Infrastructure-Led Development: The Reciprocal Asymmetry of the Urban Growth Machine

Abstract

Road construction is a catalyst of land-use change. In this analysis, drawing on the growth machine framework, the investigators assess whether there is any reciprocal feedback in this process, specifically asking if there is a bidirectional association between road construction and land development. To answer that question, the investigators use recently released satellite imagery from the National Land Cover Database’s Impervious Descriptor dataset. The analysis covers the years 2010 to 2024, during a period of sustained growth for the 3,203 census tracts in the four major metropolitan areas of the “Texas Triangle.” The investigators compare results from two types of longitudinal models using first differences: cross-lagged structural equation models and spatial regression models. Results from these models reveal that the impact of road construction on land development is proportionately greater than the reciprocal impact of land development on road construction, providing an example of an asymmetric reciprocal socioecological association. From the growth machine perspective, the investigators argue that road construction and land development do not function as equal, twin engines of growth but rather as primary and subordinate catalysts of urban expansion, implying an infrastructure-led strategy of the growth coalition elite.

Keywords

growth machine land development road construction reciprocal association

In this study, we set out to test whether there is an asymmetric reciprocal association between road construction and land development across the 3,203 census tracts in the four major metropolitan areas of Texas, covering the years 2010 to 2024, during a period of sustained urban growth in the state. Over this 15-year period, the population of the metropolitan areas of Austin, Dallas–Fort Worth, Houston, and San Antonio grew by more than 30 percent, a growth rate greater than four times the national increase of approximately 8 percent during the same time frame. These four metropolitan areas are part of an emerging macro-metropolitan system, sometimes referred to as the “Texas Triangle” (Fulton et al. 2021). This region represents a unique experience in the recent history of urban growth, with projections indicating that it will soon be home to four of the nation’s 10 most populous cities, with a collective economic output that would rank as the fifteenth largest global economy (Fulton et al. 2021). Meanwhile, Texas’s metropolitan growth has been facilitated by relatively permissive land-use regulations and state-level policies that enable aggressive land development (e.g., Brannan 2010; Martinez 2025). Yet this high-magnitude growth has resulted in various urban infrastructure challenges; for instance, the pace of road construction has not been able to keep up with the demand that arises from aggressive land development, resulting in more traffic congestion and increased accidents (Barfield 2022; Saginor et al. 2011). According to the Texas Department of Transportation (2024a), between 1975 and 2015, the state’s number of registered vehicles grew by 172 percent, but its highway capacity (in lane miles) grew by only 19 percent. And, this infrastructure challenge continues into the present, particularly for the “Texas Triangle”; as of 2024, according to a standardized ranking of 2,100 congested roadways across the state, measured in terms of delay per mile, “94 of the top 100 . . . are in Austin, Dallas-Fort Worth, Houston and San Antonio” (TxDOT 2024b).

Thus, in Texas’s major metropolitan areas, because of all this growth, there has been a mismatch between infrastructure capacity and the demand for infrastructure. In this study, using recently released land cover data from satellite imagery (USGS 2024), we draw from growth machine theory to assess this mismatch, asking whether there is an asymmetric association between road construction and land development. As discussed later, growth machine theory provides a comprehensive framework to study not only whether this association is asymmetric but also whether it is reciprocal. Thus, we ask whether the association between these two socioecological dynamics of urban growth is bidirectional and asymmetric, such that the impact of road construction on land development is proportionately unequal to the reciprocal impact of land development on road construction. These questions draw from theoretical and empirical research in the growth machine literature. Indeed, for empirical studies of the growth machine, Logan, Whaley, and Crowder (1997) emphasized the importance of “studying effects over time, and estimating causal reciprocal relationships” (p. 624). We take up this call, following in the path of more recent empirical research on the topic using satellite imagery on land cover (e.g., Clement et al. 2023; see also Aljoufie et al. 2011; Tong and Qiu 2020). For example, Clement et al. (2023) examined the reciprocal association between core features of the growth machine, including land development. Yet although Clement et al. (2023) claimed that the reciprocal association was asymmetric, they did not report the results of any statistical techniques to test for asymmetry, for example, using a likelihood ratio test to compare unconstrained versus equality-constrained structural equation models (e.g., Ambrona and López-Pérez 2014; Kenny 1979).

For this study, drawing from the reciprocal dynamics of the growth machine theory, we conduct formal, statistical tests of the mismatch between road construction and land development, the results of which would signal whether infrastructure-led development, as a core dynamic of the growth machine model, is unbalanced or asymmetric. In this context, we assess whether there is a bidirectional association between road construction and land development and whether this association is asymmetric. By evaluating the potential for an asymmetric reciprocal association between these two key socioecological dimensions of urban growth, and by identifying which of these forces might exert a dominant influence on the other, we can help clarify the degree of symmetry or asymmetry in the operation of the growth machine. Moreover, from the growth machine perspective, given the significant roles played by private real estate interests as well as the state and local governments in Texas, testing whether this association is symmetric or asymmetric can also help clarify the functional relationship between public and private actors of the growth coalition, in terms of their support for road construction and land development. Specifically, this analysis can provide evidence about whether public subsidies and private investment function equally as twin engines of growth, or whether one serves as the primary catalyst that triggers urban expansion.

Theoretical Framework

In sociology, growth machine theory offers a multipurpose sociological lens for addressing the socioecological dimensions of community change, from population growth and changes in unemployment to the physical reconfiguration of the landscape (Logan and Molotch 2007; Molotch 1976). With the notion that growth is a “syndrome of associated events” (Molotch 1976:310), the growth machine draws from the human ecological tradition (Molotch 1967), adopting the view that the socioecological dimensions of community change are connected through interdependent, reciprocal processes (see also Elliott and Frickel 2015). To this human ecological idea, the growth machine adds a political economy critique, claiming that “any given piece of real estate has both a use value and an exchange value” (Logan and Molotch 2007:1). Yet it is the pursuit of exchange value that serves as the primary motive of urban growth; indeed, as Logan and Molotch argued, urban growth functions as a mechanism for increasing real estate value, an objective that unites a broad coalition of stakeholders. The exchange value potential of real estate generates a widely shared interest in land-use intensification (e.g., Gerber and Phillips 2003); all the same, despite the broad support it may receive, the economic gains of land development remain concentrated among the wealthy and powerful of the growth coalition elite (e.g., Dye and McMillen 2007; Guerrieri, Hartley, and Hurst 2013). In this light, land development, and the way it economically benefits real estate developers and wealthy residents, represents a core dynamic in the growth machine perspective, making it a fundamental, interdependent part of urban growth.

Despite the theory’s emphasis on the physical intensification of land, foundational empirical work by Molotch (1976) and subsequent quantitative research often relied on more accessible demographic and economic measures, such as population growth or unemployment (e.g., Fodor 2012). Although scholars produced insightful growth machine studies on land use, this research tended to focus on the demographic, economic, and political correlates of land development, rather than relying on direct measures of land cover (e.g., Logan and Zhou 1989; Warner and Molotch 2000). To be sure, qualitative sociologists have provided detailed historical accounts of landscape transformation from the growth machine perspective (Rudel 2009). Nevertheless, of relevance to the present analysis, the expanding availability of high-resolution satellite imagery has enabled quantitative sociologists to develop concrete measures of land cover, which has contributed to a flourishing body of research for testing growth machine propositions on landscape transformation (Clement and Podowski 2013; Ergas, Clement, and McGee 2016; Landis 2017; Pravin 2019; Smiley and Hakkenberg 2020; Steele and Luloff 2003; Wongso and Göçmen 2025). Meanwhile, scholars have long pointed out that road construction is a primary catalyst of land development; the construction of roads penetrates undeveloped areas, enabling the flow of human activity and facilitating the transformation of this land (e.g., Bernier and McGlinchy 2023; Greenberg, Wishart, and Danielsen 2024; Rudel 2009). Indeed, Logan and Molotch (2007) argued that “transportation doesn’t just serve growth, it creates it” (p. 74). As such, viewed through the lens of the growth machine, road construction can be framed as the catalytic mechanism for the “increasing intensification of the land use of the area” (Molotch 1976:309), signaling its role in enhancing exchange value by fostering land development. Yet in the growth machine framework, land development serves a dual purpose: it maximizes exchange value for private real estate interests while simultaneously providing the state with the fiscal capacity to fund further road construction, which effectively socializes the costs of the infrastructure required for continued urban expansion (e.g., Harvey 1982). Indeed, through various channels, such as local taxes, developer impacts fees, and tax increment financing (TIF), more land development generates more public revenue, which is dedicated specifically for more road construction (e.g., Briffault 2010; Jeong and Feiock 2006; Kirkpatrick 2016; Saginor et al. 2011; Sroka 2024). In this way, although road construction is a primary catalyst of land-use change, the subsequent land development then reciprocally funds new road construction, thereby reigniting the catalyst of urban growth.

Yet as mentioned in the introduction, the continued challenges of congestion and traffic accidents in Texas’s metropolitan areas suggest an imbalance or asymmetry in the association between road construction and land development, an observation that informs our research questions, which we evaluate in the following analysis.

To develop a testable idea of an unbalanced bidirectional relationship, we use the framework of asymmetry as established in the quantitative social science literature (e.g., Allison 2019; Lieberson 1985; York and Light 2017). In examining the link between road construction and land development, we specifically test for reciprocal asymmetry, which we treat as a distinct concept from directional asymmetry. Indeed, the notion of reciprocal asymmetry (the impact of X on Y ≠ the impact of Y on X) is related to the notion of directional asymmetry (the impact of ↑X on Y ≠ the impact of ↓X on Y), but they ultimately offer different insights. On one hand, directional asymmetry not only can be helpful in identifying differences in the magnitude of impact, but it can also reveal how the mechanism through which ↑X affects Y is different than ↓X affecting Y. On the other hand, the concept of reciprocal asymmetry is primarily helpful in identifying differences in the magnitude of impact rather than revealing new information about the mechanisms of influence (e.g., Ambrona and López-Pérez 2014).

To demonstrate this idea for our study, we will use the X variable for road construction and the Y variable for land development. So, with reciprocal asymmetry, because road construction (X) and land development (Y) are different variables, the mechanism through which X affects Y is plainly different from the mechanism through which Y affects X. Thus, a test of reciprocal asymmetry will not reveal any new information about the mechanisms through which X and Y influence each other; but, with the appropriate statistical techniques, an assessment of reciprocal asymmetry reveals differences in the proportional influences of these two variables on each other. In this context, reciprocal asymmetry points to an imbalance in the bidirectional system of change, indicating that one variable acts as a primary stimulus of change while the other variable is still a significant yet subordinate force of feedback.

On that note, if the interdependent factors within the growth machine model show reciprocal asymmetry, it means their proportional influences on each other are unequal, reflecting an overall imbalance in the operating forces of urban growth. Specifically, with respect to the association between road construction and land development, reciprocal asymmetry would point to which variable is the primary instrument and which variable is the subordinate feedback force in catalyzing local, urban growth. Such a finding can contribute to a more nuanced discussion about the theoretical and practical implications of urban growth as well as alternatives to the current form of development in the United States, especially given its car-centric history. For instance, if there is an asymmetric reciprocal relationship between road construction and land development, then we argue that road construction serves as a primary instrument of land-use change, but land development still operates as a significant, yet secondary, force of feedback. This reciprocal asymmetry would point to a systemic pattern where road construction stimulates infrastructure-led land development, which then in turn subsequently feeds back into, albeit at a lower magnitude, more road construction, maintaining an unbalanced system of car-centric urban growth.

In other words, reciprocal asymmetry, consistent with the growth machine perspective, would help clarify the state’s role in the growth coalition. Through state support, road construction acts as a lead but interdependent socioecological force that contributes to subsequent private development, which then reciprocally contributes to road construction as a form of urban infrastructure that is designed around the private automobile (e.g., Angelo and Wachsmuth 2020; see also Harvey 1982). By providing the infrastructure that precedes private investment, the state does more than facilitate urban growth; it effectively helps socialize the initial costs and risks of urban expansion. If there is a reciprocal association between road construction and land development, then this would create a cycle where the public subsidizes the infrastructure which then ensures the intensification of private land development. In growth machine terms, if road construction is found to be a primary catalyst of expansion, conceptually, it is helping maximize exchange value and, theoretically, road construction emerges as a structural cornerstone of the growth machine perspective. Moreover, if urban growth is calibrated for infrastructure-led development, then a finding of asymmetry could also be seen as making a contribution to the vast literature describing how public transit has been marginalized in urban America (e.g., Lee and Sener 2017; Mattioli et al. 2020). With evidence of asymmetry, the reciprocal forces of urban growth could be described as structurally designed to favor projects that maximize exchange value through prioritizing private automobile access, instead of public transit, even if road construction projects, in terms of traffic congestion and accidents, come at a higher social and environmental cost than public transit (e.g., Warner and Molotch 2000).

To that end, we test for reciprocal asymmetry between road construction and land development across 3,203 census tracts in Texas’s major metropolitan areas, evaluating whether these dynamics function in a balanced feedback loop or in an unbalanced system where public infrastructure serves as the primary catalyst for private exchange value. In an unbalanced system, such a relationship would, by extension, imply that alternative development models are marginalized in favor of a car-centric landscape.

Data and Analytic Techniques

For our analysis, we focus on the 3,203 census tracts in the four major metropolitan areas of Texas: Austin, Dallas–Fort Worth, Houston, and San Antonio. As mentioned, the study covers the years 2010 to 2024, during a period of sustained growth in the state of Texas. To account for the multiyear construction time frame required to complete infrastructure projects (Glaeser and Poterba 2021), we use four waves of data from 2010, 2015, 2019, and 2024; furthermore, these four- and five-year intervals generally align with the windows for major infrastructure planning in Texas (e.g., NCTCOG 2022). Variable names, descriptions, and sources are provided in Table 1.

Table 1.

Variable Names, Descriptions, and Sources (n = 3,203 Census Tracts).

Variable Name	Short Name	Description	Years	Notes	Source
Road construction	Road	The percentage of the census tract’s total area covered by all road types. All road types include primary roads (e.g., interstates), secondary roads (e.g., state highways), tertiary roads (e.g., residential streets), and thinned roads (e.g., unpaved rural roads).	2010, 2015, 2019, 2024	We used 2010 census tract shapefile boundaries and the Tabulate Area tool in ArcGIS. For the analyses, we calculate the first differences of these values. In Table 2, with the univariate statistics, we report both the raw levels and the specific first differences used in the analyses.	National Land Cover Database’s Impervious Descriptor dataset (USGS 2024)
Land development	Land	The percentage of the census tract’s total area covered by all nonroad, human-built impervious surfaces, which includes, for example, residential structures, service and commercial centers, and industrial facilities.	2010, 2015, 2019, 2024	We used 2010 census tract shapefile boundaries and the Tabulate Area tool in ArcGIS. For the analyses, we calculate the first differences of these values. In Table 2, with the univariate statistics, we report both the raw levels and the specific first differences used in the analyses.	National Land Cover Database’s Impervious Descriptor dataset (USGS 2024)
Control variables
Population density	Density	The total resident population of the census tract divided by the total area of the tract (in 1,000 m²).	2010, 2015, 2019	To calculate the denominator for total land area, we use USGS (2024). For the analyses, we compute the natural logarithm of each tract’s density, for each year (after adding 1), then we calculate the first differences of these logged values. In Table 2, with the univariate statistics, we report both the raw levels and the specific first differences used in the analyses.	American Community Survey; National Land Cover Database’s Impervious Descriptor dataset (USGS 2024)
Percentage white	White	The percentage of the census tract’s total resident population that is non-Hispanic white.	2010, 2015, 2019	For the analyses, we calculate the first differences of these values. In Table 2, with the univariate statistics, we report both the raw levels and the specific first differences used in the analyses.	American Community Survey
Median household income	Income	The median income (adjusted for inflation, in 2010 constant dollars) of all households in the census tract.	2010, 2015, 2019	For the analyses, we compute the natural logarithm of the income for each census tract, for each year (after adding 1), then we calculate the first differences of these logged values. In Table 2, with the univariate statistics, we report both the raw levels and the specific first differences used in the analyses.	American Community Survey

The data for the road construction and land development variables come from the National Land Cover Database’s Impervious Descriptor satellite imagery (USGS 2024). With a tract-level shapefile, based on 2010 census boundaries, we use the Tabulate Area tool in ArcGIS to quantify, for each census tract, and for each year 2010, 2015, 2019, and 2024, the area (in square meters) covered by roads and the area (in square meters) covered by all other nonroad, human-built impervious surfaces, including residential structures, commercial centers, industrial facilities, and the like. For the analysis, both variables are turned into percentages according to the total area (in square meters) of the census tract; the road construction variable equals the percentage of the total area covered by all road types, which includes everything from unpaved rural roads to residential streets to interstate freeways.¹ The land development variable equals the percentage of the total area covered by all nonroad, human-built impervious surfaces. The use of percentages for the analysis standardizes the slope estimates, which indicate the percentage point change in the dependent variable for every 1 percentage point change in the predictor variable, holding the rest of the equation constant.²

Drawing from previous social science research on land-use change using satellite imagery (e.g., Clement et al. 2023; Smiley and Hakkenberg 2020), we also include three variables to control for local demographics and socioeconomic status: population density, percentage white, and median household income. Population density is the total resident population of the census tract divided by the tract’s total land area (in 1,000 m²); percentage white is the percentage of the tract’s population that is non-Hispanic white; and median household income is the middle income of all households in the census tract, adjusted for inflation, in 2010 constant dollars. Before computing first differences, we calculate the natural logarithm for population density and median household income after adding a constant of 1.³ These variables are used as controls to assess whether their impact may confound the reciprocal association between road construction and land development; on that note, for the control variables, we use only data from 2010, 2015, and 2019 as the lagged predictors of road construction and land development.

To test whether there is an asymmetric reciprocal association between road construction and land development, we use two types of modeling techniques: structural equation models and spatial regression models. For both, we estimate models using two waves of first-differenced values from the four time periods: 2010, 2015, 2019, and 2024; because the variables are percentages, these first differences are interpreted as percentage point changes. Using lagged first differences from four times periods of data yields three waves of first differences: Δ2010–2015, Δ2015–2019, Δ2019–2024, from which we can discern whether slope estimates change over time. All structural equation and spatial regression models are estimated using the robust variance-covariance estimator to account for potential heteroskedasticity (the “robust” vce option in Stata) (White 1980).

For the structural equation models, we use a cross-lagged structure with autoregressive parameters. See Figure 1, which does not include the lagged predictors for the control variables. The cross-lagged associations are between road construction and land development: ΔRoad_t−1 → ΔLand_t and ΔLand_t−1 → ΔRoad_t. The series of cross-lagged models we estimate involve comparing results from an unconstrained model, where the cross-lagged estimates are free to vary, and an equality-constrained model, where the cross-lagged estimates are set to be equal. Because the only parameters that are different between the two nested models (i.e., unconstrained and constrained models) are the cross-lagged slopes, then comparing their model fits with a likelihood ratio test would tell us whether we should treat the cross-lagged effects as equal or unequal; if the likelihood ratio test is not significant, then the model with equal slopes exhibits better fit, which provides evidence that the reciprocal association is symmetric; if the likelihood ratio test is significant, then the model with unequal slopes exhibits better fit, which provides evidence that the reciprocal association is asymmetric (e.g., Ambrona and López-Pérez 2014; Kenny 1979).

Figure 1.

Cross-lagged first-differenced structural equation model.

Covering the years 2010, 2015, 2019, and 2024, using two waves of first-differenced values, we estimate two models: an unconstrained versus a constrained model. Following are the generic equations for these models.

Unconstrained structural equation model:

Δ Y_{i t} = β_{1} Δ Y_{i, t - 1} + β_{2} Δ X_{i, t - 1} + u *_{i t}

Δ X_{i t} = β_{3} Δ X_{i, t - 1} + β_{4} Δ Y_{i, t - 1} + v *_{i t}

u *_{i t} = u_{i t} - u_{i, t - 1}

v *_{i t} = v_{i t} - v_{i, t - 1}

with:

C o v (u *_{i t}, v *_{i t}) = σ_{u v}

Equality-constrained structural equation model:

Δ Y_{i t} = β_{1} Δ Y_{i, t - 1} + β_{2 e q u a l} Δ X_{i, t - 1} + u *_{i t}

Δ X_{i t} = β_{3} Δ X_{i, t - 1} + β_{2 e q u a l} Δ Y_{i, t - 1} + v *_{i t}

u *_{i t} = u_{i t} - u_{i, t - 1}

v *_{i t} = v_{i t} - v_{i, t - 1}

with:

C o v (u *_{i t}, v *_{i t}) = σ_{u v}

In these equations, ΔY_it and ΔX_it are the first-differenced values of Y and X, and ΔY_i,_t−1 and ΔX_i,_t−1 are the lagged first-differenced values of Y and X. β₁ and β₃ are the autoregressive slope estimates. In the unconstrained model, β₂ and β₄ are the cross-lagged slope estimates for ΔY_i,_t−1 and ΔX_i,_t−1, which are allowed to vary; in the equality-constrained model, the cross-lagged slope estimates β_2equal are made equal. The terms u*_it and v*_it capture the unexplained variation in ΔY_it and ΔX_it. The term σ_uv is the covariance between the unexplained residuals in ΔY_it and ΔX_it.

For the spatial regression analysis, our analysis begins with a temporally lagged first-differenced model structure, on the basis of the approach recommended by Allison (2009). Allison (2009) demonstrated that with three time periods of data, two separate ordinary least squares models of lagged first differences can be run, effectively creating a reciprocal association fixed-effects model to assess bidirectional relationships. We applied this strategy using first-differenced data from three time periods covering the years 2010, 2015, 2019, and 2024, which yielded two models where road construction is the dependent variable (and land development is the independent variable) and two models where land development is the dependent variable (and road construction is the independent variable), giving us a total of four cross-lagged models. We then added a spatial lag term ρ and spatial error term λ to these reciprocal association fixed-effects models. With both ρ and λ, the model is a spatial autoregressive model with spatial autoregressive disturbances or a spatial autoregressive combined model (see Anselin and Florax 1995). The generic equations for these models are as follows:

Spatial regression models:

Δ Y_{i t} = β_{1} Δ X_{i, t - 1} + β_{2} Δ Y_{i, t - 1} + ρ_{1} (W Δ Y_{i t}) + m_{i t}

m_{i t} = λ_{1} (W_{m i, t - 1}) + Δ ζ_{i t}

Δ X_{i t} = β_{3} Δ Y_{i, t - 1} + β_{4} Δ X_{i, t - 1} + ρ_{2} (W Δ X_{i t}) + n_{i t}

n_{i t} = λ_{2} (W_{n i, t - 1}) + Δ η_{i t}

In these equations, β₁ and β₂ represent the (temporally) lagged slope estimates; W is a row-standardized, queen first-order contiguity spatial weights matrix that is the same for all t; ρ₁ and ρ₂ are the spatial lag parameters; and m_it and n_it are the unexplained variances for ΔY_it and ΔX_it and have two parts: the spatial error terms λ₁ and λ₂ and the remaining unexplained variation Δζ_it and Δη_it.

Results

Table 2 reports univariate statistics for 2010, 2015, 2019, and 2024 for the road construction and land development variables, and for 2010, 2015, and 2019 for the control variables. We use the first differences for the analysis. Looking at the first differences, between 2010, 2015, 2019, and 2024, there were consistent increases for both road construction and land development (all first differences were positive; p < 0.001, paired t tests). In terms of the control variables, the results show inconsistent variation over time. On one hand, population density, on average, increased between 2010, 2015, and 2019 (as log differences, these numbers are interpreted, roughly, as 4.1 percent and 2.7 percent increases in population density). On the other hand, the typical tract became more white and less affluent between 2010 and 2015; but between 2015 and 2019, we observe the opposite: the typical tract became less white and more affluent.

Table 2.

Univariate Statistics (n = 3,203 Census Tracts).

	Road		Land		Density		White		Income ($)
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Raw levels
2010	40.442	20.014	36.567	18.205	1.456	1.358	67.740	21.291	59,659.490	31,980.980
2015	40.966	19.850	37.487	18.019	1.565	1.438	70.639	20.008	57,889.390	31,019.430
2019	41.322	19.713	38.331	17.843	1.637	1.484	69.955	20.137	61,436.330	32,100.920
2024	41.761	19.713	39.456	17.457
First differences
Δ2010–2015	.524***	1.165	.920***	2.177	.040***	.088	.065***	.263	−.033***	.242
Δ2015–2019	.356***	.811	.844***	1.849	.027***	.027	−.684***	8.860	.060***	.322
Δ2019–2024	.439***	.541	1.125***	2.799

Note. The values for population density and median household income were log transformed before calculating the first differences.

***

p < .001 (one-tailed paired t tests).

To assess whether road construction and land development are reciprocally associated, we begin with the first-differenced cross-lagged structural equation modeling results in Tables 3 and 4. The unconstrained results are also displayed in Figure 1, which does not display the control variables.

Table 3.

Unconstrained and Constrained Cross-Lagged Structural Equation Models (n = 3,2023 Census Tracts).

			Unconstrained		Constrained
Endogenous Outcome	Lagged Predictor		Estimate	Robust SE	Estimate	Robust SE
ΔLand (2015–2019)	Cross-lagged	ΔRoad (2010–2015)	.159**	.057	.041***	.011
	Autoregressive	ΔLand (2010–2015)	.454***	.031	.496***	.027
	Control variable	ΔDensity (2010–2015)	.323	.300	.507^†	.299
	Control variable	ΔWhite (2010–2015)	−.203**	.077	−.212**	.078
	Control variable	ΔIncome (2010–2015)	.029	.073	.049	.074
		Constant	.344***	.028	.361***	.026
ΔRoad (2015–2019)	Cross-lagged	ΔLand (2010–2015)	.026*	.012	.041***	.011
	Autoregressive	ΔRoad (2010–2015)	.364***	.035	.332***	.029
	Control variable	ΔDensity (2010–2015)	.036	.133	.074	.131
	Control variable	ΔWhite (2010–2015)	−.04	.033	−.041	.033
	Control variable	ΔIncome (2010–2015)	−.023	.034	−.018	.033
		Constant	.141***	.013	.144***	.013
ΔLand (2019–2024)	Cross-lagged	ΔRoad (2015–2019)	.368*	.152	.001	.009
	Autoregressive	ΔLand (2015–2019)	.86***	.068	.956***	.059
	Control variable	ΔDensity (2015–2019)	−.524	.558	−.138	.571
	Control variable	ΔWhite (2015–2019)	−.003	.004	−.005	.004
	Control variable	ΔIncome (2015–2019)	−.08	.102	−.054	.1
		Constant	.285***	.044	.322***	.04
ΔRoad (2019–2024)	Cross-lagged	ΔLand (2015–2019)	−.004	.009	.001	.009
	Autoregressive	ΔRoad (2015–2019)	.267***	.041	.254***	.042
	Control variable	ΔDensity (2015–2019)	−.255^†	.153	−.254^†	.153
	Control variable	ΔWhite (2015–2019)	.002^†	.001	.002^†	.001
	Control variable	ΔIncome (2015–2019)	.051*	.023	.053*	.023
		Constant	.353***	.011	.353***	.011
ΔDensity (2015–2019)	Autoregressive	ΔDensity (2010–2015)	.045^†	.023	.045^†	.023
ΔDensity (2015–2019)		Constant	.025***	.001	.025***	.001
ΔWhite (2015–2019)	Autoregressive	ΔWhite (2010–2015)	−3.923*	1.961	−3.923*	1.961
ΔWhite (2015–2019)		Constant	−.429*	.178	−.429*	.178
ΔIncome (2015–2019)	Autoregressive	ΔIncome (2010–2015)	−.189^†	.107	−.189^†	.107
ΔIncome (2015–2019)		Constant	.054***	.007	.054***	.007

†

p < .10, *p < .05, **p < .01, and ***p < .001 (two-tailed tests).

Table 4.

Cross-Lagged Structural Equation Model Fit Indices and Asymmetry Test (n = 3,2023 Census Tracts).

		Unconstrained		Constrained
Model Fit Index		Estimate	Robust SE	Estimate	Robust SE
χ²		169.662***		226.286***
RMSEA		.051		.057
AIC		83,591.208		83,643.832
BIC		83,791.579		83,832.059
CFI		.975		.967
TLI		.947		.935
SRMR		.033		.036
CD		.522		.520
Asymmetry tests
Likelihood ratio test	χ² statistic	56.620***
Slope estimate equality test
β_Road − β_Land (Δ2010-2019)	z statistic	4.403***
β_Road − β_Land (Δ2015-2019)	z statistic	6.141***
Spatial autocorrelation of the residuals
Road (Δ2015–2019)	ρ	.381***	.025	.384***	.025
Land (Δ2015–2019)	ρ	.334***	.026	.333***	.026
Road (Δ2019–2024)	ρ	.374***	.025	.379***	.025
Land (Δ2019–2024)	ρ	.389***	.025	.405***	.025

Note. The model fit indices are derived from structural equation models that do not include population density as an additional control. Although adding population density as a control does not change the substantive findings of the analysis, including the results of the asymmetry tests, it does seriously degrade model fit in terms of the TLI, which decreases from 0.947 to 0.838 after including population density. AIC = Akaike information criterion; BIC = Bayesian information criterion; CD = coefficient of determination; CFI = confirmatory fit index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; TLI = Tucker-Lewis index.

***

p < .001 (two-tailed tests for slope equality and residual spatial autocorrelation).

Using the 2015–2019 interval for the endogenous outcomes, the significant slope estimates from both the unconstrained and equality-constrained models show that there was a positive feedback loop between road construction and land development. Because the variables are percentages, the slope estimates are interpreted as a percentage point change; for instance, in Table 3, looking at model 1, across the 3,203 census tracts, for every 1 percentage point increase in the rate of change of road construction between 2010 and 2015, the rate of change of land development between 2015 and 2019 increased by 0.159 percentage points (p < .01), all else being equal (including the prior rate of change of land development, which is based on the significant slope estimate for the land development parameter in Table 4). Likewise, for every 1 percentage point increase in the rate of change of land development between 2010 and 2015, the rate of change of road construction between 2015 and 2019 increased by 0.026 percentage points (p < .05), all else being equal (including the prior rate of change of road construction).

However, using the 2019–2024 interval for the endogenous outcomes, there was not a significant bidirectional association between road construction and land development. In fact, only the unidirectional impact of road construction on land development was significant (b = 0.368, p < .05).

For the unconstrained model in Table 3, the slope estimates for road construction were larger than the slope estimates for land development. Thus, in Table 4, to assess whether these differences were significant and whether there is evidence of asymmetry, we ran two types of tests: a likelihood ratio test to check for differences in model fit, and two z tests to check if the differences between the slope estimates were significant for both first-differenced waves. The results for both the likelihood ratio test and the two z tests were statistically significant (p < .001). Taken together with the slope estimates, these results highlight three points for our study. First, that allowing the cross-lagged estimates to vary freely significantly improved model fit, providing evidence of asymmetry. Second, the slope estimates for road construction were significantly greater than the slope estimates for land development; moreover, this difference in slope estimates, on average, increased between the 2010–2015 and 2015–2019 first-differenced waves; this is evidence that the asymmetry became more pronounced over time. Third, according to the structural equation modeling results, the relationship between road construction and land development was asymmetric yet still reciprocal between 2010 and 2019; however, between 2015 and 2024, there was no bidirectional association between these two variables. This suggests a shift in the growth dynamic, where the early interdependence between these variables evolved into a more linear process in the most recent time interval.

Looking at Table 4, although the traditional structural equation model fit indices (i.e., χ², root mean square error of approximation, Akaike information criterion, Bayesian information criterion, Tucker-Lewis index, confirmatory fit index, standardized root mean square residual, and coefficient of determination) all show that the unconstrained model exhibits better fit than the constrained model,⁴ we also test whether the residuals from these models are spatially autocorrelated. If they are, this spatial dependence might be biasing slope estimates, suggesting the need for a spatial regression approach. At the bottom of Table 4, we report these results; we find that the residuals from the primary endogenous outcome variables (i.e., road construction and land development) do exhibit highly significant, positive spatial autocorrelation. Figure 2 displays a map of the residuals from the unconstrained model when land development Δ2015–2019 is the endogenous outcome. We do this only for Harris County, given that it would become difficult to see tract boundaries if we presented residuals for all census tracts in all the counties across all four metropolitan areas. In this map, consistent with the estimate for the spatial autocorrelation of the residuals (ρ = 0.334, p < .001), we observe a discernible spatial pattern in the clustering of the residuals. Although some tracts outside the urban core did exhibit lower than predicted land development, the spatial clustering of higher residuals outside the urban core reflects extra development the model could not account for, which is likely the above-average pace of land development characteristic of suburban sprawl. On that note, to check whether this spatial dependency might be biasing the slope estimates, we next ran a series of cross-lagged spatial regression models.

Figure 2.

Harris County residuals from the unconstrained structural equation model in Table 3, with land development (2015–2019) as the endogenous outcome.

Table 5 reports the results from these spatial regression models. We first note that the estimates for ρ and λ are significant in all models (at least p < .05), and the residuals from these models no longer exhibit significant spatial autocorrelation. Again, only for Harris County, Figure 3 displays the residuals from model 1 of Table 5, in which land development Δ2015–2019 is the dependent variable. Now, in this spatial regression model, we do not observe a discernible spatial clustering in the residuals; instead, we see more randomly distributed residuals. Thus, although suburban sprawl is occurring, the inclusion of the spatial parameters successfully minimized any potential bias in the slope estimates (of the predictor variables) that stems from such spatial clustering processes. Indeed, looking at the slope estimates in Table 5, we do observe changes in the slope estimates. First, in the spatial regression models, the cross-lagged slope estimates for both road construction and land development are now all positive and significant, which refines the results from the structural equation models and now suggests a significant reciprocal association for all three first-differenced waves: Δ2010–2015, Δ2015–2019, and Δ2019–2024. Second, in Table 5, the results from the Z-tests of the equality of slope estimates indicate that the estimates for road construction were significantly greater than the estimates for land development; yet this asymmetric association did not intensify over time like it did in the structural equation modeling analysis. Ultimately, these findings suggest that while the intensity of the imbalance persisted, the bidirectional link between road construction and land development is more enduring than the structural equation modeling analysis initially indicated.

Table 5.

Cross-Lagged Spatial Regression Results (n = 3,203 Census Tracts).

	Δ2015–2019					Δ2019–2024
		Model 1		Model 2			Model 3		Model 4
		ΔLand		ΔRoad			ΔLand		ΔRoad
Variable	Δ2010–2015	Estimate	Robust SE	Estimate	Robust SE	2015–2019	Estimate	Robust SE	Estimate	Robust SE
Cross-lagged	ΔRoad	.126***	.028			ΔRoad	.210***	.055
Cross-lagged	ΔLand			.012^†	.007	ΔLand			.023***	.006
Autoregressive	ΔRoad			.314***	.013	ΔRoad			.202***	.013
Autoregressive	ΔLand	.373***	.016			ΔLand	.643***	.025
Control variable	ΔDensity	.035	.287	−.044	.13	ΔDensity	−.478	.530	−.168	.12
Control variable	ΔWhite	−.089	.091	−.016	.041	ΔWhite	.000	.004	.000	.001
Control variable	ΔIncome	.057	.101	−.017	.046	ΔIncome	−.010	.112	.011	.024
Spatial lag	ρ	.394***	.032	.417***	.036	ρ	.509***	.025	−.523***	.06
Spatial error	λ	−.125*	.052	−.131*	.056	λ	−.230***	.048	.718***	.03
Constant		.083*	.033	.029^†	.015		−.100**	.037	.595***	.04
Spatial autocorrelation of the residuals	ρ	.004	.031	.009	.031	ρ	.004	.031	−.029	.031
Asymmetry test	z statistic	4.011***				z statistic	3.385***

Note. The slopes for the cross-lagged and control variables are the direct effects from the spatial regression model.

†

p < .10, *p < .05, **p < .01, and ***p < .001 (two-tailed tests).

Figure 3.

Harris County residuals from spatial regression model 1 in Table 5, with land development (2015–2019) as the dependent variable.

Before we discuss these results and make some concluding comments, we briefly report the results for the control variables. Across the structural equation and spatial regression models, the slope estimates for the control variables were not consistent. For the structural equation modeling analysis, we will focus on the unconstrained model because it exhibited significantly better fit than the constrained model. In the unconstrained structural equation model, an increase in the percentage of the white population was significantly associated with lower rates of land development in the earlier first-differenced wave, but then, in the later first-differenced wave, percentage white was positively associated with road construction, along with median household income. Meanwhile, population density was negatively associated with road construction in the later first-differenced wave in the unconstrained structural equation model. However, we caution making any conclusive generalizations about the control variables given that, in the spatial regression models, the slope estimates for the control variables were no longer significant.

Discussion and Conclusion

Drawing from growth machine theory, the foregoing analysis of 3,203 census tracts across the four major metropolitan areas of the “Texas Triangle” (Fulton et al. 2021) reveals that road construction and land development are characterized by a persistent asymmetric association. Although initial structural equation modeling results suggested a shift in the later waves toward a unidirectional relationship, the spatial regression models suggest that there is a persistent bidirectional link that is significant across the entire study period. Although only the spatial regression model shows that the reciprocal relationship persists across the entire study period, the association between these two socioecological processes is consistently asymmetric in both the structural equation modeling and spatial regression analyses, as the slope estimates for road construction significantly exceed those for land development for all first-differenced waves. Thus, the spatial regression results identify a feature of reciprocal asymmetry in the process of urban growth, which helps refine growth machine theory: while both road construction and land development are socioecological dynamics that function as twin engines of urban growth, these twins are not equal; rather, infrastructure provision acts as the primary catalyst of change. Ultimately, in describing these key socioecological dynamics of urban growth in the “Texas Triangle,” we argue that road construction dictated the pace of land development but in the presence of a subordinate yet significant feedback from land development.

To elaborate on these findings and connect them to the actual on-the-ground experience of urban growth in Texas, we display two sets of images (Figures 4 and 5) based on an analysis of the residuals from the spatial regression models 1 and 2. Again, focusing on Harris County as an example, we highlight two census tracts in that county, where the rates of land development and road construction were far higher than predicted on the basis of the equations from models 1 and 2. In Figure 4, we display imagery for census tract 48201343200, which is in southeast Harris County, containing much of the Fairmont Park neighborhood in La Porte. Although development in Fairmont Park began in the late 1950s, and the neighborhood was relatively built out by the 1980s, the area, like much of greater Houston, has continued to experience development. For example, this tract experienced a rate of land development between Δ2015–2019 that was 13.652 percentage points higher than predicted on the basis of the equation used to estimate model 1. Then, in Figure 5, we display imagery for census tract 48201241103, which is in north central Harris County, containing a portion of the Timber Lane area of Spring. Spurring the development of the Breckenridge Forest neighborhood, this tract experienced a rate of road construction between Δ2015–2019 that was 10.653 percentage points higher than predicted based on the equation used to estimate model 2. In Figure 5, we can see the sequence of infrastructure-led development, where prior road construction penetrates undeveloped land, opening it up for later development, which would be the Breckenridge Forest neighborhood.

Figure 4.

Comparison of land development over time in census tract 48201343200.

Figure 5.

Comparison of road construction over time in census tract 48201241103.

Before we offer some concluding thoughts on the theoretical and practical implications of the study, we highlight some limitations of the foregoing analysis, with an intention to stimulate future scholarship.

First, although the National Land Cover Database’s Impervious Descriptor satellite imagery distinguishes between road area and nonroad area, it does not distinguish between the different types of roads (e.g., unpaved rural roads, residential streets, highways) and the different types of land development (e.g., residential, commercial, industrial). As such, this study is not able to test whether the road construction-land development reciprocal association varies depending on the type of road or the type of land development. To be fair, although deriving both variables from a single satellite imagery dataset ensures consistency in how land cover categories are defined, future research might consider whether merging different data products is feasible.

Second, for the purpose of this study, which was focused on whether there was an asymmetric reciprocal association between two key socioecological processes of urban growth, the control variables were not treated as endogenous outcomes but rather simply as predictors of road construction and land development. For instance, as we discussed, adding population density to the SEM models, simply as an exogenous control variable, resulted in a deteriorated model fit, in terms of a lower Tucker-Lewis index; however, there is scholarship showing that both land development and road construction can reciprocally influence changes in population size (Baum-Snow 2007; Clement et al. 2023). Future research, in more elaborate models, may assess whether these key socioecological processes of urban growth feed back into the demographic and socioeconomic variables we examined here.

Third, although the spatial regression models successfully accounted for residual autocorrelation, the structural equation modeling results exhibited spatial clustering of high residual values in suburban tracts outside the urban core, which is a pattern reflecting the rapid pace of development characteristic of suburban sprawl. Consequently, given the highly significant, positive spatial lag estimates (ρ) for land development in the spatial regression models, all these results taken together suggest that this sprawling development is driven significantly by a spatial contagion process, where the intensification of land use in one suburban tract significantly increases the probability of development in adjacent tracts. Thus, although our spatial regression models successfully accounted for this contagion, future research might directly address this spatial process, perhaps asking how networks of growth coalition elites coordinate land development strategies across suburban tracts.

Fourth, our findings about census tracts in the major metropolitan areas of Texas are situated within a unique period of urban growth, between 2010 and 2024. Although this time frame provides a clear window into the operations of a high-magnitude form of urban growth, it may also reflect specific land use policies and economic conditions that are not universal. Future scholarship can explore whether and how the finding of reciprocal asymmetry may manifest in different political contexts or in areas of contraction or stagnation.

To conclude, we elaborate on the theoretical and practical implications of the results of our analysis. First, this analysis adds to a burgeoning body of research using satellite imagery on land cover to test growth machine propositions (Clement and Podowski 2013; Ergas et al. 2016; Landis 2017; Pravin 2019; Smiley and Hakkenberg 2020; Steele and Luloff 2003; Wongso and Göçmen 2025). Furthermore, although Clement et al. (2023) claimed that the reciprocal association they observed was asymmetric, they did not use any formal statistical tests for asymmetry, such as a likelihood ratio test comparing unconstrained versus equality-constrained structural equation models; our study redresses this gap (see also Ambrona and López-Pérez 2014; Kenny 1979). Thus, our study, using these techniques, showed that road construction created arterial corridors through which human activity moved and then expanded its impervious footprint (see Figure 5), and reciprocally, land development fed back into more road construction, although its proportional influence was smaller. These results are consistent with but add a reciprocal-asymmetry nuance to how the growth machine framework describes infrastructure-led development (e.g., Bernier and McGlinchy 2023; Greenberg et al. 2024; Rudel 2009).

As previously mentioned, in the state of Texas, through various direct and indirect channels, land development can spur further road construction. With land development comes more property taxes; while state and federal taxes fund major highway construction, property taxes are used for local road construction. Municipalities in Texas also have specific policy mechanisms they can use to fund road construction through land development; as two examples: (1) developers pay impact fees on new projects, creating dedicated road construction funds, and (2) tax increment reinvestment zones create revenue to fund public improvements, including major road construction (e.g., Briffault 2010; Jeong and Feiock 2006; Kirkpatrick 2016; Sroka 2024). In the first instance, although developer impact fees require private interests to contribute to the cost of infrastructure for new projects, these fees generally cover only a fraction of the total expenditure, leaving much of the financial burden for road expansion to be subsidized by the public through broader tax revenues. In the second instance, TIF allows a portion of the future property tax revenue generated by a specific development to be captured and reinvested directly back into that area’s infrastructure. This mechanism essentially uses anticipated public funds to pay for the immediate road construction necessary to make the private development viable, thereby allowing developers to realize the benefits of increased land values while the public assumes the long-term financial risk.

On that note, this study further reinforces the supportive role of the government in the growth coalition. Not only does the government fund road construction, which, as a primary catalyst, opens up land to new development, but it also can then generate revenue from that newly developed land, reciprocally, to fund more road construction, although this feedback has a proportionally smaller impact. Practically, as mentioned, developers’ impact fees do go to road construction, but they represent a smaller source of funding for road construction than public tax dollars. Yet consistent with the political-economy of the growth machine, developers and growth coalition boosters also openly advocate for TIF. For instance, Near Southside, Inc. (NSI), is a nonprofit organization that works with the real estate industry to promote the further development of neighborhoods immediately south of downtown Fort Worth. During the time period of this study, NSI was contracted with the city to administer a TIF district, working with various stakeholders to support “public infrastructure constructed in conjunction with dozens of private redevelopment projects, providing an incentive that defrays development costs for projects that advance district goals and, without the TIF’s assistance, wouldn’t be financially feasible” (NSI 2026). Thus, developers and growth coalition boosters, like NSI, advocate for state support, like the creation of TIF districts, to bridge the financing gaps for the public infrastructure necessary for their projects. In this way, with state assistance, the real estate industry can have risk transferred to the public, yet they can still capture the profit from the subsequent increase in land value that comes from the development (e.g., Briffault 2010; Kirkpatrick 2016). Whether through large-scale greenfield projects or the development of smaller parcels, this state-led infrastructure investment would lower the barrier to land intensification, pointing to a growth coalition strategy to incentivize development and maximize exchange value. And, our study clarifies that road construction, as car-centric public infrastructure, plays a significant role in this process.

Drawing from the growth machine framework, our finding of reciprocal asymmetry adds detail to how local land development strategies are unequally intertwined together with car-centric infrastructure projects, validating arguments that the ascendance of the private automobile has happened at the expense of public transit (e.g., Angelo and Wachsmuth 2020; Lee and Sener 2017; Mattioli et al. 2020). In other words, the government is socializing the risk and cost of road construction reciprocally in service of private land development and the private automobile. This dynamic reveals the asymmetric reciprocity that links the socialization of public cost to private, car-centric development. Although this gives us a socioecological example of an asymmetric reciprocal association, conceptually, it demonstrates that road construction, as infrastructure-led development, is a functional yet unbalanced component in the interdependent framework of the growth machine.

Footnotes

ORCID iD

Matthew Thomas Clement

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

1

The National Land Cover Database Impervious Descriptor satellite imagery does not distinguish among the various road types; the raster file simply signals whether the cell pixel is covered by a road. We discuss this as a potential limitation in the conclusion of the study.

2

Among 3,203 census tracts, 2 have no roads and 3 have no (nonroad) developed land, representing 0.06 percent and 0.09 percent of the sample size, respectively.

3

Nine census tracts had no resident population between 2010 and 2024, representing about 0.28 percent of the sample size. These include wildlife refuges, industrial areas, and airports.

4

With the exception of the χ² statistic, which is likely the result of the large sample size. As noted in , the model fit indices are derived from structural equation models that do not include population density as an additional control. Although adding population density as a control does not change the substantive findings of the analysis, including the results of the asymmetry tests, it does seriously degrade model fit in terms of the Tucker-Lewis index, which decreases from 0.947 to 0.838 after including population density.

Author Biography

Matthew Thomas Clement is an associate professor in the Sociology Department at Texas State University, where he teaches courses on statistics and environmental sociology. His recent research has been using satellite imagery on land cover to develop alternative measures of urbanization and assess their implications for social and environmental change.

References

Aljoufie

Mohammed

Zuidgeest

Mark

Brussel

Mark

van Maarseveen

Martin

. 2011. “Urban Growth and Transport: Understanding the Spatial Temporal Relationship.” Pp. 315–28 in Urban Transport XVII: Urban Transport and the Environment in the 21st Century, edited by Pratelli

Brebbia

C. A.

Southampton, UK: WIT Press.

Allison

Paul D

. 2009. Fixed Effects Regression Models. Thousand Oaks, CA: SAGE.

Allison

Paul D.

2019. “Asymmetric Fixed-Effects Models for Panel Data.” Socius. doi:10.1177/2378023119826441.

Ambrona

Tamara

López-Pérez

Belén

. 2014. “A Longitudinal Analysis of the Relationship between Positive and Negative Affect and Health.” Psychology 5:859–63.

Angelo

Hillary

Wachsmuth

David

. 2020. “Why Does Everyone Think Cities Can Save the Planet?” Urban Studies 57(11):2201–21.

Anselin

Luc

Florax

Raymond J.

1995. “Small Sample Properties of Tests for Spatial Dependence in Regression Models: Some Further Results.” Pp. 21–74 in New Directions in Spatial Econometrics, edited by Anselin

Florax

R. J.

Berlin: Springer-Verlag.

Barfield

Rebekah F.

2022. “Infrastructure Consequences of Exurb Growth in Texas.” Mosbacher Institute for Trade, Economics, and Public Policy, The Bush School of Government & Public Service, Texas A&M University. Retrieved April 24, 2026. https://bush.tamu.edu/wp-content/uploads/2022/03/TXL_Infrastructure_white_paper.pdf.

Baum-Snow

Nathaniel

. 2007. “Did Highways Cause Suburbanization?” Quarterly Journal of Economics 122(2):775–805.

Bernier

Nathan

McGlinchy

Audrey

. 2023. “Highway to Sprawl: How I-35 Shapes Where People Live in Austin.” KUT 90.5, June 29. Retrieved April 24, 2026. https://www.kut.org/transportation/2023-06-29/highway-to-sprawl-how-i-35-shapes-where-people-live-in-austin.

10.

Brannan

Ryan

. 2010. “Limited County Land Use Authority: An Avenue to Protecting Private Property Rights.” Texas Public Policy Foundation. Retrieved April 24, 2026. https://www.texaspolicy.com/wp-content/uploads/2018/08/2010-12-PB06-CountyAffairs-rb.pdf.

11.

Briffault

Richard

. 2010. “The Most Popular Tool: Tax Increment Financing and the Political Economy of Local Government.” University of Chicago Law Review 77(1):65–95.

12.

Clement

Matthew T.

Dede-Bamfo

Nathaniel

DeWaard

Jack

Kim

Seoyoun

. 2023. “The Interdependent Forces of Local Growth: A County-Level Study, 2001–2011.” Sociological Quarterly 64(4):564–586.

13.

Clement

Matthew T.

Podowski

Elizabeth

. 2013. “Intensifying the Countryside: A Sociological Study of Cropland Lost to the Built Environment in the United States, 2001-2006.” Social Forces 92(2):815–38.

14.

Dye

Richard F.

McMillen

Daniel P.

2007. “Teardowns and Land Values in the Chicago Metropolitan Area.” Journal of Urban Economics 61(1):45–63.

15.

Elliott

James R.

Frickel

Scott

. 2015. “Urbanization as Socioenvironmental Succession: The Case of Hazardous Industrial Site Accumulation.” American Journal of Sociology 120(6):1736–77.

16.

Ergas

Christina

Clement

Matthew T.

McGee

Julius

. 2016. “Urban Density and the Metabolic Reach of Metropolitan Areas: A Panel Analysis of per Capita Transportation Emissions at the County-Level.” Social Science Research 58:243–53.

17.

Fodor

Eben

. 2012. “Relationship between Growth and Prosperity in the 100 Largest U.S. Metropolitan Areas.” Economic Development Quarterly 26(3):220–30.

18.

Fulton

William

Clark

J. H. Cullum

Hendricks

David

Cisneros

Henry

. 2021. The Texas Triangle: An Emerging Power in the Global Economy. College Station: Texas A&M University Press.

19.

Gerber

Elisabeth R.

Phillips

Justin H.

2003. “Development Ballot Measures, Interest Group Endorsements, and the Political Geography of Growth Preferences.” American Journal of Political Science 47(4):625–39.

20.

Glaeser

Elisabeth

Poterba

Justin

. 2021. “Economic Perspectives on Infrastructure Investment.” Pp. 53–74 in Rebuilding the Post-pandemic Economy, edited by Kearney

M. S.

Ganz

Washington, DC: Aspen Institute.

21.

Greenberg

Pierce

Wishart

Ryan

Danielsen

Sabrina

. 2024. “Driving Environmental Inequality: The Unequal Harms and Benefits of Highways.” Environmental Sociology 10(4):443–56.

22.

Guerrieri

Veronica

Hartley

Daniel

Hurst

Erik

. 2013. “Endogenous Gentrification and Housing Price Dynamics.” Journal of Public Economics 100:45–60.

23.

Harvey

David

. (1982). The Limits to Capital. London: Blackwell.

24.

Jeong

Moon-Gi

Feiock

Richard C.

2006. “Impact Fees, Growth Management, and Development: A Contractual Approach to Local Policy and Governance: A Contractual Approach to Local Policy and Governance.” Urban Affairs Review 41(6):749–68.

25.

Kenny

David A.

1979. Correlation and Causality. New York: John Wiley.

26.

Kirkpatrick

Lucas O.

2016. “The New Urban Fiscal Crisis: Finance, Democracy, and Municipal Debt: Finance, Democracy, and Municipal Debt.” Politics & Society 44(1):45–80.

27.

Landis

John D.

2017. “The End of Sprawl? Not so Fast.” Housing Policy Debate 27(5):659–97.

28.

Lee

Richard J.

Sener

Ipek N.

2017. “The Effect of Light Rail Transit on Land Use in a City Without Zoning.” Journal of Transport and Land Use 10(1):541–56.

29.

Lieberson

Stanley

. 1985. Making It Count: The Improvement of Social Research and Theory. Berkeley: University of California Press.

30.

Logan

John R.

Whaley

Rachel B.

Crowder

Kyle

. 1997. “The Character and Consequences of Growth Regimes: An Assessment of Twenty Years of Research.” Urban Affairs Review 32(5):603–630.

31.

Logan

John R.

Molotch

Harvey L.

2007. Urban Fortunes: The Political Economy of Place. Berkeley: University of California Press.

32.

Logan

John R.

Zhou

Min

. 1989. “Do Suburban Growth Controls Control Growth?” American Sociological Review 54(3):461–71.

33.

Martinez

Erica

. 2025. “Texas Property Owner Rights and the Future of Senate Bill 2038—Land Development Shift in Texas.” St. Mary’s Law Journal 57(1):155–84.

34.

Mattioli

Giulio

Roberts

Cameron

Steinberger

Julia K.

Brown

Andrew

. 2020. “The Political Economy of Car Dependence: A Systems of Provision Approach.” Energy Research & Social Science 66:101486.

35.

Molotch

Harvey L

. 1967. “Toward a More Human Ecology.” Land Economics 43(3):336–41.

36.

Molotch

Harvey L.

1976. “The City as a Growth Machine: Toward a Political Economy of Place.” American Journal of Sociology 82(2):309–32.

37.

NCTCOG (North Central Texas Council of Governments). 2022. “Introduction and Background.” Chapter 1 in 2023–2026 Transportation Improvement Program. Retrieved April 24, 2026. https://www.nctcog.org/getmedia/46ed0d6c-66d8-48db-ad07-cac71191ff21/_Chapter-1-for-web.pdf.

38.

NSI (Near Southside, Inc.). 2026. “Tax Increment Financing (TIF) District #4.” Retrieved April 24, 2026. https://www.nearsouthsidefw.org/tax-increment-finance.

39.

Pravin

Avni D.

2019. “Environmental Justice and Flood Adaptation: A Spatial Analysis of Flood Mitigation Projects in Harris County, Texas.” University of Oregon. Retrieved April 24, 2026. https://scholarsbank.uoregon.edu/items/63eb7cef-7c26-400e-9528-847e53effbc2.

40.

Rudel

Thomas K.

2009. “How Do People Transform Landscapes? A Sociological Perspective on Suburban Sprawl and Tropical Deforestation.” American Journal of Sociology 115(1):129–154.

41.

Saginor

Jesse

Dumbaugh

Eric

Ellis

David

Minjie

. 2011. “Leveraging Land Development Returns to Finance Transportation Infrastructure Improvements.” Final Report No. UTCM 09-13-12. University Transportation Center for Mobility, Texas Transportation Institute. https://utcm.tti.tamu.edu/publications/final_reports/Saginor_09-13-12.pdf.

42.

Smiley

Kevin T.

Hakkenberg

Christopher R.

2020. “Race and Affluence Shape Spatio-temporal Urbanization Trends in Greater Houston, 1997 to 2016.” Land Use Policy 99:105093.

43.

Sroka

Robert

. 2024. “Subsidizing a Sports Arena in a Bankrupt City: Detroit’s Little Caesars Arena.” Social Science Quarterly 105(5):1691–1705.

44.

Steele

Jennifer

Luloff

A. E.

2003. “The Timing of Adoptions of Land-Use Plans and Regulations in Pennsylvania Localities.” Social Science Quarterly 84(1):210–18.

45.

TxDOT (Texas Department of Transportation). 2024a. “Texas Statewide Multimodal Transit Plan.” Alamo Area Metropolitan Planning Organization, January 22. Retrieved April 24, 2026. https://alamoareampo.org/Committees/TPB/minutes/2024/1-Jan/CMays_TxDOT_TEXAS%20TRANSIT%20PLAN_AAMPO%20Policy%20Board_1_22_24.pdf.

46.

TxDOT (Texas Department of Transportation). 2024b. “2024 Top 100 Congested Road Segments.” Texas A&M Transportation Institute, November 8. Retrieved April 24, 2026. https://static.tti.tamu.edu/tti.tamu.edu/documents/TTI-2024-1.pdf.

47.

Tong

Qingmeng

Qiu

Feng

. 2020. “Population Growth and Land Development: Investigating the Bi-directional Interactions.” Ecological Economics 169:106505.

48.

USGS (U.S. Geological Survey). 2024. “Annual NLCD Collection 1 Science Products.” Version 1.1, June 2025. Retrieved April 24, 2026. https://www.usgs.gov/data/annual-national-land-cover-database-nlcd-collection-1-products-ver-11-june-2025.

49.

Warner

Kee

Molotch

Harvey

. 2000. Building Rules: How Local Controls Shape Community Environments and Economies. Boulder, CO: Westview.

50.

White

Halbert

. 1980. “A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity.” Econometrica 48(4):817–38.

51.

Wongso

Samto

Göçmen

Z. Aslıgül

. 2025. “Assessing Ordinances Promoting Compact Development in Fast-Growing Wake County, North Carolina.” Journal of Urban Affairs. doi:10.1080/07352166.2025.2560957.

52.

York

Richard

Light

Ryan

. 2017. “Directional Asymmetry in Sociological Analyses.” Socius. doi:10.1177/2378023117697180.