Next Article in Journal
Managerial Challenges in Implementing European Rail Traffic Management System, Remote Train Control, and Automatic Train Operation: A Literature Review
Previous Article in Journal
Identification of Individual Mobility Anchor Places and Patterns Based on Mobile Phone GPS Data
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Capturing the Value of Walkability

Schar School of Policy and Government, George Mason University, Arlington, VA 22201, USA
Future Transp. 2024, 4(4), 1334-1349; https://doi.org/10.3390/futuretransp4040064
Submission received: 16 September 2024 / Revised: 24 October 2024 / Accepted: 29 October 2024 / Published: 3 November 2024

Abstract

Capturing the value infrastructure investments add to the residential market is a longstanding policy to defray their expense. Unfortunately, estimates of the added value of infrastructure, generally, and estimates of the added value of walkability, specifically, are scarce. Novel, multiscale models free independent variables to manifest simultaneously at different scales of analysis to greatly improve specifications to precisely estimate walkability valuation. Results from analysis of years of transactions within walking distance of heavy rail stations suggest that walkability adds value available for capture locally, not systemically. Stakeholders confront myriad problems to replicate the accessibility characteristics shown to add value given the distinct cluster where walkability adds value available for capture in a heavy rail system.

1. Introduction

Value capture is a popular policy for defraying the capital and the operational costs of infrastructure investments. The rationale is to capture value attributable to the accessibility characteristics of residential markets [1,2,3]. The empirical literature on value capture is longstanding [4,5]. Generally, the empirical literature supports the argument that infrastructure investments increase proximate residential property values [6,7,8]. Interestingly, the magnitude of the increase is sensitive to the approach to modelling the increase as well as the method to select the context of the increase in the guise of catchment areas versus control areas [9]. Unfortunately, empirical evidence on the value available for capture from infrastructure for pedestrians is uneven, at best. Unevenness in the empirical literature is problematic because reliable estimates of total revenues available for capture are scarce. One methodological obstacle to reliable estimation is the difference in the scale at which infrastructure and walkability manifest in the travel behavior of pedestrians.
Conceptually, walkability is more than infrastructure, more than behavior [10], and more than service [11]. Walkability encompasses the quality of the conditions for pedestrian trips as well as the degree to which pedestrian trips are safe, comfortable, and convenient [12]. An example from the Borough of Manhattan in New York City is illustrative. Sidewalks are ubiquitous in the Borough of Manhattan, but not all streets and not all avenues are safe, comfortable, and convenient for pedestrian trips. Put succinctly, infrastructure for pedestrians is global in scale, but walkability is local in scale. Here, multiscale, local models of residential market transactions control for the scale mismatch in the manifestation of infrastructure versus the manifestation of walkability. Such models yield accurate and precise estimates of the value walkability adds to the residential market at the point of sale.
The following section reviews the theoretical literature on value capture and the empirical literature on walkability. The Data Section and the Methodology Section, respectively, list the dependent variable and the independent variables and specify the multiscale geographically weighted regression (MGWR) models. The Results Section interprets the estimates from the models. The Discussion Section contextualizes the results from the models. The Conclusions Section highlights the limitations of this study, highlights the contribution of the results, posits an explanation for the results, suggests a trajectory for future research, and cautions restraint in the generalization of the results elsewhere.

2. Literature Review

2.1. Value Capture

Accessibility characteristics as well as environmental characteristics, physical characteristics, and public sector characteristics are known to affect prices in residential markets [13,14,15]. Infrastructure investments increase accessibility characteristics via decreases in trip costs [16,17,18], so infrastructure investments add value. Reviews of the empirical literature by Armstrong and Rodríguez [19] and by Duncan [20] support theory on the additional value available for capture from infrastructure investments.
The present study extends the hedonic approach [21] to valuation where price is a function of a bundle of characteristics. Specifically, the models bundle such characteristics as well as temporal covariates and spatial covariates [22,23] known to buyers and to sellers such as infrastructure investments. With regard to the covariates, the models are representative of the longstanding trend [24,25] and the ongoing trend [26,27] in the empirical literature on value capture to accurately and to precisely contextualize transactions at the point of sale.

2.2. Walkability

The integration of pedestrian infrastructure with infrastructure for public transportation is vitally important to the attainment of future sustainable transportation goals [28]. In the United States, the majority of passengers need to walk to and from public transportation stations [29]. Concomitantly, public transportation infrastructure, particularly heavy rail systems, which by definition transports large volumes of passengers at high speeds on dedicated rights-of-way [30], expands the scale of accessible destinations for passengers [31].
On the one hand, the 2023 Community & Transportation Survey by the National Association of Realtors [32] reveals 78% of respondents who said walkability is very important or somewhat important were willing to pay more to reside in a walkable neighborhood and 85% of respondents said sidewalks are very important or somewhat important. On the other hand, 65% of respondents said proximity to public transportation is very important or somewhat important. The above empirical evidence on consumer preferences highlights the future benefits of walkability for heavy rail ridership. The Washington Metropolitan Area Transit Authority (WMATA), the public agency that owns and operates the Metro heavy rail system, estimates an annual increase of 42,000 trips and an annual increase of USD113,000 in revenues from a proposal to increase pedestrian access to the Naylor Road Metro station [33]. Such a ridership boost is consistent with estimates in the public transportation literature on ridership increases from accessibility improvements for nonmotorized modes [34].
Unfortunately, the integration of plans for pedestrian infrastructure and plans for public transportation infrastructure is rare. The failure to integrate such plans hampers monetization of the value of walkability to buyers and to sellers in the residential property market. Examples of the unevenness of the empirical evidence on the price effect of infrastructure for pedestrians are as follows. Guo et al. [35] and Yang et al. [36] found that the price effect of accessibility for pedestrians depends on the destination. Petheram et al. [37] and Gilderbloom et al. [38] found that the price effect of walkability is positive. More empirical evidence on the unevenness of the price effect of walkability is as follows. Methodology affects the magnitude of the price effect of walkability [39]. Effects of proximity to infrastructure for public transportation are also inconsistent. Network distance to commuter rail stations [40] is not statistically significant. Pedestrian accessibility to bus stops or to bus rapid transit stops is also not statistically significant [35]. Euclidean distance to bus rapid transit stops is statistically significant [36]. Network distance to heavy rail stations is also statistically significant. Counterintuitively, more sidewalks are not statistically significant [40]. More stops are also not statistically significant [36]. Finally, Yang et al. [41] found that the rent effect of walkability in an MGWR model of housing is positive near Transit-Oriented Development (TOD) stations in Wuhan, China.
The following sections list the dependent variable and the independent variables and specify the MGWR models, respectively.

3. Data

Data for transactions are from Metropolitan Regional Information Systems, Incorporated (MRIS). The temporal context for the MRIS data is from 1 January 2002 to 31 December 2014. Five Metro stations (McLean, Tysons Corner, Greensboro, Spring Hill, and Wiehle-Reston East) on the new Silver line were open from 26 July 2014, so the Metro system here includes 86 stations, not 91 stations. The spatial context for the MRIS data is the counties (Arlington County, Fairfax County, Montgomery County, and Prince George’s County) and the county equivalents (Alexandria city, Fairfax city, Falls Church city, and the District of Columbia) in the Metro service area. Deletion of foreclosure transactions, missing data transactions, non-market transactions, repeat-sale transactions, short sale transactions, and transactions in the tails of the price distribution left a subsample of transactions for detached homes and for townhomes in the Metro service area. The latter rank first and rank second by type in the Metro service area. Random selection without replacement by coordinates (X, Y) left 29,422 transactions in the Metro service area. Most trips to Metro stations are on foot, and most pedestrian trips to Metro stations are less than 0.80 km in distance, though distances vary by Metro station [42]. For example, most passengers walk from 0.80 km to 1.61 km to the Medical Center Metro station and most passengers walk more than 1.61 km to the Farragut West Metro station. Differences in walk distances and data availability constraints on the origins of the pedestrian trips to the respective Metro station destinations create a problem in how best to select transactions so as to match the behavior of passengers to the infrastructure for pedestrians. One solution is to create custom, Euclidean-distance buffers in 0.80 km increments to match the walk distance threshold to Metro stations. The spatial scale of the analysis is the Metro system, so a 0.8 km buffer, a 1.61 km buffer, or a 2.41 km buffer minimizes overlap to preserve, to the extent possible, the spatial resolution of walk distance buffers from Metro stations (Figure 1). Selection of transactions within the respective walk distance buffers left 1470 transactions.
Data for the accessibility characteristic distance are from ArcGIS 10.8 [43]. The Address Locator geocodes transactions as point features. The Near tool calculates the Euclidean distance in meters from a point feature to a Metro station.
Data for the accessibility characteristics infrastructure and walkability are from the Smart Location Database [44] and the National Walkability Index [45], respectively. Data for the environmental characteristic density are from the Smart Location Database [44]. The Spatial Join in ArcGIS 10.8 [43] appends the polygon (block group) feature data for infrastructure, for walkability, and for density to a point feature.
The dependent variable is the inflation-adjusted price [46]. The independent variables capture the physical characteristics, the accessibility characteristics, and the environmental characteristics known to add value in residential markets [13,14,15]. Physical characteristics approximate the interior with the number of full baths, the number of half baths, and the number of bedrooms as well as the quality with the age in years. Accessibility characteristics approximate the supply of infrastructure with the minimum Euclidean distance in meters to a Metro station, the network density of infrastructure for pedestrian trips in the block group per square kilometer [44], and the walkability score of the block group [45]. The environmental characteristic approximates the concentration of activity with the density of jobs and housing units in the block group per square kilometer [44].

4. Methodology

The ordinary least squares (OLS) model of prices is as follows:
Yi = β0 + ∑jβjXij + εi.
The Ordinary Least Squares tool in ArcGIS 10.8 [43] fits the OLS model. The Spatial Autocorrelation (Moran’s I) tool in ArcGIS 10.8 [43] measures spatial autocorrelation in the OLS model residuals.
The geographically weighted regression (GWR) model of prices is as follows:
Yi = β0(ui, vi) + ∑jβj(ui, vi)Xij + εi.
MGWR 2.2 [47] fits the GWR model. Here, (ui, vi) are the coordinates of the point feature i and βj(ui, vi) captures the continuous function βj(u, v) at point feature i [48]. The density of point features is random so the spatial kernel to calibrate the GWR model here is adaptive; that is, the spatial kernel adapts in size to variation in the density of the point features. The price distribution is approximately symmetrical, so the GWR model type is Gaussian. The corrected Akaike Information Criterion (AICc) [49,50] is the optimization criterion to calibrate the GWR model. Default GWR model fit is with (dependent and independent) variable standardization (0, 1) to conserve iterative computation time. The Spatial Autocorrelation (Moran’s I) tool in ArcGIS 10.8 [43] measures spatial autocorrelation in the GWR model residuals.
The MGWR model of prices is as follows:
Yi = β0(ui, vi) + ∑jβbwj(ui, vi)Xij + εi.
MGWR 2.2 [47] fits the MGWR model. Here, (ui, vi) are the coordinates of the point feature i and bbw in βbw is the bandwidth to calibrate the conditional relationship between the jth independent variable and the dependent variable [51]. The density of point features is random, so the spatial kernel to calibrate the MGWR model here is adaptive; that is, the spatial kernel adapts in size to variation in the density of the point features. The price distribution is approximately symmetrical, so the MGWR model type is Gaussian. The AICc [49,50] is the optimization criterion to calibrate the MGWR model. Default MGWR model fit is with (dependent and independent) variable standardization (0, 1) to ease interpretation of the different bandwidths, or scales, for the independent variables. The Spatial Autocorrelation (Moran’s I) tool in ArcGIS 10.8 [43] measures spatial autocorrelation in the MGWR model residuals.
Significance level corrections to mitigate falsely positive local t in GWR and in MGWR, respectively, due to multiple hypothesis tests and dependence between local estimates [52] are as follows:
α = ξ⁄(ENP/P)
and
αj = ξ/ENPj.
α is the significance level, ξ is the standard Type I error rate (for example, 0.01), ENP is the Effective Number of Parameters in the GWR model, and P is the number of parameters in the OLS model. If ENP = P, then ξ = α so the t in the OLS model and in the GWR model are equivalent. αj is the significance level for the jth set of parameter estimates, ξ is the standard Type I error rate (for example, 0.01), and ENPj is the Effective Number of Parameters for the jth model term.

5. Results

Descriptives for the dependent variable and the independent variables appear in Table 1. The left column presents descriptives for the transactions within the Metro station buffers (n = 1470). The right column presents descriptives for the transactions within the Metro service area (n = 29,422). The results in the right column help to contextualize the results in the left column. Overall, prices are lower within the Metro station buffers because residences are smaller and older than residences within the Metro service area. Such a result is consistent with the modernization trajectory of the residential stock from the core of the Metro service area to the periphery of the Metro service area. Metro stations are expectedly closer within the Metro station buffers than within the Metro service area. The block groups within the Metro station buffers have more infrastructure suitable for pedestrians, have higher walkability scores, and have a higher density of jobs and housing units than the block groups within the Metro service area. Such a result is consistent with the context of residential development in the core of the Metro service area (Figure 1). The Great Recession from the fourth quarter of 2007 to the second quarter of 2009 overlaps the temporal context of the MRIS data, but no obvious difference in sales prices is evident from 2007 to 2009 (Table 2). Indeed, the mean sales price is greatest in 2009 (USD526,637.69) and the maximum sales price is in 2009 (USD1,056,000.00).
Parameters from the OLS model appear in Table 3. Diagnostics for the GWR model and for the MGWR model appear in Table 4. Parameters from the GWR model and from the MGWR model appear in Table 5. The parameters and the diagnostics for the respective models yield the following. First, multicollinearity is not present in the OLS model. The variance inflation factors in the OLS model are all below threshold [47]. Second, the MGWR model best fits the data. The AICc in the MGWR model (2266.79) is less than the AICc in the GWR model (2437.37) or the AICc in the OLS model (3583.96). Third, the MGWR model accounts for more of the variation in the dependent variable than the GWR model or the OLS model. The Adjusted R-Square in the MGWR model (0.76) is higher than the Adjusted R-Square in the GWR model (0.75) or the Adjusted R-Square in the OLS model (0.34). Fourth, spatial autocorrelation is present in the residuals from the OLS model, but spatial autocorrelation is not present in the residuals from the GWR model or the residuals from the MGWR model. On the one hand, Moran’s I is statistically significant in the OLS model (z-score = +38.06, p = 0.00). On the other hand, Moran’s I is not statistically significant in the GWR model (z-score = +0.83, p = 0.41) or the MGWR model (z-score = −0.91, p = 0.36). Fifth, on the one hand, the relationships between the dependent variable and all of the independent variables in the OLS model are not consistent due to nonstationarity or to heteroscedasticity. Koenker (BP) [53,54] is statistically significant (Koenker (BP) = 63.19, p = 0.00). On the other hand, all of the independent variables in the GWR model are nonstationary. The Monte Carlo tests for spatial variability are all statistically significant (p = 0.00). Interestingly, two of the three independent variables that capture the physical characteristics of the interior of the residence (half baths and bedrooms), one of the three independent variables that capture the accessibility characteristics proximate to the residence (infrastructure), and the independent variable that captures the environmental characteristics of the space proximate to the residence (density) in the MGWR model are stationary. The Monte Carlo tests for spatial variability for half baths (p = 0.28), for bedrooms (p = 0.12), for infrastructure (p = 0.92), and for density (p = 0.87) are not statistically significant (p > 0.10). The latter result suggests that the relationship between the dependent variable and the accessibility characteristic infrastructure is consistent, or stationary.

5.1. Global Model

The OLS model parameters (Table 3) yield the following. The physical characteristics full baths (+99,604.63), half baths (+68,053.14), and age (+986.85), the accessibility characteristics infrastructure (+12.34) and walkability (+8369.69) as well as the environmental characteristic density (+2.25) all increase prices. If the dependent variable is the natural log of the inflation-adjusted price, then the magnitudes of the statistically significant (p < 0.01) parameters for the accessibility characteristics show that infrastructure increases prices by +0.005% and walkability increases prices by+2.99%. The latter result on the magnitude of the effect of the accessibility characteristic walkability is higher than the magnitude of the effect in the empirical literature from Hamidi et al. [55]; that is, +2.99% versus +1.10%.

5.2. Local Models

The GWR model diagnostics and the MGWR diagnostics (Table 4) yield the following. The scale (global versus local) of the independent variable effects differ between characteristics and within characteristics. The bandwidth of the accessibility characteristic infrastructure (1469) shows that the infrastructure effect on prices is global in scale; 1469 is the maximum bandwidth in the MGWR model (n − 1 = 1469). The bandwidth of the accessibility characteristic distance (339) and the bandwidth of the accessibility characteristic walkability (142) shows that the distance effect on prices is local in scale and the walkability effect on prices is local in scale; 339 is close to the mean bandwidth in the GWR model (107) and 142 is close to the mean bandwidth in the GWR model (107). The GWR model parameters and the MGWR model parameters (Table 5) and the exploratory data analyses [56] on all of the accessibility characteristic parameters (Table 6 and Figure 2) yield the following. First, the MGWR model parameters with different bandwidths, or scales, are more precise than the GWR model parameters without different bandwidths, or scales. The MGWR model standard deviations are lower than the GWR model standard deviations for all of the accessibility characteristics. Second, the five-number summary of the minimum (Min), the 25th percentile (p25), the 50th percentile (p50), the 75th percentile (p75), and the maximum (Max) for the respective model parameters and side-by-side (GWR versus MGWR) boxplots of the respective model parameters show interquartile ranges (IQR = p75 − p25) are lower for the MGWR parameters than for the GWR parameters. Second, the magnitude of the walkability effect on prices is slightly higher in the MGWR model than in the GWR model. The walkability p50 in the MGWR model is slightly higher than the walkability p50 in the GWR model.

5.3. Exploratory Spatial Data Analyses

Exploratory spatial data analyses [57] of the GWR model parameters for walkability versus the MGWR model parameters for walkability yield the following. Figure 3 maps adjusted critical t for walkability from the GWR model (3.12) versus adjusted critical t for walkability from the MGWR model (3.01), respectively. The left panel of Figure 3 maps the GWR adjusted critical t of 3.12. The point features in the left panel with an adjusted critical t greater than or equal to +3.12 are red. The point features in the left panel with an adjusted critical t lesser than or equal to −3.12 are blue. The right panel of Figure 3 maps the MGWR adjusted critical t of 3.01. The point features in the right panel with an adjusted critical t greater than or equal to +3.01 are red. The point features in the right panel with an adjusted critical t lesser than or equal to −3.01 are blue. The exploratory spatial data analyses in Figure 3 highlight the value of a different bandwidth, or scale, for the walkability independent variable in the GWR model (107) versus in the MGWR model (142). First, the MGWR model shows more positive (red) point features and more negative (blue) point features than the GWR model—53 red in the GWR model versus 80 red in the MGWR model and one blue in the GWR model versus 18 blue in the MGWR model. Second, the statistically significant (p < 0.01) point features in the GWR model and the statistically significant (p < 0.01) point features in the MGWR model all appear in clusters except for the one blue point feature in the GWR model. The red clusters in the GWR model are in the District of Columbia and in Montgomery County. The red cluster in the MGWR model is in Montgomery County. The blue clusters in the MGWR model are in Alexandria city. Third, the different bandwidths for the walkability independent variable in the GWR model (107) versus in the MGWR model (142) subtract 6 point features from the red cluster in the District of Columbia and add 33 point features to the red cluster in Montgomery County. The loss of 6 red point features from the GWR model to the MGWR model and the gain of 33 red point features from the GWR model to the MGWR model increases the mean price in the red cluster by +USD5705.91 consistent with an increase in the mean number of bedrooms by +0.23 bedrooms and a decrease in the mean age by −7.69 years. Interestingly, the increase in bandwidth, or scale, increases the mean, minimum Euclidean distance to a Metro station by +31.49 m which, at the same time, decreases the mean infrastructure suitable for pedestrians by −567.00 per square kilometer, decreases the mean walkability score by −0.72, and decreases the mean density of jobs and housing units by −1527.00 per square kilometer. Fourth, the different bandwidths for the walkability independent variable in the GWR model (107) versus in the MGWR model (142) also subtract the one blue point feature in Alexandria city and add a blue cluster of 10 point features and a blue cluster of 8 point features in Alexandria city. The loss of one blue point feature from the GWR model to the MGWR model and the gain of 18 blue point features from the GWR model to the MGWR model helps to understand why the difference in the mean price from the red cluster in the MGWR model to the blue clusters in the MGWR model is so great. The difference is an enormous −USD201,125.74, consistent with a decrease in the mean number of bedrooms of −0.99 bedrooms. The mean, minimum Euclidean distance to a Metro station increases by +440.12 m, which, at the same time, decreases the mean density of jobs and housing units by a staggering −2566.60 per square kilometer. Therefore, the mean infrastructure suitable for pedestrians in the block group and the mean walkability score in the block group ought to decrease; however, unexpectedly, both increase instead by +1017.29 per square kilometer and by +1.12, respectively. Such results highlight the costliness of noise as a negative externality in the context of air travel to and from Reagan National Airport even if walkability scores in the block group increase in Alexandria city.

6. Discussion

The selection of transactions within Euclidean-distance buffers representative of the pedestrian trip behavior of Metro passengers helps to realistically contextualize local models of walkability. The local models highlight the value of novel multiscale, local models, which free independent variables to manifest at different scales of analysis rather than restrict all of the independent variables to manifest at the same scale of analysis. The development of such methods is consistent with the trend in the social sciences toward realistically complex models [58]. The application of such methods is consistent with the trend in the valuation literature toward methodological sophistication [59].
The application of such methods also answers a longstanding call in the empirical literature to improve model specification [39] so as to accurately and to precisely estimate the price effect of walkability. Overall, results show how scale affects walkability valuation estimation in residential markets and where the market succeeds and/or fails to capitalize the value of walkability into prices. The magnitude of the price effect of walkability is probably high here because transactions, not assessments [55], best measure outcomes in residential markets. Methodology indeed matters [39], especially when infrastructure manifests on prices at a global scale and walkability manifests on prices at a local scale. The ubiquity of the former effect helps to understand why the empirical evidence characterizes the additional value available for capture from infrastructure as modest [40].

7. Conclusions

The limitations of this study are as follows. First, the data availability constraints on the origins of the pedestrian trips to the respective Metro station destinations limits the spatial resolution of the walk distance buffers from Metro stations [33] to select transactions. By implication, the walk distance buffers imprecisely match the behavior of passengers to the infrastructure for pedestrians in the selection of transactions. Second, the temporal context for the MRIS data is from 2002 to 2014. However, the walkability scores for the block groups are only from 2010 [45]. By implication, one walkability score for block groups precludes the exploration of how changes in walkability scores within the temporal context of the transaction data interact with prices to precisely estimate the value of walkability. Third, total square footage measures how interior size affects property prices. Unfortunately, MRIS data on total square footage are unreliable. By implication, the number of full baths, the number of half baths, and the number of bedrooms imprecisely measure interior size, so the physical characteristics of each property in this study only approximate interior size.
Nonetheless, the contributions of this study are as follows. Models whose dependent variable pool transactions proximate to a heavy rail system show how to improve specifications to estimate walkability market valuation. A +2.99% premium from a global model result is beyond the outstanding estimate in the empirical literature. Specification improvements from local models notwithstanding, the contribution of the local model results centers on hyperlocal walkability valuation. The dramatic cluster of premiums contradicts the argument that walkability serves as a systemic price buffer in residential markets, especially within walking distances of infrastructure like heavy rail stations. Nonetheless, future research to explore the spatial variation in the value of walkability is important. To that end, exploratory research on how noise within the flight paths for arrivals and for departures to and from Reagan National Airport offsets the premium walkability confers to local residential property elsewhere is ongoing.
Such a conclusion on hyperlocal valuation may disappoint walkability advocates, but the context for price changes in the residential market is different today [60]. In the past, price changes were a function of infrastructure investments. Presently, the public values residences more so as an investment and less so as a domicile than in the past. Speculation in an intranational market or in an international market suggests that the infrastructure for pedestrian trips or the infrastructure for rail trips not elsewhere available adds to the amenity bundle future renters will value after the transaction. Speculators may personally prefer less accessible developments with less infrastructure for pedestrian trips, as is the case on the periphery of the Metro service area, but future renters of the residence may need to forego internal (household) costs to own and to operate private vehicles and so may personally prefer more accessible developments with more infrastructure for pedestrian trips in the core of the Metro service area.
The policy implication of the results relates to the problem stakeholders confront to replicate the worldwide demand for the residential stock proximate to the Medical Center Metro station on the Red line. The evident buffer is a hyperlocal cluster of transactions in proximity to institutions like the National Institutes of Health whose work is not immune to macroeconomic shocks but is certainly less vulnerable and so continues to attract human capital worldwide. Indeed, self-imposition of an ad valorem tax by developers in the Red line corridor north of the Medical Center Metro station to redevelop land proximate to Metro stations so as to advantage walkability will need to accommodate passenger walk distances of at least 1.61 km to succeed.
The results of this study with regard to the magnitude of the price effect of walkability are not unquestionably generalizable for the following reasons. First, the labor market in the Metro service area is unlike labor markets elsewhere given the presence of federal government employment and the concentration of professional services and business services [61]. To that end, the majority of federal government employment is in the District of Columbia, home to the hub of the Metro system. Besides the first core of employment in the District of Columbia in the public sector, Fairfax County, the jurisdiction in Virginia on the west border of the District of Columbia, is the second core of employment in the public sector. Indeed, the location quotient for professional services and for business services (1.83) supportive of federal government activity is highest in the Metro service area and in the surrounding metropolitan area amongst the twelve largest metropolitan areas nationwide. Second, the Metro system is unlike heavy rails systems elsewhere. Half of the stations in the Metro system serve a federal facility and one in two passengers are, directly or indirectly, federal government employees [30]. However, WMATA receives no dedicated funds for capital costs or for operational costs. Rather, member jurisdictions subsidize WMATA annually. WMATA also generates annual revenue from internal operations such as passenger fares. Indeed, WMATA generates 44.33% of operational funds from fare revenues. For the transit agencies in areas with populations greater than one million, the percentage is 35.53%, and for the nation, the percentage is 12.46%. Further, the Metro heavy rail system ranks second amongst heavy rail systems nationwide in fare revenues per total operational expenses, also known as cost recovery ratio; WMATA recovers 61.6% of operational costs for the heavy rail system from fares. The overreliance on fare revenues is problematic economically because fiscal pressure to increase fares foreshadows a future disequilibrium when fare increases decrease ridership, which ultimately decreases operations, which further decreases ridership in a vicious cycle. For the above reasons, restraint in the generalization of the results from the Metro service area on the price premium for walkability is prudent.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The raw data on transactions analyzed in this manuscript are not readily available because of proprietary limitations.

Acknowledgments

The raw data on transactions analyzed in this manuscript were donated by the Center for Regional Analysis at George Mason University.

Conflicts of Interest

The author declares no conflicts of interest.

References

  1. Wise, D. Public Transportation: Federal Role in Value Capture Strategies for Transit Is Limited, but Additional Guidance Could Help Clarify Policies. 2010. Available online: https://www.gao.gov/products/gao-10-781 (accessed on 28 October 2024).
  2. Levinson, D.; Istrate, E. Access for Value: Financing through Land Value Capture. 2011. Available online: https://www.brookings.edu/research/access-for-value-financing-transportation-through-land-value-capture/ (accessed on 28 October 2024).
  3. Vadali, S. Using the Economic Value Created by Transportation to Fund Transportation: A Synthesis of Highway Practice. 2014. Available online: https://nap.nationalacademies.org/read/22382/chapter/1 (accessed on 28 October 2024).
  4. Knight, R.; Trygg, L. Evidence of land use impacts of rapid transit systems. Transportation 1977, 6, 231–247. [Google Scholar] [CrossRef]
  5. Higgins, C.; Kanaroglou, P. Forty years of modelling rapid transit’s land value uplift in North America: Moving beyond the tip of the iceberg. Transp. Rev. 2016, 36, 610–634. [Google Scholar] [CrossRef]
  6. Debrezion, G.; Pels, E.; Rietveld, P. The impact of railway stations on residential and commercial property values: A meta-analysis. J. Real Estate Financ. Econ. 2007, 35, 161–180. [Google Scholar] [CrossRef]
  7. Mohammad, S.; Graham, D.; Melo, P.; Anderson, R. A meta–analysis of the impact of rail projects on land and property values. Transp. Res. A-Pol. Pract. 2013, 50, 158–170. [Google Scholar] [CrossRef]
  8. Hamidi, S.; Kittrell, K.; Ewing, R. Value of transit as reflected in U.S. single-family home premiums: A meta-analysis. Transp. Res. Rec. 2016, 2543, 108–115. [Google Scholar] [CrossRef]
  9. Yen, B.; Mulley, C.; Shearer, H. Different stories from different approaches in evaluating property value uplift: Evidence from the Gold Coast light rail system in Australia. Transp. Res. Rec. 2019, 2673, 11–23. [Google Scholar] [CrossRef]
  10. Herrmann, T.; Boisjoly, G.; Ross, N.; El-Geneidy, A. The missing middle: Filling the gap between walkability and observed walking. Transp. Res. Rec. 2017, 2661, 103–110. [Google Scholar] [CrossRef]
  11. Tuydes-Yaman, H.; Karatas, P. Evaluation of Walkability and Pedestrian Level of Service. In Engineering Tools and Solutions for Sustainable Transportation Planning; Knoflacher, H., Ocalir-Akunal, E., Eds.; IGI Global: Hershey, PA, USA, 2017; pp. 30–57. [Google Scholar]
  12. Litman, T. Economic value of walkability. Transp. Res. Rec. 2003, 1828, 3–11. [Google Scholar] [CrossRef]
  13. Brigham, E. The determinants of residential land values. Land Econ. 1965, 41, 325–334. [Google Scholar] [CrossRef]
  14. Ball, M. Recent empirical work on the determinants of relative house prices. Urban Stud. 1973, 10, 213–233. [Google Scholar] [CrossRef]
  15. Stull, W. Community environment, zoning, and the market value of single-family homes. J. Law Econ. 1975, 18, 535–557. [Google Scholar] [CrossRef]
  16. Alonso, W. Location and Land Use; Harvard University Press: Cambridge, MA, USA, 1964. [Google Scholar]
  17. Haig, R. Towards an understanding of the metropolis. Q. J. Econ. 1926, 40, 402–434. [Google Scholar] [CrossRef]
  18. Muth, R. Cities and Housing; University of Chicago Press: Chicago, IL, USA, 1969. [Google Scholar]
  19. Armstrong, R.; Rodríguez, D. An evaluation of the accessibility benefits of commuter rail in east Massachusetts using spatial hedonic price functions. Transportation 2006, 33, 21–43. [Google Scholar] [CrossRef]
  20. Duncan, M. The synergistic influence of light rail stations and zoning on home prices. Environ. Plan. A 2011, 43, 2125–2142. [Google Scholar] [CrossRef]
  21. Rosen, S. Hedonic prices and implicit markets: Product differentiation in pure competition. J. Polit. Econ. 1974, 82, 34–55. [Google Scholar] [CrossRef]
  22. Can, A. The measurement of neighborhood dynamics in urban house prices. Econ. Geogr. 1990, 66, 254–272. [Google Scholar] [CrossRef]
  23. Can, A. Specification and estimation of hedonic housing price models. Reg. Sci. Urban Econ. 1992, 22, 453–474. [Google Scholar] [CrossRef]
  24. Slater, P. Spatial and temporal effects in residential sales prices. J. Am. Stat. Assoc. 1973, 68, 554–561. [Google Scholar] [CrossRef]
  25. Slater, P. Disaggregated spatial-temporal analysis of residential sales prices. J. Am. Stat. Assoc. 1974, 69, 358–363. [Google Scholar] [CrossRef]
  26. Dubé, J.; Legros, D. Spatial econometrics and the hedonic pricing model: What about the temporal dimension? J. Prop. Res. 2014, 31, 333–359. [Google Scholar] [CrossRef]
  27. Fotheringham, A.; Crespo, R.; Yao, J. Exploring, modelling and predicting spatiotemporal variation in house prices. Ann. Reg. Sci. 2015, 54, 417–436. [Google Scholar] [CrossRef]
  28. Walk21. Integrating Walking + Public Transport. 2024. Available online: https://walk21.com/resources/walking-and-public-transport/ (accessed on 28 October 2024).
  29. Le, V.; Dannenberg, A. Moving towards physical activity targets by walking to transit: National household transportation survey: 2001–2017. Am. J. Prev. Med. 2020, 59, e115–e123. [Google Scholar] [CrossRef] [PubMed]
  30. Puentes, R. Washington’s Metro: Deficits by Design. 2004. Available online: https://www.brookings.edu/wp-content/uploads/2016/06/20040603_puentes.pdf (accessed on 28 October 2024).
  31. Institute for Transportation & Development Policy. 2024. Better Together: Walkable Cities and Public Transport. 2024. Available online: https://itdp.org/2024/08/15/better-together-walkable-cities-and-public-transport/ (accessed on 28 October 2024).
  32. National Association of Realtors. 2023 Community & Transportation Preference Survey. 2023. Available online: https://www.nar.realtor/infographics/2023-community-transportation-preference-survey (accessed on 28 October 2024).
  33. Washington Metropolitan Transit Authority. Metrorail Station Investment Strategy. 2016. Available online: https://planitmetro.com/uploads/MISIS_Report_August_2016.pdf (accessed on 28 October 2024).
  34. Litman, T. Evaluating Public Transit Benefits and Costs. 2024. Available online: https://www.vtpi.org/tranben.pdf (accessed on 28 October 2024).
  35. Guo, Y.; Peeta, S.; Somenahalli, S. The impact of walkable environment on single-family residential property values. J. Transp. Land Use 2017, 10, 1–20. [Google Scholar] [CrossRef]
  36. Yang, L.; Wang, B.; Zhou, J.; Wang, X. Walking accessibility and property prices. Transp. Res. D-Transp. Environ. 2018, 62, 551–562. [Google Scholar] [CrossRef]
  37. Petheram, S.; Nelson, A.; Miller, M.; Ewing, R. Use of the real estate market to establish light rail station catchment areas. Transp. Res. Rec. 2013, 2357, 95–99. [Google Scholar] [CrossRef]
  38. Gilderbloom, J.; Riggs, W.; Meares, W. Does walkability matter? An examination of walkability’s impact on housing values, foreclosures and crime. Cities 2015, 42 Pt A, 13–24. [Google Scholar] [CrossRef]
  39. Boyle, A.; Barrilleaux, C.; Scheller, D. Does walkability influence housing prices? Soc. Sci. Quart. 2014, 95, 852–867. [Google Scholar] [CrossRef]
  40. Li, W.; Joh, K.; Lee, C.; Kim, J.-H.; Park, H.; Woo, A. Assessing benefits of neighborhood walkability to single-family property values: A spatial hedonic study in Austin, Texas. J. Plan. Educ. Res. 2015, 35, 471–488. [Google Scholar] [CrossRef]
  41. Yang, S.; Peng, C.; Hu, S.; Zhang, P. Geospatial modelling of housing rents from TOD using MGWR and implications on integrated transportation-land use planning. Appl. Geogr. 2024, 170, 103356. [Google Scholar] [CrossRef]
  42. Washington Metropolitan Area Transit Authority. Time for Those Walking Shoes, Part 1. 2013. Available online: https://planitmetro.com/2013/07/30/time-for-those-walking-shoes-part-1/ (accessed on 28 October 2024).
  43. Environmental Systems Research Institute. ArcGIS 10.8. 2020. Available online: https://www.esri.com/en-us/arcgis/about-arcgis/overview (accessed on 28 October 2024).
  44. United States Environmental Protection Agency. Smart Location Database. 2021. Available online: https://www.epa.gov/smartgrowth/smart-location-mapping#SLD (accessed on 28 October 2024).
  45. United States Environmental Protection Agency. National Walkability Index. 2021. Available online: https://www.epa.gov/smartgrowth/smart-location-mapping#walkability (accessed on 28 October 2024).
  46. Federal Housing Finance Agency. Home Price Index. 2016. Available online: https://www.fhfa.gov/DataTools/Downloads/Pages/House-Price-Index.aspx (accessed on 28 October 2024).
  47. Oshan, T.; Li, Z.; Kang, W.; Wolf, L.; Fotheringham, A. MGWR: A Python implementation of multiscale geographically weighted regression for investigating process spatial heterogeneity and scale. ISPRS Int. Geo.-Inf. 2019, 8, 269. [Google Scholar] [CrossRef]
  48. Fotheringham, A.; Brunsdon, C.; Charlton, M. Geographically Weighted Regression: The Analysis of Spatially Varying Relationships; Wiley: Hoboken, NJ, USA, 2002. [Google Scholar]
  49. Akaike, H. Information Theory and an Extension of the Maximum Likelihood Principle. In 2nd International Symposium on Information Theory; Petrov, B., Csáki, F., Eds.; Akadémiai Kiadó: Budapest, Hungary, 1973; pp. 267–281. [Google Scholar]
  50. Hurvich, C.; Simonoff, J.; Tsai, C.-L. Smoothing parameter selection in nonparametric regression using an improved Akaike information criterion. J. R. Stat. Soc. B 1998, 60, 271–293. [Google Scholar] [CrossRef]
  51. Fotheringham, A.; Yang, W.; Kang, W. Multiscale geographically weighted regression (MGWR). Ann. Am. Assoc. Geogr. 2017, 107, 1247–1265. [Google Scholar] [CrossRef]
  52. Breusch, T.; Pagan, A. A simple test for heteroscedasticity and random coefficient variation. Econometrica 1979, 47, 1287–1294. [Google Scholar] [CrossRef]
  53. Yu, H.; Fotheringham, A.; Li, Z.; Oshan, T.; Kang, W.; Wolf, L. Inference in multiscale geographically weighted regression. Geogr. Anal. 2020, 52, 87–106. [Google Scholar] [CrossRef]
  54. Koenker, R. A note on studentizing a test for heteroscedasticity. J. Econom. 1981, 17, 107–112. [Google Scholar] [CrossRef]
  55. Hamidi, S.; Bonakdar, A.; Keshavarzi, G.; Ewing, R. Do Urban Design qualities add to property values? An empirical analysis of the relationship between Urban Design qualities and property values. Cities 2020, 98, 102564. [Google Scholar] [CrossRef]
  56. Tukey, J. Exploratory Data Analysis; Addison-Wesley: Reading, PA, USA, 1977. [Google Scholar]
  57. Mennis, J. Mapping the results of geographically weighted regression. Cartogr. J. 2013, 43, 171–179. [Google Scholar] [CrossRef]
  58. Best, N.; Spiegelhalter, D.; Thomas, A.; Brayne, C. Bayesian analysis of realistically complex models. J. R. Stat. Soc. A Stat. 1996, 159, 323–342. [Google Scholar] [CrossRef]
  59. Krause, A.; Bitter, C. Spatial econometrics, land values and sustainability: Trends in real estate valuation research. Cities 2012, 29 (Suppl. S2), S19–S25. [Google Scholar] [CrossRef]
  60. Shiller, R. Irrational Exuberance; Princeton University Press: Princeton, NJ, USA, 2015. [Google Scholar]
  61. Perrins, G.; Nilsen, D. Industry Dynamics in the Washington, DC, Area: Has a Second Job Core Emerged? 2006. Available online: https://www.bls.gov/opub/mlr/2006/12/art1full.pdf (accessed on 28 October 2024).
Figure 1. Metro service area.
Figure 1. Metro service area.
Futuretransp 04 00064 g001
Figure 2. Side-by-side (GWR versus MGWR) boxplots of parameter estimates for distance, for infrastructure, and for walkability.
Figure 2. Side-by-side (GWR versus MGWR) boxplots of parameter estimates for distance, for infrastructure, and for walkability.
Futuretransp 04 00064 g002
Figure 3. Positive (red) walkability t versus negative (blue) walkability t: GWR model (left) versus MGWR model (right).
Figure 3. Positive (red) walkability t versus negative (blue) walkability t: GWR model (left) versus MGWR model (right).
Futuretransp 04 00064 g003
Table 1. Descriptives.
Table 1. Descriptives.
Variable Metro Station BuffersMetro Service Area
Characteristic M 1SD 2MSD
Dependent
Price (USD1000.00)307.24150.58312.88143.13
Independent
Physical
Interior
Full baths2.060.752.390.74
Half baths0.640.610.790.58
Bedrooms3.260.833.740.79
Quality
Age53.0030.3232.9721.58
Accessibility
Distance690.06361.715763.464186.01
Infrastructure4146.481419.832840.191364.82
Walkability14.752.3111.643.63
Environmental
Space
Density5595.9811,176.461444.403062.62
n147029,422
1 M = Mean. 2 SD = Standard Deviation.
Table 2. Descriptives for price by year.
Table 2. Descriptives for price by year.
Price (USD1000.00)
YearnM 1SD 2Min 3Max 4
200283421.78195.54165.00950.00
200378455.82198.31162.20903.00
2004137453.71233.61162.001020.00
2005153502.85216.39165.001045.00
2006139467.04202.81180.00985.00
2007145478.57200.27186.381010.00
200897456.94219.10180.001050.00
2009118526.64256.97167.501056.00
2010109436.67211.75164.10930.00
201189470.39229.34165.001049.00
201281494.33233.76165.001055.00
2013124510.46245.88165.001040.00
2014117517.37247.42171.001045.00
Total1470478.75224.76186.38 51056.00 6
1 M = Mean. 2 SD = Standard Deviation. 3 Min = Minimum. 4 Max = Maximum. 5 2007. 6 2009.
Table 3. Parameter estimates from OLS model.
Table 3. Parameter estimates from OLS model.
Characteristic Coefficient 1SE 2VIF 3
Intercept−167,363.47 ***27,951.75
Physical
InteriorFull baths+99,604.63 ***5209.771.49
Half baths+68,053.14 ***5564.581.11
Bedrooms−7219.334568.551.41
Quality
Age+986.85 ***121.501.32
Accessibility
Distance+15.04 *9.011.04
Infrastructure+12.34 ***2.551.28
Walkability+8369.69 ***1549.571.25
Environmental
Space
Density+2.25 ***0.311.15
n
AICc
Adjusted R-Square
Moran’s I
Koenker (BP)
1470
3583.96
0.34
+0.47 4
63.19 5
1 * p < 0.01 and *** p < 0.10. 2 SE = Standard Error. 3 VIF = Variance Inflation Factor. 4 Moran’s I is statistically significant (z-score = +38.06, p = 0.00) so likelihood residual clustering is due to chance is less than 1.00%. 5 Chi-square (df = 8) is statistically significant (p < 0.01) so reject null hypothesis of model stationarity or model homoscedasticity.
Table 4. Diagnostics from GWR model and from MGWR model.
Table 4. Diagnostics from GWR model and from MGWR model.
Characteristic GWR 1MGWR 1
InterceptBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for spatial variability (p)
107
247.07
3.12
0.00 ***
53
57.04
3.33
0.00 ***
Physical
Interior
Full bathsBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for spatial variability (p)
107
247.07
3.12
0.00 ***
190
17.40
2.99
0.00 ***
Half bathsBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for spatial variability (p)
107
247.07
3.12
0.00 ***
977
3.23
2.42
0.28
BedroomsBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for Spatial variability (p)
107
247.07
3.12
0.00 ***
927
2.93
2.39
0.12
Quality
AgeBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo Test for spatial variability (p)
107
247.07
3.12
0.00 ***
81
35.99
3.20
0.00 ***
Accessibility
DistanceBandwidth
Effective number of parameters
Adjusted Critical t
Monte Carlo Test for spatial variability (p)
107
247.07
3.12
0.00 ***
339
6.69
2.68
0.01 ***
InfrastructureBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for spatial variability (p)
107
247.07
3.12
0.00 ***
1469
1.35
2.09
0.92
WalkabilityBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for spatial variability (p)
107
247.07
3.12
0.00 ***
142
18.98
3.03
0.00 ***
Environmental
Space
DensityBandwidth
Effective number of parameters
Adjusted critical t
Monte Carlo test for spatial variability (p)
107
247.04
3.12
0.00 *
1218
1.30
2.07
0.87
1 * p < 0.01 and *** p < 0.10.
Table 5. Parameter estimates from GWR model and from MGWR model.
Table 5. Parameter estimates from GWR model and from MGWR model.
GWRMGWR
Characteristic M 1SD 2MSD
Intercept+0.020.71−0.0020.73
Physical
Interior
Full baths+0.260.16+0.280.13
Half baths+0.140.10+0.160.02
Bedrooms+0.070.11+0.070.03
Quality
Age+0.010.19+0.020.16
Accessibility
Distance+0.010.15+0.0040.08
Infrastructure+0.050.10+0.040.003
Walkability+0.020.13+0.020.10
Environmental
Space
Density+0.100.56+0.010.01
n
Effective number of parameters
AICc
Adjusted R-Square
Moran’s I
1470
247.07
2437.37
0.75
+0.01 3
1470
144.91
2266.79
0.76
−0.01 4
1 M = Mean. 2 SD = Standard Deviation. 3 Moran’s I is not statistically significant (z-score = +0.83, p = 0.41) so residual pattern is likely not different than random. 4 Moran’s I is not statistically significant (z-score = −0.91, p = 0.36) so residual pattern is likely not different than random.
Table 6. Five-number summary of parameters for accessibility characteristics from GWR model and from MGWR model.
Table 6. Five-number summary of parameters for accessibility characteristics from GWR model and from MGWR model.
Characteristic GWRMGWR
Accessibility
DistanceMin 1
p25 2
p50 3
p75 4
Max 5
−0.37
−0.08
+0.01
+0.08
+0.56
−0.17
−0.01
+0.02
+0.04
+0.18
InfrastructureMin
p25
p50
p75
Max
−0.24
−0.01
+0.04
+0.09
+0.35
+0.03
+0.03
+0.04
+0.04
+0.04
WalkabilityMin
p25
p50
p75
Max
−0.43
−0.05
+0.01
+0.09
+0.41
−0.21
−0.04
+0.02
+0.08
+0.31
n14701470
1 Min = Minimum. 2 p25 = 25th percentile. 3 p50 = 50th percentile. 4 p75 = 75th percentile. 5 Max = Maximum.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zolnik, E. Capturing the Value of Walkability. Future Transp. 2024, 4, 1334-1349. https://doi.org/10.3390/futuretransp4040064

AMA Style

Zolnik E. Capturing the Value of Walkability. Future Transportation. 2024; 4(4):1334-1349. https://doi.org/10.3390/futuretransp4040064

Chicago/Turabian Style

Zolnik, Edmund. 2024. "Capturing the Value of Walkability" Future Transportation 4, no. 4: 1334-1349. https://doi.org/10.3390/futuretransp4040064

APA Style

Zolnik, E. (2024). Capturing the Value of Walkability. Future Transportation, 4(4), 1334-1349. https://doi.org/10.3390/futuretransp4040064

Article Metrics

Back to TopTop