1# RECORD 721bde276de0a5af885925a9444c628cfce6d18a2# TITLE: The impacts of light rail on residential property values in a non-zoning city: A new test on the Houston METRORail transit line3# YEAR: 2019.0 VENUE: Journal of Transport and Land Use DOI: 10.5198/jtlu.2019.13104# AUTHORS: Qi-Sheng Pan5# PAGES: 24 CANDIDATE TABLE PAGES (1-based): [3, 14, 18]6# ABSTRACT: The impacts of rail transit system on residential property values have been examined for many metropolitan areas in the U.S. But there are few studies on the effects of light rail in a non-zoning city. As the rail transit in the largest non-zoning city, Houston’s light rail transit line, or the so-called METRORail, has not received much attention from the planning research society since it opened to the public in 2004. A previous study by the author utilized 2007 household data to analyze the impacts of Houston’s METRORail line and found the net effects of the rail transit line change significantly at different distances from the rail stations. One limitation of that study was that the physical environment and neighborhood characteristics of the station areas may not have had notable changes over a relatively short time span, i.e., three years after the opening of the light rail. This study employs 2010 InfoUSA household data to re-examine the effects of Houston’s METRORail line. Similar to the previous studies, the author adopts a traditional ordinary linear regression (OLS) to investigate the contribution of a set of variables representing the physical, neighborhood, and accessibility characteristics of properties, and also employs a multi-level regression model (MLR) to examine the hierarchical structures of spatial data explicitly. In addition, this study tests the spatial autocorrelation in the modeling process and analyzes its effects on the results. The modeling results suggest that the METRORail line has had significant net positive effects on residential property values. The MLS model captures the difference of these effects with more spatial details. The spatial regression model improves model fit, but spatial autocorrelation is not completely eliminated.789===== PAGE 1 (score 10, numeric-paren 0, table-words 0, var-words 10) =====1011Abstract: The impacts of rail transit system on residential property val-12ues have been examined for many metropolitan areas in the U.S. But 13there are few studies on the effects of light rail in a non-zoning city. 14As the rail transit in the largest non-zoning city, Houston’s light rail 15transit line, or the so-called METRORail, has not received much atten-16tion from the planning research society since it opened to the public in 172004. A previous study by the author utilized 2007 household data to 18analyze the impacts of Houston’s METRORail line and found the net 19effects of the rail transit line change significantly at different distances 20from the rail stations. One limitation of that study was that the physi-21cal environment and neighborhood characteristics of the station areas 22may not have had notable changes over a relatively short time span, 23i.e., three years after the opening of the light rail. This study employs 242010 InfoUSA household data to re-examine the effects of Houston’s 25METRORail line. Similar to the previous studies, the author adopts a 26traditional ordinary linear regression (OLS) to investigate the contribu-27tion of a set of variables representing the physical, neighborhood, and 28accessibility characteristics of properties, and also employs a multi-level 29regression model (MLR) to examine the hierarchical structures of spatial 30data explicitly. In addition, this study tests the spatial autocorrelation in 31the modeling process and analyzes its effects on the results. The model-32ing results suggest that the METRORail line has had significant net 33positive effects on residential property values. The MLS model captures 34the difference of these effects with more spatial details. The spatial re-35gression model improves model fit, but spatial autocorrelation is not 36completely eliminated.37Keywords: Light rail, residential property values, soning, ordinary lin-38ear regression, multi-level regression model, METRORail.39The impacts of light rail on residential property values in a non-40zoning city: A new test on the Houston METRORail transit line41Article history:42Received: September 17, 201743Received in revised form: 44December 8, 201845Accepted: January 6, 201946Available online: April 15, 201947Copyright 2019 Qisheng Pan48http://dx.doi.org/10.5198/jtlu.2019.131049ISSN: 1938-7849 | Licensed under the Creative Commons Attribution – Noncommercial License 4.0 50The Journal of Transport and Land Use is the official journal of the World Society for Transport and Land Use (WSTLUR) 51and is published and sponsored by the University of Minnesota Center for Transportation Studies. 52T J T L U http://jtlu.org53V. 12 N. 1 [2019] pp. 241–26454Qisheng Pan55Tongji University56Texas Southern University57pan_qs@tsu.edu581 59Introduction60There are a large number of studies on the impact of rail transit services in many cities However, few of 61them have been conducted for a non-zoning city. Zoning has been a common method of controlling 62land use in U.S. cities since the 1920s. It is important for decision makers and planning scholars to 636465===== PAGE 3 (score 21, numeric-paren 0, table-words 4, var-words 11) =====666724368The impacts of light rail on residential property values in a non-zoning city69Table 1. Mode shares of commuting trips in Houston Metropolitan Statistical Area, 1980-200970Notes: 1. Large US metro areas have population greater than 3 million in 2000 excluding New York. Source: Census Trans-71portation Planning Package (CTPP), 1980-2000 data sets. 722. 2009 data are obtained from the 2009 American Community Survey. 73Due to the decentralization trend (often simplified via the catch phrase “urban sprawl”), Houston’s 74population density is low, suggesting that it is not feasible to support a sustainable public transit system 75to operate in the entire metropolitan area. Transit service in Houston has been mainly limited to the core 76county, Harris County. Similar to the other comparable cities, Houston has a low and continually de-77creasing transit mode share. All the evidence makes it a reasonable expectation that Houston’s METRO-78Rail has little, if any, effect on residential property values. However, most discussions on the relationship 79between density and transit system ignore the variation of density within a metropolitan area (Bertaud, 802003, 2004). Many studies do not provide details on spatial variation of transit mode share in large met-81ropolitan areas. It is possible that the light rail in Houston has significant effects on land-use change and 82residential property values in the light rail corridor due to the relatively high transit mode share in the 83corridor and its unique land-use policies. Table 1, for example, does not provide sufficient details on spa-84tial variation of transit mode share in Houston, especially for commuting trips to Downtown Houston. 85A commute survey in 2009 showed a surprisingly high public transit mode share in downtown Houston 86(Table 2). Therefore, it is important to incorporate spatial variation into the analysis.87Table 2. Mode share of downtown Houston commuting, 200988Source: Central Houston Inc. (2009), Downtown Houston Commute Survey Report89Houston90Large US Metro Areas*1 91Mode921980931990942000952009*296198097199098200099Drove Alone10069.410176.11027710378.810465.0710572.2110673.71107Carpool10822.510914.611014.211112.111218.7211313.4011412.64115Transit 1162.91173.71183.31192.21208.541216.921226.43123Walk1242.71252.31261.61271.61284.501293.391302.69131Others1322.51333.31343.91355.31363.161374.071384.53139Total Workers (1000)1401,5121411,7691422,0821432,70914427,98114534,74414638,009147Total Population (1000)1483,1201493,7311504,6701515,86715262,10115371,41715482,329155Mode156Number of Respondents157Mode Share158Drive alone 1595,97716048.40%161Carpool 1621,37716311.15%164Vanpool 1651801661.46%167Park & Ride 1683,55816928.81%170Local or express bus 1714891723.96%173Light rail 1741531751.24%176Bus and rail 1773281782.66%179Walk 1801291811.04%182Rarely travel to downtown 183661840.53%185Motorcycle or scooter 186551870.45%188Bicycle 189381900.31%191Sum19212,350193100.00%194195196===== PAGE 4 (score 13, numeric-paren 0, table-words 1, var-words 11) =====197198244199JOURNAL OF TRANSPORT AND LAND USE 12.1200The effects of Houston’s METRORail on residential property values were examined in very few 201studies. The empirical analyses by Pan and Ma (2009) and Pan (2013) employed both a Hedonic Price 202Model (HPM) with Ordinary Least Square (OLS) function and a multilevel regression (MLR) model 203to examine the impacts of Houston’s light rail system. It obtained information on home sales prices 204and physical characteristics of properties from InfoUSA’s 2007 household database. They reported that 205the opening of the light rail has had significant net positive effects on residential property values. But 206immediate proximity to light rail stations and bus stops had significant negative impacts on properties 207located within a quarter mile of rail stops. One limitation of this study was that the data collected in a 208relatively short time span, i.e., three years after the opening of the light rail, may not capture the changes 209of physical environment and neighborhood characteristics of the station areas, which usually take a long 210time to mature. 211An interesting comparison between the METRORail in Houston, Texas, with the Metro lines in 212Shanghai also showed that both rail transit systems had significantly positive effects on residential prop-213erty values, but notable variations of their effects were observed at different distance ranges and time 214spans in Houston and Shanghai (Pan, Pan, Zhang, & Zhong, 2014). 215The HPM model with the cross-sectional data is the most common methodology in the literature 216for estimating the implicit value of differences in property characteristics (Debrezion, Pels, & Rietveld, 2172007; Dubé, Thériault, & Des Rosiers, 2013). However, the inherent spatial nature of this approach 218may lead to spatial dependence and heterogeneity. Spatial dependence or spatial autocorrelation results 219from spatial correlation among observations that are assumed to be independent, while spatial heteroge-220neity comes from the spatial correlation of the error terms (Anselin, 1988). As Higgins and Kanaroglou 221(2016) pointed out, it is important to take into account the spatial dependence and heterogeneity in 222examining the effects of rail transit since the failure to address this issue may lead to the violation of the 223statistical independence assumption, heteroskedasticity, biased estimates, and unreliable test statistics, 224etc. 225Moran’s I statistics and Lagrange multiplier tests have been usually employed to examine the pres-226ence of spatial dependence and heterogeneity. Spatial methods such as spatial lag models and spatial 227error models have been adopted to control spatial dependence and heterogeneity. Spatial fixed effects 228have been used to capture spatial dependence and heterogeneity. Geographically Weighted Regression 229(GWR) has been developed to explore spatial varying relationship and examine heterogeneity between 230dependent and independent variables over space. The difference-in-differences (DID) estimator pro-231vides a means for avoiding the bias due to omitted variables. But all these methods have advantages and 232drawbacks linked to their uses (Higgins & Kanaroglou, 2016). 233The objective of this study is to conduct a new test on the Houston METRORail Transit Line to 234identify impacts on residential property values by controlling spatial dependence and heterogeneity. To 235compare with the previous studies, the author also adopts a traditional ordinary linear regression (OLS) 236using recent property transaction data to investigate the contribution of a set of variables representing 237the physical, neighborhood, and accessibility characteristics of properties, and also employs a multi-level 238regression model (MLR) to examine the hierarchical structures of spatial data explicitly. In addition, this 239study tests spatial dependence and heterogeneity in the modeling process using Moran’s I statistics and 240control the spatial dependence and heterogeneity of the variables using spatial regression models.241In subsequent sections, we review the relevant literature in Section 2, describe the methodology in 242Section 3, present the empirical study in Section 4, and summarize findings and draw conclusions in 243Section 5. 244245246===== PAGE 14 (score 34, numeric-paren 6, table-words 6, var-words 15) =====247248254249JOURNAL OF TRANSPORT AND LAND USE 12.1250Table 5. Variables used in the models2514.2 252Analysis and results253This study employs a large set of observations to examine the effects of light rail on property values. The 254MLR and OLS models are applied with 100 percent observations, which include 384,368 residential 255properties. To test the spatial autocorrelation in the OLS model, Moran’s I method is applied to test the 256residuals of the residential property values in the OLS models. Spatial regression models including a 257spatial lag model and a spatial error model are utilized to examine the spatial dependence of the variables 258with 5 percent samples from the population.259Moran’s I and spatial regression models in GeoDa or R’s spdep package are utilized to examine 260spatial dependence of spatial variables. The outputs of Moran’s I test are shown in Table 6 and plotted 261out in Figure 2. 262Table 6. Spatial autocorrelation diagnostics for the OLS residuals of the residential property values.263Moran’s Index2640.4227 (0.0000)265Moran’s I statistic standard deviate26618.4282 (0.0000)267Lagrange Multiplier (LM) Diagnostics for Spatial Lag (Spatial lag Model as alternative hypothesis)268750.7199 (0.0000)269Lagrange Multiplier Diagnostics for Spatial Error (Spatial Error Model as alternative hypothesis)270331.8029 (0.0000)271Robust Lagrange Multiplier to Spatial Lag 272529.23956 (0.0000)273Robust Lagrange Multiplier to Spatial Error 274110.3226 (0.0000)275Variables276Levels277 278Descriptions279Source280Dependent2811st level282Lgval1000283Log(Home Sale Prices ($1000))284InfoUSA285Independent286Homesize287Home Size (square feet)288InfoUSA289Homeage290Home Age (Year)291InfoUSA292Raillineop293Home Sale Before (0) or After (1) METRORail Line Open294GIS Map295RailQMI296Within one-quarter mile of METRO Rail stop297GIS Map298RailHMI299Within one-quarter to one-half mile of METRO Rail stop300GIS Map301Rail1MI302Within one-half to one mile of METRO Rail stop303GIS Map304Rail2MI305Within one to two miles of METRO Rail stop306GIS Map307Rail3Mi308Within two to three miles of METRO Rail stop309GIS Map310BusStQMI311Within one-quarter mile of METRO Bus stop312GIS Map313Hwyhmi314Within one-half mile of Highway intersections315GIS Map316Hwy1mi317Within one half to one mile of Highway intersections318GIS Map319Hwy2mi320Within one to two miles of Highway intersections321GIS Map322Hwy3mi323Within two to three miles of Highway intersections324GIS Map325Distcbd326Distance to CBD327GIS Map328Distmed329Distance to Medical Center330GIS Map3312nd Level332Medinc333Median Income ($1000)334CTPP/GIS Map335Popdens336Population density337CTPP/GIS Map338JobDens339Job density340CTPP/GIS Map341Pctminor342Percentage of minority population343CTPP/GIS Map344JobAcess345Total job accessibility346CTPP/GIS Map347348349===== PAGE 15 (score 8, numeric-paren 0, table-words 3, var-words 2) =====350351255352The impacts of light rail on residential property values in a non-zoning city353Figure 2. Moran scatter plot for the standardized logarithm of residential property values ($1000)354Table 6 shows the results of testing spatial autocorrelation in the OLS residuals. Moran’s Index is 355calculated as the slope of the Moran scatter plot between standardized logarithm of residential property 356values and their spatial lag values (Figure 2). The score of 0.4227 is highly significant, which indicates 357significant spatial autocorrelation of the residuals in the OLS model. All the other testing results, includ-358ing the Lagrange multiplier (LM) diagnostics for spatial lag and spatial error, and the robust Lagrange 359multiplier to spatial lag and spatial error, point to the rejection of the null hypothesis and indicate the 360presence of strong spatial autocorrelation in the OLS model. It also indicates the missing of lagged 361dependent variable and the existence of error dependence. So both spatial lag model and spatial error 362model need to be applied to examine the spatial dependence of the variables. 363Table 7 reports the results of MLR and OLS with 100 percent of residential properties and 5 per-364cent random samples of properties. It also shows the results of spatial regression models including spatial 365lag model and spatial error model with all the explanatory variables. 366367368===== PAGE 18 (score 133, numeric-paren 116, table-words 1, var-words 1) =====369370258371JOURNAL OF TRANSPORT AND LAND USE 12.1372Table 7. Comparison of results from MLR, OLS, SLM, and SEM373Notes: *p = 0.1, significant at the 0.1 level, ** p = 0.05, significant at the 0.05 level, and *** p = 0.01, significant at the 3740.01 level.375100 Percent of Properties3765 Percent Random Samples of Properties377Model378MLR379OLS380MLR381OLS382SLM383SEM384Fixed eff.385Level 1386 Intercept3873.2236388***3893.8603390***3913.9791392***3933.9010394***3952.9145396***39710.8554398***399 homesize4000.2331401***4020.2877403***4040.2480405***4060.2830407***4080.2474409***4100.2619411***412 homeage413-0.0092414***415-0.0097416***417-0.0099418***419-0.0101420***421-0.0086422***423-0.0099424***425 Raillineop4260.0350427***4280.0398429***4300.0301431***4320.0375433***4340.0380435***4360.0353437***438 RailQMI4390.4288440***4410.3911442***4430.6012444***4450.4246446**4470.3593448*4490.4394450**451 RailHMI4520.2462453***4540.3598455***4560.5019457***4580.4610459***4600.3226461***4620.4727463***464 Rail1MI4650.2381466***4670.3277468***4690.3740470***4710.3400472***4730.2381474***4750.3627476***477 Rail2MI4780.1791479***4800.3021481***4820.2518483***4840.2095485***4860.1417487***4880.2249489***490 Rail3Mi4910.0884492***4930.2121494***4950.1668496***4970.2175498***4990.1502500***5010.2155502***503 BusStQMI504-0.0146505***5060.0945507***5080.0416509***5100.1131511***5120.0859513***5140.0935515***516 Hwyhmi517-0.0053518-0.00155190.0067520-0.00975210.0056522-0.0144523 Hwy1mi524-0.00885250.00345260.01875270.02515280.0310529*5300.0139531 Hwy2mi5320.0136533**5340.0129535***5360.02935370.0219538*5390.0223540*5410.0164542 Hwy3mi5430.0076544-0.0055545*5460.00855470.00475480.0079549-0.0025550 Distcbd5510.0075552**553-0.0049554**555-0.0071556-0.0070557***558-0.0087559***560-0.0062561*562 Distmed563-0.0267564***565-0.0192566***567-0.0188568***569-0.0171570***571-0.0112572***573-0.0183574***575Level 2576 Medinc5770.0062578***5790.0034580***5810.0048582***5830.0033584***5850.0021586***5870.0035588***589 Popdens5900.0030591*592-0.0071593***594-0.0042595*596-0.0064597***598-0.0034599**600-0.0066601***602 Jobdens6030.0026604***6050.0034606***6070.0026608**6090.0045610***6110.0029612***6130.0038614***615 Pctminor616-0.5325617***618-0.6479619***620-0.6615621***622-0.6687623***624-0.5012625***626-0.6629627***628 Jobacess6290.0004630***6310.0001632***6330.00016340.0001635***6360.0001637***6380.0001639***640Random eff.641Level 1642 Intercept6430.3393644***6450.3256646***647Level 2648 Intercept6490.0597650***6510.0425652***653ρ (Eq. 6)6540.2420655***656λ (Eq. 8)6570.2049658***659R Square6600.43606610.44476620.46916630.4581664Pseudo R Square6650.47986660.5063667Log Likelihood668-17,236.3669-16,889.8670-17,056.9671AIC67233,923.267334,514.667433,823.667534,155.7676Breusch-Pagan Test677863.4998678***679914.6319680***681LRTErr682358.8515683***684LRTLag685692.989686***687Obs. of Level 168838436868938436869019072691190726921907269319072694Obs. of Level 26951427696 697 6981207699 700 7011207702 703 704705706===== PAGE 19 (score 14, numeric-paren 0, table-words 0, var-words 14) =====707708259709The impacts of light rail on residential property values in a non-zoning city7105 711Discussion and conclusions712This research contributes to the growing literature on the relationship between light rail and residential 713property values in three unique respects, (1) it examines the effects of light rail on residential property 714values in the largest non-zoning city, i.e., Houston; (2) the effects of the Houston’s light rail are examined 715by utilizing the 2010 residential property database, which allows the effects to be captured after a reason-716able time span, i.e., six years after the opening of the new light rail line; (3) it employs spatial regression 717models together with the widely used ordinary linear regression (OLS) and the sophisticated multi-level 718regression (MLR) model to investigate the impacts of light rail system on residential property values.719In the largest non-zoning city, Houston’s light rail is expected to have little, if any, effect on resi-720dential property values due to the decentralization of population, manufacturing, and office firms, low 721population density, and low transit mode share in the non-zoning city. However, this study shows that 722the opening of the light rail has significant positive effects on residential property values. It finds that 723Houston’s METRORail has developed significant positive effects on the values of residential properties 724near to light rail stations after six-year services, which cannot be identified within a three-year time span 725in a study by Pan and Ma (2009) and Pan (2013). These results are consistent to the findings of Lee and 726Sener (2017) that reported a significant increase of commercial development along the original light rail 727corridor in Houston, about 4 to 10 years after its opening in 2004. 728Policy makers and planners have considered light rail as a means to induce urban land-use change 729and increase land values due to the improvement of transit accessibility. However, the marginal increase 730in transit accessibility alone may not be large enough to change broad land-use patterns and thus raise 731land uses without government support (Giuliano, 1995, 2004; Cervero & Landis, 1995). The empirical 732studies in the literature have reported mixed results and a large variation for the effects of rail stations on 733property value (Debrezion et al., 2007). It is argued that most studies on the impacts of light rail have 734done too early to find consistent patterns of land development and land value change. The few studies 735that analyze the effects of light rail over a large time span have shown LRT proximity has considerable 736influence on the probability of land-use change (Cervero & Landis, 1997). 737Our study confirms that it needs times for Houston’s METRORail to develop significant positive 738effects on residential property values. Comparing to a zoning city, the time span for a non-zoning city 739to show significant change of land-use patterns and residential property values after the opening of light 740rail may be longer due to the lack of government assistance. The land-use study by Lee and Sener (2017) 741found that the spike of commercial and industrial land use in Houston’s METRORail corridor between 7422008 and 2014 after a small decline between 2005 and 2008. During the same periods, residential land 743uses had modest gains while the number of vacant and other land parcels decreased. It implies that the 744increased transit accessibility after the introduction of light rail has attracted businesses and firms to 745locate around the station areas, which increased the attraction of the neighborhoods and thus generated 746the value uplift for residential properties in the light rail corridor. 747A limitation of this research lies in its modeling approaches. A major shortcoming of OLS and 748MLR is that they have not controlled spatial dependence (or spatial autocorrelation) and heterogeneity 749though MLR takes into account of the nesting structure of the spatial data. Though OLS and MLR 750models captured the effects of the light rail on residential property values, their results were considered 751biased and the capitalization effects of light rail were overestimated. Spatial regression models improve 752the model fit but they have not completely ruled out spatial autocorrelation. The difference-in-differenc-753es (DID) method provides a simple way to avoid the bias in the estimated coefficients that results from 754the omitted variables. It is also able to treat the problem of spatial autocorrelation among residuals if it 755is generated by omitted variables. We plan to adopt the DID method using longitudinal or repeat sales 756data in our future research.757