New rail transit stations and the out-migration of low-income residents

Abstract

Chinese

This article tests the hypothesis that low-income residents disproportionately move out of neighbourhoods in close proximity to new rail transit stations. This transit-induced gentrification scenario posits that the development of rail transit will place an upward pressure on land and housing values and that higher-income residents will outbid low-income residents for this new amenity. The most transit-dependent population may therefore be displaced from the most accessible locations, forming a paradox in the investment in new transit systems. We test this hypothesis using the Panel Study on Income Dynamics (PSID) dataset to trace the out-migration of residents across the United States from census tracts within five years of the opening of a new station, between 1970 and 2014. We find that low-income individuals are more likely to move, regardless of their neighbourhood. However, we do not find significant evidence that low-income individuals are more likely to move out of transit neighbourhoods, after controlling for both individual and other neighbourhood characteristics. The odds of moving out of a transit neighbourhood for low-income residents is statistically insignificant. In other words, they do not have a heightened probability of leaving new transit neighbourhoods compared with other residents. Our results are robust across decades, when examining renters alone, for different time spans and for varying definitions of transit neighbourhoods. We further find that those living in transit neighbourhoods are not more likely to live in a crowded dwelling. Our results therefore suggest that, on average, across the nation, low-income residents do not disproportionately exit new transit neighbourhoods.

Keywords

demographics displacement gentrification public transit residential mobility transport

Introduction

In February 2018, the Los Angeles Times published an opinion piece entitled ‘Transit-oriented development? More like transit rider displacement’ (Rosenthal, 2018) in which the author argues that the decline of public transit ridership in the city is caused by the displacement of low-income families from neighbourhoods around transit lines. The article cites several studies that have associated new transit stations with increases in neighbourhood home values and rents, concluding that ‘transit-oriented development might as well be called transit-rider displacement’. Missing from the argument were any statistics on the extent to which the displacement of low-income individuals around new transit stations occurs. This is true not only in the popular discourse, but in the academic literature as well. While threats of displacement have inspired housing advocates to counter new transit initiatives across the country, to date no published study has empirically estimated how new transit developments contribute to the displacement of low-income individuals (Rayle, 2015).

Some of the resistance to new transit projects stems from an apparent paradox initiated by their development. On the one hand, transit and its associated access provides an opportunity to connect auto-less residents with jobs and urban opportunities, potentially lifting their economic prospects. On the other hand, urban economic theory predicts that increased accessibility will be capitalised into land and housing values, and so the low-income individuals who would benefit the most from the new transit investment may no longer be able to afford to live nearby and utilise these services. This phenomenon, also referred to as transit-induced gentrification, has received an increasing amount of attention in the literature, as cities have progressively turned to transit as an urban redevelopment strategy (Ferbrache and Knowles, 2017; Schwanen, 2015). While a large body of research has been devoted to analysing price capitalisation effects (e.g. Debrezion et al., 2007), largely supporting the idea that transit does place an upward pressure on home values, comparatively little research has examined how these changes affect residential sorting around transit stations.

In this article, we contribute to this literature by analysing whether low-income individuals have a higher propensity to move out of a neighbourhood as a new fixed rail transit station is put into place. Our analysis utilises the Panel Study on Income Dynamics (PSID) to model residential movements out of neighbourhoods, as proxied by census tracts, in metro areas across the United States from 1970 to 2014. We control for both individual- and neighbourhood-level characteristics in a multi-level modelling framework. At this national scale, we find no evidence that low-income individuals, on average, have a heightened out-migration rate from transit neighbourhoods. Low-income residents in general have higher mobility rates, but those living in neighbourhoods close to new transit stations have statstically identical odds of leaving transit neighbourhoods as compared with other neighbourhoods.

The article begins with a conceptual framework for our analysis, followed by a review of the empirical literature on the impacts of transit on home values and neighbourhood trajectories. The next section contains an explanation of our data and analytical strategy, followed by our results and conclusions, which include a discussion on the implications and limitations of our analysis.

Transit investments and residential mobility

Conceptual framework

Transportation, and the accessibility it enables, has long held a prominent place in neoclassical economic theories explaining the residential sorting of individuals by income across urban areas. The price of land and housing, and subsequent residential location decisions, have traditionally rested on the notion that highly accessible locations such as central business districts will generate the most demand, and therefore the highest rents, and that households will make a trade-off between commuting costs and land prices. Wealthier residents able to afford higher commuting costs will choose to maximise their housing consumption farther from the city centre, while poorer residents with less income to spend on transportation will live in smaller, more accessible locations (Alonso, 1964; Mills, 1967; Muth, 1969). However, a time–cost-based pull may attract high-income residents towards more central locations (Brueckner and Rosenthal, 2009; Glaeser et al., 2008). In addition to accessibility benefits, rail transit tends to attract commercial services such as retail which serves as a further benefit to residents, extending to those who may not directly benefit from the accessibility derived from riding the train itself (Bowes and Ihlanfeldt, 2001).

Tiebout (1956) espoused the idea that households will make location decisions based on the bundle of available public goods (and taxes) and sorting will occur based on the demand for these amenities and the willingness and ability of households to pay for them. Therefore, given that access to transit is viewed as an amenity by residents, capitalisation benefits can be expected in neighbourhoods and properties surrounding stations, thereby impacting the spatial sorting of residents by income. Per this vantage point, if housing costs rise around stations, lower-income residents may be displaced as higher-income residents outbid for these high demand locations. Even if low-income residents are not disproportionately displaced from potentially gentrifying transit neighbourhoods, the overall landscape is still impacted by rising rents in transit neighbourhoods as the number of potential locations for low-income residents to move into is reduced, again furthering the spatial concentration of disadvantage (Newman and Wyly, 2006).

Empirical studies

The existing literature has overwhelmingly focused on the relationship between transit and either land values or neighbourhood-level socioeconomic conditions. With respect to the former of these relationships, the rather large body of literature investigating the effects of new transit stations on surrounding property values has generally found supporting evidence of price capitalisation effects around transit stations. However, these effects vary across metropolitan environments and with respect to the location and type of neighbourhood within the metropolitan area (Atkinson-Palombo, 2010; Billings, 2011; Debrezion et al., 2007; Knowles and Ferbrache, 2016; Mohammad et al., 2013). Weaker housing markets and economic climates, both at the metropolitan and neighbourhood scale, have been shown to dissipate potential positive capitalisation benefits (Hess and Almeida, 2007).

These property-level effects have translated to some observable residential sorting outcomes at the neighbourhood scale, as several studies have concluded that neighbourhoods in close proximity to a new transit station have a heightened probability of undergoing increases in median incomes, home values, the share of college-educated residents and multi-family housing construction – or those characteristics most often associated with gentrification (Bhattacharjee and Goetz, 2016; Deka, 2017; Kahn, 2007; Nilsson and Delmelle, 2018; Pollack et al., 2010). While the probability of gentrification-type changes is greater in new transit neighbourhoods than in similar neighbourhoods in a metropolitan area, the research to date has emphasised that the majority of transit neighbourhoods do not undergo dramatic changes in their socioeconomic composition in the decade following the placement of the station (Nilsson and Delmelle, 2018). Like property value capitalisation, neighbourhood changes also vary by type of neighbourhood, metropolitan environment and station type (park-&-ride vs. walk-&-ride, or transit-oriented development vs. transit-adjacent development) (Baker and Lee, 2019; Kahn, 2007; Nilsson and Delmelle, 2018). A growing consensus among this body of literature is that neighbourhood socioeconomic ascent following the placement of a new station is highly context dependent, particularly in respect of the conditions of the neighbourhood prior to the station opening.

A lesser-studied link in the transit-induced gentrification debate is an empirical analysis on how residential movements give rise to these aggregate, neighbourhood outcomes (Rayle, 2015; Zuk et al., 2018). Few studies have explicitly investigated the effect of public transit investments on residential mobility, and they have thus far overwhelmingly focused on the characteristics of those moving into new transit corridors. One exception is a recent study by Rodnyansky (2018), who examined residential displacement in the case of rail transit in Los Angeles County. Using a dataset of tax filers in the county over 21 years, at the census block level, he found no evidence of a disproportional out-migration of residents after controlling for other personal characteristics (income changes, age, dependants, marital status). Rather, he found that residents of all incomes were less likely to move out of a block following the opening of a new transit station.

With respect to studies that have examined the characteristics of those living in a new transit corridor, Liu et al. (2016) surveyed 1023 passengers along the Hudson–Bergen light rail system in New Jersey, eight years after its opening. They found that passengers who had recently relocated to the new transit corridor were younger and had smaller household sizes, lower incomes and lower rates of car ownership. Higher-income earners tended to live nearer to the station. Lund (2006) surveyed residents who had recently moved into a transit-oriented development (TOD) neighbourhood in the San Francisco Bay area, San Diego and Los Angeles, and found that compared with the general population, TOD residents had higher incomes and were less likely to be Hispanic. Overall access to amenities either by car, foot or transit was also acknowledged as an important factor in deciding with reside there. Cao and Schoner (2013) surveyed residents who moved into the light rail corridor along the Hiawatha system in Minneapolis. They found that compared with residents who lived there before the transit line opened, newer residents had higher education levels, were younger and were more likely to be renters. These individual findings therefore support the changes observed at the neighbourhood scale where increases were observed in younger, educated and wealthier residents. Whether or not these influxes are accompanied by an out-migration of lower-income residents has yet to be determined.

While few studies have examined the out-migration of residents from transit neighbourhoods (Rayle, 2015; Zuk et al., 2018), there are studies that have examined residential mobility with respect to gentrifying neighbourhoods more generally. The causal logic behind displacement due to gentrification is as follows: gentrification increases demand for properties in the area, causing property prices to increase, which in turn increases market values of comparable properties in the area, which increases assessed values of those properties for tax purposes, resulting in increased property tax liability and rents which may force current residents to move (Martin and Beck, 2018). Limited quantitative evidence has been found to substantiate the idea that lower-income residents, who already have elevated rates of mobility owing to more precarious financial situations and a heightened risk of evictions (Newman and Wyly, 2006), are more likely to move out of gentrifying neighbourhoods than other types of neighbourhoods (Ding et al., 2016; Ellen and O’Regan, 2011; Freeman, 2005; Martin and Beck, 2018). Some evidence has suggested that lower-educated residents have higher out-migration rates from gentrifying neighbourhoods as compared with higher-educated households (McKinnish et al., 2010). Consistent with broader, national intra-urban migration statistics, Ellen and O’Regan (2011) find that younger residents, renters and minorities have a higher likelihood of moving out of a neighbourhood, whether it be gentrifying or not.

Our analysis takes a nationwide perspective on the movement of individuals out of transit neighbourhoods since the 1970s. We focus on whether or not lower-income residents have an elevated propensity to leave within five years before or after the station has opened, while controlling for other individual and neighbourhood characteristics.

Empirical approach

Data

Our analysis uses data from the Panel Study on Income Dynamics (PSID), the longest-standing representative longitudinal population survey in the United States, to trace residential movements out of transit neighbourhoods throughout the United States since 1970. Aside from Freeman’s (2005) influential study, the PSID has been used extensively by researchers to study the effects of neighbourhood conditions on residential mobility (see McGonagle and Sastry (2016) for a review).

For this study, we used a merged dataset that combines the geocoded version of the PSID data with data from the Center for Transit-Oriented Development (CTOD) on when and where fixed rail stations were opened.¹ The CTOD data does not contain opening dates for stations prior to 2000, so we updated this database with the opening years for all fixed rail stations in the US. We identified a ‘transit tract’ as a census tract that intersected a 0.25 mile buffer around a transit station. The census tract is the smallest geographic unit in which individuals are identified in the PSID dataset, necessitating their use as a proxy for an individual’s neighbourhood. The use of a 0.25 mile buffer is simply a means of selecting those census tracts in closest proximity to the transit station. Direct economic benefits around rail transit stations in the form of new developments may occur in the immediate proximity around a station, while secondary, indirect effects including property value increases stemming from spatial proximity to both the station and new developments are expected to occur beyond this immediate area. Research on price capitalisation of new transit stations has shown positive impacts to occur a mile or more from new stations (Billings, 2011; Bowes and Ihlandfeldt, 2001; Debrezion et al., 2007). Thus, while imperfect neighbourhood proxies, based on our theoretical framework whereby increases in property values and rents may spur the disproportionate out-migration of low-income residents, the use of census tracts encompasses the broader area in which economic impacts may be felt. As robustness checks, we also varied the definition of a transit tract by only considering those census tracts that intersected a 0.25 and 0.5 mile service area, as well as whether 25%, 50% and 75%, respectively, of a tract’s total area was covered by a rail transit station’s half mile network-based walking service area.

Our timeframe of interest is five years before or after the station had opened, to narrow our analysis on the impact of new rail transit stations on residential mobility. This time range was guided by findings in the literature that changes in property values in neighbourhoods often precede the opening of the transit line, as the announcement of a major public investment such as rail transit can serve as a promise of future development and hence spur private investments in affected neighbourhoods (Billings, 2011). We tested the sensitivity of this time range to examine differences in out-migration before and after the station opened; three years before and after; and with no time restrictions.

Merging the PSID and transit datasets resulted in a sample of 989 household heads who lived in a census tract (at some point) five years before or after the opening of a fixed rail transit station between 1970 and 2013. Our final panel dataset includes records of these household heads for all the years they are included in the PSID survey, resulting in a total of 11,976 person–year observations.

For all individuals, we have information on the census tract they reside in, plus their employment status, income and earnings, demographic and residential characteristics, as well as educational attainment and family situation. The literature suggests several factors that could influence residential mobility, including socioeconomic status (income, educational attainment), life cycle factors (age, marital status, family status), housing satisfaction (crowding, unit condition), tenure status and unanticipated changes (employment status shifts, financial problems) (Crowder et al., 2012; Ding et al., 2016; Martin and Beck, 2018). We also control for neighbourhood housing and population characteristics. Census tract data come from Brown University’s Longitudinal Tract Database (LTDB) which contains estimates of tract-level socioeconomic and demographic variables within 2010 tract boundaries for years 1970 to 2010 (Logan et al., 2014).

To measure affordability of neighbourhoods, we include housing prices (median home value) and socioeconomic status of residents in terms of share of college-educated residents. Similar to Ding et al. (2016), we argue that housing values and rents reflect the quality of various amenities and investment in the neighbourhood and indicate the changing affordability of neighbourhoods. We also rely on the share of college-educated residents rather than incomes to reflect the socioeconomic status of a neighbourhood for two reasons: (1) these two measures tend to be highly correlated; and (2) in the cases where they are not, education is perhaps a better marker of socioeconomic status, as income may overlook neighbourhoods experiencing an influx of young, highly educated but low-paid professionals (e.g. college students, artists, young professionals, etc.) (Ding et al., 2016; Freeman, 2005).

Estimation framework

For the analysis, our dependent variable is a residential move out of a census tract reported by an individual. To test the hypothesis that low-income residents are more likely than other residents to move out of neighbourhoods following the placement of a fixed rail station, we regress our dependent variable on a set of individual and neighbourhood factors that are likely to influence residential mobility, described in Table 1.² Since the data used in this study are hierarchically structured, with individuals constituting level one and neighbourhoods (census tracts) constituting level two, we apply a multi-level modelling approach which enables us to control for unmeasured tract-level effects by allowing the intercept to vary across tracts. Given the binary nature of the dependent variable, our models have the following logistic probability form:

\begin{matrix} \ln [\frac{P_{Moved}}{1 - P_{Moved}}] = b_{00} + b_{10} Lowin c_{it} + b^{'} X_{it} \\ + b^{″} Z_{jt} + b_{01} Transi t_{j} + b_{11} Transi t_{j} \\ \times Lowin c_{it} + u_{0 j} + e_{ij} \end{matrix}

(1)

where i indicates i the individuals, j the census tract and t time. P_Moved is the probability of an individual i moving out of census tract j in year t. Lowinc_ijt is a dummy variable equal to one if the respondent was considered low-income (had an annual income less than 60% of county per capita personal income) in year t.³Transit_j is a dummy indicating whether census tract j has recently received or is about to receive a rail transit station (within a ± 5-year time window of station opening). To explicitly test the hypothesis that low-income individuals are more likely to move out of new rail transit neighbourhoods, we include an interaction variable between Lowinc_ijt and Transit_j. X _it and Z _it are column vectors of individual-level and tract-level characteristics in year t. u and e are random error terms at the level of the census tract and individual level, respectively. We also include a set of decade dummies to control for potential trends in mobility rates over time, as well as county and decade specific annualised population growth rates to reflect demand and competition for housing in the county of residence.

Table 1.

Descriptive statistics (all monetary values in 2013 US$).

	All	Mean (std. dev.)Low income(< 60% of local PCI)	Middle–high-income(> 80% of local PCI)	Difference in means
Number of observations	11,976	5782	4254
INDIVIDUAL
Female (1 if yes)	0.50 (0.50)	0.66 (0.47)	0.28 (0.45)	−0.33***
Age	43.81 (15.55)	47.94 (17.70)	39.64 (11.06)	−8.54***
Family size	2.72 (1.87)	2.59 (1.90)	2.90 (1.84)	0.26***
Children (1 if yes)	0.46 (0.50)	0.42 (0.49)	0.51 (0.50)	0.08***
Married or permanently cohabiting (1 if yes)	0.30 (0.46)	0.15 (0.36)	0.51 (0.50)	0.31***
Black (1 if yes)	0.62 (0.49)	0.74 (0.44)	0.44 (0.50)	−0.25***
White (1 if yes)	0.35 (0.48)	0.22 (0.42)	0.53 (0.50)	0.26***
Income (US$1000)	84.85 (739.71)	5.12 (7.76)	203.24 (1170.83)	165.16***
High-school diploma (1 if yes)	0.20 (0.40)	0.21 (0.41)	0.19 (0.39)	−0.01
Bachelor’s degree or higher (1 if yes)	0.40 (0.49)	0.31 (0.46)	0.52 (0.50)	0.18***
Weeks unemployed last year	1.27 (5.42)	1.72 (6.83)	0.66 (2.99)	−0.93***
DV: moved from tract in current year? (1 if yes)	0.21 (0.41)	0.21 (0.41)	0.19 (0.40)	−0.01
Moved from rail transit tract in current year? (1 if yes)	0.09 (0.29)	0.10 (0.30)	0.08 (0.27)	−0.02**
Mobility rate (# of moves observed/years observed)	0.28 (0.22)	0.28 (0.22)	0.27 (0.20)	0
Own home (1 if yes)	0.32 (0.47)	0.21 (0.41)	0.49 (0.50)	0.23***
House value (US$1000)	68.73 (162.99)	36.02 (124.19)	121.53 (204.09)	64.81***
Rent (US$1000)	0.52 (0.33)	0.42 (0.29)	0.68 (0.34)	0.23***
Property tax (US$1000)	2.49 (2.43)	2.05 (2.59)	2.81 (2.36)	0.66***
Crowded unit (1 if yes)	0.09 (0.29)	0.11 (0.32)	0.07 (0.25)	−0.04***
NEIGHBOURHOOD
Rail transit	0.54 (0.50)	0.57 (0.49)	0.50 (0.50)	−0.07***
Recent in-movers (%)	67.24 (14.22)	66.63 (14.01)	67.97 (14.43)	1.27***
Black (%)	49.97 (41.28)	60.57 (38.84)	34.70 (39.89)	−22.06***
College graduates (%)	15.60 (15.17)	12.91 (13.76)	19.70 (16.56)	5.61***
Houses built ≥ 30 years (%)	52.59 (28.94)	57.41 (26.84)	46.12 (30.29)	−10.03***
Multi-unit housing (%)	44.70 (27.46)	47.79 (27.41)	39.86 (26.90)	−6.44***
Median home value (US$1000)	72.60 (86.56)	65.49 (82.41)	84.28 (94.72)	14.76***

Notes: Difference in means t-test significance: *** 1%, ** 5%, * 10%.

Results

Table 1 shows descriptive statistics of our variables for all observations and segmented by low income (income less than 60% of county per capita personal income) and medium–high-income (income greater than 80% of county per capita personal income) individuals. While the sample appears nationally representative with regards to gender, age and family size distribution, the share of black and low-income residents is higher than what might be expected. This bias is likely to stem from the data itself. The original focus of the PSID was on the dynamics of poverty, and hence a disproportionately large number of low-income households were sampled at the beginning of the study in 1968. This oversampling of lower-income families in the late 1960s resulted in a sizable sub-sample of blacks (Hill, 1991).

For all individuals, our dependent variable (DV), the probability of moving out of a census tract from one time period to the next, is 0.21. This value is statistically equivalent across income groups. There is a statistically significant difference in means in the probability of leaving a transit tract between the low-income and medium–high-income groups, with the former slightly more likely to move out (0.10 vs. 0.08). Low-income individuals are also more likely to live in a rail transit neighbourhood in general (0.57 vs. 0.50).

The differences reported in Table 1 do not account for other individual and neighbourhood factors which may influence a person’s decision to move. Table 2 shows the results of the logistic multi-level regression models of residential mobility between 1970 and 2013. All results are reported as odds ratios for ease of interpretation. Model 1 only includes individual-level variables, decade dummies and the rail transit tract indicator. Model 2 adds neighbourhood factors to the specification in Model 1. In Model 3, the interaction variable between the low-income dummy and the rail transit tract dummy is added, and hence Model 3 represents the full model specification outlined in equation (1). Overall, the resulting coefficients and their significance are robust across model specifications. Our results are consistent with previous research, in that younger residents, renters and those living in more crowded units⁴ have a higher likelihood of moving out of a neighbourhood (Ellen and O’Regan, 2011; Martin and Beck, 2018). Larger families, and married and female household heads are less likely to move from one time period to the next. Consistent with the literature, in general, low-income individuals are more likely to move regardless of their neighbourhood. However, we do not find significant evidence that low-income individuals are more likely to move out of transit neighbourhoods, after controlling for both individual and other neighbourhood characteristics. The odds of moving out of a transit neighbourhood for low-income residents is statistically insignificant. In other words, they do not (on average) have a heightened probability of leaving these neighbourhoods compared with other residents.⁵

Table 2.

Estimated odds ratios.

DV: moved from neighbourhood	Model 1	Model 2	Model 3
INDIVIDUAL
Age	0.9654***	0.9651***	0.9651***
Family size	0.9546**	0.9532**	0.9527**
Crowded unit	1.3837***	1.4112***	1.4127***
Own home	0.4286***	0.4259***	0.4265***
Bachelor’s degree or higher	1.0821	1.0597	1.0582
Low income (< 60% of local PCI)	1.0892*	1.1126**	1.1488**
Female	0.8793**	0.8776**	0.8773**
Minority or black	1.0615	1.1025	1.1002
Married	0.8306**	0.8158***	0.8146***
Children	1.0428	1.0423	1.0441
Weeks unemployed	1.0024	1.0020	1.0020
NEIGHBOURHOOD
Rail transit ± 5 years	1.0119	1.0264	1.1077
Rail transit ± 5 × low income			0.8666
Recent in-movers (%)		1.0057**	1.0058**
Black (%)		0.9997	0.9997
College graduates (%)		1.0053*	1.0053**
Houses built ≥ 30 years (%)		1.0018	1.0017
Multi-unit housing (%)		0.9973**	0.9973**
Median home value (US$1000)		0.9995	0.9995
Cnty population growth (%)	1.0356	1.0204	1.0205
TIME PERIOD
1980	1.0281	1.0318	1.0328
1990	1.2803***	1.3081***	1.3086***
2000	1.9265***	1.8873***	1.8887***
2010	0.8460	0.8784	0.8787
Constant	1.3882***	0.9273	0.9141
Number of observations	11,811	11,541	11,541
χ ²(17/23/24)	638.97***	641.18***	642.16***
Tjur R²	0.21	0.21	0.21
Log likelihood	–5534.69	–5403.24	–5402.57
Variance of intercept (sd)	0.58 (0.06)	0.53 (0.06)	0.53 (0.06)
LR test (vs. logistic model)	132.88***	116.74***	116.51***

Notes: Statistical significance at the *** 1%, ** 5% or * 10% level.

To further test the robustness of these results, we varied the definition of low income,⁶ the definition of a rail transit tract and the timing, to examine mobility rates three years before and after the opening, five years before and after the opening, as well as without regards to when the transit station opened in relation to their move. Table 3 reports on the two coefficients of interest (estimates of all other variables are nearly equivalent to those reported in Table 3). The results in Table 3 only differ from those in Table 2 when we do not consider the timing of the move in relation to the station opening. In that case, residents in transit tracts – regardless of income level – are significantly less likely to move. In all other cases, our results remain similar to those in Table 2, and the conclusions stay the same. In each model specification, low-income residents, regardless of how they are defined, have a higher probability of moving out of a census tract from one time period to the next. However, when looking at low-income residents in transit neighbourhoods, the odds of leaving are lower, but also statistically insignificant.

Table 3.

Estimated odds ratios of Model 3 for different low-income, timing and rail transit definitions.

Different low-income specification (keeping Rail transit ± 5 years)
Low income (< 30% of local per capita income)	1.1294**
Rail transit ± 5 years × low income	0.8399
Low income (< 80% of local per capita income)	1.1771***
Rail transit ± 5 years × low income	0.8852
Low income (< 2 × federal poverty level)^a	1.1281**
Rail transit ± 5 years × low income	1.0893
Different station opening specification (keeping Low income < 60% of local per capita income)
Rail transit ± 3 years	1.0404
Rail transit ± 3 years × low income	0.8943
Rail transit 5 years before	1.0623
Rail transit 5 years before × low income	0.8353
Rail transit 5 years after	1.1243
Rail transit 5 years after × low income	0.9293
Rail transit (regardless of time)	0.6975***
Rail transit (regardless of time) × low income	1.0096
Different rail transit tract specification (keeping Low income < 60% and ± 5 years of opening)
Within ½ mile service area	1.0422
Within ½ mile service area × low income	0.9930
Within ¼ mile service area	1.0267
Within ¼ mile service area × low income	1.0156
25% within ½ mile service area	1.0067
25% within ½ mile service area × low income	1.0095
50% within ½ mile service area	1.0008
50% within ½ mile service area × low income	0.8767
75% within ½ mile service area	0.7541
75% within ½ mile service area × low income	1.1456

Notes: Statistical significance at the *** 1%, ** 5% or * 10% level. ^aRefers to the federal poverty level in 2013, since all monetary values have been inflation adjusted to 2013 dollars.

The results from Model 3 were also tested with regards to the definition of what is considered a transit tract. Instead of whether a tract intersected a 0.25 mile Euclidean buffer around the station, we used a service area of a 0.25 and 0.5 mile walk, where the latter is the standard for rail transit while the former is often used for bus services. We also performed estimations where transit neighbourhoods are defined as being those which have 25%, 50% or 75% of the total tract area covered by a 0.25 mile service area buffer around the station. These definitions increasingly restrict the analysis to those stations located in the smallest and most walkable tracts, eliminating large census tracts that may have little relation to the transit station itself. The resulting odds ratios for the rail transit and interaction term in Table 3 vary somewhat depending on definition; however, they remain close to 1 and are consistently insignificant, as in the original Model 3 presented in Table 2. In conclusion, we find no statistically significant evidence of a difference between lower- and higher-income residents in mobility rates out of transit neighbourhoods.

To get a better understanding of the factors influencing moving decisions across different income groups, we estimate a modified version of the model in equation (1) without the interaction term and we replace the low-income variable with a continuous measure of income as we estimate the equations separately for the low- and medium–high-income groups. Overall, the results, reported in Table 4, are similar for the two income groups. The lack of significance for the crowded unit variable for the higher-income group is likely reflective of the fact that low-income individuals are more likely to live in crowded units and so there are fewer observations for the higher-income group. While being married does not appear to be a major determinant for higher-income earners, married low-income residents are less likely to move than their unmarried counterparts. With regards to the variable of interest, the presence of a rail transit station, there appears to be no significant difference between low-income earners and higher-income earners. The odds of moving out of a transit neighbourhood are not statistically different from other neighbourhoods, after controlling for individual and other neighbourhood-level variables.

Table 4.

Estimated odds ratios for low vs. middle–high-income individuals.

DV: moved from neighbourhood	Renters
INDIVIDUAL
Age	0.9643***
Family size	0.9509**
Crowded unit	1.4469***
Rent (US$1000)	1.0593
Bachelor’s degree or higher	1.0256
Low income	1.1261*
Female	0.9368
Minority or black	1.2228**
Married	0.9936
NEIGHBOURHOOD
Rail transit ± 5 years	1.0598
Rail transit ± 5 × low income	0.9518
Recent in-movers (%)	1.0069**
Black (%)	0.9993
College graduates (%)	1.0075**
Houses built ≥ 30 years (%)	1.0011
Multi-unit housing (%)	0.9956***
Median home value (US$1000)	0.9988*
Cnty population growth (%)	0.9649
TIME PERIOD
1980	1.0288
1990	1.4620***
2000	2.2596***
2010	1.0164
Constant	0.8110
Number of observations	7204
χ ²(24)	308.63***
Tjur R²	0.25
Log likelihood	−3802.64
LR test (vs. logistic model)	65.46***

Notes: Statistical significance at the *** 1%, ** 5% or * 10% level.

Some scholars have argued that transit-induced gentrification and displacement have been more prominent during the ‘transit renaissance’ that has occurred during more recent years (as opposed to earlier waves of transit investments). The coefficients of the decade dummies in the regressions in Tables 2 and 4 indicate significantly higher mobility rates during the 1990s and 2000s (only a few years of PSID data are available for the 2010s, which may explain the direction and significance of their coefficients). To test for differences across decades, we ran the same model segmented by each decade and found the results to be largely stationary through time. The propensity to leave a transit neighbourhood is not escalated in the later decades.⁷

The results in Tables 2 and 4 suggest that renters are generally more mobile than homeowners, a finding that is consistent across the residential mobility literature (Coulson and Grieco, 2013). As argued by Martin and Beck (2018), many homeowners regard their homes as a form of savings, so capital appreciation may keep them in their homes even if this increases their property tax burden in the short run, particularly in cities with property tax limitations. Renters, on the other hand, may be more susceptible to rent increases (or even evictions) and possibly displacement. Studies that find limited evidence of displacement in gentrifying neighbourhoods tend to be those that include both renters and homeowners in their analysis framework (Ellen and O’Regan, 2011; McKinnish et al., 2010). We therefore perform a separate analysis on renters alone. The results, reported in Table 5, are largely consistent with those in Table 2. One result that stands out is the change in significance in the estimated coefficient for the minority variable, suggesting that minorities who rent are (on average) more likely to move than white renters. However, the significance and magnitude of the rail transit and rail transit–low income interaction terms are similar to those reported in Tables 2 and 3. When controlling for other factors, renters are no more likely to move out of transit neighbourhoods as compared with other neighbourhoods, and this effect is the same for low-income and higher-income renters.

Table 5.

Estimated odds ratios for renters alone.

DV: moved from neighbourhood	Renters
INDIVIDUAL
Age	0.9643***
Family size	0.9509**
Crowded unit	1.4469***
Rent (US$1000)	1.0593
Bachelor’s degree or higher	1.0256
Low income	1.1261*
Female	0.9368
Minority or black	1.2228**
Married	0.9936
NEIGHBOURHOOD
Rail transit ± 5 years	1.0598
Rail transit ± 5 × low income	0.9518
Recent in-movers (%)	1.0069**
Black (%)	0.9993
College graduates (%)	1.0075**
Houses built ≥ 30 years (%)	1.0011
Multi-unit housing (%)	0.9956***
Median home value (US$1000)	0.9988*
Cnty population growth (%)	0.9649
TIME PERIOD
1980	1.0288
1990	1.4620***
2000	2.2596***
2010	1.0164
Constant	0.8110
Number of observations	7204
χ ²(24)	308.63***
Tjur R²	0.25
Log likelihood	−3802.64
LR test (vs. logistic model)	65.46***

Notes: Statistical significance at the *** 1%, ** 5% or * 10% level.

Given the reduced tendency for all residents to leave a neighbourhood following the placement of a new transit station, we explore one other plausible explanation for how low-income residents in particular manage to stay in these neighbourhoods. Our theoretical framework suggests that rents will rise in neighbourhoods that recently received or are about to receive a new rail transit station. One solution that low-income residents may adopt in order to stay in these neighbourhoods and benefit from the new transit station is to move into a more crowded housing unit to split the rent with others.

To test whether renters are more likely to live in a more crowded unit in a transit neighbourhood, controlling for other individual and neighbourhood factors, we run a regression similar to the ones reported in previous tables with the dependent variable being whether or not the individual lives in a crowded unit. The results in Table 6 suggest that, with the exceptions of neighbourhood median home value and (to a lesser extent) age of housing stock, neighbourhood characteristics are insignificant in determining whether an individual lives in a crowded unit or not. Not surprisingly, those with low incomes and larger family sizes are more likely to live in crowded quarters, while those with higher rents and higher educational attainment are less likely to live in a crowded unit. Living in a rail transit neighbourhood does not significantly affect the probability of living in a crowded unit, regardless of income status.

Table 6.

Estimated odds ratios for renters living in crowded units.

DV: crowded unit	Renters
INDIVIDUAL
Age	0.9556***
Family size	3.4226***
Rent (US$1000)	0.1833***
Bachelor’s degree or higher	0.6051***
Low income	1.3867**
Female	0.6221***
Minority or black	1.0922
Married	1.0653
NEIGHBOURHOOD
Rail transit ± 5 years	1.0648
Rail transit ± 5 × low income	1.1545
Recent in-movers (%)	0.9989
Black (%)	0.9976
College graduates (%)	0.9910
Houses built ≥ 30 years (%)	0.9922**
Multi-unit housing (%)	1.0034
Median home value (US$1000)	1.0043***
Cnty population growth (%)	0.8698
TIME PERIOD
1980	1.3111*
1990	0.7802
2000	0.6582
2010	1.1283
Constant	0.0138***
Number of observations	7219
χ ²(21)	658.40***
Tjur R²	0.10
Log likelihood	−1384.97
LR test (vs. logistic model)	179.98***

Notes: Statistical significance at the *** 1%, ** 5% or * 10% level.

Discussion and conclusions

In this article, we tested the hypothesis that low-income residents disproportionately move out of neighbourhoods following the placement of a new rail transit station. Prior studies on the impact of new transit investments have largely focused on how they have influenced land and home values, with the presumption that observed increases in these values will have a noticeable influence on residential sorting. To date, few studies have empirically examined the characteristics of those that move out of new transit neighbourhoods to provide supporting evidence to the transit-induced gentrification and displacement discussion. Our analysis used the Panel Study on Income Dynamics dataset to examine the movement out of neighbourhoods, as proxied by census tracts, across the United States from 1970 to 2013. We initially defined a transit neighbourhood as one that intersected a 0.25 mile buffer around a station five years before or after its construction, or those tracts in closest proximity to the new stations, and performed robustness checks on our results by varying this definition. We used a multi-level logistic regression to control for the effect of individual, neighbourhood and county-level characteristics on our dependent variable, a movement out of a tract from one time period to the next.

Our results suggest that on average, across the country, regardless of their income, individuals do not have statistically different odds of leaving a transit neighbourhood as compared with other neighbourhoods. Our results remained consistent when segmenting the data by decade to test that this transit-induced gentrification scenario is a more recent phenomenon, and they remained consistent when examining renters alone. Finally, the results were robust when examining moves before and after the station opened separately, when looking at a three-year time span and when altering our definition of a transit tract to include only those covered by a walking catchment area around the station. Our additional analysis did not uncover evidence that residents in transit neighbourhoods were more likely to reside in a crowded unit.

The nationwide scale of our analysis sought to determine if transit-induced displacement of low-income residents in the US is as pervasive as the popular narrative on the subject suggests. This is not to say that it does not occur or that the anecdotal evidence is incorrect, but it serves to reject the wholesale notion that investment in transit will automatically lead to the exodus of low-income residents and possibly the most transit-dependent population. The results of this analysis suggest that on average, across the country, lower-income residents have not moved out of new transit neighbourhoods at a disproportionate rate, controlling for other individual- and neighbourhood-level factors. These national results are very consistent with the recent findings of Rodnyansky (2018) for Los Angeles County who also was unable to find evidence of low-income displacement in neighbourhoods around new transit stations.

Considering previous research which has shown that, across the country, the overwhelming majority of neighbourhoods do not undergo drastic changes in their socioeconomic and housing characteristics following the placement of a new transit station (Baker and Lee, 2019; Nilsson and Delmelle, 2018), the results found in this article are supportive of those aggregate-level findings. Of the neighbourhoods that do change, one of the defining characteristics of those changes is an increase in multi-family housing that accompanies an influx of younger, educated and wealthier residents. One possible explanation behind our findings that deserves further scrutiny could be that increased housing supply in the direct vicinity of transit stations, located in previously underdeveloped land, houses this new population, but does not reduce the housing supply for existing residents in these neighbourhoods. Future research should hone in on neighbourhoods that are undergoing significant transformations and take a closer look at residential dynamics in those locations. It could also be the case that displacement is an extremely localised phenomenon that is overlooked when using the census tract as a neighbourhood unit of analysis. In some instances, the relationship between transit and property values may be non-linear whereby a negative, nuisance effect is felt in the area immediate to the station, and prices increase beyond that (Golub et al., 2012). In these instances, the use of a census tract would overlook these nuances. The coarseness of the census tract boundary prohibited a more detailed analysis on the spatial scale at which displacement may occur and is one limitation when working with the PSID dataset and this research. Another potential limitation with the PSID dataset is the overrepresentation of lower-income and black families resulting from an oversampling of low-income households in the early waves of this nationwide survey (Hill, 1991). Future research could take a more fine-grained approach with an alternative data source. We also only considered moves out of a transit neighbourhood and did not consider those who moved within the census tract itself. This may underestimate any increased mobility incurred by lower-income residents that did not necessarily result in a move of a great distance from the new transit station.

While our results suggest that low-income residents may not be disproportionately displaced from transit neighbourhoods, the overall landscape may still be impacted by rising rents in some of these neighbourhoods, effectively reducing the number of potential locations for low-income residents to move. A next step in investigating the impacts of rail transit investments is to study destination choices of low-income individuals leaving transit neighbourhoods. Another limitation in our analysis is that we treated all transit stations as identical amenities, whereas more accessible locations or higher performing stations may be expected to have different impacts on demand for nearby locations. Local affordable housing policies may also play a role in whether or not low-income residents stay in their neighbourhoods. Their effectiveness could be analysed in a more detailed local study. Our large-scale quantitative analysis is also unable to capture the reasons behind residential mobility decisions or the local population’s perceptions of the new transit station. Qualitative studies in these areas would provide complementary insight on what is occurring in these neighbourhoods and how residents are affected. Finally, more insight could be gained from an alternative modelling framework to further study the timing of moves beyond the three and five years before and after station opening times used here.

Footnotes

Acknowledgements

The authors would like to thank participants of the special session ‘Public Transit: Social and Economic Opportunities and Consequences I’ at the 2018 American Association of Geographers Annual Meeting, the editor and the anonymous referees for thoughtful comments and suggestions. All remaining errors and omissions are our own.

Declaration of conflicting interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Science Foundation under Grant No. 1759714. This work was also supported, in part, by funds provided by the University of North Carolina at Charlotte.

Notes

ORCID iD

Elizabeth Delmelle

References

Alonso

(1964) Location and Land Use: Toward a General Theory of Land Rent. Cambridge, MA: Harvard University Press.

Atkinson-Palombo

(2010) Comparing the capitalisation benefits of light-rail transit and overlay zoning for single-family houses and condos by neighbourhood type in metropolitan Phoenix, Arizona. Urban Studies 47: 2409–2426.

Baker

Lee

(2019) How does light rail transit (LRT) impact gentrification? Evidence from fourteen US urbanized areas. Journal of Planning Education and Research. 39: 35–49.

Bhattacharjee

Goetz

(2016) The rail transit system and land use change in the Denver metro region. Journal of Transport Geography 54: 440–450.

Billings

(2011) Estimating the value of a new transit option. Regional Science and Urban Economics 41: 525–536.

Bowes

Ihlanfeldt

(2001) Identifying the impacts of rail transit stations on residential property values. Journal of Urban Economics 50: 1–25.

Brueckner

Rosenthal

(2009) Gentrification and neighborhood housing cycles: Will America’s future downtowns be rich? Review of Economics and Statistics 91: 725–743.

Bureau of Economic Analysis (BEA) (2018) Personal income by county, metro, and other areas. Available at: https://www.bea.gov/data/income-saving/personal-income-county-metro-and-other-areas (accessed 12 September 2018).

Cao

Schoner

(2013) Transportation impact of transitways: A case study of Hiawatha light rail transit in Minneapolis. CTS Report 13–13, Number 7, Transitway Impacts Research Program.

10.

Coulson

Grieco

(2013) Mobility and mortgages: Evidence from the PSID. Regional Science and Urban Economics 43: 1–7.

11.

Crowder

Pais

Scott

(2012) Neighborhood diversity, metropolitan constraints and household migration. American Sociological Review 77: 325–353.

12.

Debrezion

Pelz

Rietveld

(2007) The impact of railway stations on residential and commercial property value: A meta-analysis. Journal of Real Estate Finance and Economics 35: 161–180.

13.

Deka

(2017) Benchmarking gentrification near commuter rail stations in New Jersey. Urban Studies 54: 2955–2972.

14.

Ding

Hwang

Divringi

(2016) Gentrification and residential mobility in Philadelphia. Regional Science and Urban Economics 61: 38–51.

15.

Ellen

O’Regan

(2011) How low income neighborhoods change: Entry, exit, and enhancement. Regional Science and Urban Economics 41: 89–97.

16.

Ferbrache

Knowles

(2017) City boosterism and place-making with light rail transit: A critical review of light rail impacts on city image and quality. Geoforum 80: 103–113.

17.

Freeman

(2005) Displacement or succession? Residential mobility in gentrifying neighborhoods. Urban Affairs Review 40: 463–491.

18.

Glaeser

Kahn

Rappaport

(2008) Why do the poor live in cities? Journal of Urban Economics 63: 1–24.

19.

Golub

Guhathakurta

Sollapuram

(2012) Spatial and temporal capitalization effects of light rail in Phoenix: From conception, planning, and construction to operation. Journal of Planning Education and Research 34: 415–429.

20.

Hess

Almeida

(2007) Impact of proximity to light rail rapid transit on station-area property values in Buffalo, New York. Urban Studies 44: 1041–1068.

21.

Hill

(1991) The Panel Study of Income Dynamics: A User’s Guide. Vol. 2. Newbury Park, CA: Sage Publishing.

22.

Kahn

(2007) Gentrification trends in new transit-oriented communities: Evidence from 14 cities that expanded and built rail transit systems. Real Estate Economics 35: 155–182.

23.

Knowles

Ferbrache

(2016) Evaluation of wider economic impacts of light rail investment on cities. Journal of Transport Geography 54: 430–439.

24.

Liu

Deng

Vine

(2016) Residential relocation in response to light rail transit investment: Case study of the Hudson–Bergen light rail system. Journal of Modern Transportation 24: 139–144.

25.

Logan

Stults

(2014) Interpolating US decennial census tract data from as early as 1970 to 2010: A longitudinal tract database. The Professional Geographer 66: 412–420.

26.

Lund

(2006) Reasons for living in a transit-oriented development, and associated transit use. Journal of the American Planning Association 73: 357–366.

27.

McGonagle

Sastry

(2016) Using the panel study of income dynamics to analyze housing decisions, dynamics, and effects. Cityscape 18: 185–199.

28.

McKinnish

Walsh

White

(2010) Who gentrifies low-income neighborhoods? Journal of Urban Economics 67: 180–193.

29.

Martin

Beck

(2018) Gentrification, property tax limitation, and displacement. Urban Affairs Review 54: 33–73.

30.

Mills

(1967) An aggregative model of resource allocation in a metropolitan area. The American Economic Review 57: 197–210.

31.

Mohammad

Graham

Melo

, et al. (2013) A meta-analysis of the impact of rail projects on land and property values. Transportation Research Part A 50: 158–170.

32.

Muth

(1969) Cities and Housing: The Spatial Pattern of Urban Residential Land Use. Chicago, IL: University of Chicago Press.

33.

Newman

Wyly

(2006) The right to stay put, revisited: Gentrification and resistance to displacement in New York City. Urban Studies 43: 23–57.

34.

Nilsson

Delmelle

(2018) Transit investments and neighborhood change: On the likelihood of change. Journal of Transport Geography 66: 167–179.

35.

Pollack

Bluestone

Billingham

(2010) Maintaining Diversity in America’s Transit Rich Neighborhoods: Tools for Equitable Neighborhood Change. Boston, MA: Dukakis Center for Urban and Regional Policy, Northeastern University.

36.

Rayle

(2015) Investigating the connection between transit-oriented development and displacement: Four hypotheses. Housing Policy Debate 25: 531–548.

37.

Rodnyansky

(2018) Do rail transit station openings displace low-income households? In: Household mobility and neighborhood impacts. Unpublished PhD dissertation, University of Southern California, Los Angeles, USA.

38.

Rosenthal

(2018) Transit-oriented development? More like transit rider displacement. Los Angeles Times, 20 February. Available at: https://www.latimes.com/opinion/op-ed/la-oe-rosenthal-transit-gentrification-metro-ridership-20180220-story.html (accessed 9 March 2019).

39.

Schwanen

(2015) Geographies of transport I. Progress in Human Geography 40: 126–137.

40.

Tiebout

(1956) A pure theory of local expenditures. Journal of Political Economy 64: 416–424.

41.

Zuk

Bierbaum

Chapple

, et al. (2018) Gentrification, displacement, and the role of public investment. Journal of Planning Literature 33: 31–44.