7.1 Summary of Purpose and Intended Contributions

We wrote this book with the central goal of documenting patterns and trends of racial and ethnic segregation across communities and over time in the United States using refined methods of measurement analysis, which can sometimes be expected to change what we thought we knew from past research and at other times add more to our understanding of established patterns. In making this our goal, we produced several contributions that happily build continuity with past research and set a foundation for future research, which we can expect to come in waves each time there is a decennial census data release. First and foremost, by using measures of segregation that are free of index bias and specifically employing the separation index, a measure of evenness that can dependably signal when prototypical patterns of segregation are occurring, we were able to reanalyze and describe patterns and levels of racial and ethnic residential segregation across the United States and over time in Chap. 3. We are not the first to describe patterns of segregation, here operationalized as the uneven distribution of two groups across neighborhood-level spatial units, across communities and over time in the United States. But we are the first to simultaneously use measures that are corrected for index bias, measure segregation of households rather than persons, and expand our analysis to not only metropolitan areas but also micropolitan areas and noncore counties. Our findings in Chap. 3 should be viewed as reliable benchmarks for descriptive analyses of racial and ethnic residential segregation across a broad range of communities moving forward and should also be taken instructively, as they demonstrate the application of the methodological changes that we recommend should be the standard for residential segregation measurement.

In addition to revisiting popular areas of segregation research, such as the segregation of large racial and ethnic groups in metropolitan areas, we also addressed a major shortcoming of the existing literature by measuring and analyzing segregation in understudied contexts including nonmetropolitan communities and Latino, Asian, and Black new destination communities in Chaps. 4 and 5. These are topics that have not gone ignored, but rather we think have been strategically avoided or studied with caution due to the fact that the measurement issues we address in this book are most prominent in scenarios characteristic of smaller and more homogenous communities. More specifically, index bias will be at its worst when using small spatial units and when the two groups in the analysis are very imbalanced in size. This includes nonmetropolitan communities where one must use census blocks in order to capture neighborhood-level homogeneity and that are often not very diverse. It also includes new destination communities, of which the majority are nonmetropolitan and also, by definition, are predominately White with a small but emerging minoritized racial population. In addition, the choice of segregation index for measuring evenness is more consequential in these contexts. As we carefully demonstrated throughout this book and review more below, the popular dissimilarity index is incapable of making the distinction between polarized and dispersed unevenness. The former is a pattern of prototypical segregation where two groups have very little residential contact with one another and there exist the conditions for there to be location-based inequalities, while the latter does not manifest as meaningful group separation across space. The dissimilarity index will be more prone to registering high scores under conditions of dispersed unevenness in communities where one group in the analysis makes up a much smaller share of the population than the other, which is often the case in nonmetropolitan communities and is always the case, by definition, in new destinations.

In being able to overcome these two measurement challenges, in addition to making a simple adjustment to measure the segregation of households rather than persons, we are able to provide a solid foundation for residential segregation research of smaller populations and in nonmetropolitan communities. The importance of this contribution is clear if one looks at the last decade of residential segregation research, which has demonstrated an increasing awareness that racial and ethnic diversity is no longer a feature one can only expect to find in metropolitan areas. Migration and natural demographic transitions have made the nonmetropolitan United States more heterogenous than ever (Johnson & Lichter, 2022). As we have argued throughout this book and as other researchers have claimed as well, these changes open up new opportunities to test prevailing theories of residential segregation and neighborhood inequalities that largely emerged through empirical studies of urban environments. Therefore, our substantive and methodological contributions to these areas of residential segregation research should be viewed as a path forward that is cleared of the obstructions created by segregation index bias.

In keeping with our intention of dialoguing with past residential segregation research to establish new directions, we also demonstrated how Fossett’s (2017) innovations in segregation measurement can improve and advance how we analyze locational attainments, or household-level neighborhood outcomes. Past research on locational attainments revealed much about the micro-level factors that determine residential location and how these correlations may vary by racial group. But where the literature has fallen short is in being able to draw a direct link between household-level locational attainment outcomes and overall patterns of segregation. In Chap. 6, we explain how this is due to the conventional formulas employed to calculate popular measures of segregation. Because of Fossett’s (2017) reformulation of these measures as a difference of group means, it is now possible to disaggregate any popular segregation index to a household-level outcome, which establishes the missing quantitative link between locational attainments and residential segregation. Thus, in Chap. 6 we take several liberties to demonstrate how this innovation introduces an analytical approach to residential segregation research that is commonly found in inequality studies, including regression standardization and decomposition. This approach makes it possible to more robustly test prevailing theories of residential segregation and identify the factors that are determinant of household locational attainments which shape patterns of racial residential segregation.

Finally, as we have made clear throughout this book and in this chapter so far, what should make our contributions so attractive to segregation researchers is that they establish clear continuity with past research. In many cases, what we find when applying our new methods of segregation measurement and analysis does not overturn previous findings in the literature. In cases where our findings do conflict with previous research findings, the reasons why are clear and should not be surprising, because these instances occur when studying segregation that involves communities and populations where standard measures of segregation are known to be less trustworthy as a result of the well-documented problem of index bias. Often, segregation researchers have known to avoid these cases anyways. Therefore, our contributions remove the reasons for avoidance and make it possible for researchers to expand the scope of their work. What we contribute, both methodologically and substantively, with this book should be taken as a course correction rather than starting completely from the beginning.

7.2 Establishing Continuity with Past Research

Before reviewing specific empirical developments presented throughout this book, we first want to acknowledge the foundation of work that we built on. First, key methodological contributions to segregation research dating back to Duncan and Duncan (1955) have set the standard to how we approach conceptualizing and operationalizing residential segregation as a demographic and social outcome. From work by Duncan and Duncan (1955), Zoloth (1976), James and Taeuber (1985), White (1986), Massey and Denton (1988), and Reardon and Firebaugh (2002), we have a toolbox of segregation indices that are heavily relied on to summarize and describe segregation patterns across communities. These studies put forward segregation indices such as the dissimilarity index, the Gini index, the separation index, and the Theil entropy index and showed us their various applications, refinements, benefits, and limitations. That researchers were aware from the beginning that these indices had their flaws and sought out ways to address index bias (e.g. Carrington & Troske, 1997; Winship, 1977) is exactly why we view this book as a contribution that establishes continuity with the existing literature. This is because we directly address those limitations and demonstrate how to apply the changes needed in order to advance this area of research.

Second, we acknowledge those studies that set the standard for conducting macro-level, descriptive studies of residential segregation exemplified in work by Massey and Denton (1985) and Iceland et al. (2002) which demonstrated the best conditions under which one can do segregation research using segregation indices in their original formulation without correction for index bias. These studies, which typically focus on the largest metropolitan areas and employ the dissimilarity index, developed a model for comparing segregation patterns across communities and over time using large data sources and convenient summary measures. With this book we also connect our work to this tradition by conducting the same types of analyses except with segregation indices that have been corrected for index bias and with other specifications that make it possible to identify varying patterns of uneven distribution. By correcting for index bias and addressing other issues that posed challenges for extending the scope of analysis beyond the largest metropolitan areas and populations, we have broadened the possibilities for segregation research within this tradition of macro-level analyses.

This last point brings us to also acknowledge those who pioneered segregation research in communities beyond the largest metropolitan areas in the United States, including those who have been working through the challenges of measuring residential segregation in nonmetropolitan communities and in Latino and immigrant new destination communities. This includes very early work by Hwang and Murdock (1983) and later work by Lichter et al. (2007, 2010) and Hall (2013). These studies faced numerous measurement challenges with the knowledge that standard segregation index formulas were inherently flawed in a way that becomes apparent when using small spatial units (i.e. census blocks) and measuring segregation of small populations (e.g. immigrant groups and newly emerging racial and ethnic groups). In most cases, researchers have taken safe routes through by imposing tight restrictions on case selections so that only communities with large enough populations were included or by using weighted segregation indices to down-weight the more problematic cases affected by index bias. These researchers have made important contributions to our understanding of residential segregation outcomes in nonmetropolitan settings and has raised the call for more work in this area. This book answers that call and also opens up new possibilities for research on these communities by directly remedying the measurement issues that have hindered any progress. Thus, in so many ways we see the contributions of this book to the residential segregation literature as a leap forward on the same path, encouraging established approaches to be used but with some modifications to address the problems that have severely limited what is possible to learn and know about residential segregation patterns.

7.3 Empirical Developments from the Present Work

In this section we review specific empirical findings that we have presented throughout this book. First, in Chap. 3 we measured levels of White-Black, White-Latino, and White-Asian residential segregation in addition to levels of segregation between minoritized racial groups across all community types from 1990 to 2010. Our specific methodological approach expanded what we know about patterns and trends of residential segregation in the United States by making it possible to include more communities, including many nonmetropolitan communities. For the largest metropolitan areas, we produced findings consistent with past research on a few points. First, White-Black segregation is highest among the group comparisons, follows a pattern of polarized unevenness characteristic of prototypical segregation, and is declining over time. This is true even after correcting for index bias, employing the separation index, and measuring segregation of households rather than persons. Second, White-Latino and White-Asian segregation is holding steady at the same levels over time. But this is where our findings deviate from past research.

The first indication that our approach produces different results is in finding that White-Latino segregation has been lower and more in line with a pattern of dispersed unevenness than previously understood. Across all community types, White-Latino segregation has generally followed a pattern of dispersed unevenness at the initial timepoint of 1990, but trajectories from there vary by community type. Significantly, White-Latino segregation appears to be trending towards a pattern of polarized unevenness in metropolitan areas, meaning that White and Latino households are increasingly more separated across space over time. White-Asian segregation is also quite low, but that has generally been understood to be the case. What we have learned, however, is that White-Asian segregation also tends to follow a pattern of dispersed unevenness, meaning that to the extent that unevenness is detected, it is not enough to permit location-based inequalities. White and Asian households for the most part reside in the same neighborhoods, with Asian households living in neighborhoods that have only slightly lower percentages of White households. The dissimilarity index would not make this clear, but the separation index can be relied upon to understand this important aspect of uneven distribution. As a final and related important finding from Chap. 3, by looking at the separation index and the dissimilarity index simultaneously, we are able to chart out the trajectories of patterns of uneven distribution over time based on how the two indices are changing in tandem. These findings are essential for answering questions about the changing nature of group separation and related inequalities and group interactions over time.

In Chap. 4, we went further into understanding patterns and trends of residential segregation in nonmetropolitan areas. As we reviewed above, this area of research has faced tremendous barriers due to the limitations of standard segregation indices. Thus, our findings in these communities, generated using segregation indices completely free of the troublesome issue of index bias, are foundational. In addition to what we found in Chap. 3 about the general levels of segregation observed in nonmetropolitan communities, we also found how critical it is to use the separation index to measure segregation in nonmetropolitan communities. In these contexts where the minoritized racial group is relatively much smaller compared to the size of the White population, the dissimilarity index has a high likelihood of registering high scores when in fact what is occurring is dispersed displacement from even distribution that does not at all resemble a prototypical pattern of segregation. While White-Black segregation typically looks prototypical even in nonmetropolitan communities, White-Latino and White-Asian segregation in these communities frequently registers medium-level scores on the dissimilarity index and very low scores on the separation index – indicating a pattern of dispersed unevenness. The implications here are important, because it means that, when dispersed unevenness is occurring, these groups are actually having high levels of residential contact with White households and opportunities to create location-based inequalities are quite low. But to be clear, dispersed unevenness is not a given in nonmetropolitan communities, even when the minoritized racial group is very small in number. We highlighted cases in Chap. 4 where in fact polarized unevenness occurs even when the minoritized racial group makes up less than 3 percent of the pairwise population.

For all the reasons why dispersed unevenness and related challenges with measuring segregation using the dissimilarity index are common in nonmetropolitan communities, these issues are even more pronounced in new destination communities. Given that dispersed unevenness is more likely (but not a given) when one group in the comparison is disproportionately small, new destinations are by definition communities where dispersed unevenness would be expected to be more common. Indeed, we often found this to be the case in Chap. 5, especially at the initial time point prior to the significant population growth of the minoritized racial group. This is an important finding for a number of reasons. First, scholarly interest in residential segregation in new destinations has grown over the last decade and researchers need to be prepared with the proper measurement tools to assess and evaluate levels and trends of segregation in these communities. This means not only using indices free of index bias but also considering which index is best suited to detect prototypical segregation when it is occurring. What we found is that the separation index, corrected for index bias, is well up to the task. While the dissimilarity index will pathologically give high scores when there are no visible indications of prototypical segregation occurring, the separation index will only give a high score when it is clear from reviewing spatial distributions and average levels of group contact that the two groups in the analysis are in fact having little residential contact with one another. Second, given that new destinations are by definition demographically dynamic, with one group emerging and growing rapidly over a short period of time, there is much interest in wanting to understand how segregation is shifting over time in these communities as the minoritized racial group grows. One cannot answer this question with the dissimilarity index because a high score can either signal dispersed unevenness or polarized unevenness and therefore shifts in the underlying pattern of unevenness, from polarized to dispersed or vice versa, may not be detected with the dissimilarity index. This is troubling because a shift in either direction is an important signal for how race relations in the community are changing over time as the minoritized racial group grows, with White households either being more integrated with or more segregated from the new group. The best way to accurately measure patterns of unevenness and how they change over time is to use the separation index.

An additional point from our findings on nonmetropolitan communities and new destinations, which are also often nonmetropolitan communities, is that contrary to some of the existing research, segregation is often quite low in these communities and rarely approaches the levels seen in metropolitan areas that are known for being highly segregated. This is especially the case for White-Latino and White-Asian segregation, which most often appears to demonstrate a pattern of dispersed unevenness. The somewhat exception to this point is that White-Black segregation more often follows a pattern of polarized unevenness even in nonmetropolitan communities, albeit at lower levels than in metropolitan areas. This distinction is possible to make by correcting for index bias and using the separation index, in addition to measuring segregation of households rather than persons, the latter of which contributes to the problem of index bias. Making these adjustments also produces trends over time that can be believed because one can be assured that any changes in segregation scores are the result of real shifts in population distributions across neighborhoods rather than resulting from changes in factors that contribute to index bias. Thus, we are able to conclude that segregation is rising in some nonmetropolitan communities and for some specific groups. For Latino households, this is occurring in new destinations, which is in direct contrast to other nonmetropolitan communities as well as metropolitan areas. This is also the case for Asian new destinations. Only for Black households do we see segregation generally declining in all community types, including nonmetropolitan communities and Black new destinations.

The final set of major empirical findings that we would like to review come from our micro-level analyses of segregation in Chap. 6, where we disaggregated the separation index using Fossett’s (2017) formula to predict the household-level neighborhood outcomes that underlie segregation patterns and are used to calculate the separation index. With the separation index reconstituted as a measure of group inequality on residential contact with White households, we can model neighborhood proportion White at the household level with household characteristics as predictors of the outcome and employ the methods often used in inequality studies including regression standardization and decomposition. This approach to analyzing segregation is in alignment with the level of theorizing that prevails in segregation research, where theories of spatial assimilation and place stratification emphasize resources and barriers, respectively, that affect minoritized group contact with White households. What we found was support for both theoretical perspectives, although the relevance of each theory varies by group, with Black households experiencing more pronounced place stratification effects than Latino or Asian households. Latino and Asian households also experience some place stratification effects, especially in high-segregation contexts, but they are also more likely to see returns on their gains in socioeconomic status and acculturation in increased residential contact with White households.

7.4 Methodological Developments

Finally, we review the methodological advancements in segregation research that we feature throughout this book through the empirical analyses summarized in the previous section. These technical contributions to segregation research, described in detail in Chap. 2, deserve to be mentioned again here to emphasize the impact that they can and should have on future segregation studies. Our key methodological contributions were developed from the work of one of the authors of this book, which can be found in full technical detail in New Methods for Measuring and Analyzing Segregation by Mark Fossett (2017). But the empirical applications of these methods are the impetus for this book as they demonstrate how our understanding of residential segregation patterns and trends might change, or sometimes hold strong, if we make the necessary adjustments to the tools we use to measure and analyze segregation.

The first of these contributions is the difference-of-means formulation of common segregation indices including the dissimilarity index and the separation index. Standard formulas for calculating these indices assume that what the researcher has on hand are census tabulations, and therefore these formulas are designed for convenient use with tabulated data aggregated to some neighborhood-level spatial unit such as a census tract. These formulas mask the individual (i.e. household)-level neighborhood outcomes that make up these tabulations and are ultimately used to construct a segregation index. Fossett’s (2017) revised formulas are mathematically equivalent but reconfigured so that it is clear how these segregation indices are an aggregation of individual-level outcomes. In calculating a segregation index using location-based scores assigned to individual households, many other advancements are possible, including the ability to identify and remove the source of segregation index bias.

Thus, correcting for index bias is a major feature of this book. Removing the source of index bias involves subtracting the reference household from the calculation of the group proportion for the group that that household is a member of, so that no household is counted as its own neighbor. By making this simple and effective adjustment, index bias is no longer an issue and we are able to generate new measures of segregation that both correct results from past research and open up new areas of research in communities and on populations where index bias was too problematic to produce trustworthy segregation scores. These include nonmetropolitan communities and smaller minoritized group populations such as those found in new destination communities. There were also cases where correcting for index bias had no or minimal effect on the scores that were produced, including large metropolitan areas where the conditions that lead to index bias are not present. These cases do not cause us any concern, because it leads us to make the following main point about index bias. To the extent that index bias is a problem, using the unbiased scores will completely eradicate the problem and produce scores that can be believed according to the intention of the index being used. When correcting for index bias does not change the score, there is no downside to using the unbiased index regardless. The point, therefore, is that one should always use the unbiased formulas because it never makes segregation measurement worse and, in many cases, it will be an improvement.

While index bias is a problem that segregation researchers are well aware of, there are issues particular to the dissimilarity index that researchers may be less familiar with despite this index being the workhorse of segregation research. We make another methodological contribution to the study of segregation by demonstrating how the dissimilarity index is incapable of distinguishing between polarized and dispersed unevenness. While the former refers to a pattern that we expect to find when the dissimilarity index is high – a pattern of prototypical segregation with little residential contact between the two groups – the latter is a pattern that does not look like prototypical segregation because the two groups are in fact having high levels of residential contact with one another. Under conditions of dispersed unevenness, the dissimilarity index may still take on a high score. Throughout Chaps. 3, 4, and 5 we show how relying solely on the dissimilarity index to measure residential segregation can cause the researcher to miss variations in underlying patterns of uneven distribution that produce sociologically meaningful divergences in outcomes. Dispersed unevenness is a pattern of uneven distribution that does not produce the conditions under which location-based inequalities can occur.

As an example of why this distinction between dispersed and polarized unevenness matters, consider how the concept of redlining has in recent years gained more attention as researchers have explored ways to link historical redlining to present-day location-based outcomes including health and educational disparities and racial wealth gaps. During the 1930s and 1940s, the racial makeup of the neighborhood was often an explicit reason to rate a neighborhood as hazardous for lending (color-coded as red). Neighborhoods mostly composed of Black, Mexican, or Chinese households often fell under this category, while predominately White and affluent neighborhoods were rated as the best locations for homeowner loans. However, redlining would have only been possible under conditions of polarized unevenness where neighborhoods could be distinctly identified as having predominately White households or having predominately racially minoritized households. If one were to use the dissimilarity index to identify the spatial distributions that could make redlining possible, there would be communities misidentified as having those conditions because D can take on a high score when either dispersed unevenness or polarized unevenness is occurring, despite the former not being a pattern that would support the practice of redlining.

In contrast, the calculation of the separation index makes it impossible to register a high score unless polarized unevenness is occurring. To review, the separation index is the simple difference in the average residential contact that each group has with White households. A high score on S is a direct measure of a large difference in contact, where White households have high levels of contact with White households and the minoritized group households have low levels of contact with White households. When visualized on a map of population distributions across neighborhoods, these large differences in residential contact with White households will always appear as a pattern of polarized unevenness where there are neighborhoods that are distinctly identifiable as being predominately White or being predominately of the minoritized racial group. When this pattern occurs, it is entirely possible to deny resources to neighborhoods where minoritized racial groups live and direct resources to predominately White neighborhoods, thereby revealing a link between residential segregation and location-based inequalities. The point, therefore, is that in order to identify sociologically meaningful patterns of residential segregation, it is better to use the separation index over the dissimilarity index. Similar to our argument for correcting for index bias, there is no downside to using the separation index instead of the dissimilarity index. The two indices will agree when polarized unevenness is occurring. When they disagree, it will always be when the dissimilarity index is showing higher levels of uneven distribution than the separation index, and this is a strong signal that dispersed unevenness is occurring. This distinction is critically important to be able to make as a researcher, and it can only be made by using the separation index.

Another methodological advancement featured in this book is one that we have contributed to the literature in previous empirical studies (Crowell & Fossett, 2018, 2020, 2022) and present again in Chap. 6, which is the ability to model segregation as a household-level outcome in the tradition of locational attainments research. What makes our approach different from past locational attainment studies is the difference-of-means formula that we use to calculate the separation index (and can also be used to calculate the dissimilarity index). These formulas disaggregate the index into an individual (i.e. household)-level score based on some neighborhood outcome such as neighborhood proportion White. Previous studies by other researchers have also modeled these outcomes, but without the ability to link them to aggregate measures of segregation for the larger community (i.e. the metropolitan area). By reformulating the index as a difference of group means, we are not only able to directly link locational attainments to segregation outcomes, but we are also opening up the opportunity to employ methods of analysis that are popular in inequality studies.

One of these methods is regression standardization and decomposition, where predicted values on the separation index can be generated based on matching the two groups in the analysis on group characteristics or rates of return on those characteristics using the covariates in the model and the estimated coefficients from the model. The separation index can then be decomposed into the contributions made by group differences on those two components, which allows us to better understand how segregation is driven both by differences in group characteristics (such as income, education, and nativity) and unequal rates of return on those group characteristics. These two components correspond to the two prevailing theories of residential segregation, with the former falling within the spatial assimilation framework and the latter within the place stratification framework. While the methods we use here are not new by any means, they are in many ways new to the study of residential segregation because they were not possible without the difference-of-means approach to calculating segregation indices introduced by Fossett (2017). Like all of the other methodological developments we have contributed so far, this one also establishes continuity with existing research conventions because it advances the locational attainments approach to studying segregation that has been popular in the literature over the past several decades.

Finally, in this book we demonstrated one more methodological adjustment to the study of residential segregation, which is to measure the segregation of households rather than persons. Although we did not give this issue as much attention, we do discuss it in Chap. 2 and demonstrate the impact that it can have using a case study in Chap. 5 where we analyzed segregation outcomes in Latino and Asian new destinations. The problem with measuring the residential segregation of persons instead of households comes back to the issue of index bias. While we now have the formula correction to remove index bias, it cannot be dealt with completely unless one is using households as the microunit rather than persons. This is because the source of index bias is that the reference individual is counted as their own neighbors and therefore same-group residential contact is overcalculated. This logic extends to the problem of counting persons who share a household as neighbors because one cannot assume that these individuals could be randomly redistributed across neighborhoods to achieve even distribution. In reality, individuals who share a household would likely move together as a single unit. Thus, the only way to fully eradicate index bias is to use households as the microunit of analysis rather than persons. This problem is especially pronounced in communities with the demographic characteristics that make index bias worse in general, including nonmetropolitan communities and new destination communities where the minoritized group is significantly smaller than the White population.

7.5 Future Directions for Residential Segregation Research

In the introductory chapter of this book, we explained that the purpose of this book was not necessarily to provide current data on residential segregation patterns and trends. Given that 2020 census products are being released at the time that this book is being written, that would not be a credible claim. Instead, what we have provided with this book are corrected and more comprehensive baselines for contemporary racial and ethnic residential segregation patterns in the United States leading up to the present, which will put the literature on the correct course to understand how these patterns are shifting going forward. Thus what we ultimately hope readers will take from this book are new ways to analyze segregation that will overcome many of the problems that have hindered this area of research and also open up opportunities to ask new research questions in an area that has been constrained to a narrow focus on certain communities and populations.

As the 2020 census summary files become available, we will see a new wave of residential segregation studies aiming to understand how our communities have changed in an increasingly multiracial and diverse society. As is tradition in this literature, efforts to provide broad summaries of residential segregation patterns across metropolitan areas and beyond will be made with attention given to how these patterns have changed over the decades. Interest in residential segregation in nonmetropolitan communities and destinations that are new for Latino, Asian, and immigrant populations will stay strong as these communities grow and the migrants who have settled in them over the past three decades continue to establish a presence through family, economic, and social life. Variations in household movements across neighborhoods by group and neighborhood characteristics will also continue to hold our attention because they are key to asking questions about the barriers and opportunities that can either weaken or reinforce residential segregation patterns. And there will be new and understudied questions that will come up about populations that have not received enough attention, often due to their small numbers. These include ethnic subgroup populations disaggregated from panethnic categories, multiracial populations, and immigrant populations.

For all of these focus areas in the study of residential segregation, the empirical results and methodological techniques that we provide in this book will be critically important. The issue of index bias will confound any results that come from these studies unless it is dealt with directly by removing the source of index bias in the formula and studying the segregation of households instead of persons. The choice of segregation index for measuring uneven distribution will have serious implications when studying any types of communities where the two groups in the analysis are majorly imbalanced in size. And any interest in how locational attainments drive segregation patterns will be best served by using the difference-of-means approach to calculate the segregation index so that the score can be disaggregated to individual-level outcomes and modeled as locational attainments. Thus, we encourage researchers who study residential segregation to use the results and techniques provided here to refine our understanding of residential segregation patterns, explore new questions about different communities and populations, and move the literature forward.