Ensuring Equity in Climate-Forward Cities
Social and Economic Impacts of Transit-Oriented Affordable Housing Development in Denver, Colorado
Samuel Waechter-Cass Student number: 13559109 July 14, 2022
MSc Economics, Environmental Track
Amsterdam School of Economics, University of Amsterdam
Supervisor: Dr. Ana Varela Varela
This paper aims to investigate the causal effects of transit-oriented affordable housing development projects on their surrounding communities. The strategic acquisition of properties along light rail corridors in Denver, Colorado, sets the stage for a difference in differences comparison between stations selected for transit-oriented affordable housing development and those subject only to traditional real estate market forces. Dependent variables of median household income, gross rent as a share of median household income, the share of workforce commuting in single-occupancy vehicles, and the share of the population identifying as Hispanic or Latino quantify the social and economic impacts of developments. Treatment areas
experienced decreases in median household income, the share of the population identifying as Hispanic or Latino, and the share of commuters who travel in single-occupancy vehicles compared to control areas but experienced a relative increase in gross rent as a share of median household income. These results suggest that transit-oriented affordable housing developments allow lower-income residents to reap benefits like improved urban mobility options and proximity to economic opportunities. However, the developments may still contribute to displacement in the city that disproportionately impacts Hispanic and Latino residents.
1. Introduction 2. Literature review
2.1: Evidence Supporting TOD 2.2: Evidence against TOD 3. Data
3.1 Denver Transit-Oriented Development fund 3.2 American Community Survey
3.3 Descriptive Statistics 3.4 Correlation Matrices 4. Methodology
4.1: Parallel Trend Assumption 4.2: Identification
4.3: Data setup 4.4: Model setup 5. Results
5.1: Expanded data set
5.1.1: Median income, expanded
5.1.2: Gross rent as a share of median income, expanded
5.1.3: Share of commuters using single-occupancy vehicles, expanded 5.1.4: Share of the Hispanic population, expanded
5.2: Robustness Check, Limited Data Set 5.2.1: Median income, limited
5.2.2 Gross rent as a share of median income, limited
5.2.3 Share of commuters using single-occupancy vehicles, limited 5.2.4: Share of the Hispanic population, limited
6.1: Explanation of main results 6.2: Comparison to Existing Literature 6.3: Modern policy implications
6.5: Potential next steps:
The impacts of our changing climate are more pronounced than ever. Parallel to this issue is a historic affordable housing shortage that threatens the health and safety of millions as extreme weather events become more frequent and intense. Denver, Colorado’s ambitious 80x50 Climate Action Plan accounts for both of these issues and aims to reduce greenhouse gas emissions (GHGs) in the city by 80% compared to the 2005 baseline by 2050 (Denver Department of Public Health and Environment, 2018). Part of the Climate Action Plan includes expanding low-carbon urban mobility options by adding five light rail lines to the existing network in what is currently the largest public transit expansion under construction in the United States (Urban Land Conservancy, 2013). Another pillar outlined in the plan to ease the housing crisis and reduce greenhouse gas emissions in Denver is by promoting the development of medium and high-density communities near the newly expanded transit network.
Revitalizing urban cores and designing new developments that integrate with the surrounding neighborhoods and the greater regional transportation network yields various benefits. Lowered vehicle traffic reduces fine particulate matter pollution, a health issue that Denver has struggled with for decades (Koken et al., 2003). Well-designed urban development can spur local job creation and serve as a social mobility springboard for low-income residents by providing new work opportunities that are accessible without a vehicle. However, one of the greatest opportunities is the ability of dense developments to reduce the combined transportation and housing burden that weighs heavily on personal finances (Zhao, 2015). In 2020, the average American household earned $84,352 and spent nearly 45% of that income on housing and transportation (U.S. Bureau of Labor Statistics, 2021). This figure quickly exceeds 50% for low-income households in Denver and other high-cost-of-living cities throughout the country (Airgood-Obrycki et al., 2022). High housing and transportation burdens limit households’ ability to build an emergency fund or save for other important financial matters.
A popular proposal in cities looking to reduce the carbon intensity of their community while also creating new, dense housing supply is the concept of transit-oriented development (TOD). This term was codified in the 1980s by Peter Calthorpe and was published in “The New American Metropolis” in 1993. Transit-oriented development is a mix of commercial, residential, office space, and entertainment centered around or located near a transit station (Federal Transit Administration, 2019). Transit-oriented development reduces emissions in several ways.
et al., 2020). Transit-oriented development also decreases transportation emissions by nudging residents towards low-impact transit alternatives (Ashik et al., 2022). In 2021, 28% of all daily car trips made in the United States were less than one mile (U.S. Bureau of Transportation Statistics, 2022). Driving is often the only viable option in many American cities. When citizens have the ability and financial capacity to live near their daily destinations, they are less reliant on cars, and auto emissions subsequently decline (Center for Neighborhood Technology, 2010).
Finally, transit-oriented development lowers per capita emissions from infrastructure construction and maintenance.
development increases the housing stock and eases pressure on a city’s housing supply by simply building more with less.
Dense housing is relatively scarce in Denver, with 77% of the city’s land dedicated exclusively to single-family detached housing, as seen in figure 1 (Denver Department of Community
Planning and Development, 2020). Single-family housing does not support the population density required to operate a high-capacity transportation network, nor does it allow for large-scale additions to the city’s housing stock. City planners are well aware of the distortionary impacts that too much single-family zoning can have on a city’s housing market. Minneapolis, Minnesota, was the first major American city to ban single-family zoning in 2019, and other cities have followed suit. Denver has not yet made this move. Thus, building transit-oriented developments near stations that are already zoned to allow for multi-unit residential construction is the most practical way to increase the city’s housing stock and meet the targets of the 80x50 plan.
TOD aims to create urban environments where individuals have multiple transportation options and live within a reasonable distance of public amenities like parks, schools, and shopping districts. For needs unattainable in the immediate built environment, citizens should be able to travel inter-regionally without relying on a single-occupancy vehicle. By creating these environments, developers and urban planners can lower the carbon intensity of a neighborhood
and alleviate pressure on the city’s housing supply. Despite the documented benefits of transit-oriented development, many community stakeholders remain skeptical.
Opposition to implementing transit-oriented development typically follows one of two trains of thought. The first is that TOD will unleash economic forces that provoke gentrification.
Gentrification is defined as the process whereby the character of an urban area is changed by wealthier people moving in, improving housing, and attracting new businesses, often displacing current inhabitants in the process. Residents of Denver have good reason to be concerned about potential gentrification. Denver is located at the foothills of the Rocky Mountains, a strategic transportation and logistical gateway to rapidly-expanding western states like Utah, Nevada, Arizona, and Colorado. The city’s population ballooned from 550,000 in 2000 to over 700,000 residents today, and most of the new arrivals come from places outside of Colorado.
The National Community Reinvestment Coalition determined that Denver gentrified at the second-fastest rate of any city in the United States between 2013 and 2017, second only to global tech hub San Francisco (Richardson et al., 2020). Gentrification in the United States disproportionately impacts Black and Hispanic communities (Hwang & Sampson, 2014).
Roughly 1 in 3 Denverites currently identifies as Hispanic or Latino, a figure that has held steady since 2000. Despite this steady trend, community leaders in historically Hispanic neighborhoods are sounding the alarm over displacement impacting their communities. Leaders believe that if development investments expand to lower-income areas of Denver, real estate developers will seize the opportunity to raise housing prices. Significant price increases could leave residents with no option but to relocate to a different Denver neighborhood or depart the city entirely.
On the other end of the spectrum, some individuals and neighborhood associations oppose the development of increased-frequency transit and high-density housing out of concern that it will cause the neighborhood to deteriorate. This concept, known as the NIMBY (Not In My Backyard) conflict, exists in various forms throughout the public realm. NIMBYism often has a prominent political undertone and can require political solutions to overcome (Feinerman et al., 2004). There is a prominent sentiment among some Americans that public transportation and dense housing are the last resort and only for people who cannot afford to travel in a private automobile or own their own home (Agrawal, 2015). This train of thought is complex and goes beyond the scope of this thesis, but opponents of TOD on this side of the conversation often cite
social and economic impacts of these large-scale investments. Enabling societal benefits without perpetuating existing inequalities or creating issues in otherwise healthy areas is essential. This analysis aims to understand the community effects of transit-oriented housing developments and determine if this planning style aligns with Denver’s goals for the 21st century.
2. Literature review
The increasing popularity of transit-oriented development in the past 30 years has led to a flourishing field of innovative literature. Research on the impact of transit systems and transit-oriented development on local communities is widespread and encompasses multiple academic disciplines, including economics. The analysis of this paper, however, is unique from the bulk of previous TOD literature in four primary ways. First, much of the existing literature focuses on the urban planning and physical design aspects of transit-oriented development. For example, empirical studies about the optimal frequency of transit service or how wide a bike line should be for citizens to feel comfortable enough to use it. This analysis concentrates on the economic and social implications of implementing a transit-oriented affordable housing development. Second, among popular economics-based articles about transit-oriented development, the benefits are often assumed to outweigh any externalities. Studies often seek to determine which category of TOD is best for a particular area rather than asking if the developments in question are a positive addition to a neighborhood in the first place. This analysis aims to provide a holistic view of the data to determine the net impact on a community.
Third, many studies capitalize on situations where the treatment (construction of transit-oriented developments) coincides with the construction of the transit line itself. These studies are valuable, but it can be difficult to distinguish which effects result from the transit-oriented development and which effects result from the construction of the new rail line. In this analysis, most rail lines precede the implementation of the TOD projects by 10+ years, meaning the local neighborhood had already adapted to the presence of the light rail. Thus, the effect of project development can be more accurately derived. Finally, the treatment variable of this analysis is the introduction of transit-oriented development projects with a clearly defined affordability requirement embedded into the project design. Treatment projects still contain the traditional components of transit-oriented development, such as retail space, parks, and other infrastructure amenities. However, a pre-determined amount of affordable housing units is obligatory. Many existing studies on transit-oriented development analyze projects similar to the one in this analysis but do not have affordability requirements.
This research design fills an important gap in the literature about transit-oriented development. The following sections analyze two predominant perspectives in the literature about transit-oriented development. One side suggests that TOD is beneficial and does not
2.1: Evidence Supporting TOD
Balboni et al. (2021) introduce the theory of transit-induced gentrification.
Transit-oriented gentrification suggests that the increased accessibility and economic development accompanying new transit investments will positively impact nearby property values. This increase in property values leads to gentrification and a disproportionate exodus of lower-income residents (Delmelle et al., 2020). The authors highlight that conversations surrounding infrastructure investments tend to center on the needs of current residents.
However, it is essential to note that infrastructure investments provide amenities to a place, not a specific group of individuals. Should development trends following investment cause widespread displacement, the infrastructure will remain unchanged, but those served by these public services will change. This observation appears throughout the literature as a reason to continue research on the impacts of transit-induced gentrification. It also directly relates to the situation in Denver, where residents are concerned about being displaced by new transit-oriented development.
Balboni et al. continue by augmenting a standard urban commuting model based on (Ahlfeldt et al., 2015) to quantify the welfare impacts of urban infrastructure investments. Their model includes outcome variables like the share of income spent on housing to derive these welfare calculations. In a subsequent companion paper, Balboni et al., (2020) provide real-world context to this model by analyzing the impacts of a new bus rapid transit (BRT) system in Dar es Salaam, Tanzania. Preliminary results of the study revealed that the BRT system turned out to be a pro-poor investment. In the neighborhoods near BRT stations, welfare gains for poor and wealthy residents were 3.0% and 2.5%, respectively. Residents elsewhere in the city in control areas also enjoyed welfare gains, though at lower rates of 2.4% and 2.5%. Thus, poorer individuals in the treatment area experienced the most significant net welfare improvement of any group in the city. This study demonstrates that public infrastructure investments can elevate communities without displacing their most vulnerable individuals when appropriately implemented. This paper guides the dependent variable selection and rationalizes the decision to utilize a difference-in-differences econometric model in this analysis.
Dawkins and Moeckel (2016) use an integrated land use/transportation model to analyze the effects of transit-oriented developments. The paper shares many characteristics with this analysis; most notably, the treatment policy in question includes an affordable housing requirement in new construction projects. The authors use a case study of the Washington, D.C.
metropolitan area and find that transit-oriented developments with affordability requirements can mitigate the potential effects of gentrification. More specifically, the policy protected existing
residents from being priced out of areas with new developments while improving their access to transit and economic opportunities throughout the region.
Padeiro et al. (2019) conducted a meta-analysis of literature regarding transit-oriented development and gentrification published between 2010 and 2018. The authors’ primary focus is on analyzing the research methods used in transit-oriented development studies to detect patterns that may signal bias. They conclude that some evidence exists supporting the existence of transit-oriented gentrification, but the methodology used in many articles that arrive at this conclusion is highly questionable. Furthermore, their results suggest gentrification closely relates to the local built environment and community policy. Understanding common research methodology mistakes is an important consideration to take into account when investigating geography-based policies like the one in Denver. The meta-analysis guides the structure of this analysis to ensure that these common mistakes do not occur.
2.2: Evidence against TOD
Gamper-Rabidran and Timmins (2011) investigate the impact of hazardous waste site cleanup on gentrification and residential sorting. The study accomplishes this by analyzing communities before and after the cleanup of sites deemed a priority public health hazard by the United States Environmental Protection Agency (EPA). Although this study does not directly relate to transit-oriented developments, it is a relevant example of a significant public investment executed to improve the environmental and economic health of the community. A previous paper analyzing this same investment found that hazardous waste sites selected for cleanup experienced an 18.5% and 5.6% increase in median housing values within 1km and 3km of the cleanup site, respectively, compared to those that did not receive federal funding for cleanup.
The authors quantify the community impacts of this by referencing a list of nearly 700 proposed cleanup sites ranked in order of perceived environmental urgency. The superfund only had enough capital to finance the cleanup of 400 sites, thus setting the stage for a natural regression discontinuity design comparing sites near either side of the cutoff. The RD design found evidence that “environmental gentrification” took place post-treatment. Treatment areas experienced a median household income increase of 26% and a 31% increase in the share of college graduates, though the treatment did not displace minority populations. These effects also exist at the 3 km radius level to a lesser extent. This analysis adopts this paper’s method of
occur. The combination of an increase in median household income and a consistent racial composition of the population still leaves the possibility that high-income households displaced low-income households of the same race.
Jones and Ley (2016) investigate a TOD policy that allows the construction of large, market-rate housing units near Vancouver’s SkyTrain automated transit system stations.
Significant stretches of the SkyTrain run adjacent to mid-rise apartments constructed as controlled, affordable housing in the 1980s. The analysis reveals the existence of gentrification in the surrounding areas as transit-oriented development accelerates due to policies implemented by the municipality that subsidize TOD. The authors go as far as to claim that the affordable housing units, which house many of the city’s recently arrived refugees, ‘are endangered by gentrification from a regional policy of transit-oriented development, a policy where environmental objectives and profitability trump social justice objectives.’ The pro-TOD policy implemented in Vancouver does not contain significant provisions to prevent gentrification or protect incumbent residents. It is the base case example that TOD opponents use when arguing against transit-oriented development. Gentrification is expedited in Vancouver because the city has some of the world's highest foreign real estate investments. However, Denver’s housing market is quickly closing in on Vancouver and other expensive North American cities in this aspect.
Debrezion et al. (2006) perform a meta-analysis of the impact of railway stations on residential and commercial property value. They find that proximity to a rail station increases residential and commercial property values. The authors analyze the impact of stations on real estate values at various spatial intervals. They find that the effect is most substantial within 250 meters of the station. They also discover that commuter railway stations have a higher positive impact on property values than light rail stations.
Each of these prominent papers makes significant contributions to the literature surrounding transit-oriented developments and influences the analysis of this paper to some extent. After reviewing this well-rounded selection of papers, a robust empirical design seeks to understand Denver's unique situation better.
3.1 Denver Transit-Oriented Development fund
Central to this analysis is the Denver Transit-Oriented Development Fund. Established in the aftermath of the 2008 subprime mortgage crisis, the fund sought to capitalize on a unique real estate development opportunity. Property values near existing and future rail stations, prime locations for TOD, were available at the lowest prices of the decade and ripe for development.
Building affordable housing near transit stations reduces the financial burdens of low-income households, allowing them to allocate more of their income to alternative uses. Housing and transportation are the budget categories that consume the most significant shares of Denverites’
take-home income (Shamsuddin & Campbell, 2021).
Despite this opportunity, most developers left standing after the market collapse were focused almost exclusively on lucrative, high-end development projects. The recent drop in property values meant that affordable housing units were simply not profitable at the time, rendering them an unattractive investment in the eyes of already cautious developers (Scopelliti, 2014).
Forward-looking affordable housing advocates recognized this opportunity to change the course of the city’s development future and secured $15,000,000 in revolving loan capital to put their vision into action during Q1 2010. When securing real estate development loans was particularly challenging, the TOD Fund offered several advantages, including an expedited underwriting process to facilitate rapid development, flexible loan terms, and low interest rates.
In addition, developers requesting funding assistance were required to build within ½ mile (0.8 km) of a rail station and adhere to affordable housing standards defined by the City and County of Denver.
The fund resulted in six major transit-oriented development projects near light rail stations and two housing preservation projects near bus rapid transit (BRT) stops that encompass 626 affordable housing units and more than 120,000 sq ft of retail space. This analysis focuses on rail transit, thus excluding the projects along bus routes. Additionally, one rail development is excluded due to its completion being too recent for reliable data to be available. The final analysis contains five treatment areas adjacent to rail stations financed by the Denver Transit-Oriented Development Fund.
3.2 American Community Survey
Data collected to measure the impact of the five transit-oriented developments come from official estimates published by the American Community Survey (ACS), an annual demographics survey conducted by the United States Census Bureau. The ACS is sent to roughly 3.5 million households annually and aggregates microdata responses into estimates for various community-related demographic metrics at various geographical levels. It is the most detailed and advanced survey of its kind in the nation and is the most reliable, publicly accessible data source for this analysis. Four metrics serve as dependent variables based on knowledge obtained from similar studies.
Sandoval (2018) investigates claims of the displacement of Hispanic and Latino populations caused by transit-oriented development in Oakland, Los Angeles, and San Diego, California. Denver has a similar ethnic composition and concern over racial displacement as these cities (Sachs, 2021). Thus, Hispanic or Latino Origin (ACS table ID B03003) is included as a variable to explore the impact of transit-oriented development on Denver’s Hispanic and Latino communities. Guerrieri et al. (2013) demonstrate that incomes and rental prices that rise faster than the regional average signify endogenous gentrification. Median Household Income in the Past 12 Months (table ID B19013) and Median Gross Rent as a Percentage of Household Income in the Past 12 Months (table ID B25071) are included in this analysis to provide insight into the income demographics of residents and to serve as a proxy for measuring rental burden.
Finally, From an environmental point of view, Gu et al. (2021) investigate the link between urban rail systems and road congestion, while Gendron-Carrier et al. (2022) investigate the causal link between urban rail systems and urban air pollution. Means of Transportation to Work: Single Occupancy Vehicle (table ID B08301) serves as a proxy for the environmental impact of TOD.
The variables Hispanic or Latino Origin and Means of Transportation to Work: Single Occupancy Vehicle are reported as an absolute number but then divided by the total population of the respective census tract in that specific year to convert the statistic into a percentage figure. Converting these variables to represent a population share allows for comparisons across years and accounts for population changes.
Data are collected for 2010-2019 and are analyzed on a census tract level. The United States Census Bureau defines census tracts as geographic segments representing roughly 4,000 residents per tract. The physical size of a tract varies depending on the location; rural tracts are much larger, while urban tracts tend to be more compact. Using census tracts as the level of analysis provides an identical geographic area of comparison for each year. Data for the 2020 edition of the survey are available. However, the Census Bureau modifies the boundaries
of tracts experiencing rapid growth every ten years. This redistricting means that the geographic boundaries changed for some tracts in 2020, excluding that year from the analysis. ACS data at the census tract level is only available in five-year averages, meaning that an observation for 2014 is the average of all values for that census tract from 2010 to 2014. One-year estimates are more prone to measurement errors. The Census Bureau cites the need for precision due to the official nature of ACS data as a reason for not releasing one-year estimates at this level of geographic detail. This analysis accounts for this data reporting method by excluding years when the pre and post-treatment periods overlap.
3.3 Descriptive Statistics
Tables 1 and 2 contain the descriptive statistics of the expanded data set in the pre and post-treatment periods, while tables 3 and 4 contain the descriptive statistics for the limited data set. Comparing the standard deviation and mean, minimum, and maximum values of the pre-treatment observations to the post-treatment observations provides a general overview of trends taking place city-wide for the four dependent variables.
The mean value for the median income of census tracts post-treatment is more than
$15,000 greater than the pre-treatment value. The standard deviation increases by more than
$5,000, the maximum value increases by nearly $30,000, and the minimum value increases by just under $2,000. Denver experienced rapid economic growth during the 2010s, clearly reflected in the median income figures. The increasing standard deviation and income growth concentrated at the upper end of the wage spectrum suggest that the influx of wealth to Denver is unequally distributed. As a result, there is a high potential for displacement and gentrification in the city.
The mean value of the share of individuals identifying as Hispanic or Latino increases slightly in the post-treatment period from 28% to 32%. The mean values for the share of commuters driving alone and the gross median rent as a share of household income decrease slightly. The changes in mean values and standard deviations of these three variables from the pre to post-treatment periods are not as large as the shift in income. However, there is still a story to be told by observing these disparities. There are census tracts in Denver where Hispanic and Latino residents account for more than 80% of the population, and there are tracts where that figure is less than 2%. The median household earns $163,000+ in the wealthiest neighborhoods, while the poorest areas see a median income of less than $10,000. There are tracts where nearly 90% of commuters drive alone and others where only 32% drive alone.
Some tracts report the median gross rent taking up just 12% of household income, while other tracts are extremely rent-burdened, with residents spending more than half of their income on rent alone.
The main takeaway from the summary statistics is that Denver, like most large American cities, is a place of significant discrepancies. There are diverse neighborhoods in the city, but there are also areas of highly concentrated wealth, ethnic/racial groups, and even transportation infrastructure. Understanding how much variation exists in Denver’s urban area is essential to understanding the potential for gentrification. Such extreme differences in places that are so geographically close to each other mean that any spillover effects of developments have the potential to change a neighborhood significantly. Figure 2 clearly shows how segmented some neighborhoods are with physical urban features such as Interstate 25 serving as a de facto
barrier between parts of the city with the highest and lowest shares of Hispanic residents (Beaty, 2021).
3.4 Correlation Matrices
A simple correlation matrix is a valuable tool to quickly explore the data in another way and develop a basic understanding of relationships between variables in the analysis. Similar to the previous section about descriptive statistics, the following figures display correlation matrices for the expanded and limited datasets. Table 5 shows the matrix for the expanded data set, revealing several interesting correlations. As the share of a tract’s Hispanic population increases, households tend to spend more on rent, the median household income falls, and fewer commuters drive alone to work. An increase in gross rent as a share of median household income correlates with a sharp drop in median household income. Finally, an increase in median household income correlates with an increase in the share of workers commuting in single-occupancy vehicles.
Table 6 represents the same correlation matrix as table 6 but for the limited data set that includes only the tracts that directly contain a control or treatment station. Analyzing the limited data set allows for identifying correlation trends that differ between the area next to the station and the larger area. Most of the limited correlation matrix relationships are similar to the expanded matrix with a few outliers. As the share of the population becomes more Hispanic in the limited case, the share of income the average household spends on rent decreases sharply, which is the opposite of the expanded data set. This relationship suggests that the rental burden tends to be less severe for households in neighborhoods with light rail stations and a higher share of the Hispanic population. There are many potential causes of this result. However, the important takeaway is that people in these areas experience less financial strain related to housing, and there is a risk of gentrification in these majority-Hispanic areas near rail stations.
Another drastically different result between the expanded and limited data sets is the relationship between the gross rent as a share of median household income and median household income. In the expanded result, median income decreased as rent as a share of median household income increased. In the limited case, median income increases as rent as a share of median household income increases. This relationship suggests that households living in areas with higher rent burdens earn more than households where the rent burden is lower.
This outcome could result from very high rents concentrated in a few luxury developments in certain tracts. However, considering the result of the previous paragraph, where the rent burden decreases in more Hispanic areas, there appears to be significant heterogeneity within the different tracts.
Given the availability of high-quality panel data from the American Community Survey, this analysis employs a difference-in-differences method to analyze the impact of transit-oriented developments on community characteristics. The standard difference-in-differences method involves two groups in two time periods. It uses quasi-experimental data to compare the differences in outcomes between a control and treatment group over time. DiD accomplishes casual inference by first calculating the difference between the treatment group’s outcome before and after treatment. This first difference allows the model to account for time-invariant group-specific traits. In other words, this accounts for the heterogeneity between stations that existed from the first year in the data. Next, the same calculation is made but for the control group. This calculation accounts for time-varying factors that are common across all groups. Finally, the second difference is subtracted from the first, resulting in the difference-in-differences estimator. With the group and time effects cleaned from the data, the DiD estimator indicates the impact of the treatment on the treated group in the post-treatment period.
The data used in this analysis contains multiple groups over multiple periods and thus requires a more complex application of the difference in differences design. This version of the differences in differences analysis contains an interaction term that equals one for the treatment group in the post-treatment period. Unmeasured outcomes are assumed to be either time-invariant or group invariant. This version includes time and group fixed effects.
4.1: Parallel Trend Assumption
The validity of the difference-in-differences model relies heavily on the parallel trends assumption. The parallel trend assumption requires that the treatment and control groups follow the same pattern without treatment. In other words, no time-varying differences exist between the treatment and control groups. Any deviations in this parallel path are a result of the treatment effect. There is not a statistical method to prove the parallel trend assumption.
However, a visual inspection of the pre-treatment trends of the treatment and control groups can help support this fundamental assumption.
The parallel trend assumption is tested by comparing the annual averages of
changes in the dependent variables caused as a result of the treatment will take some time to develop before they are evident. The dependent variables of median household income and Hispanic population share pass the parallel trend test as seen in figures 2 and 3. Five-year estimates of the median household income increased gradually by 3-4% annually for both treatment and control tracts between 2010 and 2015. The share of the population identifying as Hispanic or Latino decreased by about 1% annually in treatment and control tracts. Over the five-year period, the median income of an average treatment household is roughly $10,000 lower than the average control household, and the treatment household is 80% more likely to be Hispanic.
Median gross rent as a percentage of household income and the share of single-occupancy vehicle commuters do not pass the parallel trends test as clearly, as seen in figures 4 and 5. Treatment and control groups for median gross rent as a percentage of household income experience downward trends but with some variation. The share of commuters traveling in single-occupancy vehicles remains relatively flat for both control and treatment groups but with some variation. Although the parallel trends are not as evident in these two variables as in the previously mentioned variables, they are cautiously included in the analysis. This decision is justified for two reasons. First, the variation in these variables is minimal and likely influenced by the margin of error in data collection. Second, these variables are inherently more likely to fluctuate as the sample of individuals changes each year.
Additionally, these variables may provide valuable insight into residents' transportation decisions and financial situations. Despite the decision to include these variables, the results will only be used for preliminary analysis and cannot be used to infer casualty.
The precise definition of the exact positioning, construction date, and physical attributes of land parcels acquired using money from the Denver Transit-Oriented Development fund make selecting control and treatment entities relatively straightforward. Treatment stations are identified as those that received funding for land parcel acquisition for transit-oriented development with an affordable housing requirement. Control stations are all remaining rail stations within Denver outside the central business district. None of the stations have significant institutions or subsidized transit-oriented development projects within ½ mile (0.8 km). It is important to note that while RTD’s rail network spans the metropolitan region, only stations within Denver city proper are part of this analysis. Development adjacent to rail stations in suburbs such as Littleton, Aurora, and Centennial are subject to different municipal zoning and economic policies that complicate comparing these stations with those in Denver. Stations within Denver’s central business district are excluded since this area contains the most expensive and intensely developed urban land in Colorado and is economically different from the rest of the city. These precautions mean control stations are subject to similar market and bureaucratic forces and are not influenced by publically subsidized development. Sixteen stations are examined in total, as seen in figure 7.
Once control and treatment stations are identified, all census tracts within 1 mile are assigned to their respective station. One mile is accepted throughout the literature as a reliable estimate for capturing all spillover effects caused by transit-oriented development (Guerra et al., 2012).
Beyond this distance, residents are less likely to be impacted by station development due to the last-mile transportation problem (Zellner et al., 2016). This categorization results in 60 census
4.3: Data setup
Facilitating a difference in differences model design requires adding several indicators to the dataset. ‘Tod’ was added as a binary dummy variable and equals one if the respective census tract is within one mile of a rail station that received treatment and 0 for tracts near control stations. ‘Time’ is equal to 1 if the tract is in the post-treatment period, 0 if it is pre-treatment, and ‘-’ if the years contained within the five-year average overlap between the pre and post-treatment periods. Eliminating the overlap years is done to ensure that the treatment effect is not diluted by including pre-treatment years in the observation’s five-year average. For example, if a station received treatment in 2013, the first observation included in the post-treatment analysis would be 2017. The observation for 2017 then represents an average of strictly post-treatment years from 2013 to 2017, whereas the observation for 2013 includes overlapping pre and post-treatment observations from 2009 to 2013. Observations where time = ‘-’ are dropped from the dataset before running any regressions. The final addition is an interaction term to facilitate the difference in differences specification that = 1 for treatment tracts in the post-treatment period.
Intraction of the tod and time variables creates the interaction term that estimates the effects of the treatment in the post-treatment period. A variable that outlines the year of the observation allows for the inclusion of time-fixed effects in the model. Another variable that outlines the specific census tract facilitates the inclusion of group-fixed effects.
4.4: Model setup
The full model for a difference in differences setup is specified as follows
represents individual observations for each tract and year combination, as denoted 𝑌𝑔𝑡
by the g and t subscripts. α represents the time-invariant effect of group g and represents
the time-varying but group-invariant effects of year t. 𝐷 is an interaction term that equals one
for the treatment group in the post-treatment period. δis then the treatment estimate of interest and is the figure used for analysis in this paper. Finally, ϵ represents the error term for each
Interpretation of the interaction term coefficient estimate changes depending on which dependent variable is analyzed. A significant increase in the estimate in the regressions for
median household income would suggest that subsidizing transit-oriented development results in an influx of high-income individuals, a potential sign of gentrification. An increase in gross rent as a share of median income would suggest that treatment areas became relatively less affordable. A decrease in the share of the population identifying as Hispanic or Latino would imply that the developments contribute to the displacement of that community. Finally, a decrease in the estimate for the share of workers commuting in a single-occupancy vehicle would suggest that TOD encourages residents to utilize alternative means of transportation.
This shift away from cars could reduce transportation-related emissions and the cost burden of commuting.
Due to the experimental design, standard errors cluster at the station group level. The standard errors of the census tracts that contain or surround one of the sixteen individual control or treatment stations are linked because the assignment mechanism for casual treatment is clustered at this level. Clustering the standard errors in this way is essential to the validity of the results. Failing to cluster standard errors may lead to an over-representation of the casual effect of transit-oriented housing developments on the local area.
Fixed effects are included at the census tract level to control for characteristics that vary between census tracts but are constant over time. Using fixed effects with panel data allows the model to create a unique intercept for each tract. Individual intercepts are essential because of the spatial nature of this analysis and the fact that Denver is a large city with a diverse range of neighborhoods. For example, some neighborhoods may contain a well-established cultural or natural amenity like a park, thus leading to higher property values and a wealthier population than the surrounding areas. Fixed effects will account for this unobserved difference when calculating the coefficient estimate and adjust accordingly.
Time-fixed effects are added to the model as well. Time fixed effects control for variables that are constant across all of the census tracts but vary over time. This allows the model to account for broad shocks that impact all stations across the city at the same time. The concept of time-fixed effects also motivated the decision to only include stations in Denver city and exclude those in the surrounding suburbs. Policy decisions that change over time and impact the dependent variables of this analysis often vary at the municipality level. If all stations throughout RTD’s entire regional network were included, a situation could arise where one municipality introduces a significant policy while the others do not. In this situation, only the
Eight distinct differences in differences regressions provide coefficient estimates for the four dependent variables in the expanded and limited versions of the data. Regressions hinge on the assumptions and procedural steps outlined in the previous section, and the results are as follows.
5.1: Expanded data set5.1.1: Median income, expanded
Table 7 (n=355) contains the coefficient estimator for the regression on median income and indicates that the development of a transit-oriented housing development results in a nearly
$7,000 decrease in the median income of the adjacent area. This is significant at the 5% level though the source of its effect is difficult to pinpoint. It is possible that station development, and the implementation of affordability requirements, create an environment that attracts lower-income individuals to the area and allows them to enjoy the newly constructed public amenities. This hypothesis is in line with the field of literature that encourages transit-oriented development to empower lower-income households by providing them with affordable housing and alleviating their transportation expenses.
The other side states that transit-oriented development can act as a poverty magnet by concentrating low-income individuals in a specific area. This concentration of relative poverty can result from the exit of wealthier citizens from the treatment area to other parts of the city, a reduction in economic development, or poorer households moving into the area. Determining the cause of this reduction in median income requires further analysis facilitated by collecting data representing the demographics of households that move into and out of a particular area.
Dense, urban areas are popular among young college graduates who earn less at the beginning of their career but have high earning potential. In this situation, the reduction in median income post-treatment may still represent an influx of highly-educated individuals, making the area unaffordable in the future.
5.1.2: Gross rent as a share of median income, expanded
The gross rent as a share of median income, used as a proxy to measure rent burden, is displayed in table 8 (n = 356) and increases in treatment tracts by 2.8 percentage points compared to control stations. However, this is not statistically significant at the 5% level. Related literature suggests that the source of an increase in gross rent as a share of median income is often difficult to pinpoint. Some papers suggest that slightly higher rental burdens in areas near transit-oriented development may be a good sign as it indicates that lower-income individuals can occupy the new affordable housing and reap the rewards of the public investments (Boarnet et al., 2017). On the contrary, increasing rental burdens could signify rents rising faster than incomes at a more extreme level than in other areas of the city. This analysis cannot establish a causal effect between the transit-oriented treatment tracts and the relative increase in rental burden. In studies where the coefficient estimate is significant, it can still be challenging to distinguish between these possible causes unless more detailed information about individual households is available. However, given the previous result that median incomes are decreasing, examining changes in median rent would provide more context to this result in future analysis.
5.1.3: Share of commuters using single-occupancy vehicles, expanded
Table 9 (n = 360) contains the coefficient estimate for the interaction term in the regression for the share of commuters utilizing single-occupancy vehicles is insignificant at 0.022. If significant, this would mean that the treatment of constructing transit-oriented housing developments would cause more people to drive to work alone compared to control stations.
extend far beyond this distance. The geographic boundaries of tracts created by the United States Census Bureau are dictated by population, not physical size. In general, tracts are drawn to contain roughly 4,000 people. As a result, tracts that are closer to the city center tend to be much smaller geographically than suburban tracts. In the expanded version of the data, the suburban tracts are too large to be captured within the sphere of influence of the treatment stations.
Evidence shows that the transportation influence of rail stations is limited mainly to the station’s “walkshed.” A walkshed is a term used in transportation and urban planning that describes the distance the average person can walk from a location in a certain amount of time.
Rather than simply measuring the absolute distance from the station, the walkshed accounts for bridges, parks, sidewalks, and other environmental factors that determine how far an individual can reasonably go from the station without using another transportation. The influence of a station dissipates significantly within a 10 to 20-minute walk of the station, depending on the method of transportation used (Welch et al., 2018). The geographic scale of this analysis is simply too large, and this variable would be better studied on a more micro level if the data were available.
5.1.4: Share of the Hispanic population, expanded
Table 10 (n = 360) contains the interaction term in the regression on the share of the population that identifies as Hispanic or Latino is -0.029146, meaning that treatment areas experienced a decrease in the share of the Hispanic population of 2.9 percentage points compared to control stations. This is in line with the hypothesis that transit-oriented development causes gentrification. However, this result is not significant. The non-significance of this result does not negate the fact that displacement is taking place in Denver. Instead, the impacts of displacement may be occurring on a different scale, be it larger or smaller.
Figure 10 visualizes the change in the share of the Hispanic and Latino population for all Denver neighborhoods between the 2010 and 2020 Censuses. Even though the share of the Hispanic and Latino population increased in Colorado and the United States during this time, the opposite is true for Denver (Wyloge & Goodland, 2021). Many neighborhoods that have light rail lines experienced a decrease in this metric of more than 20%.
5.2: Robustness Check, Limited Data Set
After analyzing the expanded data set, which includes the sixteen stations and adjacent census tracts for a total of 60 entities, an additional check of a more limited data set allows for a zoomed-in perspective. This limited data set, mapped out in figure 9, is significantly smaller as it includes only the census tracts that directly contain stations. Using data sets with fewer observations presents several concerns, such as sensitivity to outliers or model overfitting.
Despite the challenges posed by smaller datasets, this analysis is justified because the impacts of transit-oriented development are sometimes highly localized (Swenson & Dock, 2004).
Comparing the broader neighborhood results to the immediate station area results allows for a better analysis of the intensity and scope of station treatment.
The analysis of the limited dataset includes the same fixed effects as in the expanded dataset analysis for each census tract to control for the factors that vary between tracts but are constant over time. Similarly, year-fixed effects are introduced to account for variables that are common across groups but shift over time. Each control and treatment station is represented only by the
5.2.1: Median income, limited
Table 11 (n = 96) contains the coefficient estimate for the interaction term in the regression analysis for median household income is -$17,796 and is statistically significant. This is a much more dramatic decrease than the -$7,000 estimate in the expanded dataset. Since this dataset contains individual census tracts, the geographical analysis is more refined, and the impact of the transit-oriented housing developments is likely to have a more significant impact on the surrounding area.
This estimate suggests that lower-income households are likely moving into the area due to the affordable housing project and proximity to public transportation. Wages and economic opportunities grew across the entire Denver metropolitan area from 2010 to 2019 at some of the fastest rates in the United States (Brennan & Contorno, 2020). Knowing this, it is unlikely that concentrated wage decreases or economic degrowth are plausible explanations for this negative coefficient. The time-fixed effects would also absorb possible concentrated wage changes.
5.2.2 Gross rent as a share of median income, limited
Gross rent as a share of median income, a proxy for housing or rental burden, increased in the regression using the limited dataset, as seen in table 12 (n = 96). Compared to the expanded dataset, where there was only a slight increase in this variable, the limited dataset sees a much more substantial and statistically significant trend. This result goes hand in hand with the previous result about decreasing median incomes to suggest that transit-oriented housing developments attract lower-income households to the area directly adjacent to the treated station. Once again, analysis of trends in the median rent would assist in further interpreting this coefficient.
5.2.3 Share of commuters using single-occupancy vehicles, limited
Table 13 (n=96) contains the coefficient estimate for the interaction term in the regression for the share of commuters using single-occupancy vehicles. It is negative and statistically significant. This differs from the expanded dataset analysis, where the share of commuters driving alone was slightly positive but not significant. This result is in line with the urban planning hypothesis of the effectiveness of a walkshed. The construction of transit-oriented housing development not only increases the concentration of individuals living close to the station in the subsidized housing unit but also spurs additional development nearby.
As more and more households are established within the station walkshed, the feasibility of using alternative forms of transportation becomes greater.
However, this result does not suggest a relative increase in the share of individuals using the adjacent train lines. That is a different variable, separate from this analysis. Rather, this coefficient represents a localized trend away from single-occupancy vehicles that are not only harmful to a city’s local environment but is also the mode of transportation that demands the most significant share of the commuter’s income. This coefficient may be capturing individuals who started biking, walking, carpooling, taking the bus, using ride-sharing apps like Uber and Lyft, taking the train, or any combination of these alternative modes of transportation. Although data are not available for this statistic, a decrease in the usage of single-occupancy vehicles likely corresponds with lower transportation costs, ultimately reducing households' transportation burden.
5.2.4: Share of the Hispanic population, limited
statistically significant. Compared to the expanded dataset, which suggested a negative but non-significant reduction, this is a much stronger indication that the displacement of the Hispanic community in Denver may be taking place due to transit-oriented development.
6.1: Explanation of main results
Comparing the results between the expanded dataset, which includes all tracts within one mile of a treatment station, to the limited dataset, which contains strictly tracts that contain a treatment station, allows for a better, broader analysis of the trends and spillover effects.
Median household income was negative and statistically significant in both datasets.
Gross rent as a share of median household income increased in both data sets, but the coefficient was much more extensive and statistically significant in the limited dataset. The share of commuters traveling to work in single-occupancy vehicles was slightly positive but insignificant in the expanded case, while the limited dataset showed a negative, significant effect. Finally, the share of residents identifying as Hispanic or Latino decreased in both data sets but was only statistically significant in the limited case.
In summary, the data show that the implementation of transit-oriented housing development leads to lower-income residents moving into the treatment area but also exasperates displacement of the Hispanic community. While trends are also apparent at the limited level in the share of commuters traveling to work in single-occupancy vehicles and the gross median rent as a share of household income, these two variables do not pass the crucial parallel trends test as clearly as median income and the Hispanic share of the population variables do.
6.2: Comparison to Existing Literature
This thesis contributes to the previously existing literature by reinforcing that each community is unique and that it is impossible to generalize how these developments will play out. It is, however, to draw comparisons between the results of this study and the conclusions reached by similar studies. For example, Jones and Ley (2016) discovered signs of gentrification in Vancouver after implementing pro-TOD policies. The situations in Denver and Vancouver are similar except for the affordability requirement in Denver. It is possible that the impacts of transit-oriented development in Denver were more favorable for the community due to the affordability requirement. Debrezion et al.’s 2006 conclusion that proximity to rail stations
The Washington, D.C. analysis conducted by Dawkins and Moeckel (2016) is structurally similar to the developments in Denver and yields similar results. The analysis in Washington D.C.
revealed improved access to transit and little to no displacement of low-income individuals. This similarity suggests that an affordability requirement in TOD helps avoid gentrification. While this analysis does not calculate total changes in welfare, as does Balboni et al. (2021), similarities still arise. In their analysis, transit-oriented gentrification was avoided, and residents benefited from improved public transit access. Improvements in Denver’s treatment neighborhoods represent a similar benefit: residents have more transportation options and use expensive modes like private cars less frequently.
6.3: Modern policy implications
Residents of Denver made headlines in late 2020 when they voted in favor of Ballot Initiative 2A, which added 0.25% to the local sales tax rate and established one of the first Climate Protection Funds (CPF) in the United States. Denver’s CPF generates more than $40 million annually and serves to help finance the city’s aggressive climate plan by supporting climate adaptation projects, investing in low-carbon infrastructure, and fostering a green job market, along with other endeavors. Notably, the bill approved by voters states that the CPF
“should, over the long term, endeavor to invest fifty percent (50%) of the dedicated funds directly in the community with a strong lens toward equity, race, and social justice” (Office of Climate Action, Sustainability, and Resiliency, 2021).
The average person of color lives in a census tract with a higher average surface temperature than their white, non-Hispanic counterparts in 169 of the largest 175 cities in the United States (Hsu et al., 2021, Climate Central, 2021). In Denver, the arid summers result in a situation where mid-afternoon temperatures vary by as much as +20 degrees Fahrenheit, depending on which neighborhood an individual lives in. Geologists and climate scientists predict that by 2080, Colorado will look and feel more like the state’s much warmer neighbor to the southwest, Arizona (Talsma et al., 2022). This climate trajectory is also predicted to disproportionately impact low-income and majority-minority communities the hardest, a trend that is consistent throughout the literature.
Denver’s recently published five-year CPF plan lists low-carbon housing and transportation projects as acceptable uses for the newly generated tax revenue. A record-breaking spending package on housing is currently addressing the city's affordable housing shortage. In addition, the region is simultaneously developing the country's most extensive voter-approved transit expansion program. These two construction booms provide a
unique opportunity to accomplish two goals simultaneously by investing in intelligent, transit-oriented development projects.
Despite the strong signaling from voters about their willingness to fund and support the city’s transition to a cleaner environment, many lifelong Denverites remain skeptical or even in opposition to this program. The city’s rapid growth and skyrocketing property prices caused some residents to believe that CPF investments in their neighborhoods would lead to gentrification and the eventual displacement of them and their neighbors. Alfonso Espino, a community activist in a predominantly Latino neighborhood in north Denver, responded to a New York Times interview by saying, “It’s always just felt more like it’s a whole front. Not for us, you know. It is for the people that are coming” when asked about potential CPF investments in his area (Penney, 2020). The city’s long legacy of historical injustices cannot be fixed overnight, and it must work diligently to regain the trust of its disenfranchised residents. This analysis shows that including certain economic protections, such as affordable housing requirements, may help the city on its journey towards social and environmental equity.
The most significant limitation that restricts the interpretation of the regression results is the five-year average nature of the data. The United States Census Bureau releases five-year averages for smaller geographical areas because they are based on smaller sample sizes and carry the risk of having large margins of error. Additionally, the Bureau cites privacy issues as reasons for restricting access to some data. However, structuring the pre and post-treatment periods in a way that none of the years within the five-year averages overlap at least allows the analysis to ensure that the impact of the independent treatment variable is not lost within the data. Ideally, single-year estimates for each census tract would be utilized to run the same analysis.
Identifying common pre-treatment trends is complicated because the data is subject to a relatively high margin of error. Particularly among the dependent variables of median gross rent as a percentage of household income and the share of single-occupancy vehicle commuters, the data points are so sensitive that even one skewed year may cause complications verifying the common trends assumption. These two variables generally follow similar pre-treatment trends. However, the sensitivity of the sample collection results in insignificant results for all
Testing multiple hypotheses can sometimes complicate the interpretation of regression output results. When multiple hypotheses are tested, as in this analysis, there is a possibility that at least one of the results will be significant due to chance. The probability of observing at least one significant due to chance in this analysis is:
𝑃(𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡 𝑟𝑒𝑠𝑢𝑙𝑡) = 1 − 𝑃(𝑛𝑜 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡 𝑟𝑒𝑠𝑢𝑙𝑡𝑠)
= 1 (1 − 0. 05)4
= 0.1854, an 18.54% chance of observing a falsely significant result. A future analysis that accounts and adjusts for this possibility would add additional credibility to the research design.
6.5: Potential next steps:
The study of the impacts of transit-oriented development is a relatively new field of research, and there are many opportunities for further study. There are several ways to augment this analysis to understand the results better. For example, access to data that differentiates between poor and wealthy individuals, as in the Balboni paper. We see median income decreases in treatment tracts compared to control tracts, but the exact source of this decrease is difficult to pinpoint. It could be that wealthier residents move away as transit-oriented development takes place and that TOD causes the opposite of gentrification. Conversely, it could be that the improved transportation and affordable housing options directly attract lower-income individuals who benefit from these investments the most. The ability to individually tract people who move to and from treatment tracts would assist in this analysis.
From a data perspective, this analysis could be improved by having annual data estimates instead of five-year averages. Although it was possible to isolate the pre and post-treatment effects by excluding values where these two time periods overlapped, the treatment effect may be somewhat diluted. Excluding overlapping periods required eliminating hundreds of potential data points, which annual estimates would solve—augmenting the data set with additional years and furthering the study's credibility. Since this analysis deals with development and neighborhood trends, it would be ideal to revisit this study in 5 to 10 years to understand better how the results hold up over time.
Conducting a similar analysis with affordable housing projects along significant bus routes would also be exciting and could lead to a larger dataset. Light rail stations serve a different purpose than bus stops which could impact the surrounding area in unique ways.
However, bus stations tend to be much smaller, and the impact of bus stop transit-oriented development may be more difficult to isolate than rail stations.
Running this analysis with information about combined rent and transportation burdens would provide valuable insight into the broader economic effects of transit-oriented development on households. One of the primary goals of the Denver Transit-Oriented Development Fund was to assist Denverites who were spending more than 50% of their monthly income on transportation and housing combined. These two categories are consistently the most significant monthly expense for the median American household. Therefore, implementing a policy that lowers rental burden or transportation cost burden independently is not as desirable as a policy that lowers both of these expenditure burdens at the same time.
Finally, adding an investigation into the total population changes for every tract to understand if the shrinking share of Hispanics is due to a displacement of Hispanic individuals or just a faster-growing non-Hispanic population. These two outcomes warrant very different policy responses from an equity point of view. For example, suppose the share of Hispanics is decreasing due to a rapid influx of non-Hispanic individuals, but not because of the displacement of current residents. In that case, policy should focus on improving equitable access to new opportunities in the area. On the other hand, suppose the shrinking share of the Hispanic population is indeed a result of displacement. In that case, city officials should shift their emphasis toward finding ways to promote development that does not come at the expense of long-time residents. This could be by introducing rental protection for long-term residents to ensure that they are not priced out of the market by transit-induced gentrification. This is particularly important in cities like Denver, Austin, and Miami, which have historically faced problems relating to this issue.
This paper attempts to infer the economic and social casual effects of transit-oriented affordable housing developments on the communities where they are built. A difference in differences model is employed to examine the construction of five transit-oriented development sites adjacent to light rail stations. The developments are partially financed by the Denver Transit-Oriented Development Fund on the condition that the newly-constructed developments satisfy housing affordability requirements. Median household income, gross rent as a share of median household income, commuters traveling in single-occupancy vehicles, and the share of the Hispanic or Latino population are used to gauge these social and economic impacts.
Transit-oriented development is a popular concept in urban planning and is touted as a potential solution to several of the problems facing modern cities. As cities spend more public funds to subsidize or encourage these developments, concerned citizens are rightfully asking questions about the consequences of transit-oriented developments. The findings of existing literature focusing on the economic and social implications of transit-oriented development are mixed. Some studies suggest that TOD causes gentrification and displacement, disproportionately impacting low-income and non-white populations. Other studies find that TOD results in welfare improvements across the community, with low-income individuals benefiting the greatest. Regardless of the results, most existing literature focuses on developments constructed in tandem with the adjacent rail station and regulated by natural market forces. This study is unique because it analyzes projects explicitly required to meet housing affordability guidelines developed after constructing the adjacent rail station.
Regressing for income reveals a statistically significant decrease in the median income of households in treatment areas compared to control areas. Wages rose rapidly across the metropolitan area during the timeframe of this analysis, but the increase was much higher in control areas. This indicates that low-income households are attracted and/or able to remain in the treatment areas. Regressing gross rent as a share of household income yields a significant increase at the localized level. This result suggests that individuals in the treatment areas spend more on their monthly rent than those in control areas. Given the results of the regressions on household income, it is likely that lower-income people can qualify for housing and move into treatment areas, resulting in a higher proportion of income going towards rent. However, additional research can be done to determine if this result is caused by migration or rising rents.
Regressing the share of commuters that drive alone reveals that residents in treatment areas use single-occupancy vehicles less frequently than their peers in control areas. This result is
likely due to access to alternative transportation options and proximity to jobs. However, this effect is limited to the immediate area and disappears in areas less than a mile from the development. Finally, regressing the share of the Hispanic or Latino population reveals that this figure decreases in treatment areas compared to control areas. This suggests that Hispanic and Latino residents are either leaving or being displaced from the development areas, a potentially alarming result given Denver’s past and present trends. In summary, implementing transit-oriented affordable housing developments appears to be a net positive for the community but raises questions about the displacement of the city’s Hispanic and Latino population.
Potential next steps for this analysis include collecting annualized data locally. While this analysis separated the pre and post-treatment periods, the five-year average nature of American Community Survey data complicates the regressions. Revisiting this analysis after more years have passed would also strengthen the results. These projects were completed in the early to mid-2010s, and gentrification/displacement effects can sometimes take years to play out fully. In addition, collecting data on the migration in and out of treatment areas would help clarify the results of this analysis. Finally, a more extensive, nationwide analysis of programs constructing transit-oriented developments with affordable housing requirements would help expand the dataset. Denver is just one midsize city, and it is entirely possible that similar programs would yield different results under alternate circumstances.