Guidelines for Analysis of Investments in Bicycle Facilities (2006)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

H-1 APPENDIX H COMMUNITY LIVABILITY BENEFITS The Value of Bicycle Trail Proximity on Home Purchases INTRODUCTION Many cities—through public dialogues, community initiatives, and other land use-transportation policies—are developing strategies to increase the “livability” of their communities. While “livability” is a relatively ambiguous term, there is emerging consensus on the following: the ease by which residents can travel by walking or bicycling represents a critical component of this goal. Communities well endowed with non-motorized infrastructure, either in the form of sidewalks, bicycle paths, or compact and mixed land uses are hypothesized to be more livable than those without. This is an often relied-upon argument used by advocates of bicycle paths or sidewalks. If livability is a cherished commodity among residents, and one important component of livability includes bicycle paths, then prox- imity to bicycle paths should be capitalized into the value of home purchases. Documenting this relationship would go a long way for advocates of bicycle facilities who often seek ways to monetize the value of these facilities. Such an endeavor would be especially bene- ficial since bicycle facilities are non-market goods, making it difficult to attach an economic value to them. Social or economic benefits can be measured either through stated preferences, in which users are asked to attach a value to non-market goods, or through revealed preferences. The revealed preference approach measures individuals’ actual behavior. In this study we measure homebuyers’ revealed preferences in the form of hedonic modeling to learn if and how much residents value proximity to bicycle paths. The first part of this paper reviews previous literature on hedonic modeling focusing primarily on the dimension of open space and trails. The second part describes the setting for this work, our data, descriptive statistics, and methodological approach. Part three describes the results of a hedonic regression model, and part four reports on the policy implications and relevant conclusions. REVIEW OF RELEVANT LITERATURE AND CONCEPTS Discerning the relative value of non-market goods using hedonic modeling techniques is a method with a long history. Taylor (229) used what are now called hedonic techniques to explain the price of cotton with its internal qualities, and later applications by Lancaster (128) and Rosen (129) standardized the method for consumer prod- ucts such as houses. An extensive review of this literature (130) doc- uments nearly 200 applications that have examined home purchases to estimate values of several home attributes including structural fea- tures (e.g., lot size, a home’s finished square footage, and number of bedrooms), internal and external features (e.g., fireplaces, air condi- tioning, garage spaces, and porches), natural environment features (e.g., scenic views), attributes of the neighborhood and location (e.g., crime, golf courses, and trees), public services (e.g., school and infra- structure quality), marketing, and financing. As the literature describes various methods to assign value to housing characteristics, there exist opportunities to increase the explanatory power of hedonic models. Recent contributions include accessibility, perceived school quality, and measures of environ- mental amenities. For example, Franklin and Waddell (230) used a hedonic model to predict home prices in King County, Washington, as a function of accessibility to four types of activities (Commercial, University, K-12 Schools, and Industrial). In assessing the relation- ship between public school quality and housing prices, Brasington (231) found that proficiency tests, per-pupil spending, and student/ teacher ratios most consistently capitalize into the housing market. Earnhart (232) combined discrete-choice hedonic analysis with choice-based conjoint analysis to place a value on adjacent environ- mental amenities such as lakes and forests. Our application here focuses on the relative impact of bicycle lanes and trails. To the casual observer, bicycle lanes and trails may be con- sidered as a single facility where any type of bicycle trail would have the same attraction. More careful thinking, however, suggests other- wise, especially for different types of bicycle facilities. Consider, for example, the three different types of trails/lanes shown in Figure 20. Some trails are on existing streets (demarcated by paint striping, hereafter “on-street lanes”); some trails are adjacent to existing road- ways (hereafter “roadside trails”) but are separated by curbs or mild landscaping (these facilities are sometimes referred to as “black sidewalks” because they are nothing more than blacktop in the usual location of sidewalks); other trails are clearly separated from traffic and often within open spaces (hereafter “non roadside trails”). For this last category, it is important to explain and control for the degree to which open space versus the bike trail contained within the open space contribute to a home’s value. In many metropolitan areas bike trails and open space share a spatial location and at minimum exhibit similar recreational qualities. On-street lanes or roadside trails are often on or near roads. In some cases they will be on well-used col- lector streets or trunk highways; in others they may be on neighbor- hood arterial streets. Home buyers tend to dislike proximity to busy roadways. Much of the attraction of these facilities therefore depends on the design speed of the roadway facility and the average daily traf- fic. Any research failing to account for any of these factors will mis- estimate the independent value of bicycle trails. It is therefore important to consider relevant literature estimating the value of open space. For example, Quang Do (136) found that homes abutting golf courses sell for a 7.6 percent premium over others. Other studies include measures of proximity and size of vari- ous open spaces (233, 234). Geoghegan (133) compared the price effects of the amount of permanent and developable open space within a one-mile radius. Smith et al. (235) examined the distinction between fixed and adjustable open spaces along a new Interstate high- way corridor. Other approaches further disaggregate developable and non-developable open space in terms of ownership type and land cover (236 ). Some studies seek to attach values to views of open space. Benson et al. (132) created a series of dummy variables for

four different qualities of ocean views, as well as lake and mountain views. Luttik (135) combined the vicinity and view approaches, dividing the geography into three levels of proximity. Anderson and West’s work (131) is particularly helpful for this spe- cific application. They modeled both proximity and size of six specific open space categories, comparing effects on home prices between the city and suburb. They found that proximity to golf courses, large parks, and lakes has a positive effect on home prices in the city, with no significant results in the suburbs. The effects of open space on home prices also increased with the size of the open space. Proxim- ity to small parks and cemeteries tended to reduce sale prices. To our knowledge, only one application focuses on proximity to bicycle trails. Lindsey (72) performed a hedonic analysis of 9,348 home sales, identifying properties falling inside or outside a half-mile buffer around fourteen greenways in Marion County, Indiana. This research found that some greenways have a positive, significant effect on property values while others have no significant effect. A survey in Vancouver found that the majority of realtors perceive little effect of bicycle trails on home values, either positive or negative (237 ). How- ever, two-thirds of respondents also indicated that they would use bicycle trail proximity as a selling point. Given the novelty of the application presented herein, theory is derived from a combination of sources, including existing published work (described in part above), consumer theory, and anecdotal evi- dence. Our first underpinning is derived from a local county com- missioner who claims that bike facilities—like libraries—are goods everyone appreciates (238). Such a claim maps well with the asser- tions of bicycle trail advocates. Assuming an ability to account for the possible disutility of living on a busy arterial, bicycle facilities—no matter their type—positively contribute to home value. However, this hypothesis needs to be tempered based on the findings of Anderson H-2 and West. Their analysis suggests that open spaces and by associa- tion, bicycle facilities, may be perceived and valued differently depending on whether they are located in the city or suburbs. Unlike other attributes which tend to be more universally valued (e.g., home size, number of bathrooms), we hypothesize that trails may be more appreciated by a subset of the population. Households who choose to live in the city are more likely to walk or bike, partic- ularly to work, (53, 239) and therefore more likely to value bicycle facilities. Because we specify three different types of facilities with two populations who may value such facilities differently, we present Figure 20 displaying the nature and relative magnitude of our hy- pothesized relationships. SETTING AND DATA Our investigation is based in the Twin Cities (Minnesota) Metro- politan Area which proves to be an almost ideal laboratory for a vari- ety of reasons. First, the Twin Cities boasts an almost unparalleled system of off-street bike paths for a major metropolitan area in the United States, totaling more than 2,722 kilometers (1,692 miles). While not nearly as extensive, striped on-street bike lanes are com- mon as well. The network of on- and off-street trails is accessible to most Twin Citians, with 90 percent of homes within 1,600 meters (one mile) of an off-street trail. In fact, in many communities within the metropolitan area, over 90 percent of the homes have some form of facility within 400 meters (one-quarter mile). Second, several municipalities and county governments pursue active roles in constructing and maintaining these facilities. The Grand Rounds Parkway in Minneapolis, considered by many to be the crown jewel of parks and recreational trails in Minnesota, con- Hypothesized Relationship with Home Value City Residents Suburban Residents ON- STREET BICYCLE LANE NON- ROAD SIDE BICYCLE TRAIL ROAD- SIDE BICYCLE TRAIL Figure 20. Representative photographs of an on-street bicycle lane, non-roadside trail, and roadside trail, hypothesized relationships.

sists of 69 km (43 miles) of off-street paved trails along the city’s chain of lakes, the Mississippi River, and Minnehaha Creek. Hen- nepin County, which includes the city of Minneapolis and many of its suburbs, works in cooperation with the Three Rivers Park District to build and maintain the largest network of off-street trails in the metro area (240). Many off-street trails in Hennepin and other coun- ties are located on former railroad rights of way for the dual purposes of recreation and preservation of the land for future transit corridors. Other off-street trails in the Twin Cities follow arterial and collector streets. The cities of Chanhassen, Eden Prairie, and Plymouth have extensive networks of these roadside trails, with somewhat smaller networks in Maple Grove, Roseville, Eagan, and Apple Valley. Roseville is the only inner-ring suburb with a substantial network of off-street trails. Third, Twin Citians appear to cherish such trails, particularly in the summer months. For example, Minneapolis ranks among the top in the United States in the percentage of workers com- muting by bicycle (55). Consistent with the prevailing literature, our hedonic model assumes a competitive market in which homebuyers are seeking a set of home attributes that can be tied to a location. Locations are defined by structural attributes (S ) (including internal and external attributes), neighborhood characteristics (N ), location and accessibility (L), and environmental amenities (A). We build an equilibrium hedonic price function on these assumptions, where the market price of a home (Ph) depends on the quantities of its various attributes: The Regional Multiple Listing Services of Minnesota, Inc., (RMLS) maintains home sale data from major real estate brokers in Minnesota. This database includes all home sales in Anoka, Carver, Dakota, Hennepin, Ramsey, Scott, and Washington counties in 2001, P P S N L Ah = ( ), , , H-3 totaling 35,002xxii home sale purchases, including structural attributes of each home. The address of each home was mapped and paired with GIS features for spatial analysis using ArcGIS. Table 26 lists each variable, its definition, and descriptive statistics. We measure location attributes through simple calculations of linear distance to the nearest central business district (either Minneapolis or St. Paul) (cbdnear) and the nearest major highway (hwynear). A third location variable (busy) indicates the presence of an arterial street fronting the home. Neighborhood attributes include school district and demographic variables. Standardized test scores capitalize into home sale prices and are an effective measure of perceived school quality (231). Mca5_att represents the sum of the average math and reading scores achieved by fifth grade students taking the Minnesota Comprehensive Assessment. Scores associated with suburban homes are measured at the school district level, while Minneapolis and St. Paul scores are assigned to elementary school attendance areas. Demographic vari- ables are derived from the 2000 United States Census. We include the percentage of people in the census tract who do not classify them- selves as Caucasian (Pctnonwt) and the average number of people in each household in the census tract (Avghhsize). MEASURES OF INTEREST AND METHODOLOGY Measures of Distance to Bicycle Facility The measures of interest for this application center on bicycle facil- ities and to a certain extent, open space. Examples of the facilities and trails in this setting are shown in Figure 21. Detailed GIS data allowed us to discern all bike trails in the region, separately identifying Variable Name Description Mean Standard Deviation Median Contrnr 1023.55947.901276.31CITY: distance to nearest on-street bicyble lane (meters) Cnrtrnr 711.29517.82799.42CITY: distance to nearest non-roadside bicycle trail (meters) Crstrnr 1219.16716.201293.81CITY: distance to nearest roadside bicycle trail (meters) Sontrnr 979.822240.181580.51 )SUBURBS: distance to nearest on-street bicycle lane (meters Snrtrnr 602.921732.291099.89 )sretSUBURBS: distance to nearest non-roadside bicycle trail (me Srstrnr 911.831728.011359.35 SUBURBS: distance to nearest roadside bicycle trail (meters) Cactive 315.35203.41340.15CITY: distance to nearest active open space (meters) Cpassive 633.76396.64683.10CITY: distance to nearest passibe open space (meters) Cactive 290.071176.45569.92SUBURB: distance to nearest active open space (meters) Environmental Amenities (A) Spassive 613.09641.12760.73SUBURB: distance to nearest passive open space (meters) Bedrooms 3.000.913.12Number of bedrooms Bathroom 2.000.882.14Number of bathrooms Homestea 1.000.340.86Homestead status Age 27.0028.9735.88Age of house Lotsize 968.008053.172097.98Size of lot (square meters) Finished 1708.00908.661871.01 Finished square feet of floor space Firepls 1.000.760.70Number of fireplaces Structural Attributes (S) Garagest 2.001.021.72Number of garage stalls Hwynear 1149.581821.441672.32Distance to nearest major highway (meters) Cbdnear 16374.7510409.6117558.59Distance to nearest central business district (meters)Location (L) Busy 0.000.210.05Home is on a busy street mca5_att 4836.10276.784760.46Standardized test score in school district Pctnonwt 7.8214.0212.51Percent nonwhite in census tractNeighborhoodAttributes (N) Avghhsiz 2.660.402.67Persons per household in census tract TABLE 26 Descriptive statistics of sample

on- and off-street facilities which are distributed across both major open space corridors (e.g., railway lines, rivers, and lakes) and other roadways. We pair the MLS data for every home sale in the seven county study area from 2001 with the location of these trails. Some on-street and off-street trails are located alongside busy traf- ficked streets, which is presumably a propelling characteristic for home locations. We therefore divide the off-street layer into roadside and non-roadside trails based on proximity to busy streets. We then calculate distance to the nearest roadside trail, non-roadside trail, and on-street bicycle lane for each home. As previously mentioned, we also measure distance to open space as a central variable, classifying such areas by type: active or passive.xxiii Measures of Density of Bicycle Facilities Motivated by Anderson and West’s (131) findings that proximity and size of open space matters, we also theorize it to be important to consider not only the distance to facilities but also the density of trails around a particular home. The overall density (length) of different facilities within a buffer area may also be appreciated by homebuy- H-4 ers. They might value a well-connected system of trails, which are prevalent in many areas throughout the Twin Cities metropolitan area. We therefore calculate the kilometers of trails within buffer dis- tance. See, for example, Figure 22 showing an example home in Min- neapolis and how we measured open space and density of bicycle facilities by differing radii of 200, 400, 800, and 1,600 meters. Interaction Terms Many of the structural attributes used in this application are uni- versally valued (e.g., home size, number of bathrooms). Several of the spatial attributes employed, however, are hypothesized to vary by segments of the population (urbanites versus suburbanites). Again, this distinction was found by Anderson and West’s application for the same region. We therefore generate interaction terms (e.g., city mul- tiplied by independent variable) to measure the attributes that may vary spatially. Doing so allows us to pool the sample of urban and suburban homes, thereby parsimoniously estimating a single model that preserves the integrity of the differing preferences. This single model provides coefficients that describe the effect of common attrib- Figure 21. Examples of off-street bicycle trails, on-street bicycle lanes, and open space.

utes while producing different coefficients for the spatial attributes that may vary across suburbanites and urbanites.xxiv Fixed Effects Finally, as with any analysis of this type there are omitted attrib- utes to consider. When estimating phenomena associated with the real estate market this dimension is particularly important. There are likely spatial attributes—not captured by any of our measures— which invariably affect home value. These attributes may include but are not limited to general housing stock of neighboring homes, the reputation effects of different neighborhoods or unobserved char- acteristics of the neighborhood. Without fixed effects, variation across all observations in all neigh- borhoods is used to identify the effect of interest. But given the likely spatial correlation between proximity to bicycle facility with other variables, this effect is susceptible to omitted variable bias. We con- trol for bias introduced by potential omitted variables by using local fixed effects, a dummy variable for each RMLS-defined market area (104 areas in our region). These boundaries mostly follow city limits in suburban areas and divide the central cities into several neighbor- hoods that closely follow similarly natured real-estate markets. By controlling for fixed effects we are estimating the effect of proximity H-5 to a bicycle trail, assuming a household has already decided to locate in one of the 104 MLS areas in the region. While more accurate, this process makes it difficult to identify the impact of bicycle trail prox- imity because it in effect reduces the variation of the variables of interest. Michaels and Smith (241) support this claim, showing that dividing a market into submarkets results in less robust estimates of the effects of hazardous waste site proximity. RESULTS AND DISCUSSION Our final model (shown in Table 27) is an OLS regression which determines the effect of bicycle trail proximity on home sale prices. We employ a logged dependent variable and also log transforma- tions of several continuous independent variables, indicated by an ln following the variable name. All structural and location variables are statistically significant and have the expected signs. Home val- ues increase with number of bedrooms, bathrooms, lot size, finished square footage, fireplaces, garage stalls, proximity to a central busi- ness district, and school quality. Home values decrease with age and percent non-white in the census tract. Similarly, proximity to a free- way has a negative effect on home value, which implies that the dis- amenity effects of freeways (e.g., noise, pollution) likely outweigh any accessibility benefits within particular neighborhoods. Looking Figure 22. Off-street trails and open space within 200, 400, 800, and 1,600 meters of a Minneapolis house.

at some of the location and amenity variables reveals a different story. Open space coefficients are generally consistent with Anderson and West’s (131) findings. Suburbanites value passive open space over active recreational areas. City residents also value lakes and golf courses, but active open space does not affect sale price. Focusing on the variables of interest in this application, our analy- sis of bicycle facilities reveals a relatively complex story. It fails to be crisp and clean because we measure three types of facilities for two different populations (urban and suburban). Our discussion sepa- rates the findings for city and suburban residents. First, city resi- dents clearly value proximity to non-roadside trails (after control- ling for open space). As Minneapolis is well endowed with many off-road facilities and appears to exhibit a relatively high cycling pop- ulation, this comes as little surprise. The opposite is true for trails alongside busy streets, however, even when controlling for adjacency to the streets themselves. On-street bicycle lanes have no significant effect in the city. The possible reason for this is that in general, the nature of on-street facilities differs considerably between Minneapo- lis and St. Paul.xxv As in the city, suburban homes near roadside trails sell for less than those further away, even when controlling for busy streets. The same is true for on-street bicycle lanes, for which there was not sta- tistically significant effect in the city. Suburban off-street trails appear to negatively influence home prices, unlike in the city. There are possibly several reasons for this. First, it may be the case that because of decreased cycling use, suburbanites simply do not value access to trails. Such proximity may not even factor into their use or option value of their home purchase locations. Second, counter- H-6 acting phenomena may be taking place. Some suburbanites may indeed value such trails. However, their preferences may be over- shadowed by a combination of the following factors. Some of the suburban trails are along former railway beds. If these property val- ues were formally depressed because of such an externality, such legacy effect may likely still be in effect. Uncertainty surrounding future uses of such corridors, such as commuter rail, could com- pound any legacy affect. Snowmobiling introduces additional exter- nalities common to exurban trails. Most notable, many suburbanites simply appreciate the seclusion of their settings. Proximity to trails— no matter their character—may be an indication of unwanted peo- ple passing by or other symptoms that run counter to factors that prompted their decision. One need only refer to several newspaper headlines (Figure 23) to learn of instances in which suburbanites oppose nearby trails. Similar analysis employing measures of the density of bicycle facilities did not reveal statistically significant findings in any of the models estimated. Because the policy variables of interest and the dependent vari- able are logged, the coefficients can be directly interpreted as elas- ticities. However, we provide the results of an effect analysis to more concretely estimate values. In Table 27 the last two columns present the effect of moving a median-priced home 400 meters closer to each facility than the median distance, all else constant.xxvi We find that in the city, the effect of moving a median-priced home 400 meters closer to a roadside bicycle trail reduces the sale price $2,272. Assuming a home was 400 meters closer to a non-roadside trail would net $510. While all relationships between bicycle facility proximity and home Variable Name Contrln Cntrln Crstrln Sonrln Snrtrln Srstrln Cactive Cpassive Cactive Spassive Bedrooms Bathroom Homestea Ageln Lotsize Finished Firepls Garagest Hwynear Cbdnrln Busy mca5_att Pctnonwt Avghhsiz Description Coefficient Standard Error t-statistic Effect of 400m Closer CITY: distance to nearest on-street bicycle lane (ln) 1.470.0026890.003950 CITY: distance to nearest non-roadside bicycle trail (ln) 509.85$–2.1*0.003732–0.007851 0.022772CITY: distance to nearest roadside bicycle trail (ln) 6.03**0.003777 (2,271.63)$ SUBURBS: distance to nearest on-street bicycle lane (ln) 0.003334 0.001272 2.62** $ (364.02) SUBURBS: distance to nearest non-roadside bicycle trail (ln) 2.91** $ (239.65)0.0013250.003858 SUBURBS: distance to nearest roadside bicycle trail (ln) 0.010230 0.001419 7.21** $ (1,058.73) 0.000012–0.000024 –0.000065 0.0000007 –9.08** –1.96* $ 3,860.35 0.000006 0.000001 3.88** $ (442.80) 0.033037 0.001570 21.05** 0.079976 0.002018 39.63** –0.000028 0.000002 –12.86** –52.65** $ 2,066.40 –0.027259 0.003481 –7.83** –0.092578 0.000003 0.000168 0.068749 0.075257 0.000009 –0.056065 –0.03351 0.000160 –0.004014 0.038961 11.314800 0.000000 0.000002 0.001768 0.001759 0.001268 0.000001 0.006926 0.005096 0.000010 0.000183 0.004481 0.079957 21.68** 82.14** 38.89** 59.37** 10.35** –8.09** –6.54** 15.31** –21.99** 8.7** 141.51** $ $ (637.20) 9,861.10 ** Significant at p<0.01 * Significant at p<0.05 Number of observations: 35,002 Adjusted R-squared: 07920 Constant CITY: distance to nearest active open space (meters) CITY: distance to nearest passive open space (meters) SUBURB: distance to nearest active opn space (meters) SUBURB: distance to nearest passive open space (meters) Nubmer of bedrooms Number of bathrooms Homestead status Age of house (ln) Size of lot (squaremeters) Finished square feet of floor space Number of fireplaces Number of garage stalls Distance to nearest major highway (meters) Distance to nearest central business district (ln) Home is on a busy street Standardized test score in school district Percent nonwhite in censu tract Persons per household in census tract TABLE 27 Regression results

sale prices are negative in the suburbs, the effect analysis shows significant variation in the magnitude of those relationships. The effects of moving a suburban home 400 meters closer to a road- side bicycle trail is −$1,059 compared with only −$240 for a non- roadside trail. CONCLUSION AND FUTURE RESEARCH There are several important implications for our results which confirm the hypothesis that the three types of trails influence home sale prices in different ways. They demonstrate the importance of controlling for bias induced by omitted spatial variables. Such bias is especially relevant for large complex and polycentric housing markets (such as in the Twin Cities, with two CBDs) and in areas where factors that influence home price differ tremendously by neighborhood. We use local neighborhood fixed effects to reduce spatial autocorrelation and also lead to more robust coefficient esti- mates. Of course, using this methodology—while technically sound and robust—also makes it more difficult to detect the effects of such proximity because we are now comparing homes within MLS areas. Our results are also able to robustly test for the fact that urbanites and suburbanites perceive and value bicycle facilities differently. The H-7 use of interaction terms between city and suburb reveals this differ- ence in preferences between city dwellers and suburbanites. We measure bicycle facilities in different ways. Distance to nearest facil- ity is the measure discussed in detail above. Models that were esti- mated to examine the role of trail density did not produce statistically significant findings. The comprehensiveness of the Twin Cities’ bicycle trails may contribute to a lack in variation among trail den- sities near homes. Further refinements would enhance our approach to estimating the value of bicycle facilities. Introducing a stated preference element akin to Earnhart’s (232) application could yield more robust esti- mates. Additional stratification of variables would also augment our understanding. We have divided bicycle trails into on-street, road- side and non-roadside facilities in the city and suburbs. Further data collection efforts aimed at identifying other differentiating charac- teristics among facilities, such as trail width and adjacent land cover, would allow the implementation of a hedonic travel cost model to place a value on such characteristics (242). Assigning future benefits based on a hedonic model presents complications, as new environmental amenities can take years to capitalize into housing prices. Ridell (243) shows that cross- sectional studies may underestimate the benefits of these goods, and provides an approach for capturing delayed benefits. In addition to The Seattle Times, Nov. 1, 2003. P. 1A. ìLake Sammamish rails-to-trails vision: ‘One person ’s dream, another’s nightmare.’” The Boston Globe, Aug. 26, 2004. Globe West P. 9. “Friends, foes of Nobscot rail trail take up sides.” The Washington Post, June 10, 2001. Loudoun Extra P. 5. “Trail’s appeal is lost on some; Neighbors: not in our back yards.” Figure 23. Newspaper headlines showing suburban trail opposition.

delayed benefits, future benefits also present an opportunity for refining model specification. Shonkweiler’s (244) methodology accounts for the potential conversion of rural land to urban uses, revealing that this qualitative consideration reduces estimation error. From a policy perspective, this research produces three important insights. First, type of trail matters. On-street trails and roadside trails may not be as appreciated as many city planners or policy officials think. Second, city residents have different preferences than suburban residents. Third, larger and more pressing factors are likely influenc- H-8 ing residential location decisions. Using fixed effects detects such considerations in terms of neighborhood quality and character. Over- all, our results suggest that off-street bicycle trails add value to home sale prices in the city, implying a contribution to social livability. No positive or significant relationship, however, is found for other types of facilities in either city or suburb. In fact, bicycle trails exhibit a disutility in suburban settings. This suggests that urban planners and advocates need to be aware of the consequences of providing for bicycle facilities, as the change in welfare is not necessarily positive for all homeowners.