Race-Based
Neighborhood Projection:
A Proposed Framework
for Understanding
New Data on Racial Integration
Ingrid Gould Ellen
Taub
Urban Research Center
Robert F.
Wagner Graduate School of Public Service
New York
University
October 9, 1997
Please direct any questions or comments to the author at the Robert F. Wagner Graduate School of Public Service, New York University, 4 Washington Square North, New York, NY 10003, or by email to ingrid.ellen@nyu.edu. Portions of this research were supported by the U.S. Department of Housing and Urban Development. Data were generously provided by the Urban Institute and the U.S. Census Bureau. I thank Richard Zeckhauser, Denise DiPasquale, John Kain, George Galster, and participants in both the Harvard University Urban Economics and the Fannie Mae Foundation Research Seminar for invaluable comments and suggestions.
Most of the research exploring racial patterns of settlement in this country has emphasized the extreme level of separation between blacks and whites, the dramatic division between "Chocolate Cities" and "Vanilla Suburbs," as one set of researchers has put it (Farley et al., 1993). Moreover, past researchers have also tended to assume that any neighborhoods in which some degree of racial mixing has occurred are highly unstable. Yet while the country is extraordinarily racially segregated, the evidence presented below reveals that integrated communities of blacks and whites do exist and, as one of the principal findings here demonstrates, a growing majority of them are remaining integrated over time. The core aim of this study is to explore these stable, shared communities and to try to understand how they differ from mixed communities that are merely on their way to becoming largely minority. Unlike past research about racial change, this study focuses on the residential location decisions of individual black and white households and distinguishes between entry and exit choices.
In brief, the theory offered is that in choosing a neighborhood, households care less about present racial composition than about expectations about future school quality, crime rates, property values, and the like. In other words, decisions are made on the basis of projections about future neighborhood quality. Race is still highly relevant under this hypothesis, however, because, rightly or wrongly, white households (and some black households as well) tend to associate an increasing minority presence in a neighborhood with structural decline. This "neighborhood racial projection" hypothesis generates several important predictions that empirical tests bear out, such as that white homeowners and households with children will be more averse to racial mixing than renters and childless households, and that racial considerations are more important to entry decisions than exit decisions.
Section 1 of this
paper catalogs the extent and stability of racial integration in contemporary
American neighborhoods. Section 2 introduces the neighborhood racial
projection hypothesis and describes the conditions under which the
theory predicts that neighborhoods are more likely to remain integrated
over time. Section 3 tracks a sample of mixed neighborhoods over time
to see whether the predictions of the neighborhood racial projection
theory hold up. Sections 4 and 5 focus on the key element of the analysis:
the decisions of individual households to move into and out of particular
communities. With the use of a special census data set that links
households to the neighborhoods that they live in, these sections
present results from logistic regression models of the residential
decisions of individual households that underlie racial change. Section
6 then uses the parameters from these regressions to construct a simulation
model to predict rates of racial change in different types of communities.
1. Extent and Stability of Racial Integration in the Contemporary United States
In exploring racial integration, the first question to consider is of course the extent and stability of neighborhood integration. This section assesses neighborhood racial integration, using census tract data from the Census of Population and Housing in 1970, 1980, and 1990. Table 1 shows the extent of racial mixing in neighborhoods, or census tracts, nationwide in 1990. Tracts are divided into four categories: < 1% black, 1-10% black, 10-50% black, and >= 50% black, with those between 10 and 50 percent black considered as racially integrated. Although arbitrary, this definition of integrated accounts for both the fact that African Americans make up just 12 percent of the U.S. population and the reality that most people envision a fairly even racial split when considering integration. The first column shows the total number of census tracts in each category in 1990, while the second and third show the proportion of non-Hispanic whites and blacks living in tracts of each type. As shown, 19 percent of all census tracts in 1990 were racially integrated, and they collectively housed 15.4 percent of non-Hispanic whites and nearly one third of blacks. Thus, while this table makes evident that segregation is the norm, it also demonstrates that a considerable minority of integrated neighborhoods exist, even within a highly segregated country.
Table
1
Extent of Neighborhood Racial Integration in 1990
All Metropolitan Areas
|
Total Number
of Neighborhoods
|
Share of Non-Hispanic
White Population
|
Share of Black
Population
|
|
| < 1% Black |
13,075
|
36.3%
|
0.8%
|
| 1-10% Black |
16,125
|
46.7%
|
12.9%
|
| 10-50% Black |
8,043
|
15.4%
|
32.4%
|
| >= 50% Black |
5,169
|
1.6%
|
53.9%
|
| Total |
42,412
|
148,069,316
|
25,768,131
|
Source: 1990 Census
data included in the Urban Institute's Underclass Database.
As for trends over time, Table 2 shows the share of whites and blacks living in tracts of different types in 1970, 1980, and 1990. Although the share of whites living in integrated areas remains modest, it nonetheless rose significantly over the two decades, by 14 percent between 1970 and 1980 and by 30 percent between 1980 and 1990. An even sharper shift occurred in the prevalence of exclusively white neighborhoods. The proportion of whites living in tracts where African Americans represent less than one percent of the population declined from 62.6 percent in 1970 to 35.6 percent in 1990, suggesting that far fewer communities remain totally closed to minorities today.
Table
2
Share of White and Black Population in
Tracts by % Black in 1970, 1980, and 1990
All Metropolitan Areas
|
1970
|
1980
|
1990
|
|
|
< 1% Black
|
|
|
|
| White Population |
62.6%
|
48.5%
|
35.6%
|
| Black Population |
0.6%
|
0.9%
|
0.8%
|
|
1-10% Black
|
|
|
|
| White Population |
25.0%
|
37.9%
|
47.1%
|
| Black Population |
6.6%
|
10.2%
|
12.9%
|
|
10-50% Black
|
|
|
|
| White Population |
10.5%
|
12.0%
|
15.6%
|
| Black Population |
25.7%
|
26.7%
|
32.4%
|
|
>= 50% Black
|
|
|
|
| White Population |
1.9%
|
1.7%
|
1.7%
|
| Black Population |
67.1%
|
62.2%
|
53.9%
|
| Total |
115,949,801
|
148,823,521
|
158,529,269
|
Examining stability in racially mixed neighborhoods is somewhat more of an empirical challenge, since it requires linking census tracts across different decennial censuses. The Urban Institute's Underclass Database (UDB) is used to perform this analysis. In particular, Table 3 shows the proportion of tracts in each category that underwent succession (a decline of at least 10 percentage points in the proportion non-Hispanic white), experienced displacement (an increase of at least 10 percentage points in the proportion non-Hispanic white), or remained stable between 1980 and 1990. These results are based on a sample of 34 metropolitan areas that in 1990 had over one million residents, black population shares of at least 5 percent, and Hispanic population shares of less than 30 percent. (The particular metropolitan areas, together with selected demographic characteristics, are listed in Appendix A.) As shown, the difference between integrated and non-integrated communities is pronounced. Yet more than half of mixed tracts emerged from the decade with either an identical or a larger proportion of non-Hispanic whites. And an examination of census tracts that were integrated in 1970 reveals that even after a full 20 years, over one third of these mixed tracts had not lost a significant portion of their white population share.
Table
3
Share of Neighborhoods Undergoing Change Between 1980-90 by 1980 Percentage
Black
34-MSA Sample
|
|
Succession
|
Stable
|
Displacement
|
Total
|
|
< 10% Black
|
21.2%
|
78.4%
|
0.4%
|
11,590
|
|
10-50% Black
|
46.1%
|
49.5%
|
4.4%
|
2,773
|
|
>= 50% Black
|
12.3%
|
83.3%
|
4.4%
|
2,816
|
|
All Tracts
|
23.7%
|
74.6%
|
1.7%
|
17,179
|
Most significantly, perhaps, a comparison with data from the 1970s suggests that neighborhoods are becoming more stable with time. As shown in Table 4, the mean white population loss in integrated neighborhoods was lower in the 1980s than in the 1970s, a greater proportion of mixed tracts remained mixed over the 1980s than over the 1970s, and the share of integrated tracts that did not lose whites was greater in the latter decade. Figure 1 graphically presents the difference between these two decades. Specifically, it compares the mean loss in white population share over the 1980s with that occurring during the 1970s for neighborhoods of differing black population proportions (0-10 percent, 10-20 percent, and so on). As shown, the loss during the 1970s was consistently larger. Moreover, while the white loss curve slopes upward until it reaches about 40 percent black in the 1970s; in the 1980s, it is basically flat across all mixed areas, suggesting that white loss rates over this decade were independent of the initial size of the black population.
Table
4
Comparison of 1970-80 and 1980-90 Racial Change
34-MSA Sample
|
1970-1980 Change
|
1980-1990 Change
|
|
| Mean Loss in Proportion White in Integrated Neighborhoods |
18
|
10.5
|
| Share of Integrated Neighborhoods that Remained Integrated at end of Decade |
61.0%
|
76.4%
|
| Share of Integrated Neighborhoods that did not Lose Whites Over Decade* |
44.5%
|
53.3%
|
* Tracts in which
the proportion of whites present either increased or failed to decline
by more than ten percentage points.
Figure 1
In sum, while
America remains remarkably segregated, neighborhood racial integration
appears to be growing both more widespread and more stable. The obvious
question is how these stable, mixed communities differ from those
that are merely temporarily integrated, and quickly on the way to
becoming predominantly black. The remainder of this paper addresses
this issue.
2. Towards a Theory of Racial Change
Although economists, sociologists, and demographers have been studying neighborhood racial change for decades, few have explored the roots of the differences in rates of racial change across integrated neighborhoods. The reason is simple: as an empirical matter, tipping has been considered virtually inevitable and universal -- a view that was solidified by the anecdotally-reported experience of many cities during the 1950s and 1960s and supported by many racial preference surveys. Indeed, sociologists writing about neighborhood change have most commonly employed the ecological succession model, which views racial transition as an irreversible and systematic process resulting from the expansion of the black ghetto (Duncan and Duncan 1957, Taeuber and Taeuber 1965). According to this model, differences in rates of neighborhood racial change are explained largely by differences across cities in the pace of black migration.
While the ecological model proved a useful heuristic for describing racial change in the 1950s and 1960s, it is far less applicable today. And in any case, it does not help us to understand the individual behavior that underlies such change. (To the extent that whites choose not to live in minority areas, what are their reasons?) Economists have typically focused more on illuminating these underlying motivations and dynamics. Their most common explanation for what drives racial separation is racial preference, including and simple pure racial prejudice.
Schelling's model, for instance, remains the classic economic tool for understanding racial change (Schelling 1972). In brief, he assumes that whites get disutility from sharing neighborhoods with blacks, and a white household will remain in a community only as long as its black population is under the household's particular tolerance threshold. Blacks are assumed to be much less sensitive to racial composition but to be wary of being in too small of a minority. Given these assumptions, the model generally finds a narrow band of racial mixing.
An alternative to this racial preference or prejudice explanation is the claim that what really drives neighborhood racial change is the class differences between whites and minorities (Clark 1986, Leven et al. 1976, Pascal 1967). In this view, change occurs either because whites can afford nicer neighborhoods or because whites prefer and can afford to live among more wealthy neighbors. The key is that whites do not mind living among minorities, at least as long as they are of similar income and similar class.
While the racial preference and the class difference explanations no doubt account for some of the motivations underlying racial change, they afford virtually no role to community context and characteristics. Building on the work of Taub, Taylor, and Dunham (1984), the theory presented here argues that such contextual factors, and people's expectations about their likely future course, are critical in shaping the rate of racial change.
The theory posits that residential decisions, especially those of white households, are indeed heavily shaped by racial attitudes. But they are heavily shaped not so much in the simple sense, described above, that white households are unwilling to live among black neighbors. Rather, white households tend to assume that all mixed neighborhoods quickly and inevitably become predominantly black, and it is the discomfort they anticipate feeling in that environment that plays the more dominant role in residential decision making.
This discomfort has two likely sources. First, whites may fear being left behind as a racial minority as their community becomes largely black. Second and more important, white households (and potentially black households as well) have negative preconceptions about what an all-black neighborhood will be like. Specifically, black neighbors are often thought to portend, because of their typically lower incomes and lesser political power, a deterioration in school quality, public safety, property values, and other quality of life attributes. In other words, racial change has come to serve as a sign of change in the "structural strength" of the community, to borrow a phrase from Taub, Taylor, and Dunham (1984). How such stereotyped associations should be distinguished from simple racial prejudice on moral grounds deserves lengthy discussion, but suffice it to say here that such associations are analytically distinct and have distinct policy implications.
Significantly, though this theory focuses on the role of private household preferences and predictions, this is not to say that housing market discrimination does not contribute to segregation as well. Determining the role of such discriminatory practices is beyond the scope of this paper, however. With that said, it is important to note that widespread racial discrimination in the past -- both official and private -- clearly fueled rapid racial transition and helped to cement the view held by many whites that all moderately integrated areas are rapidly on their way to becoming all-black. Moreover, racial discrimination also helps to maintain all-white communities and thus preserves an alternative for whites wishing to avoid mixed areas (Yinger 1978). Finally, as discussed further below, the fear of racial discrimination probably discourages many minority households from searching in predominantly white areas.
Consider a few implications of this proposed hypothesis -- call it the "neighborhood racial projection" hypothesis. First, it suggests that white households who are less invested in the structural strength of the community will be more open to racial mixing, and thus more likely to live in mixed communities. Households without children, for instance, are likely to be less sensitive to racial composition and change than those who have them, since they have little to lose if schools degenerate. Renters should meanwhile be less sensitive to racial mix than owners, since they obviously have no worries about property values. Note the pure prejudice theory would argue, as many past researchers have, that white renters -- who can enter and exit neighborhoods more cheaply than homeowners -- will be less likely to live in mixed communities (Caplan and Wolf 1960, Rosenbaum 1992, Steinnes 1977). One researcher has even dubbed homeownership a "vaccine against racial change" (Schwab and Marsh 1980).
Second, the proposed hypothesis suggests a distinct difference between the decision whether to move into a community and the decision to leave one, and in particular, suggests that the entry decision will be more influenced by racial concerns. For households living within a neighborhood are likely to be fairly certain about their neighborhood's quality, its schools, its crime, etc., when making their decision whether or not to leave. But this information may be quite difficult and time-consuming for outsiders to obtain. An easier rule-of-thumb is simply to observe the racial composition. Thus, households considering moving into an area are likely to place more weight on race as a signal of neighborhood quality. Moreover, for racial mixing to provoke a household to move out of a neighborhood, the household's distaste for racial integration must at least exceed the costs of moving. For in-movers, antipathy to racial mixing need not be so great, since these households have already settled on moving somewhere, it's just a matter of deciding where. Significantly, since entry decisions are likely to be far more influenced by racial composition, they should in turn be far more important to racial change than exit decisions. If true, this casts serious doubt on the conventional belief that racial transition is caused by "white flight."
Third, the hypothesis predicts that racial mixing will be more stable in a) communities which seem sheltered in some way from further black growth (because they are distant from the central area of black residence or have been racially stable in the past) or b) those whose school quality, property values, and other neighborhood attributes seem particularly secure, perhaps because of the presence of a university or some other large and stable institution that promises to bolster the housing market by generating a continuing supply of eager residents.
In more formal
terms, a household is assumed to obtain utility from its dwelling
unit, the present structural strength of its neighborhood, and its
assessment of the likely future strength as well. For simplicity,
assume that the utility a household gets from unit i in neighborhood
j may be represented as a simple linear relationship:
where Di represents the characteristics of the dwelling unit i, Sj is the present structural strength of neighborhood j, and FSj is the household's assessment of future structural strength of neighborhood j. Similarly, assume that predicted future structural strength can be represented as a function of current structural strength, past changes in that strength ( Sj), and projected future racial composition (FRj).
The expected future racial composition of the community (FRj) is captured in turn by present racial composition, recent changes in racial composition, and the composition of surrounding communities (or distance to the black population center).
It is worth emphasizing how this simple framework differs from the past models of racial change on which it builds, in particular, Schelling's model of tipping and Taub, Taylor, and Dunham's threshold theory of neighborhood change. The key difference between the model here and that of Schelling is that in Schelling's world, mobility decisions are made entirely on the basis of preferences for present racial composition. Racial composition is placed directly in the utility function, in contrast to the model here, where it influences residential utility via its effects on expected future neighborhood quality. (In the social class models, meanwhile, race has no part to play, after controlling for income and education.)
Schelling notes that an individual's tolerance level in his model could be interpreted as "the percentage black at which, projecting the trend . . . , he makes his decision to leave." But he acknowledges that the interpretation is only valid if the decision, and implicitly, the projected future growth, are "purely a function of the percentage black in the neighborhood and not any other indices that might influence his decision" (Schelling 1972). Yet here, the role of race is more complicated and the reaction to racial change can vary across different types of households, or even for the same household in different circumstances. If one household, for instance, is more concerned about future structural strength than another (say, a homeowner versus a renter), then the magnitude of its coefficient will be larger. Second, if one household believes more firmly than another that a growing minority population spells decline for the neighborhood, perhaps because the household is white (and thus more apt to hold negative stereotypes about minority communities) or because it lives in a city where this has happened in the past, then its coefficient on the expected future minority population -- c -- will be larger. Finally, the interaction term captures the possibility that people living in already weakening neighborhoods may be more apt to link racial change with neighborhood deterioration.
This final point -- how neighborhood deterioration can accelerate the rate of racial transition -- is made strongly by Taub et al. in their threshold model of neighborhood change. However, the model here differs from that of Taub et al. in its study of how individual factors, or the particular circumstances of a household, such as its stage in the lifecycle and whether or not it is a homeowner, can alter its attitudes about neighbors and neighborhood. Schelling's model cannot make these distinctions either, since in his abstract world, all members of a given racial group are assumed to be identical.
Another point of divergence from Schelling and from Taub et al. is that the decision of a particular household to enter a community as well as the decision to exit it are both explicitly modeled here. Schelling's model instead focuses largely on the decision of whites to leave a neighborhood, and these departures are what ultimately drive his model. As long as some minimum threshold is met, he assumes that their places will be taken by blacks or "by whites and blacks in proportions that reflect what is happening in the neighborhood" (Schelling 1972). As explained above, however, entry -- and in particular, the reluctance of whites to enter substantially integrated communities -- is predicted to drive racial change more than exit in this model. Taub et al. also concentrate on explaining neighborhood exit decisions. Significantly, the authors emphasize the critical importance of entry, or the ratio of black to white demand, to racial change, but they treat this demand to be exogenous, independent of the current proportion black in the neighborhood.
Up to now, the
discussion has focused primarily on the preferences of white households.
The justification is that both surveys of racial preference and the
data analyzed here suggest that black households are considerably
less sensitive to racial composition than their white counterparts.
Still, while blacks tend to be willing to live in a wide range of
neighborhoods, they are likely to be hesitant to move into all-white
areas both because of the hostility they believe will greet them there
and because of the potentially higher search costs required, due to
discrimination (Courant 1978). It is also possible that, like whites,
some black households (notably those with children) may be averse
to all-black areas as well because of their concerns about the quality
of life in such neighborhoods.
3. Tracking Neighborhoods Over Time
A variety of empirical tools were used to test the basic neighborhood racial projection theory. The first general approach is to track neighborhoods over time in order to identify correlates of racial change. In particular, Table 5 compares selected characteristics of two categories of neighborhoods that were integrated in 1980: those that maintained their racial mix between 1980 and 1990 and those that lost white residents, or were "transitional." (This table, and the rest of those discussed in this paper, are based on the 34-MSA sample described above.)
Table
5
Comparison
of Stable and Transitional Integrated Neighborhoods
| Selected Neighborhood Characteristics |
Stable
|
Transitional
|
| Racial Mix |
|
|
| % Black, 1980 |
24.2%
|
24.0%
|
| Socioeconomic Status |
|
|
| % in Professional/Managerial Occupations |
24.7%
|
24.6%
|
| Poverty Rate |
19.2%*
|
15.1%
|
| % of Blacks with Some College |
25.2%*
|
33.2%
|
| % of Whites with Some College |
31.9%*
|
29.7%
|
| Ratio of Share of Blacks Earning > $25,000 to Share of Whites Earning > $25,000 |
0.83
|
0.84
|
| MSA-Level Characteristics |
|
|
| Black-White Growth Differential in MSA |
115.1%
|
107.4%
|
| % Black in MSA |
18.6%
|
18.5%
|
| Tests of Neighborhood Racial Projection Theory |
|
|
| % Black, 1970 |
19.3%*
|
8.8%
|
| % of Whites Rating Schools as Poor, 1985 |
12.3%*
|
36.6%
|
| % of Blacks Rating Schools as Poor, 1985 |
10.1%
|
16.3%
|
| Homeownership Rate |
45.4%
|
45.7%
|
| % of Tracts Housing Major Institution |
6.8%*
|
4.4%
|
| Miles from Black Population Center |
11.5*
|
8.4
|
| N |
1,373
|
1,277
|
* Difference statistically significant (5% level).
Perhaps the most notable finding is that suggested already by Figure 1 -- the lack of difference between the mean percentage black living in the two types of areas at the outset of the decade. This certainly casts some doubt on the conventional racial tipping model, in which the share of blacks in a community alone determines its subsequent rate of racial change.
There is little support here for the class theory of racial change either. There is no difference in the share of working residents in professional occupations, and stable tracts actually have higher poverty rates. Moreover, it seems that neighborhoods where blacks and whites are of more equal status are, if anything, less stable. Consider that the proportion of black residents that have attended college is smaller in stable tracts than in transitional tracts, while the opposite holds true for whites. In other words, blacks appear less educated relative to their white neighbors in the stable communities. (There is no difference in the relative incomes of blacks and whites across the two tract types.) Finally, there is no evidence that either the overall share of blacks or the relative growth rate of the black population in a metropolitan area is related to the pace of neighborhood racial change, as predicted by the ecological succession model.
The neighborhood racial projection theory fares considerably better. First, while the share of blacks present in 1980 is identical in these two types of tracts, the share of blacks present in 1970 is markedly different: neighborhoods with a more substantial black presence in 1970 are more likely to be stable between 1980 and 1990, which is consistent with the claim that white residents feel more comfortable moving into a mixed community if it has been integrated for a longer span of time. Second, a significantly larger share of whites living in transitional communities describe local schools as substandard than whites residing in stable, mixed communities. The direction of causality here is not clear, but what is evident is the strong correlation between perceptions of poor school quality and growing black populations. Interestingly, there is no difference in the perceptions of black households across these community types, which suggests that white households may be exaggerating the deterioration that accompanies racial change (or that blacks are underestimating it).
A third piece of potential evidence is the lack of difference across the two tract types in the extent of homeownership. Given that renters move about three times as often as homeowners, this finding seems to suggest that white renters are somehow more at ease with racial mixing. Indeed, in a multivariate regression that controls for housing unit density, the homeownership rate is significantly and positively correlated with more rapid rates of racial transition (Ellen 1996). Fourth, there is some evidence here too that large institutions foster greater neighborhood stability. Using an unusually high ratio of residents to housing units as a proxy measure for an institutional presence, a greater share of stable tracts house large institutions. As predicted then, it seems that these universities, military bases, and the like may help to bolster confidence in an integrated area. Of course it is possible that it is merely the institutional populations themselves that are integrated, but analysis of Washington, DC suggests that in many areas, the population surrounding the institution is racially diverse (Ellen 1996).
Finally, data
from a more in-depth analysis of the Washington, DC metropolitan area
suggests that distance to the central area of black concentration
is also relevant, and as predicted, mixed tracts further from the
central black area are more likely to be stable. In that metropolitan
area at least, stable, integrated tracts are an average of 11.5 miles
away from the black population center, while transitional tracts are
only 8.4 miles away.
4. Household Mobility or Exit Decisions
Such analysis of neighborhood-level data is instructive, but to truly understand the nature of racial change, one ultimately needs to learn how racial composition influences the individual household decisions that underlie such change, the decisions, that is, to leave certain neighborhoods and move into new ones. And to analyze these residential decisions, an unusual data set is needed that includes information about both individual households and the neighborhoods that they live in. This study gained access to such data through a special arrangement with the Census Bureau. Specifically, it uses the Bureau's internal files of the American Housing Survey (AHS), a biannual survey that tracks about 50,000 housing units nationwide and collects information about the unit, its occupants, and their opinions regarding the neighborhood. Unlike the public use files, the internal AHS files also include the census tract in which each house is located. Using this tract identifier, the AHS is linked to a set of 1980 and 1990 decennial census tract variables obtained from the Urban Institute's Underclass Database. Since both 1980 and 1990 variables are included, it is possible to explore not only how neighborhood characteristics at a given point in time influence household preferences, but also how changes in these contextual attributes play a part.
Using data from the 1985, 1989, and 1993 AHS surveys, mobility patterns are studied between 1985 and 1989 and between 1989 and 1993. A logistic regression model of the decision to move is employed. The dependent variable indicates if a household moved during the particular four-year period, while the independent variables include household demographics, housing unit features, metropolitan area characteristics, and neighborhood attributes. Neighborhood characteristics from 1980 are linked to the 1985-1989 model; 1990 characteristics are linked to the 1989-1993 model. The specific variables, their definitions, and their mean values in the 1985-1989 data set are listed in Table 6.
Table
6
Means of Variables Used in 1985-1989 Mobility Model
| Name | Definition |
White Owners
|
White Renters
|
Black Owners
|
Black Renters
|
| Move | = 1 if household moved btw 85-89 |
0.24
|
0.64
|
0.18
|
0.57
|
| Age | Age of Household Head |
51.3
|
43.5
|
51.7
|
41.4
|
| Nev Mar | Never Married |
0.066
|
0.273
|
0.079
|
0.316
|
| Divorced | Divorced |
0.221
|
0.354
|
0.36
|
0.43
|
| Kids0-6 | Kids between 0 and 6 |
0.15
|
0.153
|
0.17
|
0.254
|
| Kids6-17 | Kids between 6 and 17 |
0.29
|
0.165
|
0.386
|
0.36
|
| HH Size | Household Size |
2.8
|
2.1
|
3.1
|
2.7
|
| Ten Length | Years in same unit |
14.3
|
5.2
|
14.4
|
5.4
|
| Bot Income | Bottom income quartile in sample |
0.139
|
0.312
|
0.265
|
0.56
|
| Top Income | Top income quartile in sample |
0.368
|
0.12
|
0.226
|
0.038
|
| SF Home | Unit is single family home |
0.898
|
0.24
|
0.894
|
0.215
|
| Poor unit | HH rates unit as 5 or less on 1-10 scale |
0.055
|
0.161
|
0.082
|
0.273
|
| Pub hsng | In public housing development |
NA
|
0.037
|
NA
|
0.21
|
| Northeast | Northeast Census region |
0.33
|
0.36
|
0.26
|
0.31
|
| South | South Census region |
0.29
|
0.27
|
0.39
|
0.36
|
| West | West Census region |
0.08
|
0.11
|
0.05
|
0.047
|
| Popgrowth | 1980-90 rate of growth in MSA |
151
|
125
|
89
|
115
|
| Segregation | MSA dissimilarity index |
70
|
70
|
71
|
71
|
| MSA Vac | Housing vacancy rate in MSA |
1.69
|
6.37
|
1.73
|
6.42
|
| NEIGHBORHOOD VARIABLES |
|
|
|
|
|
| Poor Nbhd | HH rates nbhd as 5 or less on 1-10 scale |
0.096
|
0.189
|
0.173
|
0.33
|
| % Black | % Black in tract in 1980 |
4.7
|
6.8
|
66
|
63.5
|
| Chblk 80-90 | (% black in 1990) - (% black in 1980) |
2.4
|
3.1
|
4.4
|
3.8
|
| % Other | % of non-black minorities in tract in 1980 |
5.2
|
8.1
|
5.1
|
7.8
|
| Choth 80-90 | (% other in 1990) - (%other in 1980) |
3.1
|
4.4
|
1.7
|
2.1
|
| Income Ratio | Ratio of HH income to 1980 tract mean |
1.12
|
0.74
|
1.1
|
0.66
|
| % Prof | % of working adults in prof. occupations |
30.1
|
30.5
|
20.6
|
19.4
|
| Poverty rate | Tract poverty rate in 1980 |
6.4
|
9.9
|
18.8
|
26.7
|
| Chpov 80-90 | (Pov rate, 1990) - (Pov rate, 1980) |
0.66
|
1.1
|
2.3
|
2.9
|
| % New Res | % moved into tract in last 5 years |
47.8
|
54.4
|
44.2
|
52.8
|
| Vacancy | Owner-occ or rental vacancy rate in tract |
4.8
|
5.9
|
6.7
|
8.3
|
| Homeown | Homeownership rate in tract |
72.6
|
49
|
59.5
|
34.1
|
| N |
5,438
|
2,720
|
725
|
1,001
|
The table shows means stratified by tenure and race, and it reveals some significant differences between owners and renters on the one hand and between black and non-Hispanic white households on the other. In total, nearly 10,000 households are studied; 83 percent are non-Hispanic white and 62 percent are homeowners. In general, black households have lower incomes, dwell in inferior housing, and live in structurally weaker communities than their white counterparts. The most significant difference between the black and non-Hispanic white respondents, however, is neighborhood racial composition. The average black household lives in a neighborhood that is roughly two thirds black; the average white household lives in a community that is not much more than five percent black.
Tables 7 and 8 present selected parameter estimates for the mobility model for the two time periods, again stratified by tenure and race. For the sake of clarity, only the neighborhood coefficients are shown here, and generally they suggest that neighborhood variables play a important role in moving decisions, at least for white households. Yet not all aspects of the neighborhood context appear to have strong effects. There is little evidence here, for instance, that either absolute socioeconomic status or changes in that status influence moving decisions. There is evidence, however, that relative status matters -- the coefficient on the ratio of a household's income to the tract mean (income ratio) is positive and significant in the 1985-89 model, suggesting that white homeowners whose earnings are high relative to their neighbors are more likely to move away than those who earn relatively less. Finally, judgments about structural strength also appear relevant. In particular, the coefficient on the "poor neighborhood" dummy variable is positive and significant in two cases for white households. A white household, that is, is more likely to move from a neighborhood if it believes it is sub-standard.
Table
7
Selected Parameters from 1985-89 Mobility Model
(Dependent Variable = MOVE)
| White Owners | White Renters | Black Owners | Black Renters | |||||
| Coeff | Std Err | Coeff | Std Err | Coeff | Std Err | Coeff | Std Err | |
| Poor
Neighborhood |
0.31*** | 0.1155 | 0.133 | 0.132 | 0.385 | 0.286 | 0.035 | 0.176 |
| % Black | 0.0038 | 0.0034 | 0.005 | 0.0039 | 0.001 | 0.0052 | 0.002 | 0.0037 |
| Chblk 8090 | 0.02*** | 0.0051 | 0.0068 | 0.007 | 0.0069 | 0.012 | 0.0038 | 0.0082 |
| % Other | 0.0085 | 0.0052 | 0.003 | 0.0052 | 0.0015 | 0.015 | 0.002 | 0.0078 |
| Choth 8090 | 0.007 | 0.0078 | 0.007 | 0.0079 | 0.024 | 0.026 | 0.015 | 0.0155 |
| Income Ratio | 0.145*** | 0.045 | 0.137 | 0.123 | 0.5* | 0.302 | 0.218 | 0.224 |
| % Profesional | 0.0077** | 0.0032 | 0.0049 | 0.0044 | 0.014 | 0.015 | 0.0184** | 0.0093 |
| Poverty Rate | 0.01 | 0.0093 | 0.001 | 0.0091 | 0.0115 | 0.017 | 0.005 | 0.0086 |
| Chpov 8090 | 0.003 | 0.009 | 0.0199* | 0.01 | 0.018 | 0.017 | 0.0044 | 0.0097 |
| % New Residents | 0.012*** | 0.003 | 0.017*** | 0.0046 | 0.0028 | 0.01 | 0.0046 | 0.0071 |
| Vacancy rate | 0.0069 | 0.009 | 0.0048 | 0.012 | 0.051* | 0.027 | 0.0143 | 0.014 |
| Home
ownership |
0.002 | 0.0024 | 0.0115*** | 0.003 | 0.004 | 0.007 | 0.0023 | 0.0053 |
| N | 5,438 | 2,720 | 725 | 1,001 | ||||
*, **, *** Significant at the 10, 5, and 1 percent levels respectively
Table
8
Selected Parameters from 1989-93 Mobility Model
(Dependent Variable = MOVE)
| White Owners | White Renters | Black Owners | Black Renters | |||||
| Coeff | Std Err | Coeff | Std Err | Coeff | Std Err | Coeff | Std Err | |
| Poor Neighborhood | 0.113 | 0.139 | 0.325** | 0.136 | 0.222 | 0.271 | 0.08 | 0.183 |
| % Black | 0.0019 | 0.0039 | 0.0016 | 0.0042 | 0.003 | 0.0056 | 0.002 | 0.004 |
| Chblk 8090 | 0.013* | 0.0076 | 0.0037 | 0.0084 | 0.0096 | 0.011 | 0.017** | 0.0079 |
| % Other | 0.0089 | 0.0057 | 0.005 | 0.0056 | 0.01 | 0.014 | 0.01 | 0.0085 |
| Choth 8090 | 0.002 | 0.011 | 0.025** | 0.011 | 0.029 | 0.029 | 0.009 | 0.017 |
| Income Ratio | 0.038 | 0.052 | 0.019 | 0.069 | 0.46* | 0.239 | 0.241 | 0.201 |
| % Professional | 0.01*** | 0.0034 | 0.0032 | 0.004 | 0.023* | 0.014 | 0.0065 | 0.0082 |
| Poverty Rate | 0.0052 | 0.01 | 0.0092 | 0.009 | 0.005 | 0.017 | 0.011 | 0.0091 |
| Chpov 8090 | 0.007 | 0.013 | 0.009 | 0.0013 | 0.011 | 0.018 | 0.012 | 0.012 |
| % New
Residents |
0.0045 | 0.0039 | 0.011** | 0.0052 | 0.013 | 0.013 | 0.003 | 0.0078 |
| Vacancy rate | 0.002 | 0.0083 | 0.0002 | 0.01 | 0.022 | 0.025 | 0.005 | 0.013 |
| Home
ownership |
0.0024 | 0.003 | 0.005 | 0.0034 | 0.014 | 0.0085 | 0.0083 | 0.0053 |
| N | 5,179 | 2,514 | 744 | 982 | ||||
*, **, *** Significant
at the 10, 5, and 1 percent levels respectively
To give a better sense of the magnitude of these effects, Table 9 shows how the probability of a white homeowner moving changes when selected dummy variables take on the value of zero or one, while Table 10 presents the partial derivatives of the probability of moving with respect to selected continuous variables (calculated at their mean values). As shown, the probability of a white homeowner moving between 1985 and 1989 increases from 21.8 to 27.5 percent if it rates its neighborhood as poor. This is roughly the same difference between a married and divorced household and between households with and without pre-school children.
Table
9
Estimated Probabilities when Selected Dichotomous Variables Equal
Zero or One
White Homeowners
| Independent Variable |
Estimated
Probability - 1985
|
Estimated
Probability - 1989
|
||
|
=
1
|
=
0
|
=
1
|
=
0
|
|
| Divorced |
0.267
|
0.211
|
0.227
|
0.182
|
| Kids 0-6 |
0.269
|
0.215
|
0.248
|
0.183
|
| Bottom Income |
0.305
|
0.211
|
0.229
|
0.186
|