plans. These stud- ies seemed to show that twelfth graders in high-SES schools had higher aspirations than those in low-SES schools, even with their own family's SES controlled (Michael, 1961; Turner, 1964; ~ Wilson, 1959~. Sewell and Armer (1966) argued that the influence of schools' mean SES had been exaggerated, but even they found that students from similar families with similar eleventh-grade IQ scores were more likely to plan on attending college if they were graduating from a high school that drew its students from a high-SES neighborhood. The usual explanation of this finding was that high-SES schools de- veloped a schoolwide culture that defined college attendance as both in- evitable and desirable. This schoolwide culture supposedly altered even working-class students' attitudes toward college. Conversely, schools with a working-class majority developed a schoolwide culture that defined college attendance as impractical and perhaps even undesirable. In such schools even middle-class students might not attend college. In 1966 James ~ Davis published an article ("The Campus as a Frog Pond") that challenged this line of reasoning. Drawing on Samuel Stouffer's work on relative deprivation among soldiers during World War II, Davis argued that reference groups had both a normative function (as "sources and reinforcers of standards") and a comparison function (a "point against which the person can evaluate himself and others"~. Because of this second function, he argued, advantaged classmates were not always as much of an advantage as most people assumed.

128 INNER-CITY POVERTY IN THE UNITED STATES Davis showed that the more academically selective a college was, the lower any given student could expect his or her grades to be. As a result, students who attended selective colleges were less likely to choose careers that required graduate training. Davis did not extend his analysis to high schools, but the analogy was obvious. High-SES schools usually have higher academic standards than low-SES schools. For students of any given ability, therefore, attending a high-SES school is almost certain to mean a lower rank in the graduating class, and it is likely to mean lower grades as well. This change is likely to lower students' academic self-confidence and their interest in attending college. In 1970 John Meyer published a paper that tried to reconcile these two conflicting models of how a high school's social composition might affect seniors' college plans. Meyer showed that attending a high-SES school had a positive effect on students' chances of planning to attend college but that attending school with students who scored high on standardized tests had a negative effect. Because a school's mean SES and mean test scores were highly correlated, the net effect of attending a high-SES school was quite small. Almost all subsequent studies support Meyer's findings. A school's mean SES tends to have a small positive effect on individual students' college plans, but its mean test score tends to have a small negative effect. Since the two are highly correlated, the net impact of mean SES on college plans is close to zero (see especially, Alexander and Eckland, 1975; and Hauser et al., 1976; but also Alwin and Otto, 1977; Hotchkiss, 1984; McDill and Rigsby, 1973; and Nelson, 1972~. Skeptics may wonder whether studies of college plans really tell us much about college attendance and completion. Innumerable studies have shown that college plans are not a very good predictor of subsequent behav- ior (e.g., Sewell et al., 1980~. If students from highways secondary schools were more likely than other students to realize their college plans, a school's mean SES could affect its graduates' eventual educational attainment even though it did not affect their plans at the time they graduated. When investigators have looked at high school seniors' eventual edu- cational attainment, they have found much the same thing as when they looked at twelfth graders' plans (Bible 11~. Attending a high-SES school has a small positive effect on subsequent educational attainment, and at- tending a school with high test scores has a small negative effect. The net effect is, therefore, close to zero. No study explicitly investigates whether low-SES students are espe- cially sensitive to the mean SES of their peers, as advocates of school desegregation often assume. Hauser et al. (1976) do, however, provide data relevant to this point. If low-SES students are especially sensitive to school quality, then the effect of parental SES on whether students attend

GROWING UP IN A POOR NEIGHBORHOOD 129 TABLE 4-1 Effects of Schools' Mean SES and Mean Ability on Students' Educational Attainment, With Family Background Controlled Standardized Measure of School Regression _) ~] Alexander and Eckland (1975) National sample of 10th graders Mean SES .09 in 1955 (N = 2,077) Mean ability -.09 Dependent variable: educational attainment in 1970 (5 categories) Hauser et al. (1976) Sample of Wisconsin 12th graders Mean mother's education .05 in 1957 (N = 7,052) Mean father's education -.02 Dependent variable: educational Mean father's occupation .07 attainment in 1964 (in years: Mean family income -.01 S.D. = 1.83) Mean 11th grade IQ score -.05 Jencks and Brown (1975b) Two follow-ups of a national sample of 9th graders in 1960 Dependent variable: educational Mean SES .045 attainment in 1964 (in years: Mean college plans .022 S.D. = 1.41) Mean test score -.091 Dependent variable: educational Mean SES -.053 attainment in 1968 (in years: Mean college plans .115 S.D. = 2.05) Mean test score -.064 NOTES: All estimates control sex, individual SES, and individual ability. Alexander and Eckland measure ability in 10th grade; Hauser et al. measure it in 11th grade; and Jencks and Brown measure it in 9th grade. Jencks and Brown also control 9th-grade college plans, cumculum assignment, and grades. Jencks and Brown's published beta coefficients have been restandardized using the individual-level S.D. of educational attainment. college should be weaker in high-quality schools than in low-quality schools. Hauser et al. found, however, that the strength of the relationship between family background and a student's eventual educational attainment did not vary significantly from one high school to another in a sample of 20 Milwaukee-area schools. This does not mean that the effect of parental SES is really exactly the same in every school. That is unlikely. But Hauser et al.'s findings do suggest that school-to-school differences in the impact of parental SES are fairly modest. If that is the case, high-quality schools increase high-SES students' chances of attending college about as much as they increase low-SES students' chances. There are, however, intriguing hints that black students may benefit more from attending a high-SES school than white students do. Meyer (1970) found that mean SES and mean test performance both had positive effects on southern black twelfth graders' college plans. This implies that the reduced-form effect of mean SES was fairly large and positive for

130 INNER-CITY POVERTY IN THE UNITED STATES southern blacks. The all-black southern schools covered by Meyer's 1955 data are no longer legally segregated, and most have either closed or been desegregated, so his findings may not hold for southern blacks today. But Thornton and Eckland's (1980) analysis of the High School Class of 1972 also suggests that both mean SES and mean test performance have positive effects on black students' college attendance.9 Although the effect of a school's mean SES on blacks remains uncer- tain, the basic findings for whites are so clear and robust as to leave little doubt about their approximate validity. Attending a high-SES secondary school does not appreciably alter a white student's college plans. This conclusion raises a puzzling question, however. How could mean SES have had positive effects on high school seniors' college plans in studies pub- lished before 1970 but have no effect in studies published after 1970? The obvious answer is that the world changed, but that explanation turns out to be untenable. Most of the studies conducted after 1970 relied on data collected in the 1950s and 1960s. Indeed, several of the studies conducted after 1970 used the same data as the earlier studies. We cannot, therefore, attribute the dramatic change in social scientists' conclusions to a change In the real world. The reason for the change appears to have been methodological. Stud- ies published before 1970 used two-way and three-way cross-tabulations to control the effects of exogenous student characteristics. As a result, inves- tigators could only control one measure of parental SES, typically broken down into only two or three categories, when they estimated the effect of a school's mean SES. Studies conducted after 1970 used regression equa- tions to estimate school effects. Regression equations enable investigators both to control all the family characteristics on which they have data and to treat these measures as continuous variables. Better controls for exogenous influences reduced the estimated effect of a school's mean SES almost to zero. This methodological finding has important implications not just for our understanding of college plans but for our understanding of school and neighborhood effects generally. It warns us that studies which control - 91bornton and Eckland do not present means, standard deviations, correlations, sampling er- rors, or reduced-form estimates. We estimated the reduced-form coefflcients from their struc- tural equations and used data from other samples to estimate the missing means and standard deviations. With exogenous influences controlled, we calculated that the proportion of black students who entered college fell by about .07 when they attended a high school in the bottom 10 percent of the SES distribution rather than an average high school. The sampling error of the estimate was about .03, so the true effect could have been considerably larger or smaller. The sit- uation may also have changed since 1972, when many blacks in high-SES schools were attending newly desegregated southern schools, whose effect on their subsequent educational attainment was not positive (see below).

GROWING UP IN A POOR NEIGHBORHOOD 131 only one or two exogenous family characteristics are likely to overestimate school and neighborhood effects. High School Racial Composition and College Attendance Race has a substantial effect on academic achievement, even when we compare blacks to whites whose parents have comparable jobs and incomes (Broman et al., 1975~. The same pattern recurs for teenage sexual activity and crime.~° A school's racial composition could therefore have quite different effects from its socioeconomic composition, even though the two are quite strongly correlated. St. John (1975) reviewed 25 studies that looked at the effect of desegre- gated schooling on black students' educational and occupational aspirations. Almost all these studies were conducted in the North. St. John found that desegregation seldom raised black students' aspirations significantly and often lowered them significantly. This is precisely what Davis's "frog pond" model predicts. Studies that focus on black students' actual attainment tell a rather different story, however. Crain (1971) found that among blacks who had graduated from high school in the 1940s and 1950s, those who had attended racially mixed northern schools were more likely to have attended college than those who had attended all-black northern schools. Crain and Mahard (1978) tell a similar story using the National Longitudinal Survey of the High School Class of 1972 (hereafter NLS-72~. They found that northern blacks were more likely to enter college and more likely to stay there continuously for the first three years after they finished high school if they had attended a racially mixed rather than an all-black high school. The i°Fumienberg et al. (1987) present data on racial differences in the age at which teenaged ini- tiate sexual activity. The odds that black 15- and 16-year-olds have had intercourse are 3.7 times the odds for whites of the same age. Controlling either mother's education or filmily income re- duced the black-white odds ratio to 3.3. Furstenberg et al. do not report the effect of controlling all available measures of parental SES, but if controlling a single measure only reduces the odds ratio from 3.7 to 3.3, multiple measures are unlikely to reduce it to 1.0. We do not have reliable data on criminal activity lay race and parental SES simultaneously, but both victims' reports and arrest data suggest that blacks are three to six times more likely than whites to engage in serious crimes (Bureau of Justice Statistics, 1987~. Because parental SES has a very modest effect on criminal activity in most studies (Hindelang et al., l9B1), it is hard to see how differences in parental SES could account for the racial difference in serious comes. 11 When estimating the effects of Percent White, Crain and Mahard (1978) controlled the mean SES of the blacks in a given school rather than controlling individual SES. If the mean SES of blacks in a school affects individual blacks' postseconda~y enrollment, and if blacks' mean SES is positively collated with Percent White, controlling blacks' mean SES will lower the coefficient of Percent White more than controlling individual SES would. Crain and Mahard conducted quite ingenious tests to determine whether the apparent effect of Percent White was due to selection bias and concluded that it was not.

132 INNER-CITY POVERTY IN THE UNITED STATES degree of segregation in a school district did not seem to have much effect on northern whites' chances of entering college or staying there. These findings suggest that we should be very cautious about studies of black aspirations. Many black high school students report very ambitious plans for the future, but many have taken none of the steps necessary to realize these plans. The available data suggest that this is especially common in all-black northern schools. Blacks in such schools are, it seems, likely to say they plan to attend college but less likely to do so. Thus, while desegregation seems to lower northern black high school graduates' educational aspirations, it seems to enhance their chances of actually entering and completing college. This is surely a worthwhile trade- off. lbrning to the South, however, the picture changes. Crain and Mahard (1978) found that, at least in 1972, attending high school with whites decreased both the chances that southern blacks would enter college and the chances that they would remain there continuously for three consecutive years. Once again, the degree of racial segregation in a southern school district did not seem to have much effect on white high school graduates' chances of entering college or remaining there. We do not have more recent data on this point. High School Graduation A high school's social composition is likely to have more influence on whether its entering students graduate than on whether its graduates attend college. In Chicago, to take an extreme example, many students report having quit high school because gangs controlled their school, making it dangerous to go there. Gangs have less influence in high-SES schools. We would therefore expect low-SES students to graduate more often if they attend high-SES Chicago schools than if they attend low-SES Chicago schools. We would not expect a Chicago high school's mean SES to exert an analogous effect on whether its graduates attended college, because gangs are less common on college campuses. Into studies have estimated the effect of a high school's mean SES on students' chances of graduating using data from the High School and Beyond (HSB) survey. HSB collected information on tenth graders' race, socioeconomic background, and academic background in 1980 and followed i2Both Alexander and Eckland (1975) and Jencks and Brown (197Sb) followed up high school dropouts as well as graduates, so in principle, their results indicate that a high school's mean SES does not have a major impact on high school graduation rates. But neither study was very successful in locating high school dropouts, so almost all the variance in educational attainment in both studies is attributable to variance in whether high school graduates attended and completed college.

GROWING UP IN A POOR NEIGHBORHOOD 133 up the students in 1982. Bryk and Driscoll (1988) used data from 357 of the HSB schools to investigate the effects of various school characteristics on students' chances of still being in school two years later, when they should have been in the twelfth grade. Bryk and Driscoll found that high school graduation was subject to the same contradictory influences as college entrance. As a school's mean SES fell, as its student body became more economically diverse, and as its minority enrollment increased, tenth graders of any given race, SES, and academic background were more likely to drop out. But with the school's racial and socioeconomic mix held constant, schools in which students had high scores in tenth grade tended to have high dropout rates. These findings suggest that academic competition increases attrition while high- SES peers reduce attrition. Since these two school characteristics are highly correlated, the net effect of having advantaged rather than disadvantaged classmates is probably small, but we cannot estimate the size of the net effect from Balk and Driscoll's data.l3 Mayer (forthcoming) used the full HSB sample (26,425 students from 968 schools) to investigate the impact of a high school's social mix on a tenth grader's chances of still being in school two years later. For students of average SES in average schools, the dropout rate averaged 13 percent for blacks, 14 percent for non-Hispanic whites, and 16 percent for Hispanics. For students one standard deviation below the mean on parental SES in average schools, the dropout rate rose 4 points for blacks, 10 points for whites, and 5 points for Hispanics. The finding that blacks drop out less than whites of comparable SES, and that this is particularly true for low-SES blacks and whites, recurs in other data sets (Jencks, forthcoming). Since Mayer wanted an "upper bound" estimate for the cumulative impact of schools' social mix, her estimates of school effects control only those student characteristics that schools cannot influence, namely parental SES and ethnicity. She did not control student characteristics that might depend on the social mix of the schools a student attended prior to the tenth grade. Using this approach, she found that when an average non- Hispanic white tenth grader attended high school with classmates whose mean SES was one standard deviation below average, his or her chance of dropping out rose from 14 to 17 percent. For blacks, the increase was from 13 to 15 percent. For Hispanics it was from 16 to 18 percent. The effect of a change in classmates' mean SES was thus between a third and a half as large as the effect of a comparable change in parental SES.14 13For a further analysis of compositional effects on high school graduation rates using the same HSB data, see Bryk and Thum (1988~. lain order to make the comparisons in the text, Mayer used the standard deviation for individ- uals to describe changes in both parental SES and classmates' mean SES. IhirW-four percent

134 INNER-CITY POVERTY IN THE UNITED STATES Having black or Hispanic classmates also increased the likelihood that a tenth grader would drop out prior to graduating, even with their mean SES controlled. The absolute effect of a change in classmates' mean SES on dropout rates was somewhat greater among low-SES students than among average students, because their base rate was higher. If neighborhood effects depend on homogeneity, the block on which a youngster grows up may be more important than the school he or she at- tends. Crane (forthcoming) studied the effect of a census tract's mean SES on its residents' chances of finishing high school using a unique 1970 cen- sus sample that links individual records to aggregate data on roughly 1,500 nearby families.~5 1b estimate residents' chances of finishing high school, Crane investigated whether 16- to l~year-olds who were still living at home had left school without graduating. The neighborhood characteristic that best predicted an individual's chance of quitting school was the percentage of professional and managerial workers living in the neighborhood. Crane interprets this finding not as evidence that professionals and managers are important in their own right, but rather as evidence that professionals and managers are in a good position to flee bad neighborhoods. Crane compared teenagers from families whose education, occupa- tional status, and income are at the national average but who live in different neighborhoods. Among 16- to 18-year-old whites living in neighborhoods where 30 to 35 percent of all workers have professional or managerial jobs, only 4 percent had left school without graduating. Among whites from similar families living in neighborhoods where only 5 to 10 percent of all workers had professional or managerial jobs, about 6 percent had dropped out. Among teenagers from typical black and Hispanic families in similar neighborhoods, dropout rates were about double the white rate. For blacks from typical families who lived in the very worst neighborhoods (where less of the observed variance in parental SES is between schools, so the standard deviation of mean SES is 58 percent of the standard deviation for individuals. A change of one individual-level standard deviation therefore implies a change of 1.72 school-level standard deviations, which is the difference between a school at the 5th percentile of the school-level distribution and a school at the 50th percentile. 15To maintain confidentiality, the Census Bureau did not merge individual records with regular census tract data, but instead created new "pseudotracts" from block-level data. The pseudo- tracts were about the same size as regular tracts, that is, about 4,000 people. Such tracts should be considerably more homogeneous than a high school attendance area, but we have no empiri- cal estimates of their homogeneity. 16Crane focused on 16- to 1~year-olds who lived at home because they were the only teenaged for whom parental characteristics were available and the only ones whose place of residence was exogenous. Crane reports that 84 percent of all 16- to 18-year-olds live at home. Omit- ting dropouts who no longer lived at home could bias the estimated impact of a neighborhood's median income either upward or downward.

GROWING UP IN A POOR NEIGHBORHOOD 135 than 5 percent of all workers had professional or managerial jobs), almost 20 percent of 16- to 18-year-olds were dropouts.~7 Neighborhoods' Cumulative Effect We located only two studies that linked young people's eventual ed- ucational attainment to the socioeconomic mix of the neighborhood in which they grew up. Both use data drawn from the Panel Study of Income Dynamics (PSID), and both use 1970 census data on the zip code in which teenagers lived to estimate the effect of neighborhood characteristics on educational attainment. There are 10 to 20 census tracts in a typical zip code, so zip codes, like high schools, are likely to be more socially diverse than census tracts or blocks. Datcher (19823 studied urban males who were between the ages of 13 and 22 and lived with their parents in 1968. Table 4-2 shows the estimated effects of a zip code's mean income and racial mix on the number of years of school Datcher's respondents had completed in 1978, when they were 23 to 32 years old. The estimates control parental income in 1968, the parents' educational attainment, the family head's educational aspirations for his or her children, the number of children in the family, region, community size, and age. A $1,000 increase in a zip code's mean income appears to raise edu- cational attainment by 0.103 years for whites and 0.087 years for blacks.~9 This black-white difference is too small to be either statistically reliable or i7The percentages of dropouts would all be higher if Crane had focused exclusively on 18-year- olds and if he had had data on individuals who no longer lived at home. The percentages would be lower today than in 1970, at least for blacks. i8Datcher did not control the household head's verbal score or occupational status. Nor did she control income in years other than 1968. The household head's occupation is usually the best single predictor of a son's educational attainment (Sewell et al., 1980), and a parent's verbal score is the best exogenous predictor of children's verbal scores (Jencks et al., 1972~. These omissions may have led Datcher to overestimate neighborhood ejects. Controlling parents' aspirations could, in principle, lead to some downward bias in a neighborhood's estimated ejects, but the effect of parental aspirations in Datcher's model is very small, so the bias is probably trivial. i9Table 4-2 shows that the coefficient of neighborhood income is only 1.4 times its standard error for blacks. This may lead some readem to conclude that neighborhood income has an "insignificant" effect on black men's educational attainment. That would be a mistake, for three reasons. First, in the absence of other information our best estimate of a neighborhood's eject is the value in Table 4-2, not zero. Second, since the coefficients for blacks and whites do not diner significantly and are quite close to one another, the hypothesis that the true coefficient is the same for both groups is more plausible than the hypothesis that the true coefficient is zero for blacks. Third, Datcher reports that when she entered neighborhood income but not neighborhood racial mix, the coefficient of neighborhood income was significant for blacks as well es whites. Unfortunately, she does not report these reduced-form estimates for either blacks or whites.

136 INNER-CITY POLARIS IN THE UNITED STATES TABLE 4-2 Effects of Neighborhoods' Mean lncane and Racial Composition in 1968 on Years of School Completed by 23- to 32-Year-Olds in 1978 Mean Family Income ($1.000S) Blacks Whites Regression coefficient (Standard error) Mean Mean for poorest 25% N SOURCE: Datcher (1982). (O to 10()) .087 .103 (.061) (.035) 9.18 11.97 8.53 1 1.08 196 356 Blacks Whites .0064 -.0056 (.0043) (.0082) 42.18 93.67 33.63 92.52 196 356 substantively important. The mean income for an urban zip code averaged about $11,500 in 1970; hence, 1bble 4-2 implies that a 10 percent increase in neighborhood income increased respondents' eventual educational at- tainment by about a tenth of a year regardless of their race. Unlike the effects of income, the effects of a zip code's racial compo- sition differed for blacks and whites. Blacks tended to get more education if they had white neighbors, while whites got more education if they had black neighbors. Cheering as this scenario is, the large sampling errors of both coefficients are also consistent with the hypothesis that a zip code's racial composition has no effect on anyone's educational attainment if the neighborhood's mean income is the same. A zip code whose economic and racial mix mirrored that of urban America as a whole would have had a mean income of $11,500 and would have been about 86 percent white in 1970. Poor blacks lived in zip codes that were 34 percent white and had mean incomes of $8,500. Datcher's model therefore implies that moving a poor black male to an average zip code would have raised his educational attainment by 0.6 years. Moving rich whites into these same zip codes would have had no effect on their educational attainment, since the costs of having slightly poorer neighbors would have been offset by the benefits of having slightly more black neighbors. These estimates are consistent with two quite different hypotheses: · Both blacks and whites benefit equally from living in affluent zip codes. Once we take this into account, neither group is affected by the zip code's racial composition. If this hypothesis is correct,

GROWING UP IN A POOR NEIGHBORHOOD 137 reducing racial and economic segregation would help blacks and hurt whites. · Blacks benefit both from living in affluent zip codes and from living in white zip codes. Whites benefit from living in affluent zip codes but suffer from living in white zip codes. If this hypothesis is correct, reducing racial segregation would benefit both blacks and whites. Datcher's data do not allow us to choose between these hypotheses. Corcoran et al. (1987) used more recent PSID data to investigate zip code effects on educational attainment. They used only respondents who were between the ages of 10 and 17 in 1968, so they did not miss as many college students as Datcher did. They also controlled more parental characteristics than Datcher did, and they looked at women as well as men. Their results support Datcher's finding that growing up in a "good" zip code leads to more schooling. Moving from a "typical" black neighborhood to a "typical" white neighborhood increased expected educational attainment by 0.41 years for men and 0.22 years for women.20 Unfortunately, Corcoran et al. do not present separate estimates for blacks and whites, so we cannot tell whether blacks gain more than whites from living in good neighborhoods. Conclusions about Educational Attainment Ken together, the results of Crane, Datcher, and Corcoran et al. strongly suggest that growing up in a high-SES neighborhood raises a teenager's expected educational attainment, even when the teenager's own family characteristics are the same. A high school's social composition, in contrast, has very little effect on a student's chances of finishing high school or attending college. The finding that neighborhood mix matters while school mix does not is puzzling for two reasons. First, we would expect neighborhood mix to be quite highly correlated with school mix. Thus, even if neighborhood 20The calculations in the text are for men and women whose parents received no income from welfare. Contrary to what advocates of residential desegregation usually expect, Corcoran et al. found that good neighborhoods conferred smaller educational benefits on children whose fami- lies received a significant fraction of their income from welfare than on children whose families received no such income. If these results are correct, segregating welfare &mikes (e.g., into pub- lic housing) would help nonwelfare families more than it would hurt welfare families. Without careful tests of alternative explanations, however, we are not inclined to take this result at face value. Datcher's data imply that moving from an average black neighborhood to an average white neighborhood would raise male educational attainment by an average of about 0.1 yeam com- pared to Corcoran et al.'s 0.4 years. This difference could reflect differences in the two studies' sampling strategies, random error, or the fact that Corcoran et al. used more measures of neigh- borhood characteristics than Datcher did.

138 INNER-CITY POVERTY IN THE UNITED STATES mix is critical and school mix is unimportant, we would expect school mix to appear important when neighborhood mix was not controlled. Indeed, several early studies of school mix used census tract data to characterize high school attendance areas. Second, even if neighborhood and school mix were not strongly corre- lated, we would expect them to have somewhat similar effects because most of the same social processes should be at work in both contexts. The main exception to this rule is that neighborhoods, unlike schools, do not assign students grades for their academic performance. As a result, growing up in a high-SES neighborhood may not have an adverse effect on students' academic self-concept unless it leads to attending a high-SES school. Before anyone invests too much ingenuity in explaining the difference between neighborhood and school effects, however, more effort should be invested in making sure that parallel analyses of the HSB, PSID, and 1970 census data really yield contradictory results. The apparent difference between school and neighborhood effects on educational attainment could derive from differences in analytic method rather than differences in the underlying causal connections. The racial composition of a school seems to have had different effects on blacks in the North and South, at least in the past. Racially mixed schools appear to have benefited northern blacks who attended them. But at least in 1972, southern blacks did not appear to benefit from attending racially mixed schools. We found no evidence on whether a high school's racial mix affected white students' eventual educational attainment. COGNITIVE SKILLS By "cognitive skill" we mean performance on any standardized test of mental skill or information. Test manufacturers often draw a distinction between `'aptitude" and '`achievement'' tests. But most studies suggest that both family background and school characteristics have about the same influence on aptitude scores (e.g., vocabulary scores) as on achievement scores (e.g., social studies information scores). Thus, we treat aptitude and achievement tests as interchangeable. We found no studies of whether test performance depended on the social composition of the neighborhood in which a student grew up, as distinct from the composition of the school the student attended. Thus, this section deals exclusively with the effects of a school's social composition on test performance. We look first at the effect of a school's mean SES and then turn to the effect of its racial composition. In assessing the impact of mean SES we look first at high schools, about which we know a fair amount, and then at elementary schools, about which we know far less.

GROWING UP IN A POOR NEIGHBORHOOD Effects of High Schools' Mean SES 139 The "Coleman report" (Coleman et al., 1966) was the first study of how a school's mean SES affected students' cognitive skills. The report was based on the 1965 Equality of Educational Opportunity (EEO) survey, which collected data on first, third, sixth, ninth, and twelfth graders in a more or less representative national sample of public schools. No compa- rable survey has been conducted since, so the data are still of considerable interest, even though schools have changed in many ways since 1965. Coleman et al. analyzed the EEO data at a time when regression equations were just becoming a part of sociologists' statistical tool kit, and their presentation emphasized changes in R2 rather than regression coefficients. As several economists immediately pointed out, this made Coleman et al.'s findings hard to interpret correctly. But unlike most economists, Coleman et al. also reported the means, standard deviations, and correlations of the variables they used in their analyses. Thus, we can reanalyze their data in such a way as to address the questions posed in this review without going back to the original data tapes. Coleman et al.'s matrices include only one measure of a school's mean SES, namely, the percentage of students who said their families owned an encyclopedia.2i The matrices include seven exogenous family background characteristics: parental education, material possessions in the home, reading materials in the home, family size, family structure, urbanism, and speaking a foreign language at home. The matrices do not include any direct measure of parental income or occupational status. Nor did the survey measure students' test scores when they entered a school.22 Table 13 shows the effect of mean encyclopedia ownership on test per- formance in the sixth, ninth, and twelfth grades with all seven background measures controlled.23 Measures of school quality, such as expenditures or teachers' credentials, are not controlled. The apparent effects of mean SES may, therefore, derive either from the fact that high-SES schools get more money, better teachers, and so forth, or from the fact that high-SES students create a more favorable climate for teaching and learning. 2lEncyclopedias are not an ideal measure of mean SES, but they predict achievement as accu- rately as the school mean of any other single SES measure collected from these students. 22The absence of initial scores would pose no problem if the measures of individual SES captured all school-to-school differences in initial ability, but this requirement is not fully met (Jencks, 1972a,b). Failing to control first-grade scores tends to bias the estimated effect of mean SES upward. Using the percentage of families with encyclopedias as the sole measure of mean SES tends to bias the estimated effect of mean SES downward. The net bias is uncertain. 23We cannot make analogous estimates for first or third graders, because family background measures for first and third graders are inadequate and the third-grade tests are flawed (Mosteller and Moynihan, 1972~.

140 - to U. U) C, CJ ha Cal ~ U. o - U) C, ED U' U' o - of U) ~ Hi: ~ m C) .~ en m .° ~ _ C.) _ C) .5 o 3 O ~ 8 . EN cat .e o _ ~ 8 3 C: O C) ~ 8 ED cot a_ a. m _% a .~ Cot ~ O ~ ~ ON _ O ~ ~ ~ 00000 00000 .. 1 ..... _% _% a_ _% hi 00 0 ~ \0 ~ 00 ~ _% _% _% a_ 00 Cal ~0 00 ~O-C'3 ~2 O O ~- O Cn _ _ 1 C~ ~ C~ - s090 '_ _ '_ ~ ~ e' 53~= _% ~_ ~_ ~_ ~ C~ 0. C~ _ _ _ _ . . . . _% ~_ ~ ~ 00 _ ~ ~ _ _ _ _~ %_ ~ _ ~ _ ~ C~ ~ ~-O- ~-0\-O t_ ~ t_ ~ ~wi _ 0\ ~ wi ~ C~) ~ 00 vo ~ ~ ~n ~ \0 ~ ~n ~c~ ~ c~ ~ ~ o0 X ~--C~ C~ oo ~ t-'} C ~_ ~1- ~ _ ~ ~ ~ O - ~- O O- O . . . . . . · . . . , = _% _% _-~ C~ ~ ~ O. C~ ~-C~ ~ C~ 0\ ~ ~ O ~ ~-C ~ ~ ~ C~ O C~ C~ - ~ ~- - - ~ _ m~0 - . . . . ~ ~ 0~ _ ~_ _% ~_ ~_ ~ ~ oO 00 . . . . ~ ~ C~ C~ _ _ _ _ t-'~, C'1 C'] C ~O~-O O C~ _ Oi C~ ~O ~ ~ ~ '~ C~ C O . _, _ C~ ~ .5 ~ ~ ~ ~ C o4 ~ ~ ~ C c o U] c .O o o c .:t CD .= c ___ ~ oo o ~C) o, _ _ . . - - C~ - c ~ C) .. C~ . o C~ C) oo ~ ~ _ ~ ~ as _ o ~ o . - - - ~ o_ - mmo o o ,, 3 ~ ~ ~ . . . _I ~ o oo ~0 P? _ . 5 ~ c, a5~ O~ C _ `_ _ o oo ~ =_ ~ C~ ~ ~ ~ C V) ~ ~ a ~ - 3o ~o o ~=D~ =o -5 C a ~ ~ ~ -5 C C ~ ~ == C c ~ _ C ~vo ~ ~ ~ - ~~ o~ oc 04 c ;l C,

GROWING UP IN A POOR NEIGHBORHOOD 141 Table 4-3 includes two notable results for high school students: First, mean SES has almost no effect on northern white high school students' achievement. Averaging across all four tests, the standardized coefficients of encyclopedia ownership among whites are .038 for ninth graders and .023 for twelfth graders. Coefficients of this size are too small to be substantively significant, and none is reliably different from zero.24 The effect of mean SES on white students' test scores also diminishes between the ninth and twelfth grades. It follows that if we were to control students' ninth-grade test scores when predicting their twelfth-grade scores, the estimated effect of mean SES on white students' twelfth-grade scores would be even smaller than Table 13 implies. Second, mean SES has a sizable effect on northern black high school students' achievement. For blacks, the standardized coefficients of mean encyclopedia ownership average .083 In ninth grade and .135 in twelfth grade. These coefficients are large enough to be substantively important and almost all of them are reliably greater than zero. Note, too, that the apparent effect of mean SES increases between the ninth and twelfth grades. Thus, even if we could control black students' ninth-grade scores, the estimated effect of a high school's mean SES on black high school students' cognitive growth would be substantial. The results in Bible 4-3 are subject to three major sources of potential error. First, the measures of individual SES contain a lot of error. Second, students' initial test scores are not controlled. Third, the measure of mean SES is not very precise. The first two problems tend to inflate the estimated effect of mean SES, while the third deflates it. Jencks and Brown (19 75b) tried to deal with these three problems using data collected by Project Talent, which tested ninth graders in 1960 and retested those who were still enrolled in the same school in 1963, when they were in the twelfth grade. Jencks and Brown analyzed changes between ninth and twelfth grade In vocabulary, social studies information, reading comprehension, arithmetic reasoning, arithmetic computation, and abstract reasoning among students enrolled in 91 predominantly white, comprehensive public schools. In addition to SES, sex, and family size, they treated ninth graders' college plans, curriculum assignment, previous grades, and test performance as exogenous. Their conclusions, therefore, apply only to cognitive growth after ninth grade. 24Some readers may suspect that mean encyclopedia ownership has little impact on white ninth and twelfth graders' test scores because it is a poor measure of mean SES. Were that hypothesis correct, however, it should apply with equal force to white sixth graders and to blacks at all grade levels. Since Table 4-3 indicates that mean encyclopedia ownership makes a difference for all these groups, the tiny effect among white ninth and twelfth graders cannot be dismissed as a methodological artifact.

142 INNER-CITY POVERTY IN THE UNITED STATES After adjusting for both exogenous factors and measurement error, Jencks and Brown found that differences between high schools accounted for between 1.0 and 3.4 percent of the variation in twelfth-grade test scores. They also found that schools which did a good job raising students' scores on one test were not especially likely to do a good job on other tests. (The correlation between schools' estimated effects on different tests averaged only .17.) They concluded that variation in high school quality would have explained less than 1 percent of the variation in a composite achievement measure based on all six tests. High-SES schools had a positive effect on vocabulary and arithmetic reasoning scores and a negative effect on social studies and reading com- prehension scores, but none of the effects was large enough to rule out the possibility that it was due to chance. Jencks and Brown's analysis is therefore consistent with the EEO data for white ninth and twelfth graders. Both analyses imply that a high school's mean SES has a negligible effect on white students' academic progress. The High School and Beyond (HSB) survey provides the best current evidence on these issues. In 1980, HSB gave vocabulary, reading, writing, civics, math, and science tests to a national sample of more than 30,000 tenth graders in roughly 1,000 public and private schools. In 1982 HSB retested the same students, most of whom were in twelfth grade. Myers (1985) analyzed the effect of a school's poverty rate on tenth and twelfth graders' reading and math achievement. He did not exploit the longitudinal feature of the survey by including tenth-grade scores in the equation predicting twelfth-grade scores.25 His analyses, therefore, measure (or at least try to measure) the cumulative effect of mean SES up to tenth or twelfth grade, not the effect of mean SES on cognitive growth between tenth and twelfth grade.26 Nonetheless, Myers's results are of some interest, because they focus directly on the effects of concentrated poverty. The poverty rate in Myers's sample of HSB schools averages 21 percent, with a standard deviation of 15 points. A 30-point increase in a high school's poverty rate (e.g., from 20 to 50 percent) lowers reading scores by 0.25 standard deviations in tenth grade and 0.35 standard deviations in twelfth grade. For math scores the difference is 0.33 standard deviations in 2sMyers does present an analysis of what he calls "learning," based on changes between the tenth and twelfth grades in the regression coefficient of the school's poverty rate. For reasons discussed below in our critique of Jencks and Brown (197Sa), we do not believe that this com- parison measures the effect of the poverty rate on learning 26Because high schools draw from a wider area than elementary schools, poverty rates vary less in high schools than in elementary schools. This fact should not bias the coefficient of the poverty rate in Myers's analyses, but it could lead us to underestimate the potential benefits of moving students from poor schools to average schools because the poorest elementary schools are poorer than the poorest high schools.

GROWING UP IN A POOR NEIGHBORHOOD 143 tenth grade and 0.46 standard deviations in twelfth grade. These are very large effects, but they may be attributable to the fact that Myers did not control a number of important exogenous variables.27 Gamoran (1987) presents a technically superior analysis of the HSB data, which focuses on cognitive growth between the tenth and twelfth grades. His measures of both individual SES and mean SES are composites that give equal weight to father's education, father's occupation, mother's education, family income, and possessions in the home. He estimates the effect of a public school's mean SES on each of the six HSB twelfth- grade tests, controlling tenth-grade vocabulary, reading, and math scores plus gender, ethnicity, and individual SES.28 With these controls, the standardized coefficient of a school's mean SES ranges from a maximum of .026 for the vocabulary test to a minimum of-.011 for the science test. The arithmetic mean of the six standardized coefficients is .009, and none of the six coefficients is reliably different from this mean. Thus, Gamoran's data suggest that moving from an average school to one with an extremely high concentration of poverty in tenth grade would lower a student's twelfth- grade test performance by something like .02 of a standard deviation. An effect of this magnitude would almost never be substantively significant. Unlike Gamoran, Bryk and Driscoll (1988) concluded that mean SES had a statistically reliable effect on twelfth-grade math scores In their sam- ple of 357 HSB schools that had participated in the 1984 Administrator and Teacher Survey. Bryk and Driscoll control essentially the same exoge- nous variables as Gamoran, but they include more measures of a school's socioeconomic and ethnic mix, so their results are not strictly comparable to Gamoran's. Nor do they present enough information for us to estimate the effect of mean SES with other school characteristics omitted. Bryk and Driscoll also found that the gap between the math scores of advantaged and disadvantaged students is greater In high-SES and m~xed- SES schools than In low-SES schools. They do not present enough ~nfor- mation for us to determine whether low-SES students are actually worse off when they move from low-SES to mixed- or high-SES schools. Low-SES 27When estimating the effects of a school's poverty rate, Myers controlled a student's gen- der, ethnicity, family size and structure, mother's education, mother's employment history, and whether the student's family was poor. He did not control father's education, father's occupation, or income differences among families above the poverty line. For reasons he does not explain, his analysis covers only a quarter of the original HSB sample. His tables and text also contain several internal inconsistencies. 28Gamoran also controls a school's mean tenth-grade math or reading score and its ethnic com- position, but these variables have such small coefficients that their inclusion is unlikely to have an appreciable effect on the coefficient of mean SES. The standardized coefficient of the school's mean score on the relevant tenth-grade test averages .002; the standardized coefficient of the proportion black averages-.0003.

144 INNER-CITY POVERTY IN THE UNITED STATES students may just gain less than high-SES students when they move from a low-SES to a high-SES school. Bryk and Driscoll's findings imply that the weak effect of mean SES on math achievement in Gamoran's analysis of the HSB data may mask gains for some and losses for others. This issue requires more detailed investigation. If Bryk and Driscoll are correct that high-SES students gain more from attending high-SES schools than low-SES students do, reducing economic segregation in high schools would hurt advantaged students more than it would help disadvantaged students. As a result, economic desegregation would lower a school system's mean achievement. Before accepting this finding at face value, however, we should check it using statistical methods that focus explicitly on this issue. Our reanalyses of Coleman et al.'s EEO matrices, Jencks and Brown's analysis of Project Talent, and Gamoran's analysis of HSB all indicate that a high school's mean SES has a negligible impact on how much the average student learns in high school. Bryk and Driscoll seem to find the opposite, but we are not yet ready to conclude that the earlier consensus on this point was wrong, because their model does not provide a clean test of the hypothesis. Such evidence as we have also suggests, however, that studying the effect of a high school's mean SES on the "average" student may conceal more than it reveals. A high school's mean SES may well have a substantial impact on some students but very little on others. Coleman et al.'s matrices suggest that mean SES has more impact on blacks than on whites. Bryk and Driscoll's results suggest that mean SES has more effect on high-SES students than on low-SES students. Ikken at face value these results support policies aimed at reducing racial segregation while increasing economic segregation, but they should not be taken at face value until they are replicated.29 Effects of Elementary School's Mean SES ~ rning from high schools to elementary schools, the evidence re- garding determinants of academic achievement is much worse. Table ~3 suggests that in 1%5 mean SES had a moderate impact on the achievement scores of both black and white sixth graders. The standardized coefficient of encyclopedia ownership averages .08 for sixth graders. This coefficient does not vary significantly by race; nor does it vary significantly from one academic subject to another. Thus, the EEO data imply that moving from a school in the bottom 5 percent of the socioeconomic distribution to a 29Mayer is engaged in an analysis of HSB that focuses on precisely these issues.

GROWING UP IN A POOR NEIGHBORHOOD 145 school of average SES would raise students' test performance by one-sixth of a standard deviation. Unfortunately, this estimate is even shakier than those for ninth and twelfth graders. The EEO measures of individual SES are based on stu- dents' reports about their parents. Because sixth graders' reports about their parents are not very accurate, an elementary school's mean SES may be a proxy for unmeasured aspects of individual SES. There may also have been differences in initial achievement between sixth graders of appar- ently similar SES who attended high-SES rather than low-SES elementary schools. These considerations suggest that Table 4-3 may overestimate the impact of mean SES. On the other hand, reports of enclyclopedia owner- ship are hardly an ideal measure of mean SES. A better measure would probably have a larger coefficient.30 Partly because the EEO survey had such serious limitations, the federal government initiated a major new study of elementary education in 1976. The Sustaining Effects (SE) study collected background data directly from parents and tested first through sixth graders in both the fall and the spring of three successive academic years. In principle, these data should provide a lot more information about the impact of a school's mean SES on a child's cognitive growth than the EEO survey did. Unfortunately, the Department of Education has never made the SE data available to outside social scientists in usable form. As a result, only one study (Myers, 1985) uses the SE data to investigate whether a school's mean SES affects students' cognitive skills independent of their family background. Myers used the poverty rate for families In a school as his measure of mean SES and treated children's gender, ethnicin,r, mother's education, mother's employment status, family size, family structure, and whether their family's income was below the povert, line as exogenous. He did not exploit the longitudinal feature of the SE data, but instead conducted a series of cross-sectional analyses. When Myers used a school's poverty rate to predict reading scores, its standardized coefficient was .05 at the beginning of first grade, .08 at the end of first grade, .15 at the beginning of second grade, and .11 at the end of second grade. The standardized coefficient fell slightly after the end of second grade and was .09 at the end of sixth grade. These standardized coefficients are quite misleading, however, because they ignore the fact that the dispersion of test performance increases with age. The absolute 30Jencks (1972a) obtained somewhat larger effects using data derived from the full northern ur- ban EEO sample and a better measure of mean SES. Because the measures of family background in the original EEO survey are inadequate, Jencks's analysis probably overstates the impact of mean SES on test performance.

146 INNER-CITY POVERTY IN THE UNITED STATES effect of mean SES increased as children got older, which is what everyday experience would lead us to expect. The SE survey had better measures than the EEO survey for both individual SES and mean SES. These improvements evidently offset one another, leaving the standardized effect of mean SES in sixth grade almost unchanged. Myers reports a standardized coefficient for mean SES in sixth grade that is almost identical to the standardized coefficient in Table 4-3, which is based on the EEO. Controlling students' initial scores would have lowered the effect of mean SES in both surveys, but we doubt that such controls would have cut the effect by more than a third.3i The poverty rate in the SE schools that Myers studied averaged 18 percent (with a standard deviation of 18 points). Thus, the SE data suggest that moving a student from a school with a poverty rate of 54 percent to an average school with a poverty rate of 18 percent would raise his or her sixth-grade reading score by something like one-sixth of a standard deviation. Effects of Racial Composition on Blacks The Coleman report was also the first major study of how a school's racial composition affected its students' achievement. The effects of racial composition are, of course, somewhat confounded with the effects of mean SES, since predominantly white schools usually have more affluent students than predominantly black schools. But the correlation between mean SES and racial mix Is far from perfect.32 Some all-black schools enroll a lot of middle~lass students, and some all-white schools enroll very few. Moreover, in comparisons of black and white students whose parents have the same number of years of school, the same occupational status, and the same income, whites do far better on standardized tests. This is 3i Longitudinal surveys suggest that the standardized coefficient of fimt-grade scores in an equa- tion predicting sixth-grade scores is unlikely to exceed .60. In the absence of school ejects, therefore, an 0.05 standard deviation difference between two students entering first grade would translate into a difference of only 0.03 standard deviations in sixth grade. Since the apparent effect of mean SES in sixth grade is about 0.09 standard deviations, the "school effect" would be at least 0.09 - 0.03 = 0.06 standard deviations. We should note that this estimate differs from Myers's estimate of the relationship between a school's poverty rate and student "learning." For reasons discussed in our assessment of Jencks and Brown (197Sa), we think his analysis of learn . . mg is wrong. 32The correlations between Proportion White and Proportion Owning Encyclopedias in schools attended lay northern whites in the EEO survey were .25 for twelfth graders, .40 for ninth graders, and .43 for sixth graders. The correlations in schools attended by northern blacks were .45 for twelfth graders, .49 for ninth graders, and .51 for sixth graders. The correlations are higher for blacks than for whites because Proportion White has more variance among blacks than among whites. Better measures of mean SES would yield higher correlations.

GROWING UP IN A POOR NEIGHBORHOOD 147 true even among four-year-olds, almost none of whom had attended school (Broman et al., 1975~. A school's racial composition, therefore, is a partial proxy for its students' mean test performance, even after its mean SES is controlled. Thus, it would not be surprising if a school's racial composition affected its curriculum, its teachers' expectations about how much their students could learn, its students' expectations about how much they could learn, or the amount students really did learn. Coleman et al. reported that attending school with white classmates increased black students' test scores in the North, but they did not estimate the magnitude of the effect. Table 14 uses their correlation matrices to estimate the effect of Proportion White on black students' test scores in the sixth, ninth, and twelfth grades. These estimates do not control mean SES. Rather, they try to estimate the effect of moving a student from a predominantly black school to a predominantly white school. Once again, we control the seven exogenous family background characteristics available in the published matrices. Combining the coefficients in Table 14 with the standard deviations in Table 4-3 and averaging over all tests, we estimate that blacks in 90 percent white schools scored 0.30 standard deviation higher than blacks from similar backgrounds in all-black schools. The magnitude of this effect shows no statistically reliable trend between sixth and twelfth grade. Crain and Mahard (1978) obtained essentially identical results for northern black twelfth graders in 1972 using data from the National Longitudinal Survey of the Class of 1972 (NLS-72~.33 In the South, where desegregation had just begun in 1972, a school's racial composition had virtually no effect on black students' performance. This may be because blacks in desegregated high schools had only been there a few years or because their experiences in newly desegregated southern schools were often unpleasant. There have been no studies using more recent data. Controlling initial ability might change the picture somewhat. Jenck~ and Brown (1975a) present EEO test scores for northern urban blacks in schools that were more than 75 percent black and in schools that were less than 10 percent black. This gap averaged 0.39 standard deviations in 33Crain and Mahard report that the effect of moving from an all-black to a 90 percent white school was half a standard deviation rather than a third of a standard deviation. But the difference is 3.69 points, and their tables leave little doubt that their tests were standardized to a national mean of 50 and standard deviation of 10. Gain and Mahard controlled the mean SES of all blacks in a school rather than the SES of individual blacks. If blacks in predominantly white schools came from high-SES families, this procedure might lead to some downward bias in the estimated effect of Proportion White, since the coefficient of mean black SES is likely to exceed the coefficient of SES for individual blacks. Fortunately, mean black SES is uncorrelated with Proportion White in both NLS-72 and the 1965 EEO, so this source of bias is not cause for concern.

148 INNER-CITY POVERTY IN THE UNITED STATES TABLE 44 Effects of Proportion White on Test Scores, With Seven Background Measures Controlled, Anthem Blacks arid Whites, 1965 Blacks B Whites (S.E.) Beta B Beta 12th grade Verbal 6.1 (1.3) .147 4.6 (3.4) .040 Reading 3.9 (1~6) .079 4.7 (4~2) .034 Math 7.0 (1.7) .129 7.5 (4.3) .054 General information 5.7 (1.0) .171 5.1 (3-0) .051 Mean of four tests 5.7 .132 5.5 .045 9th grade Verbal 2.2 (1~2) .057 8.9 (3-0) .087 Reading 3.2 (1.5) .065 11.7 (3~8) .093 Math 4.1 (1.3) .098 9.9 (3~1) .097 General inforTnanon 6.0 (1.2) .156 11.9 (3.1) .111 Mean of four tests 3.9 .094 10.6 .097 6th grade Verbal 4.4 (0.9) .147 8.6 (2.3) .108 Reading 2.0 (1 3) .045 8.0 (3~4) .072 Math 3.1 (1~0) .095 7.2 (2.5) .090 Mean of three tests 3.2 .096 7.9 .090 Mean Proportion White S.D. of Proportion White Blacks Whites Blacks Whites 12th grade .413 .899 .349 .122 9th grade .311 .881 .329 .157 6th grade .243 .877 .303 .143 SOURCE: Coleman et al. (1966:Supplemental Appendix). twelfth grade' 0.38 standard deviations in sixth grade, and Oe42 standard deviations in first grade. Since the gap is as large in first grade as in twelfth grade, it is tempting to attribute the twelfth-grade gap to the first-grade gap. Indeed, that is precisely what Jencks and Brown did. Such reasoning is misleading, however. Suppose we were to find that among blacks who attended the same school twelfth-grade scores correlated .60 with their grade scores a quite plausible estimate. This would indicate that when two blacks in the same school had scores that differed by 0.42 standard deviations in first grade, their scores typically differed by only (0.60~0.42) = 0.25 standard deviations in twelfth grade. Yet when black first graders in all-black and predominantly white schools differ by 0.42 standard deviations, they differ by 0.39 standard

GROWING UP IN A POOR NEIGHBORHOOD 149 deviations in twelfth grade. The gap is thus 0.14 standard deviations larger than it would have been if the children in question had attended the same school. We do not know the actual correlation between first- and twelfth-grade scores, but we know it is less than 1.00. Thus, if a school's racial composition had no independent effect on black students' scores, the standardized difference between blacks entering all-black schools and those entering 90 percent white schools would decline as the students got older. Since this does not happen, we must conclude that something about the schools themselves helps perpetuate the initial differences among their entering black students. Unfortunately, while this logic supports our earlier conclusion that racial composition matters, it does not tell us how much it matters. We cannot resolve that question without longitudinal data. In principle, the HSB and SE surveys provide such data, but in practice, no one has analyzed either survey with this in mind.34 The First Year of Desegregation No one has studied the long-term impact of schools' racial mix on their students' achievement since the mid-1970s. Social scientists have concentrated instead on estimating the short-term effect of desegregation plans. Most such studies share three major limitations: (1) They estimate the effect of a single year in a desegregated school, not the cumulative effect of attending desegregated elementary and secondary schools. (2) They focus on the year in which a desegregation plan was first implemented, which is typically quite chaotic. (3) They ignore secondary schools. The first limitation is especially critical. There is no way of using data on the effect of 1 year In a desegregated school to infer the effect of 12 years in such schools. Yet the cumulative effect of desegregation is what matters. Studies of what happens during the first year of desegregation are so numerous that we have not tried to review them all. Instead, we have reviewed the reviews. Since the late 1970s such reviews have taken quantitative form and have been called meta-analyses. A meta-analysis records all the apparently relevant attributes of each study (grade level, type of test, use of a pretest, location, and so forth), along with the estimated effect of desegregation on student achievement. The meta-analyst then estimates the mean gain in studies with various attributes. 34Several analyses of the HSB data look at the effect of racial composition on the achievement of a pooled sample of blacks and whites, but because the bulk of the HSB sample is white, the results do not indicate much about the effect of racial composition on blacks. We discuss these studies below.

150 INNER-C1~IY POVERTY IN THE UNITED STATES Meta-analysts typically report that during the first year of desegregation a desegregated black student's reading score rises about 0.10 standard deviations more than a segregated black student's score (Cook, 1984~. Depending on the principles that meta-analysts use to select studies, they can push the estimated effect as high as 0.30 (Crain and Mahard, 1983) or as low as 0.06 (Armor, 1984~.35 The first year of desegregation does not appear to affect black students' math scores (Cook, 1984~. Crain and Mahard (1983) argue that most meta-analysts underestimate desegregation's effect on black reading scores, for two reasons. First, gains are larger in studies using random assignment, which is the preferred method for ensuring that desegregated students are initially similar to segregated students. But as Cook (1984) notes, almost all the evidence using random assignment comes from Hartford, where relatively small numbers of inner-city blacks were bused to suburban schools that had volunteered to accept them. The Hartford experiment was unusual in many respects, so its results are probably not generalizable. Second, Crain and Mahard argue that kindergarteners and first graders gain more from desegregation than older students. They interpret this as evidence that young children are especially influenced by desegregation. But whether desegregation appears to have more effect on young children than on older children depends entirely on how we measure the effect. Like almost all meta-analysts, Crain and Mahard express gains at different ages as a percentage of the initial standard deviation. They find that one year of desegregation raises black students' reading scores by an average of only 0.09 standard deviations in sixth grade compared with 0.20 standard deviations in first grade. They conclude that desegregation does more good in first grade than in sixth grade. Unfortunately, comparing standardized test scores at different ages ignores a crucial feature of test performance, namely that performance becomes more unequal as children get older. A 0.09 standard deviation difference between two sixth graders is far larger, in absolute terms, than a 0.20 standard deviation difference between two first graders. One way to make this clear is to egress differences not in terms of standard deviations but in terms of time. Myers's (1985) analysis of the SE data indicates, for example, that chil- dren's reading scores typically rise by 1.8 standard deviations between the fall and spring of first grade. If blacks gain 0.20 standard deviations more in desegregated schools than in segregated schools, therefore, they must be learning about 0.20/1.8 = 11 percent more than they otherwise would. In 1 35 Cook (1984) notes that the mean gain is larger than the median, because a few small studies show large positive gains and no studies show large losses.

GROWING UP IN A POOR NEIGHBORHOOD 151 temporal terms, attending a desegregated first grade boosts black achieve- ment as much as an extra month of schooling. Myers reports that children typically gain only 0.42 standard deviations between the fall and spring of sixth grade. Since desegregated schooling increases black children's gains by 0.09 standard deviations in sixth grade, it increases the amount they learn by about 0.09/0.42 = 21 percent, which is roughly equivalent to a two-month gain. Using this approach, therefore, desegregation is more valuable for older students, not younger students. We do not know for sure that the temporal metric is correct. If we had a true internal scale for measuring reading skills, we might find that students learned more per month at some ages than at others. But the temporal metric surely approximates an interval scale more closely than the standardized metric favored by meta-analysts. In the absence of better evidence we must therefore conclude either that desegregation helps older blacks more than younger blacks or that there is no sensible basis for making such comparisons. The crucial question, however, is how 12 years of desegregated school- ing affects black students' achievement. Krol (1980) estimates the mean effect of two or more years in a desegregated school at about 0.2 standard deviations, compared with 0.1 standard deviations for a single year. But he found so few studies covering more than one year that this difference was not statistically reliable, and none of the studies he found covered more than three years of desegregated schooling. Our best estimates of the cumulative impact of school desegregation are, therefore, still based on the 1965 EEO and NLS-72. These two surveys suggest that black students who attended overwhelmingly white rather than all-black schools in the North in the 1960s scored something like one-third of a standard deviation higher on most tests as a result. Since the overall difference between northern blacks and whites was about one standard deviation, the benefits of desegregation were substantial. Nonlinearities The discussion to this point has assumed that a 1 percentage point change in a school's racial mix has the same effect on individual achieve- ment regardless of whether the school is initially all-white, all-black, or somewhere in between. Yet many advocates of desegregation assume that racial mix has nonlinear effects. Blacks often assume that something like a 50-50 mix of blacks and whites is optimal, at least for blacks. Whites often assume that a small black minority is optimal, at least for whites. Jencks and Brown (1975a) thought they had found evidence that blacks gained most in elementary schools that were 51 to 75 percent white. Summers and Wolfe (1977) found the same thing in Philadelphia. Both

152 INNER-CITY POVERTY IN THE UNITED STATES studies have serious limitations for our purposes, but they suggest that nonlinear effects deserve more attention. Effects of Racial Composition on Whites Table 4-4 implies that whites are at least as sensitive as blacks to the effects of a school's racial composition. Indeed, among sixth and ninth graders, white classmates appear to be worth two or three times as much to white students as to black students. Readers with long memories may find this result somewhat surprising, since Coleman et al. (1966) reported that racial composition explained a much larger fraction of the variance in black students' scores than in white students' scores. The reason for this difference, which Coleman et al. evidently overlooked, was that racial composition varied far more among blacks than among whites (see the bottom of Table 4-4~. Many other analysts (including Jencks, 1969) also overlooked this fact and used Coleman et al.'s findings to argue that desegregation would help blacks more than it hurt whites. If Table 4-4 provides a realistic estimate of desegregation's true effect on whites, mixing students from all-black and all-white elementary schools would lower white students' scores far more than it would raise black students' scores. Since there are many more whites than blacks in the nation as a whole, school desegregation on a national scale would increase the proportion of whites in the typical black student's school far more than it would reduce the proportion of whites in the typical white student's school. The costs of desegregation to the typical white would therefore be smaller than the benefits to the typical black. But because there are many more whites than blacks, the aggregate cost to society as a whole would still be substantial.36 We have several reasons for doubting the validity of this conclusion, however. First, imprecise measurement of students' family background and initial ability probably inflates the estimated effect of Proportion White more for whites than for blacks. White students' measured SES was quite strongly correlated with their school's racial mix in both the 1965 EEO and the NLS-72, which suggest that high-SES white parents avoided schools with high black enrollments. Controlling measured SES reduces the estimated effect of Proportion White on white achievement by a third. Controlling unmeasured aspects of SES and initial ability would therefore be likely to 36The expected effect of desegregation on mean test performance for the nation as a whole depends on whether the unstandardized coefficient of Proportion White is larger for blacks or whites. If the coefficient is positive and larger for whites, as it is for the sixth and ninth grades in Table 4-4, desegregation will lower the national average. If the coefficient of Proportion White is larger for blacks than for whites, desegregation will raise the national average.

GROWING UP IN A POOR NEIGHBORHOOD 153 reduce the estimated effect of Proportion White on white achievement by more than a third. Among blacks, in contrast, an individual's SES was almost uncorrelated with his or her school's racial mix in the 1965 and 1972 surveys, which suggests that high-SES black parents made little effort to enroll their children in predominantly white schools. If this explanation is correct, controlling a wider array of background measures would not reduce the estimated effect of Proportion White on black students' achievement. A second reason for skepticism about the estimates in Table 4-4 is that, at least in 1965, very few northern whites attended schools that were more than 25 percent black. Thus, the estimated effect of racial composition on white students in the EEO survey is based largely on differences among schools that are overwhelmingly white. Extrapolating to predominantly black schools could be quite misleading. Our third reason for doubting that desegregation lowers white students' achievement as much as Table 44 implies is that in Gamoran's (1987) HSB sample, which was 90 percent white, a school's racial composition had no consistent impact on cognitive growth between the tenth and twelfth grades. With mean SES, mean tenth-grade scores, and individual attributes in tenth grade held constant, students in all-white schools learned significantly more science but significantly less civics than students in schools with significant black enrollment. Averaging across all six HSB tests, students gained only 0.015 standard deviations more between tenth and twelfth grades if they attended all-white schools than if they attended all-black schools. Even allowing for the fact that all-white schools are also high-SES schools and have above-average tenth-grade scores, students in all-white schools only appear to have gained about 0.03 standard deviations more than initially similar students in all-black schools.37 Those who have reviewed studies of the first year of desegregation also claim that those studies show no effect of desegregation on white achieve- ment. St. John (1975), for example, reviewed 23 studies. lbro reported statistically reliable increases for whites, 3 reported statistically reliable de- clines, and 18 reported no statistically reliable change. This evidence is not 37Using the subsample of HSB schools that participated in the Administrator and Teacher Sur- vey, Bryk and Driscoll (1988) tell a somewhat different story about math scores than Gamoran tells. After controlling the same variables as Gamoran, plus two additional measures of socio- economic and racial diversity, Bryk and Driscoll found that Proportion Black had a statistically reliable positive effect on twelfth-grade math scores and that Mean SES had a reliable negative effect. Gamoran obtained the same signs in his math equations, but his coefficients were smaller and were not reliably different from zero. This apparent difference could reflect differences in the control variables included in each study, or it could reflect differences in the samples or esti- mation procedures. Hotchkiss (1984) examined the effect of mandatory busing on achievement in the HSB survey and concluded that the effect was small. He did not look at blacks and whites separately.

154 INNER-CITY POVERTY IN THE UNITED STATES as convincing as it sounds, however. Most of the studies cover relatively small samples. Thus, if one year of desegregation typically lowered white students' reading scores by, say, 0.10 standard deviations, most studies would find that the estimated effect was "statistically insignificant." We would need a meta-analysis that pooled all such studies to estimate the true effect. No one has done such an analysis. Conclusions About Cognitive Skills What we currently know about the effects of a school's mean SES on individual achievement can be summarized as follows: A high school's mean SES has very little effect on white ninth graders' subsequent cognitive growth. Bryk and Driscoll's (1988) work is the major exception here. A high school's mean SES may have an appreciable effect on black ninth graders' subsequent cognitive growth, but we would need a study using longitudinal data to be sure about this. An elementary school's mean SES appears to have an appreciable effect on both black and white students' cognitive growth, but again, longitudinal analysis is needed to be sure about this. We can summarize what we currently know about the effects of racial composition on test performance as follows: . The first year of school desegregation usually has small positive effects on black elementary school students' reading skills but not on their math skills. · Twelve years in a predominantly white northern school probably has a substantial positive effect on black students' achievement. · We do not know anything reliable about the cumulative impact of desegregated schooling in the South. The effect of desegregated schooling on white students is uncertain. This is a rather thin harvest from a quarter century of research. School desegregation has been one of the most controversial political issues of the past generation. Assumptions about its effect on both black and white achievement have influenced the behavior of judges, legislators, educators, and parents. Court-ordered desegregation has profoundly altered the lives of millions of children, especially in the South. Yet there has been no serious effort to assess its cumulative impact since the mid-1970s. The dearth of work in this area is particularly disturbing in light of the fact that the federal government has already collected data that could significantly advance our knowledge in this area. The High School and Beyond survey could tell us a lot more than we currently know about how

GROWING UP IN A POOR NEIGHBORHOOD 155 a high school's racial and economic mix affects the academic achievement of various kinds of students. If the Sustaining Effects study were readily available to scholars, it would be even more valuable, since it would allow us to estimate the cumulative impact of an elementary school's racial and economic mix on cognitive growth from first through sixth grades. The fact that neither the federal government nor individual social scientists have used the HSB and SE surveys to address these questions suggests that we need to rethink the way in which we organize social research on politically controversial topics. CRIME Regardless of whether we look at official police statistics or residents' reports about the frequency with which they have been victimized, we find more serious crime in poor neighborhoods than in affluent ones. Indeed, one major reason people move out of poor neighborhoods is to escape crime. Social scientists have invented a multitude of theories to explain why there is more crime in poor neighborhoods. Many of these theories assume that living in a poor neighborhood increases an individual's chances of committing serious crimes. But very little research has tried to test this assumption. School Effects: Nashville in the 1950s We located only two studies of how a school's social composition affected its students' chances of engaging in serious crime, and only one of these studies is useful for our purposes. Reiss and Rhodes (1961) studied 9,238 white males over the age of 12 who were attending Nashville-area schools in 1958. After excluding traffic offenses, Reiss and Rhodes found that the Davidson County juvenile court had judged 5.9 percent of these boys delinquent at some time between 1950 and 1958. Reiss and Rhodes then cross-tabulated delinquency rates by individual SES and a school's mean SES.38 Table 4-5 summarizes their results. Overall, 8.1 percent of low-SES boys had been judged delinquent, compared to 3.0 percent of high-SES boys. The socioeconomic mix of a boy's school also had a substantial effect on his chances of having been judged delinquent, independent of his own SES. Eking Table 4-5 at face 38Reiss and Rhodes assigned each boy to one of three SES levels based on his father's occu- pation. They assigned each of their 39 schools to one of seven SES levels, based on the mix of fathers' occupations reported by students. In order to increase cell sizes and reduce sampling error, in Table 4-5 we combined schools in their top two SES groups and in their middle two SES groups, leaving five school SES levels instead of seven.

156 INNER-CITY POVERTY IN THE UNITED STATES TABLE 4-5 Percentage of White Males Judged Delinquent in Nashville SMSA, 195(~1958 Father's Occupational Status School SES High Middle Low All High 1.8 1.4 0 1.6 (12 schools) (681) (419) (41) (1,141) Upper-Middle 3.9 3.7 4.4 3.8 (7 schools) (567) (1,039) (372) (1,978) Middle 3.6 5.6 6.5 5.5 (8 schools) (277) (892) (511) (1,680) Lower-Middle 3.4 7.5 8.4 7.4 (7 schools) (237) (1,026) (914) (2,177) Low 6.0 13.5 14.5 13.4 (5 schools) (289) (1,449) (1,267) (3,005 Total 3.0 6.0 8.1 5.9 Hi) (1,814) (3,799) (2,191) (7,804) NOTE: Number of cases in parentheses. SOURCE: Reiss and Rhodes (1961): Table 2, with traffic offenses excluded. value, a school's socioeconomic mix had far more impact on low-SES students than on high-SES students. For low-SES boys, moving from a low-SES school to an average school lowered the probability of being judged delinquent from .145 to .065. For high-SES boys, moving from an average school to a high-SES school only raises the probability of being judged delinquent from .018 to .036. Economic segregation thus increases delinquency among low-SES boys far more (.080) than it reduces delinquency among high-SES boys (.018~.39 Unfortunately, Bible 4-5 is likely to exaggerate the impact of a school's mean SES on individual behavior. Reiss and Rhodes controlled only one measure of parental SES, the father's occupation. Part of what looks like a "school effect" is therefore likely to be an effect of unmeasured socioeconomic differences between boys in high-SES and low-SES schools. This may, in other words, be another instance in which multiple regression equations would tell a different story from cross-tabulations. 39Reiss and Rhodes view the low delinquency rate of low-SES boys in high-SES schools as evi- dence that the dominant socioeconomic group in a given school, be it high-SES or low-SES, has the highest delinquency rate. The rest of Table 4-5 does not support this view, however. We therefore assume that the absence of any delinquency among the 41 low-SES boys in high-SES schools is due to sampling error.

GROWING UP IN A POOR NEIGHBORHOOD 157 If a school's mean SES really had the effect on crime that liable 4-5 implies it had, eliminating economic segregation would have reduced deliquency, but not by much. If all high-SES schools had had a represen- tative socioeconomic mix, and if that had produced individual delinquency rates comparable to those in Nashville's middle-SES schools, the overall delinquency rate would have fallen by less than one-tenth (from 5.9 to 5.4 percent).40 If Table 4-5 overestimates the impact of a school's mean SES, as seems likely, economic desegregation would have reduced delinquency by even less. The second study that provides data on schools' social composition and crime (D. Gottfredson et al., 1987) covered students enrolled in 20 junior and senior high schools located in Baltimore, Charleston, Chicago, Christiansted, and Kalamazoo. Unfortunately for our purposes, the schools were selected because they had high crime rates. Because Gottfredson et al. did not adjust their results to take account of the fact that they sampled only high-crime schools, their findings do not tell us much about the effects of mean SES in representative samples of schools.4i Neighborhood Effects: Chicago in 1972 Reiss and Rhodes's work has two major limitations: inadequate con- trols for individual SES and an exclusively white male sample. Johnstone's (1978) work overcomes both these limitations, but it covers only 1,124 teenagers who were between the ages of 14 and 18 and lived in the Chicago 40We do not know for sure that the mean SES of what we have called "middle-SES" schools is the same as that for the sample as a whole, because Reiss and Rhodes do not report schools' mean SES. Assuming that moving all students to middle-SES schools left the delinquency rates for high-SES, middle-SES, and low-SES boys the same as those that Reiss and Rhodes found in middle-SES schools, the overall delinquency rate would be [~.036~1,814) + (.056~3,799) + (.065~2,191~]R,804 = 0.054. The expected eject of economic desegregation on crime derives from the fact that a school's mean SES appears to have nonlinear ejects on crime, even with individual SES controlled. We cannot test this hypothesis rigorously without knowing the mean SES of schools at each SES level. 4iD. Gottfredson et al. used factor analysis to construct two orthogonal indices of community characteristics, which they label "Disorganization" and "Affluence and Education." Unfortu- nately, the Disorganization index does not measure community disorganization in the vernac- ular sense, because it is constructed so as to be uncorrelated with "Affluence and Education." This is achieved by giving it strong negative loadings on education and family income and strong positive loadings on welfare use, poverty, female-headed households, and the like. In any event, the Affluence and Education measure had an average correlation of-.001 with respondents' reports of their involvement in interpersonal aggression, theft and vandalism, and drug involve- ment. Given the sampling procedure, this result does not prove that mean SES has no eject on crime. D. Gottfredson et al. also estimate standardized coefficients with individual-level vari- ables controlled, but because those equations control endogenous as well as exogenous factors, they are not useful for our purposes.

158 INNER-CITY POVERTY IN THE UNITED STATES metropolitan area in 1972. The survey asked them whether they had com- mitted various crimes, how often they had done so, whether they had been arrested, and whether they had been the victim of various crimes. Johnstone categorized individuals into three SES groups based on parental education, the occupation of the household head, and the inter- viewer's assessment of the household's material standard of living. He also categorized the 221 census tracts in which the respondents lived into three SES levels, based on the percentage of high school graduates in the tract, the percentage of labor force participants in professional or managerial occupations, and median family income. Johnstone defined "serious" crimes as those listed in Part I of the Federal Bureau of Investigation's Uniform Crime Reports. Ibble 16 shows the mean number of serious crimes and the mean number of arrests that teenagers reported, broken down by their own SES and their neighbor- hood's mean SES.42 As we would expect, low-SES teenagers reported hav- ing committed more serious crimes than did high-SES teenagers. Because the ratio of arrests to self-reported serious crimes did not vary consistently with individual SES, we cannot attribute differences in arrest rates to police prejudice against low-SES teenagers. Teenagers in low-SES neighborhoods also reported committing more serious crimes than those in high-SES neighborhoods. Arrest rates, in contrast, did not vary appreciably by neighborhood SES. In theory, this discrepancy could mean that teenagers in high-SES neighborhoods were less willing to report serious crimes than teenagers in low-SES neighbor- hoods. In order to accept this explanation, however, we must be willing to claim that either (1) there is no more crime in low-SES than in high-SES neighborhoods or (2) high crime rates in low-SES neighborhoods are at- tributable to high-SES teenagers who journey to low-SES neighborhoods to commit crimes. Neither of these theories is plausible. We conclude, there- fore, that the residents of low-SES Chicago neighborhoods did, in fact, commit more serious crimes than the residents of high-SES neighborhoods. It follows that self-reports are a better guide to neighborhood differences in criminal behavior than arrest rates are, at least in Chicago. The low ratio of arrests to crimes in low-SES neighborhoods could indicate that the police make less effort to solve a given crime if it occurs in a low-SES neighborhood, that the police have greater difficuytv identifvin~ culprits in low-SES neighborhoods, or both.43 , -a--A ~r 42The questionnaire did not cover all Part I offenses, so both the means and the differences between neighborhoods are biased downward. Johnstone also gave repeat offenders a maximum score of 2 when calculating means, so if the ratio of repeat offenders to total offended varies across neighborhoods, his results may be misleading. 43Smithts (1986~ analysis of police behavior in the Rochester, St. Louis, and Tampa metropolitan areas shows that when the police actually come in contact with someone whose behavior might

GROWING UP IN A POOR NEIGHBORHOOD TABLE 4-6 Self-Reported Serious Crimes and Arrests Per Person Among 14- to 1 8-Year-Olds in the Chicago SMSA, 1972 Census Tract SES Individual SES High Middle Low All High Serious cranes 0.62 0.69 1.80 0.81 Arrests 0.40 0.34 0.98 0.46 (N) (209) (1 1 1) (52) (372) Middle Serious crones 0.60 0.81 1.72 1.00 Arrests 0.14 0.38 0.72 0.41 (O (125) (207) (124) (456) Low Serious crimes 0.38 1.45 1.39 1.30 Arrests 0.08 0.44 0.50 0.43 At) (42) (125) (21 1) (378) AD Serious crones 0.59 0.96 1.55 1.03 Arrests 0.28 0.39 0.63 0.43 (N) (376) (443) (387) (1,206) NOTES: Number of weighted cases in parentheses. The total number of unweighted cases is 1,124. SOURCE: Johnstone (1978). 159 In Chicago as in Nashville, neighborhood effects appear to be nonlin- ear. Indeed, with Johnstone's detailed controls for individual SES, living in a high-SES rather than a middle-SES neighborhood does not seem to have any consistent effect on an individual's chances of committing serious crimes.44 Only low-SES neighborhoods appear to make a difference. Reiss and Rhodes's work leads us to expect that living in a low-SES neighborhood will raise the amount of serious crime among all sorts of teenagers, but that it will have its largest effect on low-SES teenagers. In Chicago, the story is more complex: lead to an arrest and all other circumstances are the same, they are more likely to make the arrest if they are in a low-SES neighborhood. We infer, therefore, that the low ratio of arrests to self-reported serious crimes in low-SES neighborhoods reflects the fact that the police identify fewer suspects per serious crime in low-SES neighborhoods, not a tendency to treat suspects more leniently. 44Johnstone does not report the variances of these measures, so we cannot do formal significance tests, but given his small sample sizes and what he says about how his measures were constructed, we doubt that the differences between high-SES and middle-SES neighborhoods are significant.

160 INNER-CITY POVERTY IN THE UNITED STATES · There are not enough high-SES teenagers in low-SES neighbor- hoods to draw any strong conclusions about how neighborhoods affect their crime rate. · Among middle-SES teenagers, living in a low-SES neighborhood appears to increase serious crime, just as it did in Nashville. · Among low-SES teenagers, living in a low-SES neighborhood ap- pears to lower the amount of serious crime by about a fifth. This is exactly the opposite of what Reiss and Rhodes found for low-SES whites in the Nashville-area schools. How are we to explain the fact that low-SES teenagers reported committing fewer serious crimes (and also reported fewer arrests for such crimes) if they lived in "bad" neighborhoods than if they lived in relatively "good" neighborhoods? Sampling error is one possibility. Johnstone does not provide enough information to determine whether the differences are sta- tistically reliable. Relative deprivation is a second possible explanation. Low-SES teenagers may commit more crimes when they are in constant contact and competition with higher-SES neighbors. Another question is how Johnstone could have found that low-SES teenagers committed fewer crimes when they lived in low-SES neighbor- hoods when Reiss and Rhodes found the opposite. Three possibilities come to mind. First, Johnstone studied census tracts whereas Reiss and Rhodes studied secondary school attendance areas. Second, Johnstone's low-SES sample is likely to have been mainly black, whereas Reiss and Rhodes's sample was all white. Third, Nashville in the l950s may have been different from Chicago in the 1970s. If living in a high-SES neighborhood really does increase low-SES teenagers' chances of committing serious crimes, as Johnstone's findings imply, redistributing low-SES teenagers more evenly across a metropolitan area would raise the crime rate rather than lower it. Simcha-Fagan and Schwartz (1986) also studied neighborhood effects on teenage crime, but their work is not useful for our purposes. They studied 553 males between the ages of 11 and 18 who were living in 12 New York City neighborhoods in 1982. Unfortunately, their sample excluded neighborhoods whose mean SES was more than 0.75 standard deviations above or below the citywide mean. Their results do not, therefore, say anything about neighborhoods with high concentrations of poverty. They also selected their neighborhoods so as to minimize the correlation between neighborhood SES and other neighborhood characteristics, so even within the narrow range covered by their sample, their high-SES and low-SES neighborhoods are atypical. Thus, we cannot draw any conclusions from their work about the effects of a neighborhood's mean SES on crime.

GROWING UP IN A POOR NEIGHBORHOOD Rearrests Among Ex-Offenders: Baltimore in 197~1980 161 Although studies of recidivism among ax-offenders deal with adults rather than teenagers, we include them here for three reasons. First, most ax-offenders are quite young. Indeed, some are still teenagers. Second, given the paucity of evidence about how neighborhoods affect teenagers, it seems foolish to ignore potentially relevant evidence on young adults. Third, theories about how a neighborhood's mean SES ought to affect teenage crime appear to make similar predictions about adult ax-offenders. Epidemic models suggest, for example, that both teenagers and ax-offenders should commit more crimes if they live in neighborhoods that already have high crime rates. Furthermore, since the police are more overworked in neighborhoods with high crime rates, crime is less likely to be punished in such neighborhoods, which should lead to still more crime both among teenagers and among ax-offenders. If it is harder to find unskilled jobs in low-SES neighborhoods, as many assume, this should also encourage both teenagers and ax-offenders to turn to crime. Relative deprivation theories, in contrast, suggest that both teenagers and ax-offenders may be more resentful if they live in high-SES neighborhoods. (Ex-offenders are, however, freer to move if they do not like having rich neighbors.) Both teenagers and ax-offenders should also have more lucrative nearby opportunities to commit crimes if they settle in high-SES neighborhoods. S. Gottfredson and Taylor (1986) tried to predict whether men released from Maryland prisons between 1978 and 1980 would be rearrested before January 1982, and if they were rearrested, how serious the charges would be. Their sample Included approximately 500 men who moved into 67 different Baltimore neighborhoods. Once they controlled for a man's personal characteristics and criminal history, neighborhood characteristics had no effect on his chances of being rearrested. Nor did neighborhood characteristics affect the seriousness of the offenses with which the police charged the men they rearrested.45 S. Gottfredson and Taylor (1987) extended their earlier analysis by using a somewhat different sample and including three direct measures of a 45s. Gottfredson and Taylor classified neighborhoods along three dimensions: racial composi- tion, a cluster of locally visible "incivilities" (such as graffiti) that predicted the neighborhood crime rate, and industrial versus residential land use. They did not include a neighborhood's mean SES, but both racial composition and neighborhood "incivilities" are strongly correlated with mean SES. None of their three neighborhood measures had a significant effect on recidi- vism. They also used the neighborhood's mean recidivism rate to predict individual recidivism. This raised R2 by .055. They describe this increase as highly significant, but their significance test ignores the fact that they used 55 degrees of freedom in computing neighborhood recidivism rates. There were only eight respondents in a typical neighborhood, so roughly one-eighth of the unexplained variance in recidivism should have fallen between neighborhoods by chance. That is roughly what they found, so there is no evidence of neighborhood effects.

162 INNER-CITY POVERTY IN THE UNITED STATES neighborhood's economic characteristics: household income, percentage of workers with white-collar jobs, and housing prices. They found no evidence that these three neighborhood characteristics affected either the likelihood that ax-offenders would be rearrested or the seriousness of the offenses with which they were charged. One important limitation of Gottfredson and ~ylor's studies is that the ax-offenders had not spent very long in their new neighborhoods at the time the authors assessed neighborhood effects. A third of the ax-offenders were rearrested within six months of release and half were rearrested within a year. None had been out of prison much more than three years when he was followed up. The sample size is also quite small, so small effects are unlikely to be statistically significant even if they are present. Conclusions about Crime The two best studies of teenage crime appear to show the following: Attending an affluent high school in the Nashville area in the 1950s lowered the likelihood that white teenagers would engage in behavior that led the county juvenile courts to judge them delinquent. This was especially true for low-SES whites. These effects are probably overestimated, however, because the authors only controlled one measure of parental SES. · Living in a poor Chicago-area neighborhood in 1972 increased the likelihood that middle-SES teenagers would report having com- mitted serious crimes, but it reduced the likelihood that low-SES teenagers would report having committed such crimes. We cannot separate the effects of race and SES in this study. · The Nashville study implies that reducing the geographic isolation of the poor would reduce the overall crime rate. The Chicago study implies that reducing the geographic isolation of the poor would increase the crime rate. · Studies in Baltimore suggest that the neighborhoods in which ex- offenders settle have no effect on recidivism rates. We badly need better studies of neighborhoods' impact on teenage crime. We especially need studies that focus on the effects of very poor neighbor- hoods, including large public housing projects. We also need studies that follow families as they move in and out of very poor neighborhoods and examine how such moves affect teenagers' behavior. TEENAGE SEXUAL BEHAVIOR Our concern in this section is with the factors that influence young people's chances of having children before they are sufficiently mature to

GROWING UP IN A POOR NEIGHBORHOOD 163 make good parents. There is no general agreement about the optimal age for parenthood, but almost everyone agrees that teenagers are "too young." The likelihood that teenagers will become parents depends on a number of distinct factors: the age at which they initiate intercourse, the frequency with which they have intercourse, the care with which they use contraceptives, and the proportion who get abortions when they conceive a child. The well-being of the children also depends on the proportion of teenage parents who get married and stay married. The social composition of a neighborhood or school can, in principle, influence all these factors, but it is likely to have more effect on some than on others. If we want to understand how growing up in a poor neighborhood influences teenagers' chances of having children, we need to look at the way neighborhoods influence each step along the way. Unfortunately, we cannot do this with currently available data. We located only three studies that dealt with neighborhood effects on teenage sexual behavior and only two that dealt with school effects on such behavior. Each of these studies .. looks at a different measure of sexual behavior. We begin by looking at the determinants of teenage motherhood, then look at teenage pregnancy, contraception, and age of first intercourse. Single Teenage Motherhood Crane (forthcoming) used the 1970 census sample described above to investigate the effects of a census tract's social composition on whether unmarried 16- to 18-year-olds had had a child. Because Crane wanted to separate neighborhood effects from family background effects, and because the census only provides data on teenagers' family background if they still live with their parents, he looked only at 16- to l~year-old girls who were still living at home.46 Crane found that living in a poor neighborhood substantially increased the probability that 16- to 18-year-old girls had had a child out of wedlock, even after controlling parental education, occupation, and income. This pattern recurred for both blacks and whites. Among black 16- to 18- year-old girls of average SES who were living at home, the proportion who had had a child was 7 percent in the best neighborhoods, 10 percent 46Crane reports that 85 percent of all 1~ to 18-year-old girls lived with their parents in 1970, but only a third of all 16- to 18-year-old mothers were living at home. Among those still living at home, Crane found no relationship between parental income and having had a child out of wedlock. This startling result suggests that girls who have children out of wedlock are more likely to leave home if their parents are poor. This kind of sample selection bias would not affect Crane's estimates of neighborhood effects so long as it was based entirely on family income, not neighborhood characteristics. Crane did not try to determine whether neighborhoods influenced the probability that teenagers would live at home.

164 INNER-CITY POVERTY IN THE UNITED STATES in average neighborhoods, and 16 percent in the worst neighborhoods. Among whites, the rate was less than 1 percent in all but the very worst neighborhoods, where it rose to 3 percent. The effect of living in the very worst neighborhoods was especially marked in big cities. Including teenage mothers who no longer lived at home would have raised the proportions of 16- to 18-year-old mothers substantially, but would not necessarily raise the estimated effect of growing up in one neighborhood rather than another. Abrahamse et al. (19~) have also investigated the effect of social context on teenage girls' chances of becoming unwed mothers, but they looked at the effect of a high school's social mix rather than the effect of a neighborhood's social mix. They studied tenth-grade girls who participated in the HSB survey in 1980. Leo years later 4.3 percent of the whites, 9.5 percent of the Hispanics, and 13.0 percent of the blacks had become mothers. Of these mothers, 70 percent of the whites, 53 percent of the Hispanics, and 6 percent of the blacks had married. The HSB survey asked tenth graders, "Would you consider having a child if you weren't married?" l~venty-two percent of the white, 26 percent of the Hispanic, and 42 percent of the black tenth graders said yes or maybe. Abrahamse et al. defined schools as "accepting" if the proportion of students answering yes or maybe exceeded the national median for respondents of a given race. After controlling girls' own race, family background, and academic ability, attending an "accepting" school raised a typical tenth grader's probability of having a child out of wedlock during the next two years from about 1.1 to 4.5 percent among whites, from 3.4 to 6.1 percent among Hispanics, and from 9.4 to 11.9 percent for blacks. Abrahamse et al. report a modest relationship between family back- ground and the probability that a teenager would have a child out of wedlock, but they do not report the relationship between a school's mean SES and peer acceptance of having children out of wedlock Thus, while their findings are broadly consistent with the view that low-SES schools encourage out-of-wedlock childbearing, we cannot estimate the strength of the relationship. Mayer (forthcoming) used the HSB data to estimate the effect of a high school's social mix on tenth graders' chances of having a baby prior to expected graduation. Among students of any given ethnic background and SES, attending school with either low-SES or minority classmates increased the likelihood of having a baby before graduating. A one standard deviation reduction in parental SES increased the average non-Hispanic white tenth grader's chances of having a baby from 4.2 to 7.4 percent. A comparable change in the mean SES of her classmates increased her chances of having a baby from 4.2 to 6.4 percent. These effects were larger for low-SES students. For blacks, a one standard deviation decline in parental SES raised the probability from 7.9 to 12.3 percent, while a one standard

GROWING UP IN A POOR NEIGHBORHOOD 165 deviation decline in a school's mean SES only raised the probability from 7.9 to 8.6 percent. Pregnancy Hogan and Kitagawa (1985) studied 1,078 unmarried black Chicago women who were between the ages of 13 and 19 in 1979. Hogan and Kitagawa used a census tract's racial composition, median family income, percent poor, male-female ratio, children per ever-married female, and juvenile police contacts to construct an index of neighborhood quality. They then classified the best quarter of all tracts as high-SES, the middle half as middle-SES, and the bottom quarter as low-SES neighborhoods. Their low-SES neighborhoods are the kind that social critics have in mind when they talk about "concentrated poverty" and "the underclass." Four out of five respondents in these neighborhoods lived in public housing. In this sample of unmarried Chicago blacks, 28 percent had become pregnant by the age of 19. When Hogan and Kitagawa controlled the parents' marital status, fertility, and SES (a trichotomy based on parental education, occupation, income, employment status, and housing character- isticsy, they found that teenage pregnancy rates in medium- and high~uality neighborhoods did not differ in statistically reliable or substantively signif- icant ways. Living in a low-SES neighborhood did, however, raise black teenagers' chances of becoming pregnant in a given month by a third. This pattern is consistent with Johnstone's findings for teenage crime in Chicago, which also show negligible differences between middle-SES and high-SES neighborhoods. Hogan and Kitagawa also compared girls living on Chicago's West Side with girls living elsewhere in the city mostly on the South Side. The West Side was not settled by blacks until after World War II. It has fewer community organizations and churches, more first-generation migrants from the South, and a reputation for being an especially "bad" area. After controlling both family characteristics and the demographic attributes of census tracts, they found that living on the West Side still raised the pregnancy rate in a given month by two-fifths. This may be evidence that institutions matter, that culture matters, or both. It certainly suggests that if we want to understand neighborhood effects on teenagers' sexual behavior we need to look at more than just neighborhoods' demographic attributes. Hogan and Kitagawa did not investigate whether poor blacks were more sensitive to neighborhood quality than middle-income blacks, but the strong effect of very bad neighborhoods is what we would expect if low-SES black girls were more sensitive than high-SES black girls to their neighborhood's mean SES. Whatever the explanation, their findings imply

166 INNER-CITY POVERTY IN THE UNITED STATES that distributing poor blacks more evenly across the city would somewhat reduce the black teenage pregnancy rate. Contraception Hogan et al. (1985) used the Chicago survey described above to in- vestigate contraceptive use among black teenagers. They found that with family background controlled black girls in very poor neighborhoods were only half as likely as those in better neighborhoods to use contraception at the time of first intercourse. This seems likely to explain a good part of the neighborhood effect on pregnancy rates. The data do not indicate whether teenagers in better neighborhoods were more likely to use contraception because they had better information about it or for other reasons. Age at First Intercourse Furstenberg et al. (1987) used a national sample of 15- and 16-year- olds to investigate the effect of a school's racial mix on whether its students had had sexual intercourse. Among the 33 black teenagers in classrooms in which at least four-fifths of the other students were black, 67 percent reported having had intercourse. Among the 60 blacks in classrooms in which less than four-fifths of the other students were black, only 40 percent reported having had intercourse. This difference persisted with mother's education controlled. Because black parents' SES is not strongly correlated with their chil- dren's chances of attending school with whites, the fact that Furstenberg et al. control only one measure of parental SES does not seriously threaten the validity of their findings.47 Sampling error is a more serious problem. The 95 percent confidence interval for the difference in sexual activity between black teenagers in all-black versus racially mixed classrooms runs from 7 to 47 points. The former effect would be trivial, the latter huge. Thus, Furstenberg et al.'s findings suggest that school segregation encour- ages early sexual activity among black teenagers, but they do not indicate whether the effect is large or small. Among whites, being in a classroom that was more than one-fifth black raised the likelihood of having had had intercourse from 17 to 24 percent. Because of small sample size, however, the true effect of having 47Another potential problem is that low-sconng blacks in racially mixed schools are often tracked into overwhelmingly black classes while high-scoring blacks more often end up in racially mixed classes. The racial composition of black students' classrooms may, therefore, be a proxy for their academic aptitude as well as their school's racial composition. Low test scores sharply increase the likelihood that teenage girls will become pregnant (Sum, 1986~. Low scores may also be associated with early sexual activity.

GROWING UP IN A POOR NEIGHBORHOOD 167 black classmates could be to raise the proportion of whites who had had intercourse by as much as 23 points or to lower it by as much as 9 points. This range of uncertainty is too wide to justifier any firm conclusions. Hogan and Kitagawa (1985) also estimated the effect of neighborhood characteristics on the age at which black teenage girls become sexually active. They found that girls in low-SES Chicago neighborhoods became sexually active earlier than those in higher SES neighborhoods. They do not present multivariate results suitable for estimating neighborhood effects with family background controlled, but the evidence they do present suggests that neighborhood effects on age at first intercourse and pregnancy are quite similar.48 Conclusions The evidence on how neighborhoods and schools influence teenagers' sexual behavior is thin, but it suggests that teenagers' sexual behavior is quite sensitive to their classmates' and neighbors' SES and race. Holding race and family background constant, 1~ to l~year-old girls were substantially more likely to have had children out of wedlock if they lived in poor neighborhoods than if they lived in average neighborhoods. Black girls from very poor neighborhoods were also less likely to use contraception and more likely to become pregnant than black girls from similiar families who lived in better neighborhoods. Black 15- and 16-year-olds in largely black schools were also more likely to have had sexual intercourse than those in predominantly white schools. Despite its limitations, the available evidence suggests to us that neigh- bors and classmates probably have a stronger (or at least more consistent) effect on sexual behavior than on cognitive skills, school enrollment deci- sions, or even criminal activity. LABOR MARKET SUCCESS The literature relating neighborhood and school characteristics to ado- lescents' eventual labor market success is extremely sparse. We located only five studies that threw light on this question. 48Hogan and Kitagawa present an equation that controls family background, whether the girl's parents supervised her dating behavior, whether the girl had a sister who had been a teenage mother, and the girl's career aspirations. The last three measures depend partly on a neighbor- hood's mean SES, so controlling them biases the neighborhood coefficients towards zero. With these three variables controlled, a neighborhood's mean SES does not affect either the age at which girls became sexually active or whether they become pregnant.

168 INNER-CITY POVERTY IN THE UNITED STATES TABLE 4-7 Effects of Neighborhoods' Mean income and Racial Canposiiion in 1968 on Black and White Men's Hourly and Annual Earrungs in 1978 Neighborhood Characteristic Mean Family income ($1,000s) Dependent Variable Percent White (O to 1~) Blacks Whites Blacks Whites annual earnings B .030 .003 .0049 .0071 (SeE.) (.035) (.012) (.0025) (.0027) hourly earnings B (S.E.) hours worked (B1 - B2) Sample sizes Annual earnings Hourly earnings -.005 .001 (.020) (.007) .035 177 348 147 292 .0024 .0052 (.0014) (.0018) .002 .0025 .0019 NOTE: See Table 12 for means of neighborhood charactensiics. SOURCE: Datcher (1982). Neighborhood Effects Datcher (1982) used the Panel Study of Income Dynamics to estimate the effect of the neighborhood in which an urban male grew up on his earn- ings in early adulthood. Her sample covered 525 men who were between the ages of 13 and 22 and living at home in 1968. She used 1970 census data to estimate the percentage of whites and the mean family income of families living in the respondent's postal zip code. When estimating the effect of these two neighborhood characteristics, she controlled region, to- tal family income in 1968, whether the family received government transfer payments in 1968, the number of children in the family, the family head's education, whether the head expected his or her children to attend college, and whether the head liked to do "difficult and challenging things"; she did not control the family head's occupation or the family's income in years other than 1968. She used these variables to predict men's earnings in 1978, when they were between the ages of 23 and 31 Bible 4-7 summarizes her findings. Whereas a neighborhood's mean income appeared to be its most salient characteristic when predicting how much education its residents would get (see Bible 4-2), a neighborhood's racial composition has more effect on how much money its residents make when they grow up. Young urban whites typically lived in zip code areas that were 94 percent white in 1970,

GROWING UP IN A POOR NEIGHBORHOOD 169 while poor young urban blacks typically lived in zip code areas that were only 34 percent white. When all else was equal, growing up in an area that was 34 rather than 94 percent white lowered a man's expected 1978 earnings by 35 percent if he was white and by 27 percent if he was black.49 Once we control for a neighborhood's racial composition, the effect of its mean family income appears to be quite small. Urban whites typically lived in areas with mean incomes of $12,000 in 1970, whereas poor urban blacks lived in areas with mean incomes of $8,500. A $3,500 drop in the neighbors' mean income lowered a young man's expected 1978 earnings by 1 percent if he was white and 10 percent if he was black. The estimated effect on blacks is large enough to be of substantive interest, but its sampling error is very large, so the true value could either be huge or negligible. Datcher's results are striking but not definitive. Her sample was small, and it underrepresented those who left home at an early age. She omitted a number of parental characteristics whose inclusion might have altered neighborhoods' apparent effects. Her respondents were very young when she measured their earnings, and we know that the effect of family background on earnings increases with age, both in the PSID and in other samples (Jencks et al., 1979-~. Corcoran et al. (1989) use a more recent PSID sample to solve some of the above problems. Their sample covers about 800 men who were between the ages of 10 and 17 in 1968. They control more exogenous family characteristics and fewer potentially endogenous ones than Datcher did. They use a zip code's median family income, unemployment rate, pro- portion of families receiving Aid to Families with Dependent Children, and proportion of families headed by women to predict respondents' average economic position between the ages of 25 and 32. Because Corcoran et al. used four neighborhood characteristics that are highly correlated with one another, the estimated eject of each specific characteristic has a large sampling error and none is reliably different from zero. Nonetheless, their cumulative effect is quite large. Ibble 4-8 summarizes their results. Just as in Datcher's study, we can summarize Corcoran et al.'s findings by calculating the effect of growing up in a "bad" black area rather than an average white area. In this case we define "bad" black areas as those whose characteristics fall one standard deviation below the mean for all blacks. Bible 4-8 shows that these "bad" areas had median family incomes that were only 59 percent of the median in the typical white area. Bad black areas also had unemployment rates twice those in the typical white area, female headship rates almost four times those in the typical white 49The estimate in the text is calculated as 1-expt-60~.0071) forwhites and 1-expel-60~.0052) for blacks.

170 INNER-CITY POVERTY IN THE UNITED STATES TABLE 4-8 Effects of Growing Up in Poor Black Neighborhoods on Economic Status at Ages 25 to 33 Ln Median Percent Percent Family Males Female income Unemployed Heads Percent Families on Welfare Total White meant 9.21 4 9 4 Poor black meant 8.68 8 34 18 Difference -.53 4 25 14 Effect of one unit change in neighborhood mean if:2 Mother got no welfare Mother got $5,000/yr in welfare .018 -.017 .001 .010 .018 -.017 .001 .0035 Estimated effect of difference between average white neighborhood and poor black neighborhood if: Mother got no welfares -.010 -.068 Mother got $5,000/yr in welfares -.010 -.068 .025 -.140 -.193 .025 .049 -.004 1Taken from Corcoran et al. (1987), with proportions converted to percentages. Estimates for poor black neighborhoods are one standard deviation below the overall black mean. Waken from Corcoran et al. (1989), Table 2. For mothers receiving $5,000 a year from welfare, the estimated effect of a one point increase in the percentage of families receiving welfare in the zip code area is -.010 + (.027) ($5,000/$10,000) = .0035. 3(Line 3) (Line 4). Total is sum of columns 1 through 4. 4aine 3) (Line S). Total is sum of columns 1 through 4. area, and welfare recipiency rates more than four times those in the typical white area. Looking first at men whose parents had no income from welfare, we find that growing up in a "bad" black area rather than a typical white area lowers their expected earnings by 18 percent.50 Datcher's work, in contrast, implies that growing up in a bad black neighborhood can lower young men's earnings by 27 to 35 percent.5i Still, even the estimate based on Corcoran et al.'s work is sizable.52 Among men whose families got a lot 50The value in the text is calculated from Table 4-8, row 6. Expel-.193) = .82, so growing up in a bad black neighborhood lowers earnings by 1 - .82 = 18 percent. 51Corcoran et al. do not include racial composition in their model. The difference between their results and Datcherts could derive entirely from this omission, or it could derive from the fact that Corcoran et al. have better background controls and a larger sample. 52We cannot calculate sampling errors for our estimates of overall neighborhood effects from the data that Corcoran et al. present, but they are very large.

GROWING UP IN A POOR NEIGHBORHOOD 171 of income from welfare, in contrast, neighborhoods had almost no effect.53 Unfortunately, small sample size and large sampling errors mean we cannot have much confidence in either estimate. School Effects Mean SES Jencks and Brown (197Sb) estimated the effect of a high school's mean SES on students' subsequent occupational status and career plans using data from 91 predominantly white comprehensive public schools surveyed by Project Talent. Their initial sample included all ninth graders in those schools in 1960. The follow-up was conducted in 1968, five years after most of the sample had graduated. An increase of one standard deviation in a high school's mean SES raised men's subsequent scores on the Duncan scale (which runs from O to 96) by one point. The effect on women was equally small, as was the effect on career plans. Altonji (19~) estimated the effect of a school's racial mix and mean SES on male high school graduates' earnings, using data collected between 1972 and 1986 from the High School Class of 1972. His analysis included so many school characteristics that we could not estimate the reduced- form effect of mean SES, but he reestimated the model for us using only parents' education to measure mean SES. Holding men's family background constant, a one-year increase in mean parental education raised a man's expected wage by less than 1 percent. Racial Composition Crain (1970) surveyed 1,624 urban blacks who had attended high school between the late 1930s and the early 1960s. Roughly one in five had attended a racially mixed northern high school. Three results are noteworthy: Blacks were more likely to work in predominantly white occupations if they had attended racially mixed northern high schools than if they had attended all-black northern schools. Black women entered somewhat higher prestige occupations if they had attended racially mixed rather than all-black northern high schools, but black men's occupational prestige was only trivially higher if they had attended racially mixed schools. 53The coefficient of the interaction between neighborhood welfare recipiency rate and parental welfare income is only 2.1 times its standard error.

172 INNER-CITY POVERTY IN THE UNITED STATES · Black men earned 7 percent more if they had attended racially mixed northern high schools than if they had attended all-black northern schools. Younger northern blacks were more likely to have attended all-black schools.54 After controlling for this fact, Crain found that attending a racially mixed school raised black men's earnings only 4 percent.55 Altonji's (1988) work on the High School Class of 1972 tells much the same story as Crain's study. Attending an all-black rather than an all-white school lowered men's expected earnings by only 1 percent. (Altonji does not present separate analyses for blacks and whites, but he does control the respondent's race.) In the late 1960s, Project Concern offered a random sample of low- income black elementary schoolchildren from Hartford's inner-city schools an opportunity to enroll in white suburban schools outside Hartford. Crain and Strauss (1985) followed up these students in 1983, along with a sample of students from inner-city schools who had not been offered a chance to move to the suburbs. Most blacks who had been offered a chance to move had done so, but in order to avoid selection bias Crain and Strauss included all students invited to participate in the "treatment" group, regardless of whether they had actually participated. The follow-up covered approximately 700 students. Crain and Strauss do not document the differences between the sub- urban and inner-city schools in detail, but it seems fair to assume that blacks who attended suburban schools not only had more white classmates but had more affluent classmates, better teachers, smaller classes, and so forth. Those blacks invited to attend suburban schools were more likely than the control group to be in college at the time of the follow-up. Among respondents who were working, 30 percent of those invited to attend sub- urban schools held white-collar jobs, compared with 13 percent of those who attended inner-cicy schools. This was a highly significant difference. Suburban schooling did not affect unemployment rates. Crain and Strauss provide no data on earnings. 54Younger northern blacks were more likely to have attended all-black schools because the num- ber of all-black schools rose as the absolute number of blacks in the North rose. 55The 4 percent earnings advantage of blacks who attended racially mixed schools could be at- tributable to the fact that they got more schooling (Crain, 1971~. Crain does not report sampling errors, but assuming a typical dispersion of earnings, the 95 percent confidence interval for the effect of racially mixed schooling would run from a positive effect of about 21 percent to a neg- ative eject of about 13 percent.

GROWING UP IN A POOR NE GHBORHOOD Conclusions About Labor Market Success 173 The literature on neighborhoods, schools, and labor market success suggests two tentative conclusions: . Attending a racially mixed high school increases blacks' chances of working in "white" occupations. Attending racially mixed schools does not seem to have much effect on young men's eventual earn- ings. Growing up in a black zip code area or in one with high welfare dependency probably reduces both black and white men's eventual earnings. · A high school's mean SES does not have much effect on its grad- uates' economic prospects. Nor does a neighborhood's median income have much effect on young men's economic prospects once we control racial composition and welfare dependency. SUMMARY AND CONCLUSIONS In this review we have tried to determine how much effect the social composition of a neighborhood or school has on children's life chances. That is not a simple question. We limited ourselves to five outcomes: educational attainment, cognitive skills, criminal activity, sexual behavior, and economic success. But each of these outcomes has several components, and there is no reason to expect schools or neighborhoods to affect each component in the same way. A neighborhood's social composition may have more effect on teenage childbearing than on teenage sexual activity, for example. Or a school's racial mix may have more effect on sixth-grade test scores than on twelfth-grade scores. In practice, therefore, we ended up looking at more than a dozen outcomes. We also looked at the effects of four compositional measures: a neighborhood's mean SES, a neighborhood's racial mix, a school's mean SES, and a school's racial mix. With a dozen dependent variables and four independent variables, we would need at least 48 coefficients to summarize our findings. But that is just the beginning. We want to know not only the average strength of the 48 relationships, but also whether their strength varies with the race or SES of the child's family. For that, we need estimates of something like 200 relationships. If we had accurate estimates of all these relationships, we might be able to summarize them in a few elegant generalizations. But no estimates are available for many of the relationships that concern us, and sample bias, random sampling error, measurement error, and specification error distort the estimates that are available. Thus, even if neighborhood and school effects followed a simple underlying pattern in the real world, our chances of detecting it would be low.

174 INNER-CITY POVERTY IN THE UNITED STATES Our first and strongest conclusion is that there is no general pattern of neighborhood or school effects that recurs across all outcomes. Before offering any other generalizations we must therefore review what we have learned about each specific outcome that concerns us. Educational Attainment Studies of schools and studies of neighborhoods yield superficially contradictory conclusions about the determinants of educational attainment. In comparisons of white high school graduates who had the same test scores in ninth or tenth grade and came from families with the same SES, the mean SES of their classmates had almost no effect on their chances of planning to attend college, actually attending college, or graduating from college. There is some evidence that a high school's mean SES may have more impact on college attendance among blacks than among whites, but that evidence is not conclusive. A high school's mean SES does appear to affect entrants' chances of graduating, even after we control family background, but we do not know if this effect persists with entrants' test scores and plans controlled. Teenagers who grow up in affluent neighborhoods end up with more schooling than teenagers from similar families who grow up in poorer neighborhoods. This is probably partly because teenagers in affluent neigh- borhoods are more likely to finish high school than teenagers from com- parable families in poorer neighborhoods. We do not know whether high school graduates from similar families are more likely to attend college if they grow up in affluent neighborhoods. The effects of a school's racial composition on students' educational attainment are even less certain than the effects of its socioeconomic composition. Whites who graduated from racially mixed high schools in 1972 were as likely to attend college as those who graduated from all-white schools. Northern blacks who attended all-black high schools during the 1960s and early 1970s were more likely than those who attended racially mixed schools to plan on attending college, but they were less likely to enter college and less likely to remain there. In the South, attending a racially mixed high school reduced a black student's chances of attending college in 1972, when school desegregation was just beginning. No data on this point are available for more recent years. Cognitive Skills Studies of how a school's mean SES affects students' academic achieve- ment yield mixed results, depending on the students' race and grade level.

GROOVING UP IN A POOR NEIGHBORHOOD 175 A high school's mean SES does not seem to affect the amount white stu- dents learn between ninth and twelfth grade, but it may have an effect on how much black students learn. An elementary school's mean SES appears to have a substantial effect on how much both black and white students learn, but we cannot be sure of this without longitudinal studies. A school's racial composition has different effects from its socio- economic composition, even though the two are highly correlated. Data collected in 1965 and 1972 suggest that northern blacks at all grade levels learned more in predominantly white schools than in predominantly black schools, even with family background controlled. This was not true in the South, at least in 1972. We found no studies of this issue using more recent data. Most experts believe that a school's racial mix does not affect white students' achievement, but the evidence for this view is not conclusive. The first year of school desegregation usually has a small positive effect on black elementary school students' reading skills but not on their math skills. Unfortunately, the numerous studies covering one year of desegregation provide no useful information on the cumulative impact of attending racially mixed schools from first through twelfth grade. Crime Despite the existence of many complex theories about the ways in which neighborhoods affect teenage crime, the evidence for such effects is thin and contradictory. Regardless of their SES, white Nashville-area teenagers were more likely to have been arrested for serious crimes in the l950s if they attended school with low-SES classmates than if they attended school with high-SES classmates. In a study that controlled a broader array of background characteristics and pooled blacks with whites, middle-SES Chicago teenagers also reported committing more serious crimes in the early 1970s if they lived in poor neighborhoods. But contrary to what most people assume about the effects of concentrated poverty, poor Chicago teenagers reported committing fewer serious crimes if they lived in poor neighborhoods. Teenage Sexual Behavior Among unmarried 1~ to l~year-old girls living with their families in 1970, those living in very poor neighborhoods were considerably more likely so have had a child than those living in more affluent neighborhoods. This was true even with parental SES controlled and it was true for both blacks and whites. Black teenagers who lived in very poor Chicago neighborhoods in 1979 were also more likely to have become pregnant than those who lived in more affluent neighborhoods, and this remained true with parental

176 INNER-CITY POVERTY IN THE UNITED STATES SES controlled. The effect of poor neighborhoods on pregnancy appeared to derive both from the fact that girls from poor neighborhoods initiated intercourse younger and from the fact that they were less likely to use contraception. With mother's education controlled, blacks in classrooms that were more than four-fifths black also reported having initiated sexual intercourse earlier than blacks in classrooms that were less than four-fifths black. Labor Market Success Growing up in an urban neighborhood that is either predominantly black or has a high rate of welfare dependency reduces men's chances of finding well-paid jobs in adulthood. A neighborhood's median income does not appear to affect young people's economic prospects independent of its racial mix or welfare recipiency rate. Blacks who attend racially mixed schools are more likely to work in white-collar occupations than blacks who attend all-black schools. We found no evidence that a school's racial mix or mean SES affected its students' economic success independent of their own family background. Empirical Generalizations Social scientists need to be very cautious about estimates of neigh- borhood or school effects that control only one or two family background characteristics. As a rule, the more aspects of family background we control, the smaller neighborhood and school effects look. Initially, for example, we thought that attending a low-SES high school substantially reduced twelfth graders' chances of attending college. ~day, using more elaborate background measures, we are reasonably certain that the effect is trivial. The same pattern may hold for other outcomes. The literature we reviewed does not, therefore, warrant any strong generalizations about neighborhood effects. Based on what we now know, however, we offer two tentative hypotheses: · When neighbors set social standards for one another or create institutions that serve an entire neighborhood, affluent neighbors are likely to be an advantage. When neighbors compete with one another for a scarce resource, such as social standing, high school grades, or teenage jobs, affluent neighbors are likely to be a disadvantage. . Because the balance between these two kinds of influence varies from one outcome to another, there is no general rule dictating that affluent neighbors will always be an advantage or a disadvantage. Nor is there any

GROWING UP IN A POOR NEIGHBORHOOD 177 general rule about how large the advantage or disadvantage will be relative to other determinants of children's life chances. Our best guess is that better data would support the following empirical generalizations: · Advantaged classmates encourage both rich and poor children to learn more in elementary school, finish high school, and delay sexual intercourse. Advantaged classmates lower both rich and poor students' grades. Advantaged classmates have no effect on high school seniors' chances of attending college. Advantaged neighbors discourage teenagers from having children out of wedlock, encourage teenagers to finish high school, and increase teenagers' future earnings. Advantaged neighbors discourage crime among affluent teenagers but encourage it among poor teenagers, at least if they are also black. The evidence we reviewed does not allow us to draw even tentative conclusions about whether the poor gain more from residential or school desegregation than the rich lose. There is some reason to think that blacks may gain more from school desegregation than whites lose, but the evidence on this point would not convince a skeptic. Methodological Implications If social scientists want to make research on neighborhoods useful to public officials and legislators, they need to alter their analytic methods in at least three ways: · Future research should pay more attention to the most politically salient and easily understood differences between neighborhoods and schools, such as their poverty rate and racial composition. The effects of a school or neighborhood's poverty rate and racial mix should be estimated with no other neighborhood characteristics controlled.56 · Future research should report whether the effects of racial com- position and poverty rates are linear. If the effects are roughly linear, as social scientists tend to assume, moving the poor to more 56Reporting reduced-form results of the kind we described above does not rule out estimat- ing multivariate models that look at the effects of many different neighborhood characteristics simultaneously. In most cases, however, the number of neighborhoods is too small and neighbor- hood characteristics are too highly correlated with one another to separate the effects of specific advantages or disadvantages with much confidence.

178 INNER-CTIY POVERTY IN THE UNITED STATES affluent neighborhoods will redistribute the cost of having poor neighbors from the poor to the more affluent, but it will not reduce the costs to society as a whole. Such a change is unlikely to win broad political support. · Future research should investigate whether poor families are more sensitive than affluent families to neighborhood and school char- acteristics. If poor families gain more from living in a richer neighborhood than affluent families lose from living in a poorer neighborhood, reducing economic segregation can yield significant benefits to society as a whole. If affluent families lose more than poor families gain, reducing economic segregation will have signif- icant overall costs. The same logic applies to race. Implications for the Organization of Research Everyone believes- that both residential segregation and school segre- gation have important social consequences. Home buyers believe it, which is why they are willing to pay more to live in a good neighborhood. Judges believe it, which is why they turn cities upside down in order to desegregate their schools. Even committees of the National Research Council believe it, which is why they become concerned when the Census Bureau releases data suggesting that more people were living in very poor neighborhoods in 1980 than In 1970. Given the central role that everyone assigns to residential and school segregation, we were surprised by how little effort social scientists had made to measure the effect on individual behavior of either neighborhood or school composition. The subject is, of course, quite difficult to study. On reflection, however, we found this explanation for its neglect unconvincing. All social science problems are difficult, almost by definition. The easy questions were answered long ago. Compared with most of the problems that currently concern social scientists, estimating neighborhood and school effects is not especially difficult. The reason we don't know more is not that the questions are so hard to answer but that we have not invested much time or money in looking for answers. Efforts to estimate the effect of a high school's socioeconomic composition on graduating seniors' educational plans and subsequent at- tainment are the exception that proves this rule. Sociologists invested a lot of time and money in this problem, and the eventual convergence of their findings was remarkable. This is a case in which sociologists can truly claim to have learned something nobody knew to begin with, namely that a high school's socioeconomic mix has very little net effect on whether graduating seniors plan to attend college, actually attend college, or graduate from

GROPING UP IN A POOR NEIGHBORHOOD 179 college. Sociologists have also developed quite plausible explanations of why this is so. One obvious reason why social scientists have learned less about the other consequences of having low-SES classmates is that they have collected less data on those outcomes. Every follow-up of high school seniors asks about their educational attainment. Many follow-ups also ask about labor market experiences, but few studies follow graduates long enough to get meaningful estimates of how much they are likely to earn when they grow up. Few follow-ups ask about sexual behavior or criminal activity. None tests high school graduates to see how much they remember of what they studied in school. In principle, it should be easier to follow elementary school students through secondary school to see whether their elemen- ta~y school's social composition has long-term effects on their cognitive development, but no one has done this either. We know less about neighborhood effects than about school effects because collecting data on neighborhoods is more expensive than collecting data on schools. Only the Census Bureau has enough money to collect data on the socioeconomic composition of large representative samples of neighborhoods, and it has released only one data tape that includes both individual records (cleansed of identifying information) and data on the individual's neighbors. The only way to link individual characteristics and neighborhood characteristics, therefore, is to conduct private surveys of individuals and then add census data on the neighborhoods in which respondents live. Because data have been so scarce, there has never been an "invisi ble college" of social scientists grappling with the problems of estimating neighborhood effects, encouraging one another to use the best available analytic methods, criticizing questionable results before they reach print, or replicating important results after they are in print. Without such an invisible college, no field of inquiry makes much progress. If funding agencies wanted to encourage research on problems of this kind, the first step would be to make money available for collecting appropriate data. But while data collection is a necessary first step, it will not suffice. Funding agencies must also create more incentives for talented scholars to analyze the data in ways that are useful to policy analysts. At the moment, scholars cannot expect many rewards for doing such work Like all scholars, economists, sociologists, and social psychologists tale mainly to one another. As a result, economists are interested in problems that interest other economists, sociologists are interested in questions that interest other sociologists, and social psychologists are interested in prob- lems that interest other social psychologists. Furthermore, these scholars' careers depend mainly on their success in finding answers to questions that

180 INNER-C~ POPERY IN THE UNWED STATES interest other members of their discipline. 1b worry about questions that only interest public officials and policy analysts is quite risky. If legislators and public officials want first-rate work on policy ques- tions, they will have to ensure that people who work on such issues can survive in universities. At present, their survival is problematic. A handful of public policy schools reward their faculty for doing such work but they are too few in number to provide a clear career line for young scholars. Despite widespread cynicism about the value of social science, we believe that research on neighborhood and school effects could tell us a lot if it were properly organized. This would mean a number of major changes: · Funding agencies would have to make a long-term commitment (e.g., 10 years) to research in this area. Social science research, like most other research, involves a lot of false starts. Funding agencies must expect this and must be willing to wait for better answers. When slow progress is politically or institutionally unacceptable, as it often is, investing in social science research is a mistake. · Funding agencies must make money available for collecting new data on a regular basis. · Funding agencies must find ways to create a group of technically competent scholars with a long-term commitment to understanding neighborhood and school effects. This means they cannot rely entirely on contract research firms to do their work. They must also involve university-based social scientists. 1b attract good university- based social scientists, funding agencies must give them enough time to do what they and their colleagues regard as professionally respectable work. Funding agencies also need more social scientists on their own staffs. Funding agencies without such staff members seldom spec- ify in appropriate empirical terms the policy-related question they want answered. Nor do they usually negotiate acceptable compro- mises between their agency's policy agenda and the disciplinary agenda of university-based scholars. Nor are they likely to make realistic judgments about how long it will take to answer a question correctly though even social scientists are almost always overly optimistic on this score. None of the above conditions is currently met. Those who fund applied social science research seldom stay interested in any question for more than a few years. Little money is available for data collection. Partly as a result, few scholars have shown sustained interest in the field over the past generation. Thus, while much could be learned, there is little prospect that much will be learned unless we alter the way we organize our efforts.

GROWING UP IN A POOR NEIGHBORHOOD 181 Public concern about geographically concentrated poverty and home- lessness is currently high. As a result, the federal government may spend substantial sums for low-income housing during the l990s. The way we make these expenditures could either increase or decrease the current level of housing segregation. If the government tries to "save" existing public housing projects, extreme concentrations of poverty will persist. If the government builds scattered-site housing or provides housing vouchers, residential segregation might decline, but less housing might also be built. At the moment, we have no way of knowing how changes in residential segregation would affect either adults or children. Nor is there any way we can answer such questions in the next year or two. This means that social science cannot provide reliable evidence to inform near-term changes in government policy. But it does not follow that there is no point in doing research on such questions. If we begin now, we might have some fairly reliable findings by the turn of the century. If we procrastinate, we will be as ignorant a generation hence as we are now. ACKNOWLEDGMENTS We are indebted to Georg Malt for his assistance in reviewing stud- ies of schools' effects on college plans and academic achievement, to Karl Alexander, Thomas Cook, Robert Crain, Roberto Fernandez, Adam Gamoran, Bennett Harrison, John Meyer, and Michael Wiseman for help- ful comments on earlier drafts, and to Anthony Bryk, James Davis, Frank Furstenberg, Stephen Gottfredson, Dennis Hogan, and Philip Morgan for checking our summaries of their work. Needless to say, any errors that remain are our own. The Center for Urban Affairs and Policy Research at Northwestern University provided financial assistance. REFERENCES Abrahamse, Allan F., Peter Morrison, and Linda J. Waite 1988 Beyond stereotypes: Who becomes a teenage mother? Santa Monica, Calif.: Rand Corporation. Alexander, Karl L., and Bruce K. Eckland 1975 Contextual effects in the high school attainment process. American Socio- log~cal Review 40:402-416. Altonji, Joseph G. 1988 The Effects of Family Background and School Characteristics on Education and Labor Market Outcomes. Department of Economics, Northwestern University. Evanston, Ill.

182 INNER-Cm POVERTY IN THE UNITED STATES Alwin, Duane F. 1976 Assessing school effects: Some identities. Sociology of Education 49:294-303. Alwin, Duane F., and Luther B. Otto 1977 High school context effects on aspirations. Sociology of Education 50:259-273. Armor, David J. 1984 The evidence on desegregation and black achievement. Pp. 43 67 in School Desegregation and Black Achievement. Washington, D.C: The National Institute of Education. Asch, Solomon E. 1951 Effects of group pressure upon the modification and distortion of judgment. In Groups, Leadership and Men, Harold Guetzkow, ed. Pittsburgh: Carnegie Press. Bott, Elizabeth 1957 Family and Social Network. London: Tavistock. Broman, Sarah H., Paul L. Nichols, and Wallace A. Kennedy 1975 Preschool IQ: Prenatal and Early Developmental Correlates. Hillsdale, NJ.: Lawrence Erlbaum. B~yk, Anthony S., and Mary Erina Dnscoll 1988 The High School as Community Contextual Influences and Consequences for Students and Teachers. National Center on Effective Secondary Schools, University of Wisconsin, Madison. Ask, Anthony S., and Stephen W. Raudenbush 1987 Application of hierarchical linear models to assessing change. Psychological Bulletin 101:147-158. Idward a more appropriate conceptualization of research on school effects: A three-level hierarchical linear model. American Journal of Education 97:65-108. Bryk, Anthony, and Yeow Meng Thum 1988 The Effects of High School Organization on Dropping Out: An Exploratory Investigation. Department of Education, University of Chicago, Chicago, Ill. Bureau of Justice Statistics 1987 Sourcebook of Cruninal Justice Statistics 1986. Washington, D.C.: U.S. Government Printing Office. Coleman, James S., E. Q. Campbell, C J. Hobson, J. McPartland, A. M. Mood, F. D. Weinfeld, and R. L York 1966 Equally of Educational Opp~ni~. Washington D.C.: U.S. Government Printing Office. Cook, Thomas D 1984 What have black children gained academically from school integration? Examination of the meta-analytic evidence. Pp. 7~2 in School Desegregation and Black Achievement. Washington, D.C.: The National Institute of Education. Corcoran, Mary, Roger Gordon, Deborah Laren, and Gary Solon 1987 1988 1989 Intergenerational Transmission of Education, Income and Earnings. Political Science Department, University of Michigan. Ejects of Family and Community Background on Men's Economic Status. Working Paper 2896, National Bureau of Economic Research, Cambridge, Mass. Crain, Robert 1970 School integration and occupational achievement of Negroes. Amencan Journal of Sociology 75:593~}6.

1~ 1~1 Scat Con and lbe amde~c acb~vemeD1 ~ Aft. ~ =:1-~. CmiD, Act, and ~m Rabat 1~8 3~1 meal ~m-11ion and bat Whew anenden~ and a~i~en1 ~ ~ ~ ~ 51~14~. 1~3 We act of ~- melb~ol~ on d~gm alion-a~menl Judy: ^ meu-ana~i~ _= ~' ~ ~ ~:8~54. Aid, Ace, and ~ Sag_ 1~5 SO ~g=ahon Ad Blat O~palionaI ~~inmen~: ~1~ tom a Angst ~dmenl. waler far SO O~an~don of tools Johns Hopkins Unknit. Cmne, ]ona~an Cab- be pallem of neigh each on nodal pmNems. Amino ~ At. Dalcher, Linda 1~2 Each of ~mmuni~ and Emit Sigmund on amens. ~ ~ =d ~~ ~:~1. O^, Amp ~ 1` he campus as a ~ And: ~ applimlion of lhe Get ~ Bait depart {a ==er decisions of hinge men. ~= ~' ~- 7~17~1. =tenbu~, hank lo, an, S. Philip hogan, -~in ^ ~m, and James Pellet 1~7 Dam di~=n~s id 1be liming of ad~=nl influx. -~ ~ SZ511-318. Oamomn, Adam 1~7 he s~liO=1ion of bigb ~b-1 learning ~podunili=. ~_ ~- ~ -133-133. Coll~dson, Denise C., Dim ]. McNeil, and Oat D. 00l1~d~n 1S7 Immune fluent on Ind~du~ DeUngu=~. Pair pawed al Be Manual haling of lbe Sweden Stem ~ ~inol~ Oman Coll~dson, Slepben D., and Ralpb B. Valor 1 ~1 ~ ~ ~ ~ ~ am. ~ 1~ in ~e ~' ~~ ~ At. James ad. ~ and ~ ]. Sampan, As. Nag Ah: Spdn~dag. ~mmuni~ ~nl~s and Amino oxides. ~ ~ ~ ~ Aft, ~ How and ad. Sit, A. Ados Her M-~% Stoned Off. Haunt ~ ad. 1~9 Sibyls and lbe slmliO=lion pa. ~= ~ ~ 7~- 611. 1~1 ~ ~ ~ =d ~' a. Sing, D.C.: Sedan S-ologi=1 align. Haunt -~n ad., William H. Sewell, and Duane E ID 1~6 Higb scb~1 eats on a~ie~menl. In ~~ =d ~_' = ant, William H. Shell, ~n ad. Faust and D^d ~albe~an, ads. Nag Ad: Endemic Pax. Hindelang, chisel 1., Avis Kiwi, and ~=ph O. Ells 1~1 hi. ~~ Hills, OliC: Sage. Hogan, Dennis a, and Bet ad. ~lagawa 1~3 ~e impala of nodal sla{us, amid Clue, and Neil on me Philip of black adol==nls. ~ ~ ~S 9:~. 1~7

184 INNER-CITY POVERTY IN THE UNITED STATES Hogan, Dennis, Nan Marie Astone, and Evelyn Kitagawa 1985 Social and environmental factors influencing contraceptive use among black adolescents. Famih,' Planning Perspectives 17:165-169. Hotchkiss, Lawrence 1984 Ejects of Schooling on Cognitive, Attitudinal, and Behavioral Outcomes. Technical Report. The National Center for Research on Vocational Educa- tion, The Ohio State University. Jencks, Christopher 1969 A reappraisal of the most controversial educational document of our time. The New York Times Magazine, August 10:12-13 and 33-44. 1972a The Coleman report and the conventional wisdom. Pp. 69-115 in On Equalay of Educational Opportunity, Frederick Mosteller and Daniel Patrick Moynihan, eds. New York: Random House. 1972b The quality of the data collected by the Equality of Opportunity Survey. Pp. 437-512 in On Equality of Educational Opporn~uty, Frederick Mosteller and Daniel Patrick Moynihan, eds. New York: Random House. forth- Is the underclass growing? In Christopher Jencks and Paul Peterson, eds., coming The Urban Underclass. Washington, D.C.: Brookings Institution. Jencks, Christopher, and Marsha Brown 1975a The ejects of desegregation on student achievement: Some new evidence from the Equality of Educational Opportunity Survey. Sociology of Education 48:126-140. 1975b The effects of high schools on their students. Harvard Educational Review 45:273-324. Jencks, Christopher, Marshall Smith, Henry Acland, Mary Jo Bane, David Cohen, Herbert Gintis, Barbara Heyns, and Stephen Michelson 1972 Inequality. New York: Basic Books. Jencks, Christopher, Susan Bartlett, Mary Corcoran, James Crouse, David Eaglesfield, Gregory Jackson, Kent McClelland, Peter Mueser, Michael Olneck, Joseph Schwartz, Sherry Ward, and Jill Williams 1979 Ho Gets Ahead? New York: Basic Books. Johnstone, John W. C. 1978 Social class, social areas and delinquency. Sociology and Social Research 63:49-72. Krol, Ronald A. 1980 A mete analysis of the effects of desegregation on academic achievement. The Urban Review 12:211-224. Mare, Robert 1980 Social background and school continuation decisions. Journal of He American Statistical A~socianon 75:295-305. Massey, Douglas S., and Nancy A. Denton 1987 Wends in residential segregation of blacks, Hispanics, and Asians: 1970-1980. American Sociological Review 52:802-825. Massey, Douglas S., and Mitchell L Eggers 1990 The ecology of inequality: Minorities and the concentration of poverty 1970-1980. A,nencan Joumal of Sociology 95:1153-1188. Mayer, Susan E. fonh- How much does a high school's racial and economic mix affect graduation coming rates and teenage fertility rates? In Christopher Jencks and Paul Peterson, eds., The Urban Underclass. Washington, D.C.: Brookings Institution.

GROWING UP IN A POOR NEIGHBORHOOD 185 McDill, Edward L, and Leo C. Rigsby 1973 Structure and Process in Secondary Schools: The Academic Impact of Ed~ca- tional Climates. Baltimore: Johns Hopkins University Press. Meyer, John W. 1970 High school effects on college intentions. American Joumal of Sociology 76:59-70. Michael, John A. 1961 High school climates and plans for entering college. Public Opinion Quarterly 25:585-595. Mosteller, Fredenck, and Daniel Patrick Moynihan, eds. 1972 On Equality of Educational Opportunity. New York: Random House. Myers, David E. 1985 The relationship between school poverty concentration and students' reading and math achievement and learning. Pp. D-17 to D-60 in Mary Kennedy, Richard Jung, and Martin Orland, eds., Poverty, Achievement and the Distri bution of Compensatory Education Services. Office of Educational Research and Improvement. Washington, D.~: U.S. Department of Education. Nelson, Joel I. 1972 High school context and college plans: The impact of social structure on aspirations.AmericanSociolog~calRevrew 37:143-148. Reiss, Albert J., Jr., and Albert Lewis Rhodes 1961 The distribution of juvenile delinquency in the social class structure. Amer- ican Sociological Review 26:720-732. Rosenbaum, James E., Leonard S. Rubinowitz, and Marilynn J. Kulieke 1986 Low-Income Black Children in White Suburban Schools. Center for Urban Affairs and Policy Research, Northwestern University. Evanston, Ill. St. John, Nancy H. 1975 School Desegregation Outcomes for Children. New York: John Wiley ~ Sons. Sewell, William H., and J. Michael Armer 1966 Neighborhood context and college plans. American Sociological Review 31:159-168. Sewell, William H., Robert M. Hauser, and Wendy C Wolf 1980 Sex, schooling, and occupational status. American Journal of Sociology 86:551-583. Simcha-Fagan, Ora, and Joseph E. Schwartz 1986 Neighborhood and delinquency: An assessment of contextual effects. C~m- inology 24:667-703. Smith, Douglas A. 1986 The neighborhood context of police behavior. Pp. 313-341 in Communities and Come. Albert J. Reiss, Jr., and Michael Tony, eds. Chicago: University of Chicago Press. Sum, Andrew 1986 Childbearing Behavior of Unmarried Women (20-24) in the United States and Their Relationship With AFQT Test Scores: Findings of the 1979 1981 NLS Interviews. Working paper, Center for Labor Market Studies, Northeastern University, Boston. Summers, Anita A., and Barbara L Wolfe 1977 Do schools make a difference? The American Economic Review 67(Septem- ber): 639-652.

186 INNER-CITY POVERTY IN THE UNITED STATES Thornton, Clarence H., and Bruce K. Eckland 1980 High school contextual effects for black and white students: A research note. Sociology of Education 53:2A7-252. Turner, Ralph A. 1964 The Social Content of Ambition. San Francisco: Chandler. Wilson, Alan B. 1959 Residential segregation of social classes and aspirations of high school boys. American Sociological Review 24:836-845. Wilson, William Julius 1987 The Tnuly Disadvantaged. Chicago: The Univemityof Chicago Press.

Next: 5 Residential Segregation, Job Proximity, and Black Job Opportunities »
Inner-City Poverty in the United States Get This Book
×
Buy Paperback | $80.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

This volume documents the continuing growth of concentrated poverty in central cities of the United States and examines what is known about its causes and effects. With careful analyses of policy implications and alternative solutions to the problem, it presents:

  • A statistical picture of people who live in areas of concentrated poverty.
  • An analysis of 80 persistently poor inner-city neighborhoods over a 10-year period.
  • Study results on the effects of growing up in a "bad" neighborhood.
  • An evaluation of how the suburbanization of jobs has affected opportunities for inner-city blacks.
  • A detailed examination of federal policies and programs on poverty.

Inner-City Poverty in the United States will be a valuable tool for policymakers, program administrators, researchers studying urban poverty issues, faculty, and students.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!