Tieboutian clubs revisited

Abstract

This paper examines the evolution of Tieboutian clubs over time and across metropolitan regions. We address this topic by examining a large set of socioeconomic characteristics of metropolitan regions to assess empirically whether and how those member attributes may figure into a Tiebout sorting process. The method entails factor analysis to reduce an initial set of forty-nine variables to a much smaller number of factors that represent groups of intercorrelated variables. An analysis of variance is then undertaken on the corresponding factor scores to assess whether census tracts within municipalities are more alike than those between municipalities. We do this for the Los Angeles metropolitan region using census data from 1970, 1990 and 2020, and we repeat the process for the twenty largest metropolitan regions (core-based statistical areas) in the USA for 2020. Several key results emerge from this analysis. While in most cases seven or eight factors emerge repeatedly, four are most persistent and ubiquitous; these pertain to economic class, race, age cohort and immigration status. These are, in effect, the most enduring fundamental aspects of the socioeconomic landscape in metropolitan regions in the USA. The final question addressed by our work links back to the original motivation, which is to assess the role of municipalities in the formation of these clubs. We do so for the twenty largest metropolitan regions in the USA. The results from our analyses of variance are compelling. With very few exceptions we find that census tracts with similar club characteristics have a strong tendency to be co-located within the same municipalities. This strong result holds regardless of which metropolitan region we examined, or which decade. This affirms that municipalities are indeed Tieboutian clubs.

1 Introduction

This paper examines the characteristics and evolution of Tieboutian clubs over time and across metropolitan regions in the USA. Tiebout’s (1956) seminal paper posited a sorting process by which each household within a metropolitan region would choose to reside in whichever local jurisdiction offered the bundle of local public goods and taxes that best suited the household’s budget and preferences. A somewhat more generalized formulation of the Tiebout model builds on Buchanan’s (1965) theory of clubs, and of special interest here is the case where club members are not indifferent to their fellow members’ characteristics (Sandler and Tschirhart 1997). Following Heikkila (1996), we address this topic by examining a large set of socioeconomic characteristics to assess empirically whether and how those member attributes may figure into a Tiebout sorting process. The method entails factor analysis to reduce an initial set of forty-nine variables to a much smaller number of factors that represent groups of intercorrelated variables. An analysis of variance is then undertaken on the corresponding factor scores to assess whether census tracts within municipalities are more alike than those between municipalities. We do this for the Los Angeles metropolitan region for census data from 1970, 1990 and 2020, and we repeat the process for the twenty largest metropolitan regions (core-based statistical areas) in the USA.

Several key results emerge from this analysis. While in most cases seven or eight factors emerge repeatedly, four are most persistent and ubiquitous; these pertain to economic class, race, age cohort and immigration status. These are, in effect, the most enduring fundamental aspects of the socioeconomic landscape in metropolitan regions in the USA and (at least, in the case of Los Angeles) have been so for at least half a century. While the factors themselves represent “club characteristics”, the corresponding factor scores for each census tract show how those characteristics are distributed geographically. Evidence from Los Angeles indicates that “club profiles’’ are quite rooted in their respective neighborhoods (census tracts) over many decades. This finding is consistent with the neighborhood change literature (Wei and Knox 2014; Delmelle 2015) that focuses on those census tracts that do transition from one type to another, while also finding notable stability over time for many census tracts. The final question addressed by our work links back to the original motivation, which is to assess the role of municipalities in the formation of these clubs. We do so for the twenty largest metropolitan regions (CBSAs, or core-based statistical areas) in the USA. The results from our analyses of variance are compelling. With very few exceptions, we find that census tracts with similar club characteristics have a strong tendency to be co-located within the same municipalities. That is, the variance of factor scores within municipalities is significantly lower than the variance between municipalities. This strong result holds regardless of which metropolitan region we examined, or which decade. This affirms that municipalities are indeed Tieboutian clubs.

The next section reviews the antecedents for Tieboutian clubs, tracing back to Tiebout (1956) and hence to Samuelson (1954) whose public goods allocation dilemma Tiebout was responding to. The lineage is then traced through Buchanan’s theory of clubs (1965) and Schelling’s (1971) microdynamics of neighborhood change and segregation. More recent work in the Tiebout tradition is reviewed, as is recent work on neighborhood change. The section following that focuses empirically on the Los Angeles metropolitan region, as it is considered to be an exemplar of Tiebout sorting and allows us to compare our results directly with those of Heikkila (1996). While his study was confined to Los Angeles County for the year 1990, ours is based on the larger metropolitan region (CBSA) for 1970, 1990 and 2020. This half-century time span provides ample opportunity to observe any urban restructuring that might characterize the evolving nature of Tieboutian clubs. In the penultimate section of this paper, we expand our scope geographically to include the twenty largest metropolitan regions of the USA for the year 2020. Doing so enables us to assess whether the marked Tieboutian tendencies exhibited for the Los Angeles region are an anomaly, or whether they are in fact broadly representative of a national metropolitan phenomenon. We find the latter to be the case, and our concluding section offers some final observations on our findings and its public policy implications.

2 Historical antecedents for Tieboutian clubs

For our purposes, a useful way to assess the abundant subsequent literature spawned by Tiebout’s seminal work is to frame it in terms of several offshoots, each with its own corresponding seminal contributions. Principal among these, at least in terms of its direct lineage, is the local public goods/public finance literature as prompted by Tiebout’s response to Samuelson (1954). A closely related branch is club theory (Buchanan 1965), which grows more distinctive as its scope is extended to include member characteristics (Sandler and Tschirhart 1997). Even works in the tradition of Coase (1937, 1960) influenced Tiebout’s own thinking about contracting out of municipal services (Ostram, Tiebout and Warren, 1961). Yet another offshoot that is relevant here is the microdynamics of neighborhood change and segregation (Schelling 1971). While these main branches are distinct, their foliage overlaps in interesting ways, as illustrated in figure 1. The remainder of this section elaborates briefly on each of these outgrowths of the Tiebout model, and that summary in turn helps position the modest contributions of this paper.

Fig. 1

Scholarly lineage of Tieboutian clubs

The allocative mechanism for public goods advanced by Tiebout (1956) was in direct response to Samuelson’s (1954) demonstration that market mechanisms were inadequate to the task. Rather than looking for markets to allocate public goods to households, Tiebout had households allocating themselves among myriad local jurisdictions offering competing bundles of public goods and taxes. He argued that this mechanism would sidestep the market failures pointed to by Samuelson. Oates (1969, p. 959) extended the public finance dimensions of this proposition to include capitalization of the net benefits of public service and fiscal offerings in local property values. The interplay between public goods provision, taxes, land values and land use regulations (including their exclusionary effects) became a central focus of much of the subsequent literature in this branch of the Tieboutian literature (Bourassa and Wu 2022; Brasington 2017; Fischel 2002, 2006; Fennel, 2006; Paulsen 2009; Saltz and Capener 2016; Scotchmer 1997; Wooder 1999).

Intertwined with the public finance branch of Tiebout’s legacy is that of the economic theory of clubs, stemming from Buchanan’s (1965) seminal work. They have long been viewed as two sides of the same coin [McGuire 1974; Sandler and Tschirhart 1980], although Boettke and Marciano (2017) conclude that Buchanan saw his own work as an alternative framework to a flawed Tiebout model. One key difference, at least superficially, is that Tiebout’s formulation was premised on a specific institutional context—that of municipalities and other local governmental jurisdictions within a larger metropolis. While Buchanan’s theory of clubs readily applies to such contexts, it is not the starting point. More fundamentally, both Tiebout and Buchanan framed their work as a direct response to Samuelson (1954), with Buchanan explicitly seeking to bridge “the awesome Samuelson gap” between purely private and purely public goods. As Buchanan put it, “The central question in a theory of clubs is that of determining the membership margin, so to speak, the size of the most desirable cost and consumption sharing arrangement” (p. 2). He also observes that “the theory of clubs is, in one sense, a theory of optimal exclusion, as well as one of inclusion” (p. 13). In sum, Buchanan’s focus is within a single club, while Tiebout scans across a metropolitan horizon of a myriad differing clubs.

An important extension of Buchanan’s club theory, especially for our purposes, allows for member characteristics to be included as attributes of the clubs to which they belong. Indeed, in the common parlance of clubs, who its members are may be the most crucial consideration—one may not care whether it is a bowling club or a book club, so long as the right people are there¹. Buchanan himself recognized this possibility, but averred that his club formulation “implies that the individual remains indifferent as to which of his neighbors or fellow citizens join him in such arrangements … To incorporate this element, which is no doubt important in many instances, would introduce a wholly new dimension into the analysis” (p. 13). Indeed it would. It is useful to distinguish between embodied personal characteristics versus financial ones. In the relatively simple case where households differ only in terms of income, the Tiebout result is an array of internally homogeneous local jurisdictions that differ from each other only in terms of their fiscal profiles (Bayer and McMillan 2012). More generally, models in the Buchanan tradition are geared to tradeoffs between production externalities and consumption externalities within clubs.

The models become more complex as race, gender, cultural dispositions or other more embodied personal characteristics come into play. This brings us to yet another distinct lineage of models—those associated with Schelling’s (1971) work on microdynamic models of segregation. This has given rise to a plethora of contemporary cellular automata and agent-based models that are likely to endure as mainstays for modeling urban dynamics (Batty 1997; Webster and Wu (2001), Heikkila and Wang (2009); Galster (2019). McGuire (1974) and others quickly recognized the interplay between Tiebout’s and Schelling’s works: “Where [Schelling] dealt with a ‘pure’ case in which all the externalities are personal and have to do with people’s preferences regarding whom they live among, [models in the Tiebout tradition] have dealt with the opposite ‘pure’ case in which nobody cares in the slightest whom he is with, as long as the commodity value of his ‘location’ yields him the greatest net utility.” Schelling also drew implicit links to Tiebout’s work, arguing that “To choose a neighborhood is to choose neighbors. To pick a neighborhood with good schools is to pick a neighborhood of people who appreciate schools.” Linking of Tiebout and Schelling-type models continues with Fossett (2011), Wang (2011), Banzhaf and Walsh (2013) and Lynham and Neary (2024). Banzhaf and Walsh conclude that sorting models in the tradition of Tiebout and tipping models in the tradition of Schelling are fundamentally connected: “any analysis of the distribution of spatially delineated public goods across demographic groups must account for endogenous sorting on those demographics, while at the same time studies of spatial patterns in demographics must account for public goods.”

Yet another related area of research examines neighborhood change. These are primarily empirical investigations, where much of the scholarship focuses on innovative spatial data modeling. While not a direct outgrowth of Tiebout’s public finance, Buchanan’s club theory or Schelling’s microbehavioral dynamics, this work provides an important basis for assessing facts on the ground. Wei and Knox (2014), for example, examine trajectories of neighborhood change at the census tract level from 1990 to 2000 to 2010 for all U.S. metropolitan areas. They use cluster analysis to identify seven distinct types of neighborhoods, and define neighborhood change in terms of socioeconomic transitions from one cluster to another. They find that most neighborhoods have been remarkably stable over that period, with the exceptions being mostly middle-class neighborhoods that were either sinking into lower-echelon categories or rising to more elite levels. In contrast, poverty-stricken neighborhoods have remained dishearteningly stable, as were wealthier enclaves—evidence of enduring inequality. In a rather nice set of papers, Delmelle (2015, 2017, 2019) undertakes similar investigations. Viewed in the aggregate, she largely confirms the findings of Wei and Knox while further advancing the methodology. In the more recent paper she finds increasing spatial fragmentation (i.e., less clustering) of neighborhood types, arguing that neighborhood partitioning is less-ordered than traditional models would suggest.

A recurring, albeit sometimes implicit, theme in these various strands of research is the role of municipalities. These were, after all, central to Tiebout’s model, but much less so in the case of Buchanan’s or Schelling’s work, or the neighborhood change literature just discussed. In Tiebout’s case, the primary focus is on the local public goods and financing packages, and not so much on the attributes of the resident households. In Buchanan’s case, municipal jurisdictions or school districts are seen as specific examples of a more general formulation of clubs. In Schelling’s base case model, each household is at the center of its own neighborhood, which in turn is defined as a kind of buffer zone centered on that household. And, as we have seen, the neighborhood change literature focuses on census tracts rather than municipalities. Heikkila (1996) asks the question directly, “Are municipalities Tieboutian clubs?”. He infers an affirmative response by conducting an analysis of variance on factors that comprise a full range of socioeconomic characteristics at the census tract level. His results strongly indicate that variance of factor scores between municipalities well exceeds the within-municipality variance. Birds of a feather flock together, and they appear to do so with reference to municipal boundaries. Galbraith (2003) suggests that ethnic enclaves, too, may best be understood as Tieboutian clubs, with culturally specific social capital. Hachadoorian (2016) takes a more direct approach by focusing on jurisdictional boundary areas to search for evidence of jurisdictional variation in service levels.

3 Data

The remainder of this paper revisits the question of municipalities as Tieboutian clubs. The methodology we use is similar to that employed by Heikkila (1996) but extending the analysis in two ways. Rather than examining a single metropolitan area in a single year, we track the changes of the Los Angeles region over the span of half a century, with observations from 1970, 1990 and 2020. We also extend the analysis for the year 2020 over the twenty largest metropolitan regions in the USA. Taken together, these two extensions allow us to assess the evolving nature of Tieboutian clubs over time and across space.

This project uses tract-level data from the 1970 and 1990 decennial U.S. censuses and the 2016-2020 iterations of the American Community Survey (ACS). At the time of this analysis, many variables from the 2020 decennial were not available, so we use 2016–2020 5-year ACS estimates as a substitute. The ACS is conducted by the U.S. Census Bureau and uses a representative sample of more than 3.5 million randomly selected addresses. We also use the census “places” boundaries for 1980, 1990, and 2020 to represent municipalities in 1970, 1990 and 2020, respectively. Census “places” boundaries include legally incorporated jurisdictions at the sub-county level and Census Designated Places that are statistical entities that circumscribe settled, unincorporated areas.

Our data span half a century. Over the course of these decades, not only are there ubiquitous changes in the measured values for any given variable, there are also changes in what was measured. Innovations in technology, economic restructuring, and our evolving sense of identity are all reflected in these data. For example, unlike previous surveys, the 2016–2020 versions of the ACS allowed respondents to select multiple racial and ethnic categories. Likewise, some occupational or industrial sector categories gradually evolved over time, as did commuting options. These and other data considerations are addressed in more detail in the sections that follow in order to provide a more nuanced interpretations of our empirical findings.

4 Los Angeles 1970–2020

4.1 Regional trends

The left-hand columns of table 1 provide summary data for the Los Angeles metropolitan region² for 1970, 1990, and 2020. The census data selected are intended to provide a broad overview of the socioeconomic characteristics of the metropolitan region at the census tract level. In this regard, our empirical approach is broadly similar to that of Wei and Knox (2014), Delmelle (2019, 2017, 2015) and other examples of the neighborhood change literature. There are two significant differences in our approach, however, reflecting the respective purposes. For those authors, census tracts are the fundamental unit of analysis because they correspond more closely to their subject of investigation, which is neighborhood. For our purposes, however, the subject of investigation is Tiebout’s municipality. Our approach also differs somewhat methodologically—again, reflecting distinctive purposes. Much of the neighborhood change literature uses cluster analysis to sort census tracts into distinct groups (clusters) of similar types. Neighborhood change is then characterized and assessed in terms of the transition over time of census tracts from one cluster type to another. In contrast, our approach begins with factor analysis, where each factor is akin to a composite variable constructed from the original data set. The original variables help define and describe our emergent factors much as they define and describe the emergent clusters in the neighborhood change literature. These two important distinctions in our approach are both brought to bear in addressing our motivating question, “Are municipalities Tieboutian clubs?”. Supporting evidence in the affirmative is provided below through an analysis of variance applied to the scores of each factor.

Table 1 Comparison of mean values for selected socioeconomic variables for the Los Angeles metropolitan region for 1970, 1990 and 2020 and for the aggregate of the top twenty CBSAs in 2020

Several trend lines are evident over the half-century span in table 1, and these trends represent the evolving context within which Tieboutian clubs might arise. One observes a notable aging of the resident population in the Los Angeles region, with the share of 24-and-under falling from 43.0 percent to 30.4 percent from 1970 to 2020. Meanwhile, the proportion of 45-and-over rose from 30.2 percent to 40.2 percent over the same interval, as the baby boom bulge traversed the decades. The data trends not only reflect changes in ethnic and racial composition over the decades, but also changes in how we self-identify. Thus, in 1970, 96.4% of respondents were reported to be either White (86.0%) or Black (10.4%). By 1990, this stark bifurcation had begun to dissolve, with 62.9% identified as White and the remainder as Black, Asian, or other—where the latter category includes Hispanics who in prior Census years might have reported as White or Black. By 2020, a scarce majority of 50.8% reported as White, and a new Mixed race category was included. During this same period, the proportion of foreign-born respondents increased from 11.2% in 1970 to 32.6% by 2020, with much of that increase evident already by 1990.

Other trends evident in the data include shifting occupation and employment sector shares. Economic restructuring is seen in falling shares of transport-related³ and sales-related occupations. Those occupational categories gave way to a rise in professional and managerial class workers and to service-related occupations. Likewise, as part of the same economic restructuring process, we see a relative shift of employment away from manufacturing, construction and primary sectors in favor of entertainment and personal services. Another aspect of restructuring is the steady increase in respondents with at least some post-secondary education, with the share doubling from 29.3 to 60.8% from 1970 to 2020. As for commuting modes, there is a modest decline in both private automobile and public transit, thus pointing to the gradual emergence of other travel modes.

Economic restructuring is also reflected in increasing income inequality over the past half-century. The proportion of households with income⁴ below $35,000 rose from 16.4 to 23.9% while the top echelon, with incomes above %150,000 increased even more dramatically, from 7.2 to 20.3%. As both the bottom and top tiers of the income distribution grow, the middle tiers shrink commensurately, from 76.4 to 55.8%. During this same period, per capita income rose in the aggregate. Not only did people’s employment and income change, so did their housing stock. Whereas fifty years ago, almost two-thirds of households resided in single-family detached units, by 2020 scarcely over one-half did. There was also a modest but steady increase, from 45.7 to 50.1%, in the proportion of households who were renting the housing units in which they resided.

5 Factor analysis 1970–1990–2020

Factor analysis is a data reduction technique commonly used in the social sciences (Bandalos and Finney 2018). It is helpful to think of factors as composite variables derived from the original set of variables. The new factors are designed (“extracted”) to convey much of the same information in the original data set with a smaller set of newly constructed factors. These factors, in turn, can be interpreted in terms of their degree of association with the original variables. In our case, we begin with the large set of variables set out in Table 1. We then apply factor analysis separately for each of three census years: 1970, 1990 and 2020. Those basic results are set out in Table 2, where we report all factor loadings with absolute magnitude of 0.5 and above. Factors are typically reported, as is done here for each census year, in descending order of their eigenvalues. Intuitively, an eigenvalue measures how many variables’ worth of correlation is explained by the factor in question⁵. As is common practice, we retain all factors with eigenvalues above unity.

Table 2 Factor loadings with absolute values at least 0.5 in magnitude

The factor loadings in Table 2 have been color-coded for ease of interpretation, with green (tan) shading indicating a positive (negative) correlation, and with more (less) intense shading indicating absolute factor loading values above (below) 0.75. These data are further consolidated in Table 3 with the same color-coding themes applied to the variable names as well. Each factor is now presented as a “package” comprising the factor number, the factor name and year, the eigenvalue, and the associated set of color-coded variables listed in declining order of the magnitude of their corresponding factor loadings. For example, the first factor in 2020 has an eigenvalue of 12.22, so it is not surprising that there are a relatively large number of associated variables with large factor loading values. It is common practice to assign names to each factor as an aid to their interpretation (Bandalos and Finney 2018). In the case of factor one for the year 2020 the name “Professional elite” is motivated by the very high positive factor loadings for professional management occupations, education levels, and income—among others. By way of contrast, factor two for that same year is named “Young adults”, as high the factor is associated with unmarried renters aged 25 to 34.

Table 3 Factor details for Los Angeles metropolitan region

Each of the twenty-four factor “packages” in Table 3 is interesting in its own right, and the reader is invited to examine them all at leisure. For our purposes, however, we move now to Table 4 which provides a birds-eye view of the entire set of factors and how they interrelate over the half-century time span that is represented here. Several results jump out. First, we see that each of the eight factors generated by the 1970 data reappears in almost identical form in 1990. This is quite remarkable, as it indicates that the underlying aggregate socioeconomic character of the Los Angeles metropolitan region was quite stable over that twenty-year span, despite the rapid growth of the period. The names of these eight factors in 1970 indicate the broad dimensions of the socioeconomic landscape in that year:

● F1 (1970)—Professional elite	● F5 (1970)—Young adults
● F2 (1970)—Blacks	● F6 (1970)—Immigrants
● F3 (1970)—Young families	● F7 (1970)—Blue collar workers
● F4 (1970)—Transit users	● F8 (1970)—Affluent neighborhood

For seven of the eight factors, the similarity in terms of factor loadings from 1970 to 1990 (see table 3) is so strong that the same factor names readily apply. The exception is factor 8, which in 1970 (“Affluent neighborhood”) was characterized primarily by relatively high incomes, rents and housing values. In 1990, its solid middle-class credential (“Middle income”) was indicated solely in terms of income cohort rather than housing rents or values.

The persistence of factor characteristics was not as pronounced through 2020, although half of the original eight factors from 1970 persist over the half-century. Thus, there appear to be four enduring aspects of the socioeconomic landscape of the Los Angeles metropolitan region: economic class, race, age cohort, and immigration status. Restructuring was more in evidence during the 1990 to 2020 interval. As seen in Table 4, three factors that had endured from 1970 to 1990 were no longer in evidence by 2020: “Young families”, “Transit users”, and “Blue collar workers”. This does not imply that there were no longer young families, transit riders or blue collar workers in the region; but by 2020 they are no longer the defining characteristics of any factors emerging from the data analysis. Table 4 also points to a bit of an anomaly in 2020; factor 3 in that year (“Local movers”) is characterized by individuals who previously resided in the same county, but not the same house. There is no antecedent to this factor at all in either 1970 or 1990. Time will tell whether it is part of a new trending factor or whether it will quickly fade into obscurity. Another anomaly is factor 9 (“Agriculture”) in 1990.

Table 4 Factor trajectories: Los Angeles metropolitan region

6 Municipalities as clubs?

The analysis thus far has been conducted at the relatively granular census tract level. We have determined that a rich set of forty-nine socioeconomic variables boil down to four enduring aspects that permeate census tracts throughout the Los Angeles metropolitan region. It remains to be seen whether municipalities are the geographic “containers” for distinct typologies. We address this in two complementary ways. First, we ask whether the four enduring characteristics identified above are geographically stable. After all, there is nothing in the prior analysis that tells us whether it is the same census tracts that apply in each case. Although it would seem unlikely a priori, perhaps there are indeed elite professionals, Black persons, young adults and immigrants throughout the period, but they might congregate in different census tracts from one decade to the next. With this in mind, we compute the correlation matrix of the average factor scores for each municipality for all twenty-four factors, as reported in Table 5.⁶ Of special interest are the four triplets that we have identified as enduring, defining traits:

• Professional elite: [F1-1970, F1-1990, F1-2020]

• Blacks: [F2-1970, F2-1990, F6-2020]

• Young adults: [F5-1970, F4-1990, F2-2020]

• Immigrants: [F7-1970, F3-1990, F5-2020]

Table 5 Factor correlation matrix

Because the correlation matrix is calculated across the same geographic areas, and because each factor within a triplet is a similar construct to its two siblings, we would expect the correlation coefficients to be positive and significant if indeed the geographic distribution of the attribute in question is relatively stable over time. Of course, there are two reasons that we would not expect perfect correlation among these factor siblings. First, as detailed in Table 3, they are not defined identically, nor would we expect them to be—indeed the fact that they are quite similar is what is most striking about them. Secondly, we would not expect the census tracts themselves to be completely static over time, so even if the sibling factors were identical triplets, their factor scores ought to reflect the evolving nature of the LA region’s census tracts over the decades. Table 5 shows the full pairwise correlation matrix of factor scores for all twenty-four factors, where the four sets of triplets are highlighted. Ten of the twelve pairwise correlations are significant at the 1% level; the two exceptions are interesting:

• Correlation [F2-2020, F5-1970] = 0.283** ** p<0.05

• Correlation [F5-2020, F6-1970] = 0.253* * p<0.10

Note that both cover a half-century span, so there is ample time for the census tracts to have evolved. In the first case, the factor in question is “Young adults”, where one of the defining features is a strong negative factor loading for the variable “same house”. Thus, the fact that young adults represent a segment of the population that is highly inclined to relocate, it may not be a surprise that the act of moving brings about change in the youthful landscape. A similar logic applies with perhaps even more force in the second case above, where the factor in question is “Immigrants”.

The final task now is to determine whether municipalities are the geographical units corresponding to Tieboutian clubs—and for which attributes of club members. Following Heikkila (1996), we do so by means of a straightforward analysis of variance (ANOVA) to test whether within-municipality variance of factor scores is less than between-municipality variance. The null hypothesis is that within- and between-municipality variances are equal. An F-statistic is calculated for each factor, and a sufficiently large value of the F-statistic leads to rejection of the null hypothesis. These F-statistics for the Los Angeles metropolitan region are reported in table 3. The results are remarkable: of twenty-four factors from three different Census years decades removed from each other, ANOVA tests on twenty-two of those factors decisively reject the null hypothesis⁷. Put simply, our results confirm that in the Los Angeles region, municipalities ARE Tieboutian clubs, and club membership is expressed in terms of income, race & ethnicity, age, immigration status and a host of other attributes.⁸

One might well ask, “Why?”. The original Tiebout hypothesis predicts sorting on the basis of preferences for local public goods–services that are delivered by local governments to their respective residents and paid for by those same residents through taxes. Yet, our investigation did not even include any such local public finance variables. Instead, the drivers appear to be the intrinsic attributes of club members themselves. This suggests something along the lines of a Schelling club model, except that Schelling’s geographic unit of analysis was the neighborhood rather than the municipality. The answer would appear to lie in the capacity of municipal governments to influence the delivery of local public goods—including those dominated by peer effects.

Consider the four factors describing club member attributes that persisted throughout the entire study period. Economic status is dominant, and it links most directly to Tiebout’s original hypothesis, and many Tiebout-inspired models do predict sorting by income and wealth. As for race and ethnicity, municipalities also provide official platforms for representation through the electoral process. If, ceteris paribus, voters would prefer to have elected representatives who “look like me”, this could help channel Schelling-type effects into municipal boundaries. Age cohorts, too, may have distinct preferences that may align with municipal boundaries. For example, the factor “Young adults” has positive factor loadings in 1990 and in 2020 for the variable “renter”, and a strong negative loading in 2020 for “detached house”. These appear to be indicative for certain types of rental accommodation, the availability of which is in part determined by land use regulations and building codes promulgated at the municipal government level. As for the fourth crucial enduring factor, “Immigrants”, the San Gabriel Valley provides many examples of local governments that make a point of providing a full range of services in languages other than English. Tiebout may not have had such examples in mind when he authored his famous 1956 paper, but they do fit comfortably within the expanded scope of Tieboutian clubs—at least they do so within the Los Angeles region.

While Tieboutian theory offers an explanation grounded in household decision making, recent research has highlighted the role of both state and non-state actors in enforcing persistent neighborhood and municipal uniformity along racial and economic lines. Works by Gene Slater (2021), Richard Rothstein (2017), and Keeanga-Yamahtta Taylor (2019) detail the actions and policies of realtors, neighborhood associations, individual sellers and buyers, judges, and elected officials that explicitly restricted the housing and location options for racial minorities and economically disadvantaged groups while preserving White residential neighborhoods throughout the twentieth century. Such policies had lasting effects on neighborhood segregation through at least the early 1980s (Aaronson et al. 2021) and could help to explain the existence and stability of race and economic status within municipalities.

7 US metropolitan regions, 2020

We turn now to investigate whether Los Angeles is an anomaly in this regard, or whether similar phenomena can be found in other metropolitan regions throughout the country. The twenty largest core-based statistical areas (CBSAs) shown in Table 6 comprise nearly 125 million residents in metropolitan regions across the USA. With the smallest of these metropolitan regions having in the order of 3 million residents, each CBSA individually is sufficiently large to be a potential venue for Tiebout sorting. In the aggregate, they represent “Metropolis USA”.

Table 6 Twenty largest core-based statistical areas (CBSAs) in the USA

To assess the extent to which municipal-level Tiebout sorting is actually occurring, we begin with a factor analysis of the entire pooled set of census tracts using the data summarized in the right-hand column of Table 1. The detailed factor analysis results for the pooled set of twenty CBSAs are depicted in Table 7. As before, the factor analysis identifies patterns of correlation within the initial set of variables and produces a smaller number of summary factors that can be interpreted with reference to those initial variables.

Table 7 Factor analysis results for “Metropolis USA”, 2020

The eight factors emerging from this “Metropolis USA” sample are remarkably similar to those uncovered already for the Los Angeles metropolitan region, as summarized here (with eigenvalues indicated):

● F1—Professional elite—USA 2020 (9.00)	● F5—Blacks—USA 2020 (1.69)
● F2—Young families—USA 2020 (5.81)	● F6—Local movers—USA 2020 (1.47)
● F3—Immigrants—USA 2020 (3.73)	● F7—Medium house values—USA 2020 (1.21)
● F4—Transit users—USA 2020 (2.57)	● F8—Blue collar workers—USA 2020 (1.02)

Table 8 puts these “Metropolis USA’’ factors for 2020 into further perspective by drawing links to their counterparts for the Los Angeles region for 1970, 1990 and 2020. Three of the four primary factors that have persisted for half a century in the Los Angeles region are also characteristic of the country’s metropolitan regions as a whole: “Professional elite”, “Immigrants”, and “Blacks”. Another factor that appears throughout, and that is also indicated above in boldface font, “Medium house values”, pertains to slightly shifting versions over time of a “middle class” as expressed through income or housing characteristics. Thus, economic status, race, and immigration status appear to be persistent and pervasive defining features of the metropolitan socioeconomic landscape of the USA. A whole host of other socioeconomic indicators fall into line, as it were, through their correlations with these factors. They do so not just in the Los Angeles metropolitan region, which Tiebout himself viewed as exemplary, but for “Metropolis USA” as a whole.

Table 8 Factor comparison: Los Angeles (1970, 1990, 2020) and “Metropolis USA” (2020)

Table 8 also shows that three of the other eight factors for “Metropolis USA” in 2020 correspond directly to factors from the Los Angeles region for 1970 and 1990, but NOT for 2020. These three are “Young families”, “Transit users”, and “Blue collar workers”. It is interesting to speculate why these three factors were discontinued, as it were, in the Los Angeles metropolitan region while they remain present elsewhere throughout the country’s metropolitan regions. One possibility, impossible to verify yet at this stage, is that Los Angeles is the harbinger or leading edge of a trend that will soon become evident in the rest of the country. For example, the absence of “Young families” (the factor, not the people) in the Los Angeles region may point to a reconfiguration of the family unit and its correlates. Conversely, the notable absence of a factor for “Young adults” for the nationwide sample may reflect an inverse relationship with the presence of “Young families”. This might arise, for example, if young adults elsewhere in the country were more prone to start families at a younger age than their counterparts in Los Angeles. Likewise, it is possible that the Los Angeles region is relatively quick to adapt a plethora of emerging transit modes, prompting a reconfiguration of the correlates for “Transit users”. And economic restructuring due to technological change and evolving worker profiles in the Los Angeles region may have begun to dissolve the “Blue collar workers” factor. Whether these changes portend similar restructuring of socioeconomic correlates in the rest of the country remains to be seen. Finally, there is one factor from “Metropolis USA” that has a direct counterpart for the Los Angeles region in 2020 but NOT in 1990 or 1970. “Local movers” are those who reside in the same county but in a different residence than in prior years. It is reasonable to assume that many of these local movers are young adults leaving their parental homes but remaining nearby. If so, this may be part and parcel of the changes evident among the correlates of young adults finding their way in the context of tight real estate markets, COVID, and other challenges.

8 Tiebout sorting at the national level

Following Heikkila (1996), we have been asking whether municipalities are Tieboutian clubs. It may be, however, that the question should also be turned around: Are Tieboutian clubs municipalities? The factors that we have been extracting represent, in effect, the defining characteristics of clubs. We have seen that these clubs are persistent over decades and across a national geographic span. The units of analysis that we have been working with are census tracts, which correspond to “neighborhoods” in much of the recent work on neighborhood change. In the case of the Los Angeles metropolitan region, our analysis of variance demonstrates quite convincingly that these club characteristics do indeed coalesce at the municipal scale. The final step now is to assess whether the same result holds more generally across metropolitan regions throughout the country.

Table 9 shows the ANOVA results for all factors for each of the twenty largest metropolitan regions (core-based statistical areas) in the USA⁹. We have already established that, in the aggregate, these factors embody nationally the same club characteristics that we uncovered for the Los Angeles region. In Table 8, therefore, we do not undertake a detailed examination of each factor for each CBSA, although that may be a fruitful line of investigation for further research. For our purposes, it suffices simply to run the ANOVA test on all factors for each CBSA, recognizing that there may be a different number of factors for each metropolitan region. The results are remarkable. Out of a total of 149 distinct factors distributed across 20 metropolitan regions, only 19 (30) failed to reject the null hypothesis at the ten (one) percent level. For half of these 20 CBSAs, every single F-statistic was significant. In contrast, three metropolitan regions had three or more factors for which the null hypothesis was not rejected: Philadelphia, Atlanta and St. Louis¹⁰. What makes these metropolitan regions anomalous in this regard is not clear. What is clear, however, is that “Metropolis USA” is awash in Tieboutian club sorting at the municipal level, and this sorting process applies robustly to a full range of club—and club member—characteristics. We are, ourselves, in very large measure, the local “goods” that we seek. We embody the municipalities we reside in.

Table 9 ANOVA results for twenty largest CBSAs

9 Conclusions

The research presented here leads to several important findings regarding the nature of Tieboutian clubs, their persistence over many decades, and their ubiquitous presence at a national scale. We find four defining characteristics of the socioeconomic landscape of “Metropolis USA”; these pertain to economic status, race & ethnicity, age cohort, and immigration status. Of course, the emergence of these factors is largely reflective of what data are collected by the Census Bureau. If we collect data on race or age, for example, one should not be surprised to find that race and age should emerge as defining characteristics of the data set. The significant aspect of these emergent factors is that they point to persistent patterns of correlation within the original set of forty-nine variables. Factor analysis seeks out redundancies in the original data set and constructs factors that represent these principal tangles of intercorrelation. So, when we find a persistent factor such as “Professional elite”, it is not just telling us that this characteristic is present. It is telling us that a host of variables with different names are in fact differing aspects of this same factor—variations on a theme, as it were. A very complex socioeconomic landscape of forty-nine variables is summarized quite efficiently in our case with just a handful of factors. Four of these factors are especially important because of their persistence over many decades and across a vast metropolitan terrain.

Much of the empirical story is told at the census tract level, as those are the units of analysis for the factor analyses. This level of granularity matches recent work by other scholars analyzing neighborhood change, and our factor analytic findings resonate with theirs, notwithstanding the somewhat different methodological approaches. A key difference, however, is our focus on the Tiebout sorting process. A first step is to assess whether the “club” characteristics (as measured by factor scores) are rooted to the same neighborhoods (census tracts) over large spans of time. We do so by examining the correlation between like factors in the Los Angeles metropolitan region over a span of decades: 1970, 1990, and 2020. In almost all cases, we find that if a given municipality had high/low average factor scores for a given factor in 1970, it did so also in 1990 and 2020. This confirms that club characteristics not only persisted over time, but that they were also tightly rooted to their respective geographic places. The second step, then, was to assess whether similar neighborhoods tended to group together within municipal boundaries. This was confirmed convincingly through the analysis of variance applied to each factor from each of twenty of the largest metropolitan regions (CBSAs) in the USA. The overwhelming evidence is that variance of factor scores between census tracts within municipalities was significantly lower than the variance between municipalities. Thus, municipalities are Tieboutian clubs, and Tieboutian clubs tend to form at the municipal level. The evidence strongly supports the conclusion that this condition holds over time and across metropolitan regions.

This strong nexus between municipalities and Tieboutian clubs is significant in several aspects. First, and perhaps most importantly, it means that municipalities matter. Households are indeed voting with their feet; they do so to seek out clubs with certain definitive characteristics, and they co-locate with reference to municipal boundaries. Moreover, while ours is not a causal model, it is clear that we, the People, who reside in these clubs and who are represented by these data, are a strong causal factor. Our behaviors are both the cause and the effect. Our results thus provide meaning and applicability that extend well beyond what Tiebout likely had envisioned. It does, however, also resonate with the seminal works of Buchanan and of Schelling in a very meaningful way. While we have only begun to scratch the surface here, our results suggest that there may be a deeper, more comprehensive model—yet to be articulated—of urban evolution in contemporary American metropolitan regions that retains the essential elements of Tiebout, Buchanan and Schelling¹¹, while bringing about something meaningful and distinctive through the synthesis.

Finally, it is useful to reflect on the implications of these findings for public policy. One important implication is that if municipalities matter in this context, it is reasonable to expect that municipal management also matters. Decades of research geared to the original Tiebout sorting mechanism has focused primarily on the local public finance and land economy dimensions. The results presented here suggest that there is ample scope for broadening that focus to address other club characteristics more systematically. Do Tieboutian clubs align with municipal boundaries in response to—or in spite of—local government policies? One might also inquire into the efficiency and equity implications of the results presented here. To what extent does Tiebout sorting promote efficient production and allocation of local public goods when the latter are embodied in those who consume them? In a similar vein, how can we model Tiebout sorting between clubs when the very act of sorting alters the characteristics of those clubs? With reference to the supply side, given that metropolitan regions are the essential venues for Tiebout sorting, is there a role for metropolitan planning organizations to promote more equitable and efficient arrays of municipal clubs? While the research presented here does provide useful insights into the nature and extent of Tieboutian clubs, it raises many additional questions. Just like Tieboutian clubs themselves, the relevance of Tiebout’s work is extensive and enduring.