PSC. Research Reports. Population Studies Center. David R. Harris. The Flight of Whites: A Multilevel Analysis of Why Whites Move. Report No.

Transcription

1 David R. Harris The Flight of Whites: A Multilevel Analysis of Why Whites Move Report No Research Reports PSC Population Studies Center University of Michigan

2 The Population Studies Center at the University of Michigan is one of the oldest population centers in the United States. Established in 1961 with a grant from the Ford Foundation, the Center has a rich history as the main workplace for an interdisciplinary community of scholars in the field of population studies. Today the Center is supported by a Population Research Center Core Grant from the National Institute of Child Health and Human Development (NICHD) as well as by the University of Michigan, the National Institute on Aging, the Hewlett Foundation, and the Mellon Foundation. PSC Research Reports are prepublication working papers that report on current demographic research conducted by PSC associates and affiliates. The papers are written by the researcher(s) for timely dissemination of their findings and are often later submitted for publication in scholarly journals. The PSC Research Report Series was begun in 1981 and is organized chronologically. Copyrights are held by the authors. Readers may freely quote from, copy, and distribute this work as long as the copyright holder and PSC are properly acknowledged and the original work is not altered. PSC Publications Population Studies Center, University of Michigan S. University, Ann Arbor, MI USA

3 The Flight of Whites: A Multilevel Analysis of Why Whites Move by David R. Harris Research Report No April 1997 Abstract: This paper considers the question, Why do whites move? Geocoded data from the Panel Study of Income Dynamics is used to examine the effects of neighborhood racial composition, neighborhood socioeconomic status, household traits, dwelling characteristics, and geographic setting on white mobility incidence. Findings indicate that individual-level characteristics are consistently strong predictors of whether whites will move. In contrast, racial and class composition are rarely reliable predictors of mobility incidence. A notable exception is the case of white families with children approaching school age. These families are almost six times more likely to move if they live in a neighborhood that is at least 20% black instead of one that is less than 6% black. My findings suggest that while white flight may be an inappropriate characterization of whites moves in general, there is evidence to support the claim that whites with children are seeking to escape black neighbors. Data used: Geocoded Panel Study of Income Dynamics (PSID): U.S., The Author David R. Harris is Assistant Professor, Department of Sociology, and Research Associate, Population Studies Center, University of Michigan. Direct correspondence to David R. Harris, 1225 South University Avenue, Population Studies Center, University of Michigan, (drharris@umich.edu). Portions of the data were made available by the Inter-university Consortium for Political and Social Research at the University of Michigan. An earlier draft of the paper was presented at the 1997 Meetings of the Population Association of America, Washington, DC. In preparing this paper I have benefited greatly from comments by Christopher Jencks, Greg Duncan, Art Stinchcombe, and participants in the Survey Research Center Workshop at the University of Michigan.

4 1 The average American moves about once every five years. Moves occur for myriad reasons, including marriage, divorce, childbirth, and dissatisfaction with neighbors. The frequency of moves has consequences for neighborhoods. One way mobility incidence 1 affects neighborhoods is by altering their racial compositions, as the rate and character of outmigration partially determines levels of residential segregation. Segregation, in turn, increases poverty, school dropout, joblessness, teenage pregnancy, and crime (Massey and Denton 1993; Cutler and Glaeser 1997). Thus, information about mobility incidence is valuable because it enhances our understanding of the segregation process and improves our ability to fight concomitant social problems. In this paper I assess the relative importance of various determinants of white mobility. My analysis is limited to whites for two reasons. First, I want to encourage a comprehensive approach to integration policy. Many initiatives attempt to increase racial diversity by opening neighborhoods to blacks. The idea is that with fewer impediments to black mobility, integration can be achieved. Such policies may succeed in the short run, but they will not lead to stable integration unless they also consider white responses to integration. 2 If whites react by moving, then integration will lead to racial succession. A second reason for focusing on whites is to address the debate over why whites move. Many believe that whites move to escape black neighbors. The link between racial composition and white mobility incidence is so widely accepted that white flight has become a part of the American vernacular. Nevertheless, there is reason to believe that most whites moves are not racially motivated. Research suggests that a host of nonracial elements also affect mobility decisions. What remains unclear is the relative importance of each factor. By considering multiple determinants of white mobility incidence, this paper strives for a more complete response to the question, Why do whites move? Deciding to Move Whites move for many reasons. The most salient factors fall into one of five categories: neighborhood racial composition, nonracial neighborhood traits, household characteristics, dwelling attributes, and geographic setting. The first of these, neighborhood racial composition, is commonly forwarded as a primary reason why whites move. 3 Schelling s (1971) tipping model argues that whites have varying preferences for integration. When a black family moves into a neighborhood, it disturbs the racial equilibrium. In response, whites who were already in the most integrated neighborhood they could tolerate decide to move. These openings create opportunities for more black families to join the neighborhood. When the second wave of blacks arrives, their presence prompts the departure of whites who are slightly more tolerant than the whites who initially fled. Again, some of the newly available homes are acquired by blacks, and a third wave of whites leaves. This process continues, provided there are enough blacks who want to live in the neighborhood, until even the most tolerant whites feel uncomfortable and flee. One level at which the relationship between white flight and mobility decisions has been examined is that of expressed preferences. White survey respondents in several urban areas consistently voice a desire to live in predominantly white neighborhoods. On average, the optimal 1 Mobility incidence is the probability of moving from one s current home. 2 An example is the integration policy Massey and Denton advocate in American Apartheid (1993: ). The only initiative in their nine-point plan that addresses white residents is a recommendation that hate crimes be prosecuted more aggressively. 3 Most research examines white reactions to African-Americans. Work that does consider nonblack minorities finds that whites are not as averse to these groups as they are to blacks (Bobo and Zubrinsky 1996; Clark 1992; Ottensmann and Gleeson 1992; Schuman and Bobo 1988).

5 2 neighborhood is approximately 75% white. Desirability then falls precipitously, with no more than 16% of whites voicing a preference for neighborhoods that are at least 50% black (Clark 1992; Farley, Steeh, Jackson, Krysan, and Reeves 1993). Though white aversion to highly integrated areas remains strong, some evidence suggests it may be declining. Compared to the 1970s, whites now report greater comfort with integration, and fewer whites indicate that they would definitely move if a black person became their next door neighbor (Farley et al. 1993; Bobo, Schuman, and Steeh 1986). Though these studies suggest that whites tolerate low levels of integration, the link between whites words and their actions has been questioned. Clark (1992) wonders whether preference studies are observing an increase in the prevalence of code words and social censoring, what he terms socially modified responses (p. 463), rather than a genuine change in attitudes (see also Schuman, Steeh, and Bobo [1985]). A second approach to white flight is to observe the relationship between an area s racial composition and the probability that whites will move. The idea is to examine what whites do, not what they say they will do. The work of Frey (1979) represents the best example of this method. Examining data for the 39 largest metropolitan areas, Frey observes no relationship between the relative size of a central city s black population and the probability that its white residents will move. This prompts his conclusion that the term white flight is an inappropriate description of the suburbanward movement of city whites (p. 444; italics in original). While some work confirms Frey s results (Lee, Oropesa, and Kanan 1994; Goodman and Streitwieser 1983; South and Deane 1993), others disagree. Galster (1990) examines mobility patterns for Cleveland and finds both linear and nonlinear effects of percent black on white mobility incidence. Massey, Gross, and Shibuya (1994) also dispute Frey s work. Their examination of a national sample of white households yields clear evidence of racially motivated moves. As the contradictions in this literature suggest, biased assessments of white flight are common. Much work that purports to study whites responses to black neighbors employs problematic measures of racial aversion. One such tool is neighborhood diagrams (Farley et al. 1993). This method presents whites with diagrams depicting varying levels of racial integration and then asks them to select their ideal neighborhood. The diagrams do not explicitly comment on the nonracial aspects of these hypothetical neighborhoods. Similarly, many regression analyses that seek to understand the relationship between racial composition and white mobility fail to employ sufficient controls for nonracial factors (Galster 1990; Massey et al. 1994; Bobo and Zubrinsky 1996). The problem with these approaches is that they cannot distinguish between pure discrimination and white aversion to factors that are correlated with an increasing black population (e.g., declining school quality and higher poverty rates). Disentangling these effects is important if we are to truly understand the relationship between integration and mobility incidence. In addition to racial composition, a second aspect of neighborhoods that concerns whites is their neighbors social class (South and Crowder 1997; Gramlich, Laren, and Sealand 1992). This includes neighbors income, education, and occupation. The importance of these factors appears in response to survey questions. Schuman and Bobo (1988) asked whites how much they would mind if a black family moved in next door. Seventy-one percent of whites said they would not mind at all. Next, they asked a comparable sample of whites how much they would mind if a black family with the same income and education as them moved in next door. This time 79% of whites said they would not mind at all. When presented with a question that controlled for neighbors socioeconomic status (SES), aversion to blacks declined 8 percentage points. This change is

6 3 statistically significant, and strongly supports the claim that white mobility is affected by neighborhood SES. A third set of factors affecting mobility incidence reflect the demographic characteristics of households, and are thoroughly examined in Rossi s classic work, Why Families Move (1955). Writing before the urban rebellions, civil rights protests, and fair housing laws that would dramatically change the urban landscape in the 1960s, Rossi emphasized the role of life cycle changes in mobility decisions. By life cycle changes he means factors associated with age and family composition that lead families to conclude that their current home is no longer adequate. In a multimethod study of Philadelphia, Rossi observes more frequent moves by young households and families with many children. Young people experience job changes, marriage, and the start of a family, all of which lead them to reevaluate their housing needs. Large families place a premium on space and, due to concerns about influences on their children, are highly sensitive to neighborhood and school climates. In the years since Why Families Move was first published, several studies have confirmed the book s findings (Speare 1974; Lee et al. 1994; South and Crowder 1997; Wurdock 1981; South and Deane 1993), and others have extended Rossi s analysis to include additional household factors. Recently, researchers have been examining the effect of a household s SES on its propensity to move. Massey and his colleagues (1994) find that poor whites are twice as likely as more affluent whites to move in a given year. South and Deane (1993) consider total income, rather than poverty status, but similarly conclude that moving is more common among the less affluent. However, South and Deane do not present consistent results about the role of SES, as they also show a positive relationship between educational attainment and mobility incidence. Finally, Duncan and Newman (1976) present a nuanced discussion of SES effects by separating moves that occur for employment reasons from those that are driven by other motives. They find that family income is positively associated with moves undertaken to improve employment opportunities, but unrelated to moves motivated by dwelling or neighborhood characteristics. However, because only about 4% of households expected to move for employment-related reasons, Duncan and Newman s results should not be interpreted as implying that, in the aggregate, poorer whites move less often. A fourth set of factors that affect mobility incidence is dwelling characteristics. The dwelling characteristic that best predicts moving is whether the property is rented or owned. Nearly all studies of mobility patterns find that renters are much more likely than owners to move (e.g., Lee et al. 1994; South and Crowder 1997; Duncan and Newman 1976; Frey 1979; Galster 1990; Goodman and Streitwieser 1983; South and Deane 1993). One reason for this difference is that owners often decide to modify their current homes rather than seek new ones. If an owner feels his house is too small or realizes that its interior is outdated, he can build an addition or remodel. Leases bar most renters from undertaking such projects, so renters solve housing deficiencies by moving. Moving is also easier for renters. To move, homeowners must find a buyer and agree on a price. By contrast, a renter can move at the end of his/her lease without any concerns about finding someone to occupy the property. Also, because leases tend to cover fixed periods, renters are regularly presented with the opportunity to move. Finally, renters move more frequently because they often have other characteristics that are associated with high rates of mobility incidence. My examination of PSID data shows that renters are younger, poorer, and more likely than homeowners to have children. Fifth, geographic setting affects mobility incidence. Most studies overlook this relationship and either pursue case studies of a single urban area (Galster 1990; Taub, Taylor, and Dunham

7 4 1984), or sample from multiple areas, but fail to consider regional differences (Frey 1979; Goodman and Streitwieser 1983). The absence of regional controls is especially troubling in light of recent work that suggests mobility processes may be place-specific (South and Crowder 1997; Farley and Frey 1994; Lee and Wood 1991; Clark 1991). The danger of ignoring this variation is that factors whose effects differ between regions may be misinterpreted with respect to their direction and/or magnitude. Data and Methods This paper seeks to understand why whites move by evaluating data from the Panel Study of Income Dynamics. The PSID is a longitudinal survey conducted annually by the Survey Research Center of the University of Michigan. Initiated in 1968, the PSID now includes data on 37,500 individuals who resided in one of 4,800 initial sample households, were the offspring of those individuals, or were their coresidents. Due to an initial oversampling of low-income families, the later incorporation of a Latino subsample, and attrition, the PSID sample is not representative of the United States population. To correct for this discrepancy, PSID sample weights are applied. 4 The PSID focuses on economic and demographic behavior. Respondents are queried about sources of income, changes in family structure, the acquisition of job skills, residential mobility, and a host of other related issues. This paper uses a special geocoded version of the PSID that was recently prepared in response to growing interest in neighborhood effects. The file contains aggregate data from the 1970 and 1980 U.S. Censuses on the areas where respondents live, as well as codes representing respondents addresses at twelve geographic levels. The geocoded PSID allows me to examine both individual and contextual determinants of behavior. Though the PSID contains data from 1968 to the present, I only use files for 1980 through This constraint results from my desire to examine contemporary data, and the fact that no addresses after 1985 have been geocoded. I also eliminate families that experienced changes in head or spouse between 1980 and This second constraint is meant to facilitate the tracking of families across years (Hill 1992). To see why this is necessary consider the following example. At time 1 a family has head, H, and spouse, S. By time 2, H and S have divorced and formed new households, with H moving once over the interval and S remaining in the family home. The question is how many moves occurred between time 1 and time 2. Is this the same as a family that remains intact and moves once? My solution to this puzzle is to drop families that change their head or spouse. This action excludes about 19% of the 1980 white households, resulting in an underestimation of the role of family dynamics in moving decisions. However, because my primary focus is on the role of racial composition in moving decisions this inability to precisely estimate family effects in not a major concern. The geocoded PSID has two special features that make it an excellent source for studying mobility incidence. First, as has been discussed already, it contains information on individuals, households, dwellings, and geographic contexts. This allows for a multilevel analysis of the decision to move. Second, the availability of data for small areas facilitates a more precise examination of contextual effects. Numerous studies consider the role of metropolitan area (South and Deane 1993), county (Clark and Nieves 1994), or city (Frey 1979; Goodman and Streitwieser 1983) characteristics in white residential mobility. Often the conclusion is that contextual effects are unimportant. One reason for these weak effects may be that there is significant heterogeneity 4 The result of using sample weights is that standard errors must be computed with a Huber/White estimator (Greene 1993; StataCorp 1995).

8 5 within these areas. Whites may not care how many people in their metropolitan area are black, but might well be sensitive to the racial composition of the area close to their homes. Thus, data for smaller areas reduces the noise in contextual measures and facilitates a more precise analysis of areal effects. For this reason I use census tracts as proxies for neighborhoods. 5 Table 1 contains the names and descriptions of the variables in this paper. In my analysis the dependent variable is mobility incidence. It has a value of 1 for families that moved at least once between 1980 and 1985, and a value of 0 for those occupying the same residence throughout the five-year period. Table 1 also lists the predictors of mobility incidence. Each belongs to one of the five classes of explanations discussed in the previous section. Household traits include the head s age, whether he/she is married, his/her educational attainment, family income, and whether there are children living with the family. Dwelling characteristics are whether the home is a single-family residence, whether it is owned, and the ratio of occupants to rooms. The third group, geographic setting, identifies region and whether the home is in a large metropolitan area. Percent black is the sole measure of neighborhood racial composition. The last class of predictors, neighborhood SES, is represented by percent affluent, percent poor, percent highly educated, and percent unemployed. In response to evidence that the relationship between racial composition and white mobility incidence depends on the proportion of neighbors who are black (Galster 1990), I allow for nonlinear racial effects by using dummy variables. 6 Each dummy variable represents a segment of the percent black distribution. After considering several coding schemes, I settled on three intervals that capture the pattern of racial effects. Least integrated areas are less than 6% black, moderately integrated areas are between 6% and 19% black, and the most integrated areas are at least 20% black. 7 The distribution of white households across these three neighborhood racial compositions is reported in Table 2. Before turning to my findings, one additional methodological comment is warranted. It is important to note that despite the dynamic nature of white mobility incidence, none of the predictors are operationalized as change variables. Rather, each reflects 1980 levels of household, dwelling, or neighborhood characteristics. There are three reasons why I have opted to examine mobility incidence in this manner. First, I argue that a static measure of racial composition is sufficient to capture racially motivated moves. Neighborhoods with a significant black presence are more unstable than those that are predominantly white and so are prone to tipping (Schelling 1971) and racial succession (Taeuber and Taeuber 1965; Clark 1991). Support for this claim appears in recent work by Lee and Wood (1991), which finds that a majority of 1970 tracts with black residents experienced a significant influx of blacks over the ensuing decade. The implication is that the relationship between level and increase in percent black is sufficiently high that one can proxy for the other. 5 Tracts vary in size, but on average contain about 4,000 people and are the geographic unit social scientists most commonly use to represent neighborhoods (Brooks-Gunn, Duncan, Klebanov, and Sealand 1993; Gramlich et al. 1992; Massey et al. 1994). 6 I also estimated all of the models in this paper using a linear measure of percent black. In no model were any of the racial composition coefficients statistically significant. 7 Of the 114 whites in the most integrated areas, 71 live in tracts that are between 20% and 40% black, while only 12 live in tracts that are more than 70% black. Due to the small number of whites residing in highly integrated tracts, I am unable to assess the effect of variations in racial mix within the at least 20% black category.

9 6 Second, I employ static predictors of mobility incidence for reasons of comparability. Most existing work that examines the relationship between white mobility and racial composition uses static measures of neighborhood composition (Massey et al. 1994; South and Crowder 1997; Frey 1979; South and Deane 1993; Goodman and Streitweiser 1983; Galster 1990). South and Crowder (1997) claim that static measures are acceptable because neighborhood conditions change little in the short run (six years in their study). Other authors do not present an argument for omitting dynamic independent variables. Nevertheless, I believe that it is important to maintain comparability with this work so it will be clear what is gained by adopting a multilevel approach to white flight. A third reason why I construct static measures of the predictor variables is the result of data limitations. The PSID does not yet include tract data from the 1990 Census, so the only two time points available for constructing dynamic measures are 1970 and Given that I observe moves between 1980 and 1985, change in neighborhoods may not be the ideal interval. A further problem is that the 1970 tract data for percent black is missing for about onethird of white families. The reason given for this suppression is respondent confidentiality, though it is not clear that all tracts with suppressed data can be assumed to have small black populations (Adams 1991). 8 Results Table 3 reports correlation coefficients, means, and standard deviations for all variables. Overall, the direction and strength of relationships agree with expectations. Initial evidence supports the hypothesis that household characteristics, dwelling type, geographic setting, and neighborhood factors all explain why whites move, with the strongest predictors being characteristics of the household and dwelling. Table 3 also provides information about relationships within classes of predictors. Though some of these relationships are highly significant, the groups with the greatest internal consistency are household characteristics and nonracial neighborhood factors. Nearly all of the coefficients within these two categories have magnitudes greater than.30, and several exceed.50. The strength of these relationships indicates that advantages tend to be concentrated within households and neighborhoods. In Table 4, I present a series of logistic regression equations predicting mobility incidence. True to the central focus of my inquiry, the models concentrate on the role of racial composition in whites decisions to move. In each successive model the effect of percent black net of other factors is presented. Model 1 provides a clear baseline for the analysis, with nothing other than percent black included as a predictor of mobility incidence. 9 This model replicates work that assesses white flight absent of controls for individual characteristics or nonracial contextual attributes (Farley et al. 1993; Schuman et al. 1985; Clark 1991, 1992). Model 1 provides mixed support for the white flight hypothesis. It does show that white families in moderately integrated tracts are 78% more likely to move than are whites in predominantly white neighborhoods, but does not conclusively find that the 8 I did repeat all of my analyses with change variables for neighborhood factors and family income, though missing data on 1970 percent black led to a substantially smaller sample. The results differed little from those obtained using static measures. Given the problems with the 1970 percent black data, I have opted to present the static results. The remaining predictors are either implicitly dynamic (head s age), largely fixed among heads (education), fixed until the family moves (single-family home, tenure, size of metropolitan area), constrained not to change (marital status), or examined in a dynamic perspective later in the paper (number of children, people per room). 9 Percent black less than 6% is the omitted category in all equations.

10 7 odds of moving are higher in the most integrated neighborhoods. 10 Whites apparent aversion to blacks, at least in neighborhoods where blacks presence is not substantial, might be explained by the absence of controls for nonracial factors in Model 1. To assess whether this is true I now turn to a more complete model of mobility incidence. Model 2 presents the relationship between racial composition and mobility incidence controlling for household, dwelling, and geographic attributes. Here I consider whether the white flight apparent in Model 1 is spurious (i.e., present only because essential exogenous factors have been excluded). Essentially Model 2 corrects for the implicit assumption in Model 1, and the work it replicates, that white families are homogeneous with respect to their mobility response to black neighbors. The results presented in Model 2 show that this implicit homogeneity assumption is false. Controlling for household, dwelling, and geographic differences yields large reductions in observed race effects. The coefficients for moderately and highly integrated neighborhoods are, respectively, 43% and 69% smaller than they appeared in Model 1, and neither is now reliably different from zero. However, caution must be exercised before concluding that white flight is nonexistent because the coefficient for moderately integrated neighborhoods suggests that moves are 39% more likely there than in predominantly white neighborhoods. Furthermore, while not significant, the p-value associated with this effect (p=.08) is near conventional levels of statistical significance. Earlier I noted that the white flight observed in Model 1 might be due to the omission of important predictors of moving. Model 2 added household, dwelling, and geographic factors to the baseline model, and found much smaller race effects. In Model 3 I test the racial proxy hypothesis by considering nonracial neighborhood characteristics. I am interested in whether what remains of the initial racial effects can be explained by accounting for neighborhood SES. The coefficients in Model 3 provide a clear answer to this question. Adding neighborhood SES to Model 2 leads to trivial changes in the percent black coefficients, and none of the measures of neighborhood SES affects whites odds of moving in a substantively or statistically significant way. 11 While this means that I am still unable to definitively claim that whites are insensitive to their neighbors race, Model 3 does allow me to strongly conclude that whites do not move to escape lower-class neighbors. Thus far I have discussed the role of neighborhood factors in whites moving behavior and been unable to show any strong effects. This is not to say that my models have provided no indications of why whites move. On the contrary, Table 4 identifies several household and dwelling factors that greatly affect whites moving behavior. Model 3 indicates that each year of increase in a household head s age is associated with a reduction in the odds of moving of almost 5%. A second important predictor of moving is income. Whites whose family income is one standard deviation above the mean ($72,403 vs. $31,888) are only 80% as likely to move. Third, whites exhibit a clear taste for single-family homes. Even after controlling for whether the dwelling is rented or owneroccupied, Model 3 still reports odds of moving that are 77% higher when whites do not live in a single-family home. Fourth, Model 3 reveals a tremendous effect of tenure on moving behavior. White renters are more than 4 times as likely to move than are white homeowners. Taken together, 10 The difference between the dummy coefficients for moderately and most integrated neighborhoods is also not statistically significant (p=.20). Throughout my analysis this is most often the case. When the two dummy coefficients are reliably different from one another that fact is noted in the text. 11 The high degree of correlation between nonracial neighborhood factors does not account for their lack of significance in Model 3. Adding the nonracial neighborhood variables separately to Model 2 yields results similar to those reported in Model 3.

11 8 my results strongly support the conclusion that whites move primarily because of issues related to their family or dwelling, rather than in response to neighborhood conditions. Tenure, Region, and Household Composition Differences To this point my analyses have constrained the relationship between racial composition and moving behavior to be constant across all whites. In this section I divide whites into subgroups by tenure, region, and household composition, and examine racial effects on moving within each subgroup. There are many reasons why I expect whites responsiveness to black neighbors will vary across these subgroups. First, homeowners have a much greater financial stake in their residences than do renters, and as a result are more concerned about the direction of property values. Given that black residents are sometimes believed to negatively impact the value of nearby properties (Farley et al. 1994), I expect that racial composition will be a strong predictor of moving among homeowners, but not necessarily among renters. Second, regions have different cultures, norms, and histories. These differences have led to quite diverse levels of segregation and patterns of racial conflict (Taeuber and Taeuber 1965; Farley and Frey 1994; Massey and Denton 1993). I suspect that regional differences will also have implications for the way whites react to black neighbors. Third, the arrival of children forces adults to rethink their choice of a neighborhood. Many places that seem fine for adults may not be the kind of environments where people want their children to live. Thus, I expect the effect of racial integration on whites propensity to move will depend on whether there are children in their families. To test whether the effect of black neighbors varies across white subgroups, I estimate the full model of mobility incidence (Table 4, Model 3) separately by tenure, region, and household composition. Unfortunately, several of the subgroups contain a small number of cases. This means that standard errors will be inflated, and it will be difficult to estimate precise effects. Consequently, my examination of the role of racial composition within white subgroups should be viewed as an exploratory analysis. Table 5 presents equations for renters and owners. There is some evidence of a discrepancy in the way renters and owners respond to racial composition. While both are more likely to leave moderately integrated neighborhoods, only owners demonstrate flight from the most integrated neighborhoods. These differences are consistent with the hypothesis that renters and owners are differently invested in their properties, and as a result homeowners are more resistant to integration. However, because these racial effects are not statistically significant for owners or renters, or between the two groups, caution must be exercised in inferring too much about tenure differences in white flight. Table 6 presents equations for the region subgroups. With respect to white flight, three findings are of particular interest. First, all whites are more likely to move from moderately integrated neighborhoods, but the magnitude of this effect varies by region. At one extreme is the South, where living in a 6%-19% black neighborhood increases the odds of moving by 20%. At the other extreme is the Midwest, where the odds of moving are 2.8 times higher if whites live in 6%- 19% black neighborhoods. This finding concurs with work by Farley and Frey (1994), which concludes that the Midwest was home to 8 of the 15 most segregated metropolitan areas in 1980, and 10 of the 15 most segregated metropolitan areas in Given the saliency of race in the Midwest, it is not surprising that Midwestern whites in moderately integrated neighborhoods move so frequently. Second, there are notable, though not statistically significant, differences in whites sensitivity to neighborhoods that are at least 20% black. Northeastern and Southern whites in these

12 9 neighborhoods are more likely to move, while those in highly integrated Midwestern and Western neighborhoods are less likely to move. The magnitude of this effect is greatest in the Midwest, where the odds of moving are only 42% as high if whites live in neighborhoods that are 20% or more black. That Midwestern whites are the most likely to move if they live in moderately integrated neighborhoods and the least likely to move if they live in highly integrated neighborhoods is curious, especially because the difference between the two coefficients is statistically significant (p <.05). The reason for this contradiction is not clear. Third, it is important to note that even though the magnitude of racial composition coefficients differs by region, neither of the racial composition coefficients is statistically significant either within or between regions. Thus, the regional analysis confirms findings based on the full sample and also those reported for the tenure subgroups. In no model is whites probability of moving significantly affected by the proportion of their neighbors who are black, once other factors are considered. The final subgroup analysis examines differences in mobility incidence separately depending on whether families had children in their homes in First I look at the predictors of moving for whites whose homes were childless in This group is further divided into those who had children by 1985, and those who remained childless. I make this distinction because many who had their first child between 1980 and 1985 were anticipating a birth in 1980, and I suspect that they and those who were childless throughout the period likely evaluated neighborhoods in quite distinct ways. Results for these two groups appear in the first two columns of Table 7. As has been true in prior models, some of the racial composition coefficients are sizable, but none are statistically significant. The remainder of Table 7 examines predictors of moving for whites who already had children in This group is further disaggregated into families whose oldest child was less than 6, 6-13, or years old in Unlike the racial effects that have been presented thus far, results for whites with children provide conclusive evidence of white flight even after other factors are controlled. Whites whose oldest child was 6-13 in 1980 were 2.8 times more likely to move by 1985 if they lived in moderately integrated neighborhoods. 12 Further evidence of white flight appears for white families whose oldest child was under 6 in Most of these children would have attended school for the first time between 1980 and When these families live in neighborhoods that are moderately integrated, their odds of moving are twice as high as when they live in predominantly white neighborhoods. While seemingly substantively important, this effect is not statistically significant. A much stronger response appears when white families with young children live in neighborhoods that are at least 20% black. Their odds of moving are 5.8 times higher than would be expected if they lived in predominantly white neighborhoods. Here the race effect is significant, both statistically and substantively, and is also significantly larger than the comparable race effect for other white families with children (p <.05). My findings for whites with children support the hypothesis that whites use one set of standards to rate neighborhoods for themselves, and another to identify satisfactory environments for their children. That white flight is prevalent among families with school-age children, as opposed to families with infants or toddlers (see the second column of Table 7), suggests that parents use moving as a way to control their children s interactions. This strategy is unnecessary 12 This effect is significantly different from zero and reliably larger than the race effect for similar white families in the most integrated neighborhoods (p <.05).

13 10 when children are very young because parents already define the limits of their children s worlds. However, when children go to school, parents are no longer able to choose their children s playmates. The results in Table 7 suggest that parents react to this loss of control by removing their children from environments that contain undesirable playmates (i.e., blacks). Another possible explanation for these results is that parents enjoy the diversity and opportunities of integrated neighborhoods, but move because schools are better in predominantly white neighborhoods (Reardon 1993). This hypothesis posits that parents are not averse to their children interacting with blacks, but value good schools over interracial experiences. Assessing the veracity of these alternative hypotheses should be the subject of future work. Unfortunately this is an area where the PSID is of little assistance. Finally, it should be noted that not all whites with children flee integrated neighborhoods. Quite the opposite behavior is exhibited by whites whose oldest child was in When they live in the most integrated neighborhoods, these families are actually less likely to move than when they live in neighborhoods that are less than 6% black. 13 Presumably these are whites who either have a strong commitment to integration or are for some reason unable to move. In either case this group deserves closer attention, as their response to integration dramatically differs from that of other whites. Conclusion I began this paper with a simple question, Why do whites move? To answer this question I examined data from the geocoded Panel Study of Income Dynamics. This dataset is particularly well-suited to studying mobility incidence for two reasons. First, it contains individual, family, dwelling, and neighborhood measures, thus facilitating a multilevel analysis of mobility incidence. Second, it includes tract-level indicators, which permit greater precision in the assessment of contextual effects than is possible with more highly-aggregated data. My examination of the PSID yields several conclusions about why whites move. First, I conclude that their moves are related to tenure status. The odds of moving are at least twice as high for whites who rent as for white homeowners. Due to this sizable discrepancy in mobility incidence and the dissimilar constraints facing the two groups, there is reason to believe that owners and renters move for different reasons. My results are inconclusive on this point. I do find interactions between tenure and many of the other predictors, but none of these interaction effects is significant. Second, I consider the role of household characteristics in the decision to move. There is also strong evidence that these factors motivate white moves. Families that are younger or poorer are much more likely to move. Young families have changing demands for space. Poorer families lack the resources to obtain suitable housing; they move to find better housing or as a result of instability in their lives. That my research points to the prominence of household factors suggests whites reasons for moving have not changed much in the last two generations. Writing in 1955, Rossi also concluded that many moves occur for reasons related to the life cycle. Third, I examine the effect of neighborhood SES on white mobility incidence. There is concern that studies discuss racial aversion when what is actually being measured is whites preference for high-ses neighbors. This proxy argument maintains that because many social problems are overrepresented among blacks, avoiding black neighbors is a method whites use to 13 This effect is significantly different from zero and reliably larger than the race effect for similar white families in the most integrated neighborhoods (p <.05).

14 11 ensure that they live in good neighborhoods (Taub et al. 1984). My results do not support the proxy argument. Consistently the effect of neighborhood SES on whites odds of moving is small and not statistically significant. Adding these factors to the mobility incidence model does not dampen observed race effects. Based on this evidence I conclude that whites do not move as a response to their neighbors social class. That leaves one final factor neighborhood racial composition. In scholarly writings and throughout American society, it is widely believed that whites move to escape black neighbors. This process is so well accepted that people know it by a simple phrase, white flight. Despite the popularity of the term, I maintain that white flight is a theory that has not been subjected to sufficient empirical scrutiny. Most work that claims to find evidence of whites fleeing integrated neighborhoods either inadequately controls for nonracial factors or employs biased measures of racial aversion (Galster 1990; Massey et al. 1994; Bobo and Zubrinsky 1996; Farley et al. 1993). I correct for these problems by looking at actual moves and controlling for a host of life cycle, dwelling, regional, and nonracial contextual factors. My analysis yields mixed support for the white flight hypothesis. When I consider the behavior of whites in general, or almost any subgroup of whites, I find that whites are more likely to move if they have more black neighbors, but that these race effects are not statistically significant. Also, because I am not able to include measures of school quality, neighborhood deterioration, and crime in this analysis it is reasonable to believe that the true race effect is smaller than has been presented here. Nevertheless, my results do not allow me to reject the hypothesis that whites are fleeing black neighbors. While this pattern of large unreliable race effects is consistent across many of my analyses, there is one group of whites whose mobility response to black neighbors is unmistakable. That group is whites with children, a segment of the white population whose behavior is clearly consistent with the white flight hypothesis. Whites who live in moderately integrated neighborhoods and whose oldest child is 6-13 years old are almost three times more likely to move than are comparable whites in the least integrated neighborhoods. For whites with even younger children the evidence of white flight is much stronger. White families with children under 6 are nearly six times more likely to move when they live in one of the most integrated neighborhoods. Whether this flight is due to racial prejudice, a search for good schools, or some other explanation cannot be determined with my data. Establishing why whites with children are fleeing integrated neighborhoods must be a central goal of future research. So what does all of this mean for residential segregation? My findings suggest that it is unclear whether improving blacks access to predominantly white neighborhoods will cause most whites to flee. The exception is white families with children. That this group is fleeing black neighbors, while childless whites may be staying behind, is not good news for proponents of stable integration. It means that neighborhoods can sustain moderate levels of integration, but that resident whites are likely to be weakly committed to their neighborhoods because they either have few contacts with the institutions that serve children or are short-term residents. The consequence of this weak commitment is that many of the anticipated benefits of integration, such as better schools, less crime, and blacks improved access to job networks, will likely fail to materialize. However, it would be wrong to read my work as evidence that levels of neighborhood segregation must remain high. Who moves out is only part of the equation. To fully understand and predict levels of segregation we must also know who fills the vacancies that arise in neighborhoods. In other words, vacancies may occur more frequently in integrated neighborhoods, but if enough of

15 12 these openings are filled by whites, and especially whites with children, then stable integration and the benefits it entails will likely be the result.

16 13 References Adams, Terry K Documentation for 1970 and 1980 Census Extract Datasets. Survey Research Center, University of Michigan, Ann Arbor, MI. Unpublished manuscript. Bobo, Lawrence, Howard Schuman, and Charlotte Steeh Changing Racial Attitudes toward Residential Integration. Pp in Housing Desegregation and Federal Policy, edited by John M. Goering. Chapel Hill, N.C.: The University of North Carolina Press. Bobo, Lawrence and Camille L. Zubrinsky Attitudes on Residential Integration: Perceived Status Differences, Mere In-Group Preference, or Racial Prejudice. Social Forces 74: Brooks-Gunn, Jeanne, Greg J. Duncan, Pamela Kato Klebanov, and Naomi Sealand Do Neighborhoods Influence Child and Adolescent Development? American Journal Of Sociology 99: Clark, David E. and Leslie A. Nieves An Interregional Hedonic Analysis of Noxious Facility Impacts on Local Wages and Property Values. Journal of Environmental Economics and Management 27: Clark, William A.V Residential Preferences and Neighborhood Racial Segregation: A Test of the Schelling Segregation Model. Demography 28: Residential Preferences and Residential Choices in a Multiethnic Context. Demography 29: Cutler, David M. and Edward L. Glaeser Are Ghettos Good or Bad? Unpublished manuscript. Duncan, Greg J. and Sandra J. Newman Expected and Annual Residential Mobility. Journal of the American Institute of Planners 42: Farley, Reynolds and William H. Frey Changes in the Segregation of Whites from Blacks During the 1980s: Small Steps Toward a More Integrated Society. American Sociological Review 59: Farley, Reynolds, Charlotte Steeh, Tara Jackson, Maria Krysan, and Keith Reeves Continued Racial Residential Segregation in Detroit: Chocolate City, Vanilla Suburbs Revisited. Journal of Housing Research 4: Frey, William H Central City White Flight: Racial and Nonracial Causes. American Sociological Review 44:

17 14 Galster, George C White Flight from Racially Integrated Neighborhoods in the 1970s: The Cleveland Experience. Urban Studies 27: Goodman, John L., Jr., and Mary L. Streitwieser Explaining Racial Differences: A Study of City-to-Suburb Residential Mobility. Urban Affairs Quarterly 18: Gramlich, Edward, Deborah Laren, and Naomi Seeland Mobility into and out of Poor Urban Neighborhoods. Pp in Drugs, Crime, and Social Isolation: Barriers to Urban Opportunity, edited by Adele V. Harrell and George E. Peterson. Washington, DC: The Urban Institute Press. Greene, William H Econometric Analysis. Englewood Cliffs, NJ: Prentice Hall. Hill, Martha S The Panel Study of Income Dynamics: A User s Guide. Newbury Park, CA: Sage Publications. Lee, Barrett A., R.S. Oropesa, and James W. Kanan Neighborhood Context and Residential Mobility Demography 31: Lee, Barrett A. and Peter B. Wood Is Neighborhood Racial Succession Place-Specific? Demography 28: Massey, Douglas S. and Nancy A. Denton American Apartheid: Segregation and the Making of the Underclass. Cambridge, MA: Harvard University Press. Massey, Douglas S., Andrew B. Gross, and Kumiko Shibuya Migration, Segregation, and the Geographic Concentration of Poverty. American Sociological Review 59: Ottensmann, John R. and Michael E. Gleeson The Movement of Whites and Blacks into Racially Mixed Neighborhoods: Chicago, Social Science Quarterly 73: Reardon, Patrick T. More Chicagoans Find It Isn t Their Kind of Town. Chicago Tribune (28 November 1993), sec. 1: 1, Rossi, Peter H Why Families Move. New York: The Free Press. Schelling, Thomas Dynamic Models of Segregation. Journal of Mathematical Sociology 1: Schuman, Howard and Lawrence Bobo Survey-based Experiments on White Racial Attitudes toward Residential Integration. American Journal of Sociology 94: Schuman, Howard, Charlotte Steeh, and Lawrence Bobo Racial Attitudes in America: Trends and Interpretations. Cambridge, MA: Harvard University Press.