The association between residential housing prices,bicycle infrastructure and ridership volumes

Abstract

Chinese

City officials and planners have shown increased interest in pedestrian- and bicycle-friendly designs aimed at addressing urban problems such as traffic congestion, pollution, sprawl and housing availability. An important planning consideration is the economic impact associated with existing or planned infrastructure, especially in relation to home property values. In this study, we use measures of infrastructure and ridership to evaluate the relationship between bicycling infrastructure and activity and single-family home values in Tempe, Arizona. We apply a hedonic modelling approach and find that bicycle infrastructure density is positively associated with home sale price, while ridership density around home locations has no significant relationship with sale price. Our results inform discourse related to the potential economic values of residential bicycle infrastructure, especially in areas where property tax is a source of local public finance revenue. We show that the characteristics of bicycle-friendly design may be the same characteristics valued by homebuyers and the resulting increased home sale values may lead to increased property tax revenue in Tempe, Arizona.

Keywords

economic processes hedonic price analysis neighbourhood real estate transport

Introduction and problem statement

Planners and officials in urban areas are increasingly interested in addressing problems such as traffic congestion, pollution, sprawl and housing availability (Cervero et al., 2002). Solutions include compact cities initiatives, transit-oriented development (TOD) and downtown revitalisation. TOD in particular focuses on designing more compact living environments that include pedestrian- and bicycle-friendly facilities and mixed-use (i.e. business and residential) designs near transit to increase accessibility and reduce vehicle trips (Bartholomew and Ewing, 2011; Cervero et al., 2002). As these large infrastructure investments are adopted in urban areas, the associated impacts on local economies need to be evaluated and information about costs and benefits needs to be provided to taxpayers, property owners and other stakeholders.

An important concern for planners and property owners in particular is the impact of infrastructure changes on property values. Values are influenced by home characteristics (e.g. house and property size) but are also a reflection of buyers' perceived value of amenities that are both structural (e.g. nearby parks, bicycle infrastructure) and neighbourhood based (e.g. school districts) (Bartholomew and Ewing, 2011). Structural and neighbourhood considerations are important because changes have the potential to increase or decrease tax revenues through their influence on property prices, which is especially relevant in regions where property tax is a source of local public finance revenue. The concept of NIMBY (Not In My Backyard) drives home the idea that residents will oppose development if a proposal is negatively perceived. If bicycling infrastructure has a positive influence on property values and is therefore positively perceived, this can provide additional support for transit-oriented projects (Nicholls and Crompton, 2005). This paper focuses on the relationship between bicycling infrastructure, bicycle ridership and property sales prices.

Bicycle infrastructure includes bicycle lanes, buffered bicycle lanes and bicycle boulevards and all may have differing impacts on housing prices (Krizek, 2006). Bicycle lanes are typically painted lanes that share a regular vehicle traffic lane. Buffered bicycle lanes are also proximal to traffic but are separated from vehicles by a buffer such as bollards, a parking lane or a berm/landscaping. Bicycle boulevards are separated from traffic and are sometimes multi-use trails; they may be paved or unpaved. Greenbelts are open corridors that typically follow a natural landscape feature such as riverfronts or stream valleys, or manmade features such as canals or railways that are no longer in use (Nicholls and Crompton, 2005). Trails are typically defined as multi-use paths that are separated from road traffic and may be paved or unpaved.

Despite plans to improve bicycling infrastructure in many cities, the impact on property prices can be difficult to measure. Challenges include a lack of market data identifying the economic impact, or value, of bicycle facilities, as well as a lack of information on their use and how use is related to market values and economic activity (Krizek, 2007). Several studies have demonstrated that bicycle facilities are related to property values when considering measures such as the presence of bicycle lanes or proximity to local amenities by bicycle, though results are mixed as to whether the impacts are positive or negative (Liu and Shi, 2017). A limitation of prior studies is that while they account for the presence of bicycling infrastructure, they do not measure the number of riders who are using infrastructure, or ridership volume. The volume of cyclists may be indicative of neighbourhood and bicycle facility features that attract cyclists and where bicycle ridership volume may relate to property values independently from the existence of bicycle facilities themselves. For example, people are motivated to cycle when routes are removed from traffic noise or have pleasing scenery (Winters et al., 2011) and traffic noise negatively influences house values (Wilhelmson, 2000) while greenspace, urban parks or open space and trees are associated with higher home values (Brander and Koetse, 2011). Because bicycle facility presence is not a requirement for ridership, other neighbourhood features may attract cyclists and serve as a proxy for influences on home sale price. On the other hand, property owners may have concerns that increased ridership would devalue properties if that ridership is associated with negative factors such as increased crime, vehicular collisions or overcrowding of neighbourhood attractions.

This paper evaluates the relationship between bicycling infrastructure, ridership and single-family home values in Tempe, Arizona. It contributes to the literature on bicycling infrastructure and housing valuation by using crowdsourced ridership data in a hedonic pricing model to examine how bicycling infrastructure investments are capitalised into home values. This ridership measure has not been incorporated in prior modelling contexts and may give additional insight into the question of the roles that bicycling infrastructure and activity play in residential property prices and tax revenue impacts.

Literature review

Housing is a multi-attribute product with market value influenced by attributes such as environmental quality or access to public goods (e.g. greenspace, bicycle infrastructure) in addition to the specific characteristics of the home. Researchers use hedonic modelling to examine market values associated with accessibility and transportation (Bartholomew and Ewing, 2011; Parent and vom Hofe, 2013). For example, hedonic pricing analysis has been used to examine the relationship between housing price and rail/transit access (Hess and Almeida, 2007; Mohammad et al., 2013; Welch et al., 2016), walkability (Boyle et al., 2014; Li et al., 2015), access to multi-purpose trails, greenways and greenbelts (Asabere and Huffman, 2009; Campbell and Munroe, 2007; Lindsey et al., 2004; Nicholls and Crompton, 2005; Parent and vom Hofe, 2013) and bicycle-specific infrastructure (Liu and Shi, 2017; Welch et al., 2016). Each of these studies infers values for non-traded environmental or public goods by examining observed market transactions for related private goods, such as housing. Housing market transactions implicitly carry price signals about non-market attributes that are part of the bundle of attributes that consumers consider when purchasing housing (Freeman, 2003; Palmquist, 2002). Hedonic pricing analysis is therefore a revealed preference technique that uses information from market transactions of multi-attribute goods to identify premiums that buyers are willing to pay for specific non-market amenities that comprise a part of the multiple attributes. While hedonic models are difficult to generalise across regions, characteristics that are repeatedly significant in these models point to features that are more consistently valued by homebuyers (Sirmans et al., 2005).

Hedonic pricing analysis typically includes information about structural attributes of the house itself, as well as the location and surrounding neighbourhood. Structural attributes may include lot size, square footage, number of rooms, presence of a garage and/or pool and age of the house, among other attributes (Bartholomew and Ewing, 2011; Sirmans et al., 2005), finding that home characteristics, such as larger square footage, increase values. Measures reflecting amenities, including proximity to parks and greenspaces, business districts, access to transit, in addition to neighbourhood controls such as median income, school quality and crime rate, are commonly featured in model specifications (Bartholomew and Ewing, 2011; Li et al., 2015; Sirmans et al., 2005). While public events, such as town hall meetings, may suggest that residents view bicycle infrastructure as a nuisance that raises concerns for safety, transportation planners need to evaluate the impact of such infrastructure more holistically and consider empirical evidence, such as potential impacts on housing. Hedonic models are an appropriate choice for evaluating the aesthetic value of bicycling activity and bicycle facilities in particular (Krizek, 2007) because they elicit implicit values for these non-market characteristics through the observed transactions in the market for housing (Palmquist, 2002).

Several studies have used hedonic pricing analysis to estimate the economic effects of bicycle infrastructure through its impact on housing prices, though there are mixed findings in different regions and several different ways of operationalising bicycle and other pedestrian infrastructure variables. Independent bicycling infrastructure variables have typically been operationalised either as dummy variables, indicating the presence or absence of bicycle facilities, or as distance to the nearest bicycle facility. Table 1 summarises previous studies that have examined relationships between housing price and bicycle/trail facilities. For distance-based measures of bicycle facilities, several studies indicate that closer proximity to a bicycle facility is associated with higher home values (e.g. Lindsey et al., 2004; Liu and Shi, 2017; Parent and vom Hofe, 2013; Welch et al., 2016) though the type of facility differed between studies. Proximity to separated facilities (e.g. off-street or trails/greenways) tends to be associated with higher home values while on-street and roadside (i.e. those separated from the vehicular roadway by a berm) facilities are associated with lower home values. Greater bicycling infrastructure density proximal to homes is associated with higher home value (Liu and Shi, 2017). Though the size of the overall effect associated with bicycle facilities and home prices is small (Table 1), prior research concludes that bicycle facilities most commonly have a positive association with the value of homes in their proximity. In terms of the larger economic context, even small changes in individual property values can result in substantial aggregate property tax revenue at the city-level scale when many homes are affected.

Table 1.

Summary of previous research including bicycle infrastructure measures and associated value.

Study	Location	Bicycle infrastructure measure	Direction and significance	Magnitude of effect/value
Liu and Shi (2017)	Portland, OR, USA	(a) distance to nearest separated/buffered bike facility and (b) bike facility density within 0.5 mile of each single-family home	(a) negative, significant (b) positive, significant	Would expect a US$686 increase in single-family home value for each 0.25 mile closer to an advanced bicycle facility. A 0.25 mile increase in density of bicycle facilities was associated with a US$4039 increase in single-family home value
Welch et al. (2016)	Portland, OR, USA	Distance to the nearest (a) regional multi-use path (b) off-street local multi-use path (c) on-street bike facility	(a) negative, significant (b) negative, not significant (c) positive, significant	Sales price decreased by US$ 0.86 for each foot the home was located away from the nearest regional path. Sales price decreased by US$ 0.01 for each foot from the nearest off-street local multi-use path (not significant). Sales price increased by US$2.47 for each additional foot away from an on-street bike lane
Krizek (2006)	Minneapolis – St Paul, MN, USA	Distance to nearest (a) on-street lane (b) non-roadside bicycle trail (c) roadside bicycle trail in a city	(a) positive, not significant (b) negative, significant (c) positive, significant	House price would increase by US$510 if moving a home 400 m closer to a non-roadside bicycle trail. Sale price decreased by US$2272 if a median priced home were 400 m closer to a roadside bike trail
Lindsey et al. (2004)	Indianapolis, IN, USA	Sales within 0.5 mile of recreational greenway corridor	Positive, significant	14% of home sales price is attributable to proximity to the Monon trail while 2.4% can be attributed to homes being in proximity to a greenway conservation corridor, in general (overall positive relationship between proximity and sales price)
Asabere and Huffman (2009)	Bexar County (San Antonio area), TX, USA	Presence of greenbelt or trail in neighbourhood	Positive, significant	Trails have an approximate 2% impact on housing price, trails with greenbelt or greenway have 5% impact and greenbelts alone have 5% impact (overall addition to home value)
Nicholls and Crompton (2005)	Austin, TX, USA	Location on greenbelt, view of greenbelt, distance to greenbelt, greenbelt entrance within 0.5 mile	Depending on area positive, significant or negative and not significant	Barton creek: 12.2% of home value was attributable to adjacent greenbelt. Lost creek: no significant relationship with greenbelts. Travis area: 5.4% increase in average value of home. View and distance were not significant
Parent and vom Hofe (2013)	Hamilton County (Cincinnati area), OH, USA	Network distance between home and nearest trail entrance	Negative, significant	Home value increased by US$3.98 for each foot closer it was to a trailhead

Prior housing price research has also examined the impacts of features associated with pedestrian infrastructure and estimated impacts on property prices. In general, development density, land-use mix and pedestrian infrastructure are associated with higher residential property values (Sohn et al., 2012). Built environment features near residences can influence whether people engage in bicycling activity, an important consideration when valuing home prices in a neighbourhood. An increase in residential density and access to parks and attractive destinations are associated with uptake of transport-related bicycling whereas street connectivity is associated with starting recreational bicycling activity after people moved residences (Beenackers et al., 2012). Geographic features such as locations of parks, greenspace and open/abandoned land (e.g. railways), traffic volumes and speeds, land use and street patterns might influence where bicycle facilities are located (Cradock et al., 2009; Forsyth and Krizek, 2010). On-street bicycle lanes are associated with areas of highly connected streets with low speeds and lower overall traffic volume – characteristics that homebuyers may value rather than the bicycle lanes themselves. Well-connected streets with small block lengths, mixed land use and access to retail activity spur walking and bicycling activity (Cervero and Duncan, 2003; Sun et al., 2017).

While studies have examined the associations between greenways, trails, bicycle lanes and house values they do not directly measure activity; there is no guarantee that proximity to bicycle-friendly amenities is associated with an actual increase in bicycling activity (Li et al., 2015). Some bicycle activity occurs on bicycle-specific infrastructure but neighbourhood ridership volumes may relate more to street connectivity or built environment features rather than bicycle facilities alone, and riders may utilise streets that are not specifically marked as bicycle facilities. Many low-speed and low-traffic-volume streets are naturally conducive to bicycling activity even though they are not separately designated as such. In addition, bicycle ridership is associated with certain neighbourhood features that appeal to cyclists and residents alike. Bicyclists prefer areas with traffic-calming features (Broach et al., 2012) and low traffic speeds, which could also be seen as desirable neighbourhood features.

Ridership volumes are difficult data to capture, especially at a city-wide scale. Two commonly used methods for collecting street-level bicycle ridership data across a city are manual and automated bicycle counts. Both methods count bicycle activity volumes at specific locations, for all riders who pass that location. A primary limitation of these count data is that their coverage is spatially and temporally coarse. Manual counts rely on individuals positioned at a number of select locations across the study region, which limits their spatial coverage, and are typically only conducted over the course of two to three days once per year, which limits their temporal coverage. Automated bicycle counting occurs continuously throughout the year, though counts remain spatially limited because most cities only install counters at a few key geographic locations. Considering either manual or automated counting, data are limited in generating the street-by-street ridership volumes that are necessary for analyses at local levels that comprise city-wide activity.

In contrast to manual or automated bicycle counts, advances in smartphone technologies have led to new crowdsourced data sources that overcome limitations in spatial and temporal coverage. These crowdsourced data originate from apps focused on recording trips that are taken by bicycle and provide information about where, when and how often people use bicycles. Ridership volumes can therefore be generated on a street-by-street basis using these data. Strava, a route and activity tracking app, is one such source of crowdsourced data. Findings related to bicycling activity in Strava data indicate that recreational bicycle trips occur in areas nearer to residential land with shorter-length streets and high connectivity (Sun et al., 2017). These infrastructure characteristics might also be desirable neighbourhood features and therefore relate to higher property values (Bartholomew and Ewing, 2011).

Despite the benefit of providing detailed bicycle ridership data, crowdsourced bicycling data are also associated with notable limitations and potential biases. Outside of city centres, patterns of ridership in Strava data differ from those represented by conventional manual counts (Conrow et al., 2018). Riders using the app may prefer to access areas away from the urban core where bicycle facilities are located (Griffin and Jiao, 2015), which could result in misleading ridership patterns within a city, including underrepresentation of regular commuters. In general, bicycling data from smartphone apps are supplied by young, male, recreational riders which introduces bias as populations such as the very young or old, students or socioeconomically disadvantaged may not be adequately captured (Blanc et al., 2016).

This study addresses a gap in understanding the economic value of bicycle infrastructure as it relates to home values, using crowdsourced data as a new way to estimate bicycle ridership. As a first look that includes ridership volume, we specifically examine bicycle facility density alongside bicycling activity on all streets to determine whether ridership in the absence of bicycling-specific infrastructure signals economic value in the housing market. Further, the crowdsourced smartphone app data used to measure ridership is the first application of such data in a housing valuation study. We develop a ridership measure local to individual houses, which has not been possible with spatially limited ridership counts. The bicycle infrastructure and ridership measures we employ fill a gap in assessing the value of bicycle facilities and their use with respect to housing market prices and potential economic impacts. This research benefits city officials, planners, residents and other stakeholders by providing an approach to undertake data-driven decision-making in terms of potential impacts of bicycling infrastructure and activity in urban housing markets.

Data and methods

Study area

Residents and city officials in the city of Tempe, Arizona, are engaged in a debate as to the economic value of bicycling infrastructure, with some residents arguing that the presence of bicycle lanes will reduce home values. We selected Tempe as the study area for our analysis of the relationship between bicycle facilities, ridership and housing prices to help inform this debate and provide robust analysis to help city planners and citizens alike make informed, data-driven decisions on installing bicycle infrastructure.

The city of Tempe lies within Maricopa county and is part of the Phoenix Metropolitan Area (PMA). The PMA has over four million residents with 185,000 of those residing in Tempe as of 2016, an increase of 14% from 2010 (US Census Bureau, 2018). This population growth has been absorbed into the 40.2 square miles (104 km²) existing land area of the city (Figure 1), which has led to urban densification and an emphasis on mixed-use and transit-oriented development that requires mixed-modal transportation, including active transportation such as bicycling. Year-round bicycling in Tempe is made easier because of the generally flat topography, low rainfall amounts (<10 inches/year, on average) and year-round warm temperatures (the average high temperature is 87.3°F (30.7°C) and the average low is 55.3°F (12.9°C)).

Figure 1.

Mean single-family home sale price by census block group in Tempe, AZ (note: mean sales values are calculated from the 5198 single-family homes sold during the 2013–2016 study period).

One of the significant drivers of urban growth and economic development in the city of Tempe is Arizona State University (ASU). Situated in the north-central part of the city, close to the downtown area, ASU brings thousands of people to campus daily (Figure 1). ASU students and employees commute to the campus by car, public transportation (light rail and bus) or active transportation (e.g. walking, skateboarding, bicycling). Tempe's bicycle infrastructure covers 16.7% of the city’s 1049 roadway miles (1688 km) with a mix of bicycle lanes, routes, paved and unpaved multi-use paths and paved shoulders (Figure 2). The majority of Tempe’s bicycle facilities are comprised of bicycle lanes or designated portions of roadways that have preferential or exclusive use for bicycles, and bicycle routes, which are designated by signs only (TMP, 2015). Currently, a greater percentage of Tempe residents bike and walk to work compared with other cities in the PMA; 4.2% of residents bike whereas the Maricopa County average is 0.8% (TMP, 2015).

Figure 2.

(a) Distribution of bicycle facilities as of 2016, and (b) ridership density derived from 2016 ridership counts by 2010 census block group.

To support the residents of Tempe, the city’s master transportation plan includes improvements to existing bicycle lanes, adding buffered and protected lanes and developing bicycle boulevards (TMP, 2015). The city of Tempe follows a long-range transportation plan that drives investment needs on the basis that it ‘provides mobility for all, promotes clean air and conserves energy, preserves neighborhood livability, and enhances quality of life for our citizens and guests’ (City of Tempe, 2008: 1). Access and mobility are named as the two fundamental dimensions of the presently developed bicycle infrastructure, and new facilities are constructed in consideration of access to the core business areas as well as improved connectivity across the city (City of Tempe, 2008).

Data

We construct a dataset to conduct hedonic pricing analysis using information about physical property characteristics, neighbourhood features and amenities, including bicycling amenities, and socioeconomic factors. Our study area encompasses the 114 census block groups (2010) with a mean area of 0.034 square miles (0.088 km²) within the city of Tempe (Figure 1). Seventeen block groups do not contain single-family houses, instead reflecting non-residential uses such as the ASU campus, Tempe central business district (downtown) and industrial/recreational areas.

The dependent variable is actual sales price data for single-family residential property transactions over the years 2013 to 2016 inclusive, representing a period of relative market stability at the study location.¹ Sales data were provided by the Maricopa County Assessor’s Office (Maricopa County Assessor’s Office, 2018). We restricted transactions to include only sales of single-family residential properties and omitted sales that were not arms-length or included non-standard financial arrangements, resulting in 5198 unique single-family home sales over the period 2013 to 2016 within the study area.² The greatest volumes of sales occurred in the eastern and southern portions of the study area, while the highest mean house prices were in close proximity to ASU or in the south and south-central area (Figure 1). Table 2 contains summary statistics for housing sales prices by year of sale. Mean housing prices increased (in nominal dollars) over the study period, ranging from US$243,605 in 2013 to US$281,312 in 2016; median sale prices closely track the mean over the same period.

Table 2.

Summary statistics for all model variables.

Variable	Mean	Median	Standard deviation	Minimum	Maximum	Expected sign
Dependent variable
2013 Sale price (2013 USD)	243,605	225,000	98,018	74,000	714,000
2014 Sale price (2014 USD)	257,470	237,000	97,330	25,500	795,000
2015 Sale price (2015 USD)	270,423	248,750	98,197	63,000	865,000
2016 Sale price (2016 USD)	281,312	264,900	89,657	40,000	885,000
Independent variables
Property size (acres)	0.19	0.18	0.06	0.08	0.49	+
Size of house (liveable ft²)	1831	1743	553	624	3964	+
House age (years since built)	42.5	44.0	12.4	1.0	92.0	−
Bicycle facility density (total miles of all bicycle facilities/0.5 mile buffer)	39.23	39.72	16.94	0.00	103.24	+
On-street bicycle facility density (total miles/0.5 mile buffer)	32.62	33.60	15.22	0.00	83.16	+
Off-street bicycle facility density (total miles/0.5 mile buffer)	6.61	0.00	9.53	0.00	62.19	+
Ridership density (number of riders/0.5 mile buffer in 2016)	2464	243	7127	14	142,479	+
Distance to ASU (ft)	14,020.08(2.7 mi)	12,663.17(2.4 mi)	8398.87(1.6 mi)	25.22(0.005 mi)	34,306.89(6.5 mi)	−
Greenspace distance (ft)	1296.46	1174.73	814.59	0.00	4206.48	−
Middle school score	8.38	8.0	1.09	4.00	10.0	+
Elementary school score	7.74	8.0	1.37	3.00	10.0	+
Controls by Census Block Group (2016 Census)
Median household income (USD)	69,312	63,030	29,266	16,875	160,833	+
Median age (years)	36.23	34.5	8.37	20.10	58.30	+
Race (% minority)	23.70	20.26	12.20	0	67.10	−
Crime rate (number of crimes per census block population 2013–2016)	0.53	0.35	0.51	0.00	3.38	−
Number of jobs accessible in 45 min by car (2010)	270,169	271,037	25,873	25	34,305	+
Journey to work (JTW) by bicycle (%)	2.62	1.54	3.62	0.00	20.50
Dummy variables	n		Mean	Absent	Present
2013 sale year (1 = property sold in 2013)	1102		0.212	0	1	+
2014 sale year (1 = property sold in 2014)	1169		0.225	0	1	+
2015 sale year (1 = property sold in 2015)	1316		0.253	0	1	+
Pool (1 = house has a pool)	2440		0.469	0	1	+
Multiple storeys (1 = house more than 1 storey high)	678		0.13	0	1	+

Notes: Total number of observations = 5198.

Independent variables include structural attributes related to each property, neighbourhood characteristics, block group level socioeconomic characteristics, access to amenities, and bicycle infrastructure presence and usage. Structural attributes for each property include total square footage of the house, construction year, lot size, number of storeys and the presence or absence of a swimming pool and were obtained from the Maricopa County Assessor's office (Maricopa County Assessor’s Office, 2018). Number of bedrooms and bathrooms were not available in the property data, though they are common controls in hedonic modelling, so we follow prior approaches that use square footage alone (e.g. Liu and Shi, 2017; Parent and vom Hofe, 2013; Seo et al., 2014). Construction year was used to calculate the age of each house, while dummy variables were used to control for house sales in the years 2013, 2014 and 2015 (Table 2). Lastly, presence of a pool and whether a house was more than one storey high were also represented by dummy variables. Table 2 presents descriptive statistics for each property-related variable as well as expected coefficient signs, based on the findings from previous studies. The average house was built 42.5 years prior to 2016, and has approximately 1831 square feet (170 m²) of liveable space on a 0.19 acre (769 m²) lot; 47% of houses have a pool and 13% are more than one storey.

Information about bicycle infrastructure was obtained from the Maricopa Association of Governments (2014). The attributes within the bicycle infrastructure dataset did not distinguish between on- and off-street/separated facilities nor identify the dates when particular facilities were installed.³ We used the OpenStreetMaps (OSM) roads data from which the Strava ridership counts are derived, alongside visual inspection of OSM and Tempe Bikeways basemaps to determine which segments of Tempe’s bicycle infrastructure could be considered off-street dedicated cycling paths. On-street facilities comprise 83% of the total bicycle facility mileage in the study area and the off-street facilities are unevenly distributed, mainly located in the northern and southwestern extents of the study area. Bicycle facility density was computed as the length of bicycle infrastructure (total, on-street and off-street) within a 0.5 mile buffer of each property; density is therefore measured as the intensity of infrastructure within a property's individual surrounds. We chose a 0.5 mile buffer as the distance a person is willing to travel using non-motorised modes for everyday activities; this distance is consistent with previous studies examining associations between property sales and bicycle infrastructure (e.g. Lindsey et al., 2004; Liu and Shi, 2017). The median bicycle facility density surrounding each property is 39.72 miles as of 2016 (33.6 miles for on-street facilities only) and < 1% of properties have no bicycle facilities within 0.5 mile while a majority of properties (57%) had no off-street facilities within 0.5 mile of their location.

Ridership was operationalised as the number of riders on each street segment within a 0.5 mile buffer of each property. The bicycle ridership data originated from the Strava smartphone app; the counts represent the total volume of riders in either direction on each street segment in the study area during the 2016 calendar year. The mean ridership within a 0.5 mile buffer of each property is 2464 riders while the median is 243, indicating that the distribution is skewed. The total 2016 ridership count for individual street segments had a range between 1 and 14,645 riders, so it is expected that properties that are in close proximity to high-ridership segments would collect an exceedingly high number of riders within the 0.5 mile buffer area, explaining the large ridership density range (Table 2).

Figure 2 provides additional spatial representation of the bicycle-related measures, showing the distribution of bicycle facilities (Figure 2a) and ridership density (Figure 2b) by census block group. Bicycle infrastructure density varies throughout the study area, with lower facility density near the central/eastern portion of the study area and highest densities around ASU and downtown Tempe (Figure 2a). A small set of block groups that run east to west across the city and contain a canal-side path, as well as a multi-use separated path that connects to another canal section via a multi-use path, also show higher facility density. Ridership density (Figure 2b) is highest around ASU and downtown Tempe as well, though the general qualitative patterns between bicycle infrastructure and ridership density differ. Areas of highly dense bicycle infrastructure do not directly correspond to the areas with the highest ridership density and are not correlated (r = −0.009, p > 0.5).

We also measured proximal access to amenities by calculating the distance from each property to the ASU campus, downtown Tempe, the light rail line, local greenspaces and schools. Though some models control for presence of light rail and proximity to important business centres or downtown areas, distance to light rail and distance to ASU were highly correlated (r = 0.97); downtown Tempe is directly adjacent to the ASU campus thus distance to downtown and distance to ASU are also highly correlated (r = 0.98, p < 0.01). Considering the high degree of correlation between the ASU, downtown and light rail distances, we use only the distance from each sale property to the ASU campus in our models. We calculated the distance from each property to the nearest parks and golf courses, as greenspaces are associated with higher home values (Asabere and Huffman, 2009; Lindsey et al., 2004). On average, houses were less than three miles from the ASU campus and within 0.25 mile of greenspace (Table 2). School quality is another local amenity that plays into housing purchases. Each property was assigned to the elementary, middle and high school enrolment catchment in which it was located. The GreatSchools rating from greatschools.org was then used to rate each school. Ratings are on a 1–10 scale, where 10 is the highest and 4–7 is an average-performing school (‘About GreatSchools' Ratings’, 2018). The mean ratings for middle and elementary schools are 8.38 and 7.74, respectively (Table 2); middle and high school scores were correlated (r = 0.64, p < 0.01) so only middle school scores were retained in the models to reduce multicollinearity.

Sociodemographic neighbourhood control variables were derived from the US Census's American Community Survey block group five-year estimates (US Census Bureau, 2016). Each single-family residence was assigned the attribute values for the block group in which it was situated. The percentage of population using bicycling as their mode of transport to work was also computed to account for the representativeness and sampling limitations of Strava data. The descriptive statistics for median household income, median resident age, percent racial/ethnic minority population and percent of journey to work (JTW) by bicycle are shown in Table 2. Additionally, crime rate per block group population during the study period was derived based on the City of Tempe Police General Offense dataset, which documents the nature and location of criminal and city code violations (City of Tempe, 2017). Finally, we obtained the number of jobs accessible by car within 45 minutes of each block group, with network travel time-decay weighting (summarised in Table 2), from the Smart Location Database provided by the US Environmental Protection Agency (SLD, 2010). A notable limitation of the job data is that they were derived from the 2010 US Census Longitudinal Employer–Household Dynamics (LEHD) data and therefore pre-date our study period by several years, though the census block group boundaries remain unchanged; we include model results with and without the job availability measure (Table 3).

Table 3.

OLS (models 1–3) SAR (model SAR1) results.

Variable	OLS 1	OLS 2	OLS 3	SAR1
House and sale characteristics
Property size (acres)	0.8530***(0.0478)	0.7814***(0.0474)	0.7885***(0.0474)	0.7887***(0.0473)
Size of house LN (sq. ft.)	0.6717***(0.0129)	0.6609***(0.0127)	0.6579***(0.0127)	0.6577***(0.0127)
House age	−0.0052***(0.0003)	−0.0046***(0.0004)	−0.0047***(0.0004)	−0.0047***(0.0004)
Pool presence	0.0478***(0.0054)	0.0486***(0.0053)	0.0497***(0.0053)	0.0496***(0.0053)
Multiple storeys present	0.0149 (0.0085)	0.0061(0.0084)	0.0044(0.0084)	0.0045(0.0084)
Sale year 2013	−0.1866***(0.0069)	−0.1882***(0.0068)	−0.1879***(0.0067)	−0.1879***(0.0067)
Sale year 2014	−0.1156***(0.0067)	−0.1176***(0.0066)	−0.1176***(0.0066)	−0.1176***(0.0066)
Sale year 2015	−0.0613***(0.0065)	−0.0628***(0.0064)	−0.0628***(0.0064)	−0.0628***(0.0064)
Bicycling characteristics
Bicycle facility density(miles of bicycle facilities/mi²)	0.0025***(0.0002)	–	–	–
On-street bicycle facility density(miles of bicycle facilities/mi²)	–	0.0022***(0.0002)	0.0023***(0.0002)	0.0023***(0.0002)
Off-street bicycle facility density(miles of bicycle facilities/mi²)	–	0.0004(0.0003)	0.0005(0.0003)	0.0005(0.0003)
Ridership density(LN(number of riders/0.5 mile buffer in 2016))	−0.0007(0.0013)	0.0005(0.0013)	0.0008(0.0013)	0.0008(0.0013)
Amenities and access
Distance to ASU(LN(feet))	−0.0945***(0.0051)	−0.0976***(0.0053)	−0.0962***(0.0053)	−0.0961***(0.0053)
Greenspace distance(LN(feet))	0.0027(0.0023)	0.0035(0.0023)	0.0034(0.0022)	0.0034(0.0022)
LN(job availability)(number of jobs accessible in 45 min by car)	–	−0.5340***(0.0456)	−0.5454***(0.0457)	−0.5448***(0.0456)
Neighbourhood and census block group controls
Elementary school score	0.0019(0.0022)	−0.0003 (0.0022)	−0.0007 (0.0022)	−0.0007(0.0022)
Middle school score	−0.0037(0.0028)	−0.0047 (0.0028)	−0.0063*(0.0028)	−0.0062*(0.0028)
LN(Median income)	0.1366***(0.0095)	0.0954***(0.0099)	0.1021***(0.0100)	0.1021***(0.0100)
Median age	0.0023***(0.0004)	0.0009*(0.0004)	0.0012**(0.0004)	0.0012**(0.0004)
Percent racial/ethnic minority	−0.0009***(0.0002)	−0.0011***(0.0002)	−0.0009***(0.0002)	−0.0009***(0.0002)
Crime rate (number of crimes per census block population 2013–2016)	−0.0005(0.0053)	0.0187***(0.0054)	0.0183***(0.0054)	0.0183***(0.0054)
JTW bicycle %	–	–	0.0029***(0.0008)	0.0029***(0.0008)
(Intercept)	6.7386***(0.1461)	14.0451***(0.6366)	14.1181***(0.6361)	14.0241***(0.6553)
# Observations	5198	5198	5198	5198
R ²	0.7363	0.7445	0.7451
Adjusted R²	0.7354	0.7435	0.7441
Pseudo R²				0.7452
Log likelihood	1692.50(d.f. = 20)	1774.62(d.f. = 22)	1781.29(d.f. = 23)	1781.45(d.f. = 23)
AIC	−3345.01	−3505.24	−3516.58	−3514.9

Notes: Level of significance 0.01% = ***, 1% = **, 5% = *.

Hedonic pricing approach

Hedonic pricing analysis is a revealed preference approach often used to estimate the implicit prices of individual characteristics of a differentiated commodity (such as housing). It has a strong theoretical grounding in consumer theory and employs a utility theoretic framework that establishes the connections between consumer preferences for different levels of attributes associated with a differentiated good and their willingness to pay (Owusu-Ansah, 2011; Sirmans et al., 2005). Following the theoretical basis of hedonic pricing analysis, we model the contribution of bicycle infrastructure (an attribute that reflects one aspect of potential neighbourhood amenities) and ridership to sale prices of single-family homes in Tempe, AZ, using the following hedonic framework: the sale price (P) of a house is modelled as a function of a suite of differentiated property attributes such as the physical structure of the house, neighbourhood characteristics and accessibility to amenities, including features of the local bicycling context (equation 1).

P = P (H, N, A, C)

(1)

where: H is housing characteristics, for example, lot size, house size, age; N is neighbourhood characteristics, for example, median income, school quality, crime; A is amenities and accessibility in the area, for example, greenspace, distance to ASU; C is bicycling characteristics, e.g. bicycle facility density, ridership volume.

The contribution of any single attribute to the overall market price of the property can be found by taking the partial derivative of (1) with respect to any of the attributes that collectively comprise the characteristics of the differentiated good (housing in this case). This measure is the marginal price of an attribute at the given level of consumption. For example, ∂P/∂c_i represents the marginal value (marginal implicit price) of a given bicycling infrastructure variable c (i.e. the increase in expenditure required to obtain one more unit of attribute c, all else equal) (Freeman, 2003).

We form an empirical model that can be used to assess whether the current stock of the bicycling infrastructure amenity is capitalised into the Tempe housing market. The marginal implicit price for a single attribute (e.g. bicycling infrastructure) of a differentiated good (housing) can be used to estimate the value that consumers (on average) place on a change in the provision of that attribute (e.g. an increase or decrease in either the amount or quality of some amenity). Important assumptions underlying hedonic pricing are: that the analysis addresses consumer behaviour in a single housing market that contains housing stock with varying levels of each attribute (i.e. in the case of bicycling infrastructure, this would include houses with access to significant infrastructure as well as those that have little access), and that individuals have information on the alternatives available and are free to choose a house anywhere within the market. Actual purchasing behaviour will depend on individual preferences for housing attributes and related amenity values and their budget constraints.

There is no a priori functional form suggested for hedonic price functions because sale prices represent equilibrium market transactions (Freeman, 2003). Consistent with prior literature we specify a number of models using ordinary least squares (OLS) in addition to a spatial autoregressive model (SAR) (Conway et al., 2010; Cropper et al., 1988). Independent explanatory variables were selected consistent with theoretical expectations of factors anticipated to influence housing prices in addition to control variables. Our base model (OLS 1, Table 3) includes characteristics specific to the property, bicycling-related variables, amenity access and the neighbourhood. We tested for violations of OLS assumptions including assessing outliers and potential influential observations (Cook’s distance), non-normality and correlation of residuals, and we use heteroscedastic consistent standard errors. We explored the potential for omitted variable bias by adding additional candidate explanatory variables and examining whether there are large changes in coefficient values, sign and/or significance. Considering the alternative model specifications included in Table 3, the initial model was well specified with coefficient values, sign and significance remaining stable, an indicator that we have made choices that minimise omitted variable bias to the best of our ability.

Previous studies suggest that spatial dependence may occur between units or in model residuals, as we expect that housing sales prices are influenced by the price of neighbouring homes (Conway et al., 2010; Shin et al., 2007). Moran's I_i analysis was therefore conducted to determine whether spatial autocorrelation in housing prices was present in the study area; results indicated that sale prices were similar among nearest neighbours (I_i = 0.71, p < 0.01, with k-nearest neighbour (knn) = 4), and spatial autocorrelation was also present within OLS model residuals. We account for the spatial dependence between properties by specifying a spatial autoregressive model (Table 3, SAR). Robust Lagrange multiplier tests for spatial dependence in the OLS models indicated that a spatial lag model would be most appropriate for handling this spatial dependence. The spatial lag model considers spatial dependence a result of omitted variable bias and is specified as:

y = ρ Wy + X β + ϵ

(2)

where ρWy specifies a spatially lagged dependent variable parameter using a spatial weighting matrix; ρ is the spatial lag parameter, W is the weight matrix, X is the vector of independent variables consistent with (1) and ϵ is the error term. For this study we used k-nearest neighbour weighting as it can be thought of as a weighted average of the response variable in a neighbourhood, and it allows for a neighbourhood that is not of a fixed width (Owusu-Ansah, 2011). We have chosen to specify W as the four nearest neighbours to account for the spatial arrangements and reduce bias while maintaining variance.

Results

Results for three OLS model specifications (OLS 1–3) are shown in Table 3; coefficients are given with standard errors in parentheses. Building on the base model (OLS 1), the OLS 2 model included job availability and differentiated between on- and off-street bicycle facility density, whereas OLS 3 further included the JTW by bicycle control and matches the specification used in the SAR model. Results for the represented housing characteristics (property size, liveable square footage, house age, pool presence, sale year) showed the expected signs and were statistically significant, with the exception of the multiple storeys variable. Parameter estimates for the amenity and neighbourhood control variables in all models are also presented in Table 3. In terms of the bicycling-related variables of interest, higher total and on-street bicycle facility density contributed to higher home sales prices, while off-street density was not significant in any of the models. Ridership density derived from the Strava data was not significant in any of the models, though the control for JTW by bicycle was significant and positive. The SAR model results were consistent with and closely matched the OLS specifications; the coefficients of a spatial lag model cannot be directly interpreted (LeSage and Pace, 2009), so the calculated average direct, indirect and total impacts are shown in Table 4. The total impact for the SAR model is mostly comprised of direct impacts, meaning that changes in sale price for a particular house result from changes in independent variables related to that house rather than changes from variables related to other houses (i.e. spill-over effects). Total effects are in many cases similar in magnitude to those obtained from the OLS specification.

Table 4.

Direct, indirect and total impacts for model SAR1

Variable	Direct	Indirect	Total
House and sale characteristics
Property size (acres)	0.7887***	0.0056	0.7943***
Size of house LN(ft²)	0.6577***	0.0046	0.6624***
House age	−0.0047***	0.0000	−0.0047***
Pool presence	0.0496***	0.0004	0.0500***
Multiple storeys present	0.0045	0.0000	0.0045
Sale year 2013	−0.1879***	−0.0013	−0.1893***
Sale year 2014	−0.1176***	−0.0008	−0.1184***
Sale year 2015	−0.0628***	−0.0004	−0.0633***
Bicycling characteristics
On-street bicycle facility density(miles of bicycle facilities/mi²)	0.0023***	0.0000	0.0023***
Off-street bicycle facility density(miles of bicycle facilities/mi²)	0.0005	0.0000	0.0005
Ridership density(LN(number of riders/0.5 mile buffer in 2016))	0.0008	0.0000	0.0008
Amenities and access
Distance to ASU(LN(ft))	−0.0961***	−0.0007	−0.0968***
Greenspace distance(LN(ft))	0.0034	0.0000	0.0034
LN(job availability)(number of jobs accessible in 45 min by car)	−0.5448***	−0.0038	−0.5486***
Neighbourhood and census block group controls
Elementary school score	−0.0007	0.0000	−0.0007
Middle school score	−0.0062*	0.0000	−0.0063*
LN(median income)	0.1021***	0.0007	0.1028***
Median age	0.0012**	0.0000	0.0012**
Percent racial/ethnic minority	−0.0009***	0.0000	−0.0009***
Crime rate (number of crimes per census block population 2013–2016)	0.0183***	0.0001	0.0185***
JTW bicycle %	0.0029***	0.0000	0.0029***

Notes: Level of significance 0.01% = ***, 1% = **, 5% = *.

Discussion and conclusion

This study demonstrates that overall bicycle infrastructure density as well as on-street facility density are positively associated with housing sale prices in Tempe, AZ. Hedonic pricing models that relate bicycling factors to property sales prices are location- and market-specific, so these results expand the available literature to a new study area. The finding for overall bicycle facilities aligns with previous research linking higher property values with bicycle facilities (e.g. Liu and Shi, 2017), however, the finding that on-street facility density is positively associated with house price when measured as the local density around a property differs from findings when access is measured in terms of proximity (i.e. distance to on-street facilities) (Krizek, 2006; Welch et al., 2016). This difference in finding may be explained in terms of accessibility, where proximity indicates the separation between a property and bicycle facilities, and density is more indicative of the underlying spatial structure of bicycle facilities. Bicycle infrastructure is a network at the city-wide scale, so a density-based accessibility measure is more indicative of how developed the bicycle network is rather than just the proximity of the closest link. Higher density indicates a better-developed network, which property owners value in terms of their willingness to pay. The impact of bicycle facility density on house sale price is smaller than the contributions from property and neighbourhood-/access-related variables, consistent with findings from previous literature. Using sales in 2016 as an example, each additional mile of on-street cycling infrastructure within 0.5 mile of a property is expected to increase mean sale prices in that year by approximately US$619. While the effect size is small when taken individually, the potential aggregate value for the city is not trivial; bicycling infrastructure contributes positively to property tax revenues through higher property values. Property tax rates within the study area vary, but are within the range of 12.5% (Maricopa County, 2016a). A US$1 million increase in total taxable property values will increase property tax revenues by US$120,000 annually. To place this in context, the total of net assessed property values⁴ in Tempe in 2016 was approximately US$1.67 billion (Maricopa County, 2016b). Using the results from Table 3, the presence of cycling infrastructure is estimated to have contributed 0.22%, US$3.6 million, to taxable property values, yielding approximately US$459,000 of the property tax revenue in that year.⁵ This is a partial estimate of overall values to the area and does not account for other benefits such as positive health outcomes associated with physical activity, possible reduction in traffic congestion, pollution and emissions (Frank et al., 2006; Woodcock et al., 2009), and mitigation of the need for additional motorised transportation infrastructure. Access to off-street multi-use paths is an amenity that is not reflected in buyers’ willingness to pay in our models, though this is likely a reflection of the proportional lack of off-street facilities in the study area and it may nonetheless be used as an attractive selling point in some areas.

In terms of bicycle ridership reported from Strava data, we find no statistically significant relationship between ridership volume within a property’s proximal neighbourhood and house sale price within the study area. The coefficient on JTW (percentage of population using bicycling as their mode of transport to work) is positive and significant, suggesting that the ability to commute to work by bike is positively reflected in housing values, likely through location, including proximity to the ASU campus and downtown Tempe area, and neighbourhood characteristics (e.g. median age and income) (McKenzie, 2014). The highest percentages of residents using a bicycle for their JTW were located around the ASU campus and downtown Tempe area, which corresponds to areas of higher bicycle facility density. Though Census JTW data are collected more systematically than crowdsourced ridership data, they are also limited in capturing overall bicycle ridership at city-wide scales. The JTW data are not a full census of bicycle ridership because they are comprised only of workers aged 16 and over who worked during the reference week, so recreational or non-work trips are excluded and an overall low proportion of those surveyed report bicycle as their mode of transport for commuting (1%).

Our findings have implications for city officials, policy makers, community members and property owners within the city of Tempe. The results support the argument that bicycling infrastructure has a positive association with neighbourhood sales prices and property tax revenues. This information empowers homeowners and city planners in undertaking data-driven decision-making and provides additional analysis that can be called upon when proposals for new bicycle facilities are introduced in other urban contexts. While this hedonic pricing approach can be used in any housing market, the specific results for Tempe, AZ, are not directly transferrable to other urban housing markets. Future research may use alternative, quasi-experimental, approaches to further test these findings. Though causality was not the focus of our analysis, it may be possible to determine if there is a causal link between changes in ridership and home prices by comparing a ‘before-and-after’ scenario as bicycle facilities are installed, especially considering findings that changes in bicycle facilities are associated with changes in ridership patterns in crowdsourced data (Boss et al., 2018). We also recommend continued analyses of the relationships between cycling and other neighbourhood factors as the availability of ridership data increases.

A notable contribution of this work is including crowdsourced data to examine whether bicycle ridership is related to property values. These bicycling activity data, gathered from the Strava smartphone app, have added richness to available bicycling-related variables and have overcome prior limitations related to the variables available when assessing the economic value of bicycle infrastructure by allowing for examination of its practical use. In addition to addressing associations with bicycling infrastructure alone, we have included a measure of overall ridership and demonstrated that ridership and bicycle facility density do not directly correspond (Figure 2) and that increased facility density, but not ridership, is associated with higher housing sales prices. These factors, taken together, indicate that bicycle infrastructure and ridership are factors that should be accounted for separately when examining their relationship with other elements of the neighbourhood environment.

Bicycle-friendly design with compact living environments, mixed-use development and infrastructure that supports active-transport modes is of interest to planners, residents and developers in many urban areas. These design efforts aid problems associated with mobility and accessibility including congestion, pollution and urban sprawl. Prior research shows the economic valuation of such features, including potential positive influences on property prices. Results from the current study can be used to inform debates on the positive economic impacts associated with bicycle-friendly infrastructure and ridership.

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Lindsey Conrow

Notes

References

‘About GreatSchools' Ratings’ (14 March 2018) Available at: https://www.greatschools.org/gk/summary-rating/ (accessed 14 December 2018).

Asabere

Huffman

(2009) The relative impacts of trails and greenbelts on home price. The Journal of Real Estate Finance and Economics 38(4): 408–419.

Bartholomew

Ewing

(2011) Hedonic price effects of pedestrian- and transit-oriented development. Journal of Planning Literature 26(1): 18–34.

Beenackers

Foster

Kamphuis

, et al. (2012) Taking up cycling after residential relocation: Built environment factors. American Journal of Preventive Medicine 42(6): 610–615.

Blanc

Figliozzi

Clifton

(2016) How representative of bicycling populations are smartphone application surveys of travel behavior? Transportation Research Record: Journal of the Transportation Research Board 2587(1): 78–89.

Boss

Nelson

Winters

, et al. (2018) Using crowdsourced data to monitor change in spatial patterns of bicycle ridership. Journal of Transport & Health 9: 226–233.

Boyle

Barrilleaux

Scheller

(2014) Does walkability influence housing prices? Social Science Quarterly 95(3): 852–867.

Brander

Koetse

(2011) The value of urban open space: Meta-analyses of contingent valuation and hedonic pricing results. Journal of Environmental Management 92(10): 2763–2773.

Broach

Dill

Gliebe

(2012) Where do cyclists ride? A route choice model developed with revealed preference GPS data. Transportation Research Part A: Policy and Practice 46(10): 1730–1740.

10.

Campbell

Jr Munroe

(2007) Greenways and greenbacks: The impact of the Catawba Regional Trail on property values in Charlotte, North Carolina. Southeastern Geographer 47(1): 118–137.

11.

Cervero

Duncan

(2003) Walking, bicycling, and urban landscapes: Evidence from the San Francisco Bay Area. American Journal of Public Health 93(9): 1478–1483.

12.

Cervero

Ferrell

Murphy

(2002) Transit-oriented development and joint development in the United States: A literature review. TCRP Research Results Digest 52, 144 pp.

13.

City of Tempe (2008) Comprehensive Transportation Plan. Available at: https://www.tempe.gov/home/showdocument?id=557 (accessed 18 February 2019).

14.

City of Tempe (2017) Data file: Police general offense. Available at: https://data.tempe.gov/dataset/police-general-offense (accessed 4 October 2018).

15.

Conrow

Wentz

Nelson

, et al. (2018) Comparing spatial patterns of crowdsourced and conventional bicycling datasets. Applied Geography 92: 21–30.

16.

Conway

Wolch

, et al. (2010) A spatial autocorrelation approach for examining the effects of urban greenspace on residential property values. The Journal of Real Estate Finance and Economics 41(2): 150–169.

17.

Cradock

Troped

Fields

, et al. (2009) Factors associated with federal transportation funding for local pedestrian and bicycle programming and facilities. Journal of Public Health Policy 30(1): S38–S72.

18.

Cropper

Deck

McConnell

(1988) On the choice of functional form for hedonic price functions. The Review of Economics and Statistics 70(4): 668–675.

19.

Forsyth

Krizek

(2010) Promoting walking and bicycling: Assessing the evidence to assist planners. Built Environment 36(4): 429–446.

20.

Frank

Sallis

Conway

, et al. (2006) Many pathways from land use to health: Associations between neighborhood walkability and active transportation, body mass index, and air quality. Journal of the American Planning Association 72(1): 75–87.

21.

Freeman

(2003) The Measurement of Environmental and Resource Values: Theory and Methods (2nd edition). Washington, DC: Resources for the Future.

22.

Griffin

Jiao

(2015) Where does bicycling for health happen? Analysing volunteered geographic information through place and plexus. Journal of Transport & Health 2(2): 238–247.

23.

Hess

Almeida

(2007) Impact of proximity to light rail rapid transit on station-area property values in Buffalo, New York. Urban Studies 44(5–6): 1041–1068.

24.

Krizek

(2006) Two approaches to valuing some of bicycle facilities' presumed benefits: Propose a session for the 2007 national planning conference in the city of brotherly love. Journal of the American Planning Association 72(3): 309–320.

25.

Krizek

(2007) Estimating the economic benefits of bicycling and bicycle facilities: An interpretive review and proposed methods. In: Coto-Millán

Inglada

(eds) Essays on Transport Economics. Heidelberg: Physica-Verlag HD, pp. 219–248.

26.

LeSage

Pace

(2009) Introduction to Spatial Econometrics. Boca Raton, FL: Chapman and Hall/CRC.

27.

Joh

Lee

, et al. (2015) Assessing benefits of neighborhood walkability to single-family property values: A spatial hedonic study in Austin, Texas. Journal of Planning Education and Research 35(4): 471–488.

28.

Lindsey

Man

Payton

, et al. (2004) Property values, recreation values, and urban greenways. Journal of Park & Recreation Administration 22(3): 69–90.

29.

Liu

Shi

(2017) Impact of bike facilities on residential property prices. Transportation Research Record: Journal of the Transportation Research Board 2662(1): 50–58.

30.

McKenzie

(2014) Modes less traveled – Bicycling and walking to work in the United States: 2008–2012. US Department of Commerce, Economics and Statistics Administration, US Census Bureau. Available at: https://www2.census.gov/library/publications/2014/acs/acs-25.pdf (accessed 25 November 2019).

31.

Maricopa Association of Governments (MAG) (2014) Data file: Bikeways 2014. Phoenix, AZ: MAG.

32.

Maricopa County (2016a) Maricopa County 2016 Tax Rates. Available at: https://www.maricopa.gov/DocumentCenter/View/34206/Tax-Rate-2016-PDF (accessed 27 December 2018).

33.

Maricopa County (2016b) Maricopa County Tax Levy. Available at: https://www.maricopa.gov/DocumentCenter/View/34195/Tax-Levy-2016-PDF (accessed 27 December 2018).

34.

Maricopa County Assessor’s Office (2018) Assessor book series st42025. Available at: https://mcassessor.maricopa.gov/data-sales/gis.php (accessed August 2017).

35.

Mohammad

Graham

Melo

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

36.

Nicholls

Crompton

(2005) The impact of greenways on property values: Evidence from Austin, Texas. Journal of Leisure Research 37(3): 321–341.

37.

Owusu-Ansah

(2011) A review of hedonic pricing models in housing research. Journal of International Real Estate and Construction Studies 1(1): 19.

38.

Palmquist

(2002) Hedonic models. In: van den Bergh

JCJM

(ed.) Handbook of Environmental and Resource Economics. Cheltenham: Edward Elgar Publishing, pp. 765–776.

39.

Parent

Vom Hofe

(2013) Understanding the impact of trails on residential property values in the presence of spatial dependence. The Annals of Regional Science 51(2): 355–375.

40.

Seo

Golub

Kuby

(2014) Combined impacts of highways and light rail transit on residential property values: A spatial hedonic price model for Phoenix, Arizona. Journal of Transport Geography 41: 53–62.

41.

Shin

Washington

Choi

(2007) Effects of transportation accessibility on residential property values: Application of spatial hedonic price model in Seoul, South Korea, metropolitan area. Transportation Research Record: Journal of the Transportation Research Board 1994(1): 66–73.

42.

Sirmans

Macpherson

Zietz

(2005) The composition of hedonic pricing models. Journal of Real Estate Literature 13(1): 1–44.

43.

Sohn

Moudon

Lee

(2012) The economic value of walkable neighborhoods. Urban Design International 17(2): 115–128.

44.

Sun

Wang

, et al. (2017) Examining associations of environmental characteristics with recreational cycling behaviour by street-level Strava data. International Journal of Environmental Research and Public Health 14(6): 644.

45.

Tempe Transportation Master Plan (TMP) (2015) Available at: www.tempe.gov/transportationplan (accessed 29 December 2017).

46.

US Census Bureau (2016) American Community Survey 5-year Estimates, Tables B01003, B19013, B25003. Generated using American FactFinder. Available at: http://factfinder2.census.gov (accessed 4 October 2018).

47.

US Census Bureau (2018) American FactFinder ‘Community Facts’. Available at: https://factfinder.census.gov/faces/nav/jsf/pages/community_facts.xhtml?src=bkmk (accessed 14 December 2018).

48.

US EPA Smart Location Database (SLD) (2010) Smart Location Database 2.0. Accessed using EPA’s Clip and Ship Tool. Available at: https://edg.epa.gov/clipship/ (accessed 28 May 2019).

49.

Welch

Gehrke

Wang

(2016) Long-term impact of network access to bike facilities and public transit stations on housing sales prices in Portland, Oregon. Journal of Transport Geography 54: 264–272.

50.

Wilhelmsson

(2000) The impact of traffic noise on the values of single-family houses. Journal of Environmental Planning and Management 43(6): 799–815.

51.

Winters

Davidson

Kao

, et al. (2011) Motivators and deterrents of bicycling: Comparing influences on decisions to ride. Transportation 38(1): 153–168.

52.

Woodcock

Edwards

Tonne

, et al. (2009) Public health benefits of strategies to reduce greenhouse-gas emissions: Urban land transport. The Lancet 374(9705): 1930–1943.