1# RECORD 0f88c3ee1b9c25f3d6602c7c07d55400a39f254b2# TITLE: End of the line: The impact of new suburban rail stations on housing prices3# YEAR: 2023.0 VENUE: Journal of Transport and Land Use DOI: 10.5198/jtlu.2023.21994# AUTHORS: Rhea Acuña5# PAGES: 20 CANDIDATE TABLE PAGES (1-based): [11, 14]6# ABSTRACT: This study leverages the staggered opening of new Metro stations in a suburb of Washington, DC to estimate the impact of proximity to public rail transit on housing prices. Both hedonic and repeat sales models indicate that housing prices increase as distance increases, suggesting that living near public transportation in Prince George’s County is primarily viewed as a disamenity. For properties at one mile from the nearest station, the preferred repeat sales model estimates a marginal price increase of 4.6 percent for a one-mile increase in distance. I argue that the suburban environment may be key in explaining the results. In the suburbs, a greater share of the population relies on automobiles, and rail stations are typically equipped with large parking lots. The suburban environment allows households the opportunity to both benefit from public transportation access and mitigate the negative externalities associated with living right next to the station.789===== PAGE 1 (score 17, numeric-paren 0, table-words 0, var-words 17) =====10111 12Introduction13Public transportation is a ubiquitous part of urban environments. Benefits of public transportation 14are wide-ranging, such as reducing traffic congestion, improving air quality, and expanding access to 15jobs (Baum-Snow & Kahn, 2000; Bhatta & Drennan, 2003). Additionally, reducing barriers to public 16transportation can help reduce the gap between higher income households who can afford private 17modes of transportation and lower income households who rely on public transportation to access 18jobs (Baum-Snow & Kahn, 2000). However, investments in public transportation are not without 19controversy. Many cities continue to grapple with declines in public transit usage and persistent con-20cerns regarding the impact of transit-oriented development on proximate neighborhoods and residents 21(Baum-Snow & Kahn, 2000; Manville & Cummins, 2015; Padeiro et al., 2019).22As government continues to invest in public transportation, researchers have attempted to quan-23tify the relationship between proximity to public transportation and housing prices. Whether proximity 24to public transportation is an amenity or disamenity is theoretically ambiguous. For example, housing 25prices may be higher near stations as households may be willing to pay more to be closer and take ad-26vantage of accessibility benefits. On the other hand, housing prices may be lower near stations as house-27holds may be willing to pay more to avoid nuisances associated with station areas. Hedonic regressions 28that control for housing and locational characteristics have been widely used to estimate the relationship 29Article history:30Received: March 21, 202231Received in revised form: 32October 19, 2022 33Accepted: November 27, 202234Available online: February 21, 35202336Copyright 2023 Rhea Acuña37http://dx.doi.org/10.5198/jtlu.2023.219938ISSN: 1938-7849 | Licensed under the Creative Commons Attribution – Noncommercial License 4.0 39The Journal of Transport and Land Use is the official journal of the World Society for Transport and Land Use (WSTLUR) 40and is published and sponsored by the University of Minnesota Center for Transportation Studies. 41T J T L U jtlu.org42V. 16 N. 1 [2023] pp. 67–8643Rhea Acuña44Syracuse University45racuna@syr.edu46End of the line: The impact of new suburban rail stations on 47housing prices48Abstract: This study leverages the staggered opening of new Metro 49stations in a suburb of Washington, DC to estimate the impact of 50proximity to public rail transit on housing prices. Both hedonic and 51repeat sales models indicate that housing prices increase as distance 52increases, suggesting that living near public transportation in Prince 53George’s County is primarily viewed as a disamenity. For properties 54at one mile from the nearest station, the preferred repeat sales model 55estimates a marginal price increase of 4.6 percent for a one-mile increase 56in distance. I argue that the suburban environment may be key in 57explaining the results. In the suburbs, a greater share of the population 58relies on automobiles, and rail stations are typically equipped with 59large parking lots. The suburban environment allows households the 60opportunity to both benefit from public transportation access and 61mitigate the negative externalities associated with living right next to 62the station.636465===== PAGE 11 (score 30, numeric-paren 10, table-words 6, var-words 8) =====66677768End of the line: The impact of new suburban rail stations on housing prices69housing prices drove the location selection of the expansion lines. In addition to the selection of the new 70stations, I am also concerned about housing prices responding in anticipation of the new stations, which 71I address in the robustness checks. 726 73Results 74Table 3 presents the results of the main findings.7 Both the hedonic and repeat sales models estimate 75positive coefficients on distance, suggesting that housing prices increase as distance to the nearest sta-76tion increases, and a negative squared term, indicating that housing prices are increasing at a decreasing 77rate. The hedonic model estimates that for properties at one mile from the nearest station, a one-mile 78increase in distance to the nearest station leads to a marginal price increase of 2.4 percent.8 Among the 79housing characteristics, structural area, whether the house has a basement, is a split level, and in below 80average conditions were statistically significant. Not surprisingly, having a larger home, basement, and 81being a split-level increased housing prices, while being in below average condition decreased housing 82prices. The repeat sales model suggests that the hedonic model understates the effects. For properties at 83one mile from the nearest station, the repeat sales model estimates that a one-mile increase in distance 84to the nearest station leads to a marginal price increase of 4.6 percent.85Table 3. Main regression results86Variables87(1) Hedonic88(2) Repeat Sales89Estimate90SE91Estimate92SE93Intercept948.940***950.1289611.800***970.03798Distance99 Distance1000.028***1010.0071020.050***1030.011104 Distance squared 105-0.002**1060.001107-0.002*1080.001109Housing Characteristics 110 Age 111-0.0011120.001113 Age squared 114-0.0001150.000116 Structural area (ln)1170.411***1180.019119 Basement 1200.161***1210.010122 Below average 123-0.109***1240.017125 Split level1260.060***1270.012128Parking129 Parking1300.0001310.0001320.0001330.000134Fixed Effects 135 Housing 136X137 ZIP code 138X139 Time 140X141X142 ZIP code linear time trend 143X144X145Observations 14657,91214716,511148Note: The dependent variable is the natural logarithm of sales price, adjusted to 2000 dollars.149*Statistical significance level at *** p<0.001, ** p<0.01, * p<0.05, + p<0.10. 1507 See Appendix Table A2 for an alternative specification where sales price (dependent variable) and structural area are not 151logged.1528 Since I use a quadratic model and include linear and quadratic terms for distance, I calculate the marginal price effect by 153taking the partial derivative of sales price with respect to distance at one mile. For the hedonic model, ∂/∂D(β0 + β1 Di,t) + β2 154D2155i,t + β3 Parkings + αi + δt + γn T + ϵi,t) = β1 + 2β2 Di,t = .028 - 2(.002)(1) = .024.156157158===== PAGE 12 (score 3, numeric-paren 0, table-words 1, var-words 1) =====15916078161JOURNAL OF TRANSPORT AND LAND USE 16.1162Another way to illustrate the effect is by evaluating the effect size for the average property. In both 163samples, the average distance is about four miles. The predicted sales price at four miles is $254,127 in 164the full sample and $236,779 in the repeat sales sample. The hedonic model estimates a marginal effect 165of 1.4 percent at four miles, which suggests that the marginal household is willing to pay about $3,558 166to live one mile farther. The repeat sales model estimates a marginal effect of 3.5 percent at four miles, 167which suggests that the marginal household is willing to pay about $8,287 to live one mile farther. This 168further highlights that the hedonic model underestimates how much residents are willing to pay to be 169incrementally farther from the station.170Figure 4 plots the hedonic price (i.e., predicted sales price) and implicit price functions (i.e., mar-171ginal effects) for both the hedonic (red) and repeat sales (blue) models.9 Each circle represents the point 172estimate at a specific distance and the whiskers represent the associated 95 percent confidence interval. 173Although, I do not restrict the models, I restrict the figures to 10 miles since nearly 100 percent of my 174sample is located within 10 miles of the nearest station. The hedonic price functions demonstrate the 175non-linear positive relationship between distance and sales price, graphically illustrating that sales prices 176are increasing at a decreasing rate as distance increases. For both models, the predicted sales prices rapidly 177increase at lower distances then eventually level off. 178Figure 4. Hedonic and implicit price functions179Note: Sales price in 2000 dollars. 180Turning to the implicit price functions, the marginal effects are higher for the repeat sales model 181across distances, but the patterns are otherwise similar, leading to two key observations. First, the mar-182ginal effects are greatest closest to the station, suggesting that nuisance elements may be strongest right 183next to the station. This is not surprising given that certain nuisance elements, such as unsightliness and 1849 I generated the implicit price function by taking the partial derivative of price with respect to distance.185186187===== PAGE 14 (score 42, numeric-paren 23, table-words 5, var-words 9) =====18818980190JOURNAL OF TRANSPORT AND LAND USE 16.11917 192Robustness checks 193I employ several robustness checks to test the sensitivity of my results to alternative samples. Table 4 194presents the results of these robustness checks.10 I apply the repeat sales specification to each of the ro-195bustness check.196Table 4. Summary of robustness checks 197Variable198Regression Models1991200220132024203520462057a206Distance2070.050***2080.044***2090.058+2100.078***2110.062***2120.042***2130.019214(0.011)215(0.01)216(0.031)217(0.021)218(0.012)219(0.007)220(0.012)221Distance squared222-0.002*223(0.001) 224-0.001+225(0.001)226-0.000227(0.005)228-0.005*229(0.003)230-0.003**231(0.001)232-0.001*233(0.001)234-0.001235(0.001)236Marginal effects 237.046238.042239.058240.068241.056242.042243.017244Joint F test 24519.9524620.9824716.5124817.2124921.4425020.982511.16252Observations 25316,51125411,2432558,95625612,13225715,98025832,23125932,231260Note: The dependent variable is the natural logarithm of sales price, adjusted to 2000 dollars. Each column is a separate re-261gression that applied equation 2 to alternative repeat sales samples: 1=preferred repeat sales sample; 2=limit to properties built 262before 1990; 3=limit to properties less than 5 miles; 4=limit to properties less than 7 miles; 5=limit to properties to less than 26310 miles; 6=expand to include all properties sold between 1994-2011; 7=treatment date based on construction groundbreak-264ing. Controls include house fixed effects, parking is represented by the number of all-day spaces in parking lots/garages within 265one mile of the station, time fixed effects represented by sales year and quarter, and ZIP code-specific linear time trends. 266Robust standard errors in parenthesis. 267a Includes all properties that sold more than once between 1994-2011. 268*** Statistical significance level at *** p<0.001, ** p<0.01, * p<0.05, + p<0.10. 269For Model 2, I remove properties that could have potentially responded to the announcement and 270construction of the new stations by limiting the sample to properties that were built before 1990. The 271results are consistent with the main estimate. For Models 3-5, I restrict the sample to only include prop-272erties that were within the following distances – five, seven, and ten miles of the nearest station – to assess 273sensitivity to distance thresholds. I estimate the same general pattern when omitting farther properties, 274supporting the main finding that proximity to rail is considered a disamenity and effects should eventu-275ally level off. Furthermore, the greater marginal effects relative to the main estimate provide evidence 276that properties from farther distances are not driving my results. For Model 6, I extend the pre- and post- 277periods by four years to assess if the window of time being studied explains the results.11 The marginal 278effects and general patterns are consistent with the main estimate. For Model 7, I test for anticipatory 279effects to ensure that the estimated effects are in response to the opening. I use the extended sample of re-280peat sales transactions but change the treatment date from its opening to its construction date, artificially 281assigning post treatment distance to properties sold during construction years even though the stations 282are not yet in operation.12 A joint hypothesis test reveals that the study cannot reject the null hypothesis 283that both distance coefficients are zero at the 10 percent level, providing evidence that the increase in 284housing prices estimated by the main analysis is in response to the opening of the new stations.28510 For Table 4, figures of the hedonic price and implicit price functions are available upon request. 28611 I select four years given that the northern end of the Green Line expanded in 1993.28712 I artificially change the treatment date from opening, 2001, to groundbreaking, 1995, for the Green Line and change the 288treatment date from opening, 2005, to groundbreaking, 2001, for the Blue Line.289290291===== PAGE 15 (score 6, numeric-paren 0, table-words 1, var-words 4) =====29229381294End of the line: The impact of new suburban rail stations on housing prices295I also consider the robustness of my model to the quadratic distance specification. A key concern 296in using the hedonic framework is using the appropriate functional form (Yinger & Nguyen-Hoang, 2972016; Yu et al., 2017). As Yinger and Nguyen-Hoang (2016) underscored, a nonlinear hedonic speci-298fication is required for the bid function and associated envelope to be mathematically consistent and 299allow for sorting. I selected a quadratic specification due to the shape of my underlying data, but another 300commonly used nonlinear form in the literature is a categorical distance term.13 I decided not to use a 301categorical distance measure as it excludes useful identifying variations. In a repeat sales model with a 302categorical distance measure, my identification strategy would be based off houses that sold more than 303once and changed distance categories between sales as a result of the new stations; therefore, properties 304that changed distance but not categories would not be contributing to the model. However, as an ad-305ditional robustness check, I assess if my findings are sensitive to the selection of functional form by esti-306mating my hedonic and repeat sales models using a categorical distance specification. The results of the 307categorical distance models provide support for a quadratic distance specification and similarly conclude 308that distance to the station is considered a disamenity.14 309Overall, the robustness checks provide further evidence that the introduction of the new stations is 310not endogenous to recent trends in price and that sample selection and functional forms are not driving 311my results. The preferred repeat sales specification makes a strong case for causal inference as it controls 312for all time-invariant housing and locational characteristics, time-specific shocks, and time-varying char-313acteristics that follow a linear trend within a ZIP code; however, bias can stem from time-varying hous-314ing and locational characteristics that are not accounted for by the various covariates and fixed effects 315in the model. Data limitations prevented the inclusion of specific time-varying housing and locational 316characteristics and contextual limitations did not allow for the inclusion of a difference-in-differences 317strategy, which would have further reduced the threat of omitted variable bias. 3188 319Discussion and conclusions 320This study sheds light on how the suburban context may complicate or even reverse findings on the 321well-studied relationship between proximity to rail and housing prices. I use hedonic and repeat sales 322models with an unrestricted distance specification to estimate the impact of improved access to public 323transportation produced by the opening of new Metro stations in Prince George’s County. This study 324contributes to the literature by focusing on the suburban environment and employing models that are 325better suited for the car-centric suburbs.326Both the hedonic and repeat sales models suggest that being near the station is considered a disa-327menity in Prince George’s County with the preferred repeat sales model estimating larger marginal price 328increases. It appears that households are evaluating the tradeoffs between the costs and benefits of liv-329ing near a rail station and the negative elements may be offsetting accessibility and other benefits. The 330main findings are inconsistent with previous studies on central cities that find a premium for properties 331located near stations. I argue that the suburban context may be critical in explaining the overall trend of 332higher prices for houses located farther from the stations. 333When comparing the urban and suburban context, a critical element to take into account is the 334mode of access to the rail stations. A 2012 survey of riders conducted by the WMATA revealed that the 335majority of the riders from Prince George’s County rely on cars to access the Metro (Metropolitan Wash-33613 Appendix Figures A3 and A4 show that the underlying data supports a quadratic specification. 33714 Appendix Table A3 presents the regression results for models using distance categories where houses were assigned to one-338mile interval distance categories based on their distance to the nearest station. Appendix Figure A5 plots the coefficients and 339associated 95 percent confidence intervals from the hedonic and repeat sales models.340