Subjective Versus Objective: Divergency Between Subjective Walkability and Walk Score During the COVID-19 Pandemic

Abstract

Walk scores are used to evaluate accessibility by walking from a location to various meaningful destinations. Their validity remains contested because the widely available walk scores from the Walk Score® company are typically constructed using objectively measured origin–destination distances, which may be inconsistent with people’s spatial perceptions and even irrelevant to destinationless recreational walking trips. This study examines the divergency between objective walkability measured as walk score and subjective walkability assessed by perceived accessibility. Using data collected during the COVID-19 pandemic, we conducted spatial autocorrelation and multi-level regression analyses. Findings suggest that characteristics of both location and people affect the divergency between subjective–objective walkability. Regardless of their individual socioeconomic characteristics, respondents tend to match their subjective walkability assessment with the walk score better when they live in denser and older neighborhoods where more people may walk for utilitarian purposes. On the other hand, respondents with greater beliefs in environmental policies evaluate walkability in better alignment with walk score, regardless of their locational characteristics. The findings suggest that the main reasons for the subjective–objective divergency may lie in people using recreational walking for walkability assessment and their lack of lived experiences of utilitarian walking. Planners and policymakers should be mindful of the objective–subjective walkability divergency when using walk scores to support strategies and actions to improve walkability.

Keywords

pedestrians bicycles human factors accessibility planning and policy walkability

A walk score (WS) provided by the Walk Score® company is intended to show the objective walkability of a given location by providing a score on a 100-point scale, revealing a location’s accessibility to many walkable destinations. WSs are measured by analyzing the distance to various amenities with a distance decay function and the impact of population density, block length, and intersection density ( 1 ). WSs are readily available for virtually all locations in the United States and are treated as an indicator of a location’s attractiveness. They are now routinely included in real estate sales information and frequently used by planners as a predictor for non-automobile travel ( 2 , 3 ). Nevertheless, since it is constructed on the basis of walking to purposeful destinations, a WS seems only to capture a place’s walkability relevant to utilitarian walking rather than to destinationless recreational walking ( 3 ). By contrast, subjective (or perceived) walkability (SW) is often used to refers to the subjective evaluation of whether a place is walkable in general, including the design, comfort, safety, and many other factors of the walking space and their connections to walking behavior ( 4 ). Since a person’s socio-demographic characteristics, travel attitudes, and lived experience influence how they perceive the built environment characteristics, measures of SW are likely to exhibit inconsistencies with WS ( 5 ). The existence of divergency between WS and SW measures is well established in the literature. But there is limited understanding of whether and how the degrees of divergency may depend on who is making the evaluations (i.e., by population) and where the evaluations are made (i.e., by locations).

In this research, we use data about people’s perceived walkability reported during the COVID-19 lockdown period to create SW measures. Using these SW measures and WSs obtained for the same locations, we answer three research questions: (i) did divergence occur between SW measures and WSs during the COVID-19 pandemic? (ii) did those divergences vary spatially? and (iii) what individual- and neighborhood-level factors are related to the divergences? During the COVID-19 lockdown, walking or running near one’s home became a significant means for people to increase their physical activity ( 6 ). Evidence shows that some people replaced utilitarian walking with recreational walking, resulting in more overall walking trips in high-income areas than before the outbreak of COVID-19 ( 7 ). Perceived environmental walkability formulated during this period could have more to do with environmental conditions related to leisure and recreational walking, potentially increasing the divergency between WS and SW measures. Also, a recent literature review suggests a lack of studies that link walkability indicators to the COVID-19 pandemic ( 8 ).

We take advantage of the unique walking behavior and perceptions resulting from the COVID-19 disruption and study the underlying factors of objective–subjective walkability discrepancies. Specifically, we conducted an online survey of residents in the Eugene-Springfield metropolitan area and connected the survey data to WSs based on the approximate locations of respondent residences. We employed bivariate local Moran’s I analysis to examine the degree of spatial autocorrelation in the relationship between SW measures and WSs. We then developed a divergence score to measure the SW–WS divergences, which allowed us to examine the spatial variation in the divergences and how the divergences can be attributable to individual- and neighborhood-level factors. Findings from our analyses improve planners’ ability to utilize WSs in transportation planning and inform their effort to create healthy communities through interventions to the built environment.

Literature Review

Walk Score as a Valid Measure of Walkability: Mixed Evidence

WS is considered by many planners to be a robust operational measure of walkability. Planetizen offers this definition: “Walkability refers to the ability to safely walk to amenities within a reasonable distance, usually defined as a walk of 30 minutes or less” ( 9 ). The phrase “walk to amenities” indicates a clear disciplinary focus on walking as a “travel” means in urban planning. This narrow approach to walking and walkability becomes clear when we look at the definitions offered in other fields, such as public health and environmental psychology, where walking concerns researchers for its health and well-being effects regardless of whether it is carried out to reach any amenities or destinations ( 10 , 11 ). A conceptualization of walkability based on a consensus among researchers and practitioners from multiple disciplines offers us a much broader definition. It essentially expands walkability to a new term, “active living environments,” where “the emergent natural, built, and social properties of neighborhoods that promote physical activity and health and allow for equitable access to health-enhancing resources” ( 12 ). This broader “walkability” concept puts the validity of WS in doubt.

Existing research has assessed WS’s validity as a measure of objective walkability, directly or indirectly. First, research comparing WS with many other operational measures of walkability has provided some evidence for its validity. A study by Carr et al. ( 13 ) examining the characteristics of WSs shows that WSs relate to both objective (e.g., street connectivity, residential density, and access to public transit) and perceived physical activity environmental factors (e.g., summed scores of physical activity facilities). In many US cities, WS is associated with the density of various walking destinations, population density, residential density, intersection density, and median pedestrian route directness at different geographical scales, which is not surprising given how WS is constructed and the associations are more substantial in larger buffer areas ( 14 , 15 ). Studies in Japan and Canada also show that WS is positively related to environmental attributes that facilitate walking ( 16 , 17 ). Overall, many validation studies hold that WS can be a reasonably good indicator of walkability. The second area of research where WS’s validity is revealed shows significant relationships between WS and individual-level behavioral and health characteristics, including body mass index ( 18 ), obesity ( 19 ), cardiometabolic risk factors ( 20 , 21 ), and physical activity ( 22 , 23 ), as well as utilitarian, leisure, and transit walking.

Although WS has been considered a valid measurement of walkability and used for various empirical studies stated above, many studies also report WS is positively associated with factors hindering walking, such as crime, cul-de-sac count, average speed limit, and highway density ( 13 – 15 ). A recent Canadian study shows that WS is positively related to pedestrian crash rate when controlling for exposure ( 24 ). A systematic review suggests that WS is a surrogate indicator of the density of a neighborhood, which can only show the convenience of utilitarian walking ( 3 ). Moreover, more variables need to be considered for measuring leisure walkability, so WS and walkability are merely overlapping to some extent. To improve the performance of WS, scholars have developed various algorithms with new variables or statistical methods. For instance, Frank et al. ( 25 ) combined the walkability index (e.g., net residential density, retail floor area ratio, and land use mix) and WS to generate a three-dimensional Walkability Index (3D-WI). The validation results show that 3D-WI can predict physical activity better than the current network-based WS and the previous Euclidean-based WS.

Additionally, researchers have developed various methods to measure walkability at different scales apart from the WS. At the neighborhood scale, proximity and connectivity are two important indicators to represent neighborhood walkability ( 26 – 28 ). Proximity is an indicator showing the distance between residents’ locations and destinations, which is measured by density and land use mix. In general, high-density and mixed-use neighborhoods generate shorter distances between origins and destinations so that walking is a more convenient transport mode. Connectivity measures the directness of the street network between origins and destinations, which is measured by intersection density and net commercial area ratio ( 29 ). At the street scale, several urban design principles have been used to measure the street environment (i.e., street-level walkability), such as imageability, enclosure, human scale, transparency, and complexity ( 30 , 31 ). Additionally, scholars have created many environmental audit tools to evaluate environmental factors that facilitate or hinder walking, such as Systematic Pedestrian and Cycling Environmental Scan (SPACES) ( 32 ), Neighborhood Environment Walkability Scale (NEWS) ( 33 ), and Microscale Audit of Pedestrian Streetscapes (MAPS) ( 34 ). Nevertheless, not all of these methods are duplicable at a large scale because of the complexity of data acquisition and collection, so WS has its advantages in comparing walkability among different locations. Also, WS is updated every six months, so researchers and practitioners can track the change in this indicator over time.

Walk Score Versus Subjective Walkability Measures

The inherent limitation of WS to measure broad walkability predicts divergence between WS and SW measures since the latter tends to capture people’s perception of a place’s support for walking. A few studies examining the differences between WS and SW measures have offered evidence. Bereitschaft ( 35 ) compared WS with perceived walkability in Omaha, NE; perceived walkability was measured by the density of most walkable locations according to survey respondents. Results show that these two indicators tend to converge in mixed-use areas, such as the central business district and downtown. However, compared with SW measures, WS tends to underestimate areas that are suited for recreational walking, such as small entertainment districts, while overestimating the walkability of suburban and strip-mall retail corridors. Another study compared how WS and perceived walkability measures relate to different types of walking, respectively ( 36 ). Results suggest a significant relationship between WS and transport (destination-oriented utilitarian) walking, while the perceived walkability measure is related to broader walking trips, including transport, leisure, and overall walking. Nevertheless, in ultra-high-density areas of Japan, WS is related to perceived walkability consisting of several perceived built environment factors, such as access to walkable destinations, population density, access to transit, and the presence of bike lanes and sidewalks ( 37 ). Additionally, public health studies suggest that WS is more associated with GIS-derived walkability and audit-assessed walkability than self-reported walkability ( 38 ), and perceived walkability can better predict individual daily steps than WS ( 39 ).

The inability of WS to capture SW may be dependent on additional factors beyond a place’s locational characteristics. These may include an individual’s experiential and attitudinal characteristics, a place’s socio-demographic characteristics, and environmental qualities related to aesthetics and maintenance. Very little empirical research has examined those factors within the same scope. Filling this literature gap will enhance our ability to use WS properly in research or planning decision-making processes.

Methodology

Study Context

We conducted an online survey to assess how COVID-19 affected individual walking behavior and travel attitude in the Eugene-Springfield area. The survey questionnaire was divided into four sections. First, the survey included questions about travel behavior in January and February 2020, before the COVID-19 pandemic. Second, the survey asked participants about their travel behavior and subjective perception of walkability during Oregon’s shelter-in-place orders from March 16 to May 14, 2020. Third, the survey examined participants’ attitudes toward policies relating to transportation and climate change. Finally, the survey collected demographic and location information, including participants’ gender, age, household income, race, educational background, zip code and the nearest intersection to the respondent’s home.

Data Collection

The sampling frame included any resident of the Eugene-Springfield area aged 18 or older who had lived in the area continuously since at least January 2020. We adopted convenience sampling and recruited participants via listservs and social media. The research team distributed the survey through social media advertisements on Reddit and Facebook in both English and Spanish via geographically targeted paid advertising. Facebook ads targeted the area within a 20-mile radius of Eugene-Springfield. The survey link was also distributed via transportation-focused websites and lists, including the City of Eugene In Motion (monthly transportation listserv sent by Transportation Division), Downtown Eugene parking pass holders, GEARS (a bicycle club), LTD (Lane Transit District, the county’s bus and transit system), Transportation Options Program, neighborhood associations, Better Eugene Springfield Transportation transit advocacy group, CentroLatino Americano (a Latino/Latina oriented social service and advocacy organization), the University of Oregon’s communications website (“Around the O”), and the Eugene Chamber of Commerce. Survey respondents participated on a voluntary basis, but as an incentive, respondents were entered into a draw to win one of 10 $50 gift certificates to a national or regional business of their choice.

The survey was launched on June 3, 2020, three weeks after the lockdown ended, and concluded on July 31, 2020, collecting 684 responses. Table 1 compares the demographics of survey respondents with the general population of the Eugene and Springfield area. Table 2 shows the descriptions and data sources of dependent and independent variables. Specifically, we obtained the WS of the nearest intersection to the respondent’s home from Walkscore.com. To measure SW, we asked two questions in the survey about perceived walking accessibility to basic goods (e.g., groceries and restaurants) and to outdoor facilities (e.g., parks and trails) based on a seven-point Likert scale, respectively. Then, we summed these two values and normalized this index to a 0–100 scale, making the resultant SW scores comparable to WSs. Thus, the SW score (SWS) is defined as:

SW S_{i} = \frac{(WA B_{i} + WA O_{i}) - \min (WAB + WAO)}{\max (WAB + WAO) - \min (WAB + WAO)}

(1)

where $SW S_{i}$ is the SW score; $WAB$ and $WAO$ represent the rating on walking accessibility to basic goods and outdoor facilities, respectively; $i$ is the respondent. We used the differences between the SWS and WS to measure location-specific divergency between subjective walkability and objective walkability (see Table 2).

Table 1.

Demographic Characteristics of Survey Respondents Compared with the General Population of Eugene and Springfield

Gender	Region	Survey
Male	49%	23%
Female	51%	70%
Non-binary	na	3%
Prefer not to say	na	4%
Age (total population over 18)
18 to 24 years	21%	7%
25 to 34 years	18%	19%
35 to 44 years	15%	20%
45 to 54 years	13%	16%
55 to 64 years	14%	17%
65 to 74 years	11%	16%
76 to 84 years	5%	5%
85 years and over	3%	0%
Race/ethnicity
White (non-Latino/Hispanic)	79%	79%
Latino/Hispanic	10%	7%
Asian	4%	3%
American Indian/Alaska Native	1%	3%
Black	1%	1%
NHOPI	0%	0%
Other	0%	3%
Two or more	5%	na
Prefer not to say	na	5%
Income
Less than $15,000	15%	13%
$15,000 to $24,999	11%	11%
$25,000 to $34,999	12%	10%
$35,000 to $49,999	14%	13%
$50,000 to $74,999	19%	17%
$75,000 to $149,999	22%	21%
$150,000 to $199,999	4%	5%
$200,000 or more	4%	2%
Prefer not to say	na	8%

Note: na = not applicable.

Table 2.

List of Dependent and Independent Variables

Variables	Description	Data sources
Dependent variable
Divergence score	$Divergence Scores = \| Walk Score - SWS \|$	Walk Score® website and survey
Independent variables
Individual level
Age	Continuous variable	Survey
Gender	Male = 0, Female = 1	Survey
Race/ethnicity	White = 0, non-white = 1	Survey
Income (USD)	Less than 15,000 (1) 15,000 to 24,999 (2) 25,000 to 34,999 (3) 35,000 to 49,999 (4) 50,000 to 74,999 (5) 75,000 to 149,999 (6) 150,000 to 199,999 (7) 200,000 or more (8)	Survey
Education	No college degree = 0, with a college degree = 1	Survey
Time resident in Eugene	Years	Survey
Belief in environmental policy	1 = strongly disagree, 5 = strongly agree	Survey
Pre-COVID-19 routine active trips	Never, Infrequent, Frequent, Very Frequent	Survey
Neighborhood level
Population density	Number of people per square mile	2019 5-year American Community Survey (ACS) data
Commercial area	Percentage of commercial area	City of Eugene
Median year built	Median year when houses in the neighborhood were built	City of Eugene
Transit stops	Number of bus stops	City of Eugene
Parks	Number of parks	City of Eugene
Grocery stores	Number of grocery stores	City of Eugene
Bike lane	Total length of the bike lane (miles)	City of Eugene
Elderly (%)	Percentage of elderly people	2019 5-year ACS data
Bachelor (%)	Percentage of people with a bachelor’s degree	2019 5-year ACS data
Zero-car household (%)	Percentage of households without a car	2019 5-year ACS data
Non-Hispanic white (%)	Percentage of Non-Hispanic White people	2019 5-year ACS data
Median household income	Continuous variable	2019 5-year ACS data
Neighborhood safety	1 = strongly disagree, 5 = strongly agree	Survey

Data Analysis

Descriptive analysis

In this study, we first conducted a descriptive analysis to show the characteristics of respondents. Then, we ran T-tests to examine the overall difference between WS and SW. Additionally, we used Pearson correlation analysis to explore whether and how WS is related to SW.

Spatial descriptive analysis

We utilized bivariate local Moran’s I to understand whether WS and SW are spatially correlated. The local Moran’s I index for respondent $i$ is denoted by the following formula ( 40 ):

I_{i} = c \times z_{i} \sum_{j} w_{ij} z_{ij}

(2)

where $c$ is the variance of examined variable, $z$ is the deviations from the mean, $i$ and $j$ are nearby respondents, and $w_{ij}$ is the spatial weight between $i$ and $j$ . The null hypothesis is that the spatial distributions of SW and WS are independent; obtaining statistically significant local Moran’s I index from bivariate spatial autocorrelation analyses would reject the null hypothesis. A positive and statistically significant index indicates the presence of spatial clustering of similar values at a location and its neighboring sites, where either high values of SW and WS co-occur (i.e., high-high at a hot spot) or low values of SW and WS co-occur (i.e., low-low at a cold spot). By contrast, a negative and statistically significant index suggests the presence of disparities between the values of SW and WS, including a high SW value surrounded by low WS values (i.e., high-low) and a low SW value surrounded by high WS values (i.e., low-high). We used a distance of 0.5 mile from a location to define its neighboring area. We conducted separate bivariate spatial correlation analyses for SW versus WS and for WS versus SW

Regression analysis

We developed several multi-level regression models to explore what and how individual- and neighborhood-level factors are associated with divergency between WS and SW. Since respondents are nested in neighborhoods (each defined as a census block group), we included neighborhood variables in the regressions that not only control the unobservable (i.e., random) neighborhood effects but also investigate the average neighborhood effects (i.e., fixed) on the independent variables. The independent variables for regressions were standardized to make the coefficients comparable. Additionally, we conducted VIF tests, and found that the variance inflation factors (VIFs) for each variable in the final model are below 10, suggesting that any potential multicollinearity concerns are addressed in the final model specifications. The specific model is shown in the following:

y_{ij} = β_{0} + {β_{1}^{P} x_{1 ij}^{P} + \dots + β_{p}^{P} x_{pij}^{P}} + {β_{1}^{N} x_{1 ij}^{N} + \dots + β_{n}^{N} x_{nij}^{N}} + μ_{j} + ϵ_{ij}

(3)

where $x = {[(x_{1 ij}^{P} \dots x_{pij}^{P}), (x_{1 ij}^{N} \dots x_{nij}^{N})]}^{'}$ is a vector including $p$ individual-level and $n$ neighborhood-level covariates; $μ_{j}$ and $ϵ_{ij}$ are neighborhood-level random effects and errors, respectively. $y_{ij}$ denotes the aforementioned dependent variable, divergence between WS and SW. Independent variables include those associated with both subjective and objective walkability, including neighborhood built environmental variables, such as density, land uses, and presence of transportation facilities. Following the ecological model of people’s travel behaviors ( 41 ), we include individual-level and many other neighborhood-level variables associated with subjective walkability, including those measuring demographic and socioeconomic characteristics of individuals and neighborhoods. We ran the regression model block-wisely, adding individual-level variables first, neighborhood-level socio-demographic variables second, and neighborhood-level built environment variables last.

Results

Descriptive Statistics of Variables

Table 3 shows the summary statistics of dependent and independent variables. There is no statistically significant difference between the means of WS and SWS, whose mean values are 55.627 (standard deviation (SD) = 26.24) and 56.513 (SD = 36.19), respectively. The correlation analysis shows that these two variables have a statistically positive, yet weak, relationship (r = 0.246, p = 0.000).

Table 3.

Summary Statistics of Variables

Variable	Observations	Mean	SD	Minimum	Maximum
Divergence score	510	29.04	19.70	0.33	94
Age	495	47.25	15.93	18	82
Gender (dummy: male)	512	0.30	0.46	0	1
Income	516	4.64	2.23	1	9
Race (dummy: white)	519	0.20	0.40	0	1
Education (dummy: college)	519	0.38	0.49	0	1
Time in Eugene	519	31.66	18.94	1	73
Belief in environmental policy	518	33.31	5.85	7	42
Neighborhood level
Population density	470	8.65	4.97	0	52.93
Commercial area	470	18.96	23.58	0	100
Transit stops	470	3.78	3.15	0	16
Bike lane	470	3,644.06	2,751.34	0	18,828.59
Parks	470	0.24	0.50	0	3
Median household income	497	50,674.98	20,503.30	7,223.00	119,474.00
Non-Hispanic white	500	80.91	10.05	53.16	97.74
Bachelor degree	500	35.23	19.40	1.14	78.06
Median year built	496	1969.84	12.83	1944	2003
Zero-car households	499	10.04	9.68	0	50.28
Neighborhood safety	532	4.19	1.07	1	5

Bivariate Local Moran’s I Analysis

Figures 1 and 2 show the bivariate local Moran’s results for WS versus SW and SW versus WS, respectively. The red points denote “hot spots” where high WSs and high SWSs co-exist, while the blue points are “cold spots” where low WSs and low SWSs co-exist. The orange and green points represent the high-low and low-high spatial outliers, respectively.

Figure 1.

Bivariate local Moran’s I: walk score versus subjective walkability.

Figure 2.

Bivariate local Moran’s I: Subjective walkability versus walk score.

Both figures show that red “hot spots” are in relatively high-density areas (e.g., downtown Eugene and Springfield) and suburbs with relatively high accessibility to parks, trails, and other outdoor facilities, suggesting good congruency between high objective walkability and high subjective walkability reported for those areas. The blue “cold spots” do not seem to have a clear pattern in Figure 1. But they appear to occur in areas close to major transportation corridors and in the southern portion of Eugene where steep hillsides dominate the residential landscape (see Figure 2), indicating that respondents in those places perceived lower levels of walkability consistent with the low WSs assigned to those areas.

The green low-high points (Figure 1) and orange high-low points (Figure 2) indicate areas where low WSs are assigned, but high SW was reported. In both figures, they appear to concentrate in suburban areas with good accessibility mostly to public open space (e.g., parks and trails). Areas where high WSs are assigned yet low subjective walkability reported, that is, the orange points in Figure 1 and green points in Figure 2, do not seem to show a discernable distribution pattern in either figure. Nonetheless, Figure 2 shows that this type of divergency frequently occurs in downtown and its vicinity, mingling with hot spots where high SWSs converge with high WSs. This indicates that many factors can affect the perceived walkability of residents living in higher-density and mixed-use (i.e., high WS) neighborhoods. Circumstances connected to their idiosyncratic experiences are likely to affect people’s perception more than objective environmental conditions.

Regression Analysis

Table 4 reports outputs from three models examining mixed effects on the divergence score. Model 1 includes only the individual-level variables; Model 2 adds neighborhood-level socioeconomic variables; Model 3 is the full model that includes additional neighborhood built environment variables. The inclusion of additional variables improves model performance as indicated in smaller Akaike information criteria (AIC) values, supporting our argument that we should use wide-ranging and multi-scale factors to explain the objective–subjective walkability divergency.

Table 4.

Mixed Effects Regression for the Divergence Score

Variables	Model 1		Model 2		Model 3
Variables	Coef.	Std error	Coef.	Std error	Coef.	Std error
Individual factors
Age	.068	.049	.04	.049	.05	.048
Gender	–.042	.101	–.053	.099	–.054	.098
Race	.006	.116	–.043	.114	–.052	.113
Income	.083*	.048	.082*	.048	.075	.048
Education	.135	.101	.166*	.101	.151	.099
Time in Eugene	.014	.047	.041	.045	.042	.045
Belief in environmental policy	–.11**	.05	–.099*	.053	–.096*	.052
Pre routine trip	–.016	.047	–.019	.046	–.02	.046
Neighborhood socio-demographics
Housing value			.026	.059	.055	.064
Elderly			.004	.062	–.14**	.064
Bachelor degree			.076	.07	.051	.063
Zero-car household			–.184**	.082	–.22***	.076
Non-Hispanic White			–.021	.066	.007	.06
Drive alone			.079	.081	–.035	.077
Median household income			.095	.098	.03	.094
Poverty			.015	.11	.021	.102
Safety			–.059	.046	–.038	.045
Neighborhood built environment
Population density					–.14**	.06
Commercial					–.054	.061
Transit stops					.012	.057
Grocery store					.04	.047
Parks					–.018	.047
Bike lane					–.005	.052
Median year built					.268***	.065
Model statistics
	Number of observations	447	Number of observations	414	Number of observations	411
	Prob > chi2	0.075	Prob > chi2	0.000	Prob > chi2	0.000
	AIC	1,264.397	AIC	1,134.585	AIC	1,116.681

Note: ***p < .01, **p < .05, *p < .1.

Blank cells indicate variables excluded from the specfic model.

As shown in all three models, respondents with a strong belief in environmental policies are more likely to evaluate their locational walkability in a manner consistent with WS. Respondents who supported environmental policies were more likely to agree with the beliefs in reducing driving and increasing active travel. They may act on those beliefs by paying more attention to walking-related conditions than people with weaker beliefs in environmental policies. Therefore, their perceived walkability, especially to access basic goods, might be close to the real setting. Two other individual-level variables, education and household income, are associated with greater divergences in Models 1 and 2. A closer look at the data reveals that the SWSs of high-income respondents are significantly higher than WS. It is possible that these respondents live in neighborhoods conducive to recreational walking, and their walkability perceptions are primarily affected by environmental conditions disconnected from the WS computational methodology.

The regression models reveal that several neighborhood-level variables affect WS–SW divergency. Smaller divergencies between WS and SWS exist for people who live in neighborhoods with greater shares of elderly people (age 65 and above) and zero-car households, fewer new structures, or higher population densities. These are neighborhood characteristics often associated with residents having lower levels of car usage and greater levels of active travel. These residents may be those who consciously choose a neighborhood to minimize automobile use because of their environmental attitudes or those who cannot drive because of their physical conditions or lack of resources ( 5 ). Regardless of the underlying reasons, these residents tend to use walking or walking-supported trips (e.g., take public transit) for utilitarian purposes. Their lived experiences help align their walkability evaluation with the WS. Newer neighborhoods are associated with greater divergences, likely because these neighborhoods tend to be at the edge of the city where accessibility is relatively higher to open space but very low to services.

Discussion and Conclusion

This study examines the divergences between objective walkability measured as WS and SW assessed as perceived accessibility. The findings suggest that characteristics of both location and people affect the subjective–objective walkability divergences. Independently of their individual socioeconomic characteristics, respondents tend to better match SW with WS when they live in denser and older neighborhoods where more people may be walking for utilitarian purposes. On the other hand, respondents with greater beliefs in environmental policies evaluate SW in better alignment with WS regardless of their locational characteristics.

Considering that COVID-19 may have encouraged people to highlight recreational walkability more than they did before the pandemic, we believe the data collected during the COVID-19 lockdown period makes it easier for us to observe the subjective–objective walkability divergences resulting from people using recreational walking for walkability assessment. This is especially true for high-income residents. During the COVID-19 lockdown, most indoor fitness facilities were closed, causing many people to switch their indoor physical exercises to walking or running near home ( 6 ). Higher-income residents are likely the most affected group, since they are more likely to use the indoor facilities than other groups ( 42 ) and were more likely to have the work-at-home option during the lockdown. Thus, it is not surprising that nationwide pandemic data shows overall walking in high-income neighborhoods surpassing its pre-COVID level after an initial drop ( 7 ). Many consider people walking more in their neighborhood and gaining better walkability assessment a silver lining of the pandemic, which is supported by findings from this research showing higher-income residents reported SW significantly higher than the WS. But it remains unclear if the pandemic’s positive impact on people’s perception of environmental walkability could translate to greater walking for utilitarian travel as we enter the post-pandemic time. More studies should focus on whether the change in perception of walkability and actual behavior will be sustained into the future.

Results from this paper contribute to the literature and yield policy recommendations. Consistent with previous studies, our research shows that WS may not be a reliable measurement for walkability. Since both “place” and “people” affect the validity of WS, enhancing the accuracy of WS may include changing the WS algorithm based on the purpose of walking and the characteristics of people who walk. These changes will require WS to consider factors at different spatial scales (e.g., the urban design quality at the micro-scale and the socioeconomic factors at the meso-scale). Studies have shown that street view images can be used to automatically acquire various street-level factors ( 43 , 44 ), which can be combined with current WS measurements. Future research should explore how to integrate some neighborhood socioeconomic characteristics into WS to account for the impact of social norms and culture on residents. Further qualitative studies should also be used to improve our understanding of how people formulate their walkability assessment. Planners should consider the divergency in objective–subjective walkability when applying WS in transportation planning and policy-making processes.

Footnotes

Acknowledgements

We thank the following research team members who contributed to the design and analysis of the initial survey: Linda Price, Sara Hodges, Joshua Skov, Kivalina Grove, Clare Haley, Murat Kezer, and Shane Rhodes. We are grateful for funding from the University of Oregon Resilience Initiative which provided funding for this initial student and partnership with the City of Eugene.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Sian Meng, Yizhao Yang; data collection: Yizhao Yang, Rebecca Lewis; analysis and interpretation of results: Sian Meng, Yizhao Yang; draft manuscript preparation: Sian Meng, Yizhao Yang, Rebecca Lewis. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research project is partially funded by the Resilience Initiative Pilot Grant (2020-2021) from the University of Oregon and the grant from the National Institute for Transportation and Communities (grant number: NITC2016-UO-34).

ORCID iDs

Sian Meng

Rebecca Lewis

References

Walk Score®. Walk Score Methodology. 2022. https://www.walkscore.com/methodology.shtml.

Walk Score. Walk Score Customers. 2022. https://www.walkscore.com/professional/case-studies.php.

Hall

C. M.

Ram

Walk Score® and its Potential Contribution to the Study of Active Transport and Walkability: A Critical and Systematic Review. Transportation Research Part D: Transport and Environment, Vol. 61, 2018, pp. 310–324.

Arvidsson

Kawakami

Ohlsson

Sundquist

Physical Activity and Concordance Between Objective and Perceived Walkability. Medicine & Science in Sports & Exercise, Vol. 22, No. 2, 2012, pp. 280–287.

De Vos

Lättman

van der Vlugt

A-L.

Welsch

Otsuka

Determinants and Effects of Perceived Walkability: A Literature Review, Conceptual Model and Research Agenda. Transport Reviews, Vol. 43, No. 2, 2022, pp. 1–22.

Hargreaves

E. A.

Lee

Jenkins

Calverley

J. R.

Hodge

Houge Mackenzie

Changes in Physical Activity Pre-, During and Post-Lockdown COVID-19 Restrictions in New Zealand and the Explanatory Role of Daily Hassles. Frontiers in Psychology, Vol. 12, 2021, p. 410.

Hunter

R. F.

Garcia

de Sa

T. H.

Zapata-Diomedi

Millett

Woodcock

, et al. Effect of COVID-19 Response Policies on Walking Behavior in US Cities. Nature Communications, Vol. 12, No. 1, 2021, p. 3652.

Jardim

de Castro Neto

Walkability Indicators in the Aftermath of the COVID-19 Pandemic: A Systematic Review. Sustainability [Internet], Vol. 14, No. 17, 2022, p. 10933.

Planetizen. What Is Walkability? 2022. https://www.planetizen.com/definition/walkability

10.

Chandrabose

Cerin

Mavoa

Dunstan

Carver

Turrell

Owen

Giles-Corti

Sugiyama

Neighborhood Walkability and 12-Year Changes in Cardio-Metabolic Risk: The Mediating Role of Physical Activity. International Journal of Behavioral Nutrition and Physical Activity, Vol. 16, No. 1, 2019, p. 86.

11.

Wang

Zhang

Liu

Yao

Liu

The Relationship Between Visual Enclosure for Neighbourhood Street Walkability and Elders’ Mental Health in China: Using Street View Images. Journal of Transport & Health, Vol. 13, 2019, pp. 90–102.

12.

Tobin

Hajna

Orychock

Ross

DeVries

Villeneuve

P. J.

Frank

L. D.

, et al. Rethinking Walkability and Developing a Conceptual Definition of Active Living Environments to Guide Research and Practice. BMC Public Health, Vol. 22, No. 1, 2022, p. 450.

13.

Carr

L. J.

Dunsiger

S. I.

Marcus

B. H.

Walk Score™ as a Global Estimate of Neighborhood Walkability. American Journal of Preventive Medicine, Vol. 39, No. 5, 2010, pp. 460–463.

14.

Duncan

D. T.

Aldstadt

Whalen

Melly

S. J.

Gortmaker

S. L.

Validation of Walk Score for Estimating Neighborhood Walkability: An Analysis of Four US Metropolitan Areas. International Journal of Environmental Research and Public Health, Vol. 8, No. 11, 2011, pp. 4160–4179.

15.

Duncan

D. T.

Aldstadt

Whalen

Melly

S. J.

Validation of Walk Scores and Transit Scores for Estimating Neighborhood Walkability and Transit Availability: A Small-Area Analysis. GeoJournal, Vol. 78, No. 2, 2013, pp. 407–416.

16.

Koohsari

M. J.

Sugiyama

Hanibuchi

Shibata

Ishii

Liao

Oka

Validity of Walk Score® as a Measure of Neighborhood Walkability in Japan. Preventive Medicine Reports, Vol. 9, 2018, pp. 114–117.

17.

Nykiforuk

C. I. J.

McGetrick

J. A.

Crick

Johnson

J. A.

Check the Score: Field Validation of Street Smart Walk Score in Alberta, Canada. Preventive Medicine Reports, Vol. 4, 2016, pp. 532–539.

18.

Hirsch

J. A.

Diez Roux

A. V.

Moore

K. A.

Evenson

K. R.

Rodriguez

D. A.

Change in Walking and Body Mass Index Following Residential Relocation: The Multi-Ethnic Study of Atherosclerosis. American Journal of Public Health, Vol. 104, No. 3, 2014, pp. e49–e56.

19.

Chiu

Shah

B. R.

Maclagan

L. C.

Rezai

M-R.

Austin

P. C.

J. V.

Walk Score® and the Prevalence of Utilitarian Walking and Obesity Among Ontario Adults: A Cross-Sectional Study. Health Reports, Vol. 26, No. 7, 2015, pp. 3–10.

20.

Braun

L. M.

Rodríguez

D. A.

Evenson

K. R.

Hirsch

J. A.

Moore

K. A.

Diez Roux

A. V.

Walkability and Cardiometabolic Risk Factors: Cross-Sectional and Longitudinal Associations From the Multi-Ethnic Study of Atherosclerosis. Health & Place, Vol. 39, 2016, pp. 9–17.

21.

Méline

Chaix

Pannier

Ogedegbe

Trasande

Athens

Duncan

D. T.

Neighborhood Walk Score and Selected Cardiometabolic Factors in the French RECORD Cohort Study. BMC Public Health, Vol. 17, No. 1, 2017, p. 960.

22.

Thielman

Rosella

Copes

Lebenbaum

Manson

Neighborhood Walkability: Differential Associations With Self-Reported Transport Walking and Leisure-Time Physical Activity in Canadian Towns and Cities of All Sizes. Preventive Medicine, Vol. 77, 2015, pp. 174–180.

23.

Duncan

G. E.

Cash

S. W.

Horn

E. E.

Turkheimer

Quasi-Causal Associations of Physical Activity and Neighborhood Walkability With Body Mass Index: A Twin Study. Preventive Medicine, Vol. 70, 2015, pp. 90–95.

24.

Osama

Albitar

Sayed

Bigazzi

Determining if Walkability and Bikeability Indices Reflect Pedestrian and Cyclist Safety. Transportation Research Record: Journal of the Transportation Research Board, 2020. 2674: 767–775.

25.

Frank

L. D.

Appleyard

B. S.

Ulmer

J. M.

Chapman

J. E.

Fox

E. H.

Comparing Walkability Methods: Creation of Street Smart Walk Score and Efficacy of a Code-Based 3D Walkability Index. Journal of Transport & Health, Vol. 21, 2021, p. 101005.

26.

Owen

Cerin

Leslie

duToit

Coffee

Frank

L. D.

Bauman

A. E.

Hugo

Saelens

B. E.

Sallis

J. F.

Neighborhood Walkability and the Walking Behavior of Australian Adults. American Journal of Preventive Medicine, Vol. 33, No. 5, 2007, pp. 387–395.

27.

Leslie

Coffee

Frank

Owen

Bauman

Hugo

Walkability of Local Communities: Using Geographic Information Systems to Objectively Assess Relevant Environmental Attributes. Health & Place, Vol. 13, No. 1, 2007, pp. 111–122.

28.

Saelens

B. E.

Sallis

J. F.

Frank

L. D.

Environmental Correlates of Walking and Cycling: Findings From the Transportation, Urban Design, and Planning Literatures. Annals of Behavioral Medicine, Vol. 25, No. 2, 2003, pp. 80–91.

29.

Gebel

Bauman

Owen

Correlates of Non-Concordance Between Perceived and Objective Measures of Walkability. Annals of Behavioral Medicine, Vol. 37, No. 2, 2009, pp. 228–238.

30.

Ewing

Handy

Measuring the Unmeasurable: Urban Design Qualities Related to Walkability. Journal of Urban Design, Vol. 14, No. 1, 2009, pp. 65–84.

31.

Ewing

Handy

Brownson

R. C.

Clemente

Winston

Identifying and Measuring Urban Design Qualities Related to Walkability. Journal of Physical Activity and Health, Vol. 3, No. s1, 2006, pp. S223–S240.

32.

Pikora

T. J.

Bull

F. C.

Jamrozik

Knuiman

Giles-Corti

Donovan

R. J.

Developing a Reliable Audit Instrument to Measure the Physical Environment for Physical Activity. American Journal of Preventive Medicine, Vol. 23, No. 3, 2002, pp. 187–194.

33.

Cerin

Conway

T. L.

Cain

K. L.

Kerr

De Bourdeaudhuij

Owen

Reis

R. S.

, et al. Sharing Good NEWS Across the World: Developing Comparable Scores Across 12 Countries for the Neighborhood Environment Walkability Scale (NEWS). BMC Public Health, Vol. 13, No. 1, 2013, p. 309.

34.

Millstein

R. A.

Cain

K. L.

Sallis

J. F.

Conway

T. L.

Geremia

Frank

L. D.

Chapman

, et al. Development, Scoring, and Reliability of the Microscale Audit of Pedestrian Streetscapes (MAPS). BMC Public Health, Vol. 13, No. 1, 2013, p. 403.

35.

Bereitschaft

Walk Score® Versus Residents’ Perceptions of Walkability in Omaha, NE. Journal of Urbanism: International Research on Placemaking and Urban Sustainability, Vol. 11, No. 4, 2018, pp. 412–435.

36.

Tuckel

Milczarski

Walk Score(TM), Perceived Neighborhood Walkability, and Walking in the US. American Journal of Health Behavior, Vol. 39, No. 2, 2015, pp. 241–255.

37.

Koohsari

M. J.

McCormack

G. R.

Shibata

Ishii

Yasunaga

Nakaya

Oka

. The Relationship Between Walk Score® and Perceived Walkability in Ultrahigh Density Areas. Preventive Medicine Reports, Vol. 23, 2021, p. 101393.

38.

Hajna

Ross

N. A.

Joseph

Harper

Dasgupta

Neighbourhood Walkability and Daily Steps in Adults With Type 2 Diabetes. PLoS One, Vol. 11, No. 3, 2016, p. e0151544.

39.

Consoli

Nettel-Aguirre

Spence

J. C.

McHugh

T-L

Mummery

McCormack

G. R.

Associations Between Objectively-Measured and Self-Reported Neighbourhood Walkability on Adherence and Steps During an Internet-Delivered Pedometer Intervention. PLoS One, Vol. 15, No. 12, 2020, p. e0242999.

40.

Anselin

Local indicators of spatial association-LISA. Geographical Analysis, Vol. 27, No. 2, 1995, pp. 93–115.

41.

Sallis

J.F.

Owen

Fisher

E.B.

Ecological models of health behavior. In Health Behavior and Health Education: Theory, Research, and Practice, 4th ed.; Glanz

Rimer

B.K.

Viswanath

, Eds.; Jossey-Bass: San Francisco, CA, USA, 2008; pp. 465–486.

42.

IHRSA. The 2020 IHRSA Global Report: The State of the Health Club Industry. 2020. https://www.ihrsa.org/publications/the-2020-ihrsa-global-report/

43.

Long

Liu

How Green Are the Streets? An Analysis for Central Areas of Chinese Cities Using Tencent Street View. PLoS One, Vol. 12, No. 2, 2017, p. e0171110.

44.

Rundle

A. G.

Bader

M. D. M.

Richards

C. A.

Neckerman

K. M.

Teitler

J. O.

Using Google Street View to Audit Neighborhood Environments. American Journal of Preventive Medicine, Vol. 40, No. 1, 2011, pp. 94–100.