Perceived importance of context-specific built-environment factors of walking: A new perspective for prioritizing policy measures for promoting walking

Abstract Walking as one type of physical activity generates benefits for personal health and contributes to sustainability in its environmental, economic, and social dimensions. Based on a cross-sectional survey for a representative sample of German cities with at least 100,000 residents, this study investigates determinants of residents’ walking behavior. Two contexts of promenading in the city for leisure walking and the trip to the supermarket for utilitarian walking are distinguished in order to investigate differences in respondents’ perceived importance of built-environment characteristics and to bridge the gap between the individual level of residents’ behavior and the streetscape level. In addition, the analyses distinguish between frequent and less frequent walkers in order to understand differences in characteristics, behavior, preferences, and perceptions between these two groups. The results of this study with a sample of n = 4,637 respondents show that the relevance of the built environment is higher for leisure walking than for utilitarian walking and higher for frequent walkers compared to less frequent walkers. For leisure walking, “protection”-variables are ranked high (e.g. safety, security), followed by “comfort” (e.g. space, surface quality) and by “delight” (e.g. attractive buildings, greenery). Distance is key for utilitarian walking. Significant differences between frequent and less frequent walkers are identified mainly for the “Delight”-variables, this is attractive buildings and greenery for utilitarian walking, and space availability and street furniture for leisure walking. These differences should be considered in future research and street design practice which might preferably focus on the most demanding person group of frequent walkers.


Introduction
Walking for transport is beneficial from various perspectives.It is a zero emission, space-efficient, flexible and environmental-friendly transport mode.In combination particularly with public transport, it can cover all daily mobility needs.Walking as a physical activity contributes to meeting the goal of at least 150 min of moderate-intensity aerobic physical activity per week as formulated by WHO (2018) and thus significantly reduces the risk of non-communicable diseases and premature death (Guthold et al., 2018).Furthermore, a high share of pedestrians in public spaces is one important driver of the economic success of cities and contributes to social and livable public spaces (Gehl, 2010).
Studies on determinants of walking exist at two levels, this is (1) walking behavior of selected person groups at the individual level, e.g. of residents in a specific neighborhood (Gascon et al., 2019;Kerr et al., 2016;Wang et al., 2021) and (2) pedestrian volumes and behavior at streetscape level (Ewing & Clemente, 2013;Ewing & Handy, 2009;Gehl, 2010;Mehta, 2014).Dependencies between the two perspectives exist: when residents in a neighborhood walk more (individual level), more people can also be expected in the streets (streetscape level), and vice versa.Both perspectives are needed in order to develop targeted approaches for promoting walking (1) for selected person groups or (2) in specific street sections.Determinants at both perspectives can be grouped into supply side factors of the built environment at neighborhood level (meso-scale) and streetscape level (micro-scale) as well as person-related factors including mainly socio-demographic, socio-economic, and socio-psychological variables such as values, perceptions, preferences and norms (G€ otschi, Nazelle, Brand, & Gerike, 2017;Koszowski et al., 2019).Studies at streetscape level hardly include socio-demographic or other person-related variables; studies on residents' walking behavior at the individual level hardly include builtenvironment factors at streetscape level; both types of studies usually include variables describing neighborhood characteristics.
Overall, more studies on determinants of walking behavior exist at streetscape level and in addition, these are less heterogeneous with their clear focus on pedestrian volumes as dependent variable (Cambra & Goncalves, 2017;Ewing et al., 2012;Hermida et al., 2019;Lai & Kontokosta, 2018).
Studies at the individual level investigate (i) weekly walking duration for transport and for leisure based e.g. on the International Physical Activity Questionnaire (IPAQ) (Boakye et al., 2023;Cervero et al., 2009;Craig et al., 2003;Gascon et al., 2019) or (ii) trip numbers and their characteristics from travel diaries or questions on typical behavior from travel surveys (Cao, 2010;Carlson et al., 2016;Cerin et al., 2014;Cerin et al., 2009;Hsieh & Chuang, 2021;Liao et al., 2020;Sauter et al., 2016).Both types of studies at the individual level have varying foci on the whole population in specific cities or regions (Br€ uchert et al., 2020;Chan et al., 2021;Hermida et al., 2019;Larrañaga et al., 2016), on specific person groups (Guliani et al., 2015;Loh et al., 2022) or on specific trip purposes such as travel to school or to work (Guliani et al., 2015;Ribeiro & Hoffimann, 2018).Built-environment variables are mainly operationalized by objective indicators based on Geographic Information Systems (GIS).The Neighborhood Environment Walkability Scale (NEWS), developed in the IPEN-study (Saelens et al., 2003), and its abbreviated form (Cerin et al., 2013;Cerin et al., 2006;Kerr et al., 2016) are the most prominent survey instruments that use perceived built-environment characteristics.Saelens et al. (2003) compare objective GIS-based and perceived built-environment characteristics and demonstrate that the latter reliably represent the built environment.
This study is positioned at the individual level, it focuses on the built-environment determinants and particularly on the relevance of streetscape characteristics for residents' walking behavior.Our main research goal is to better understand what makes people walking with four main contributions to the existing literature: For the first time, (1) we do not only ask for general built-environment characteristics of the neighborhood but instead, we introduce two different specific situations to respondents, this is first promenading in the city for leisure (or discretionary) walking and second the trip to the supermarket for utilitarian walking.This approach is based on the hypothesis that different streetscape characteristics matter for the different walking trip contexts (Hillnh€ utter, 2021) and that these differences can only be identified with specific questions on these specific contexts.In addition, this context-specific approach also allows us to bridge the gap between the individual level and the streetscape level.This study investigates residents' walking behavior at the individual level but thanks to the context-based approach, we can ask micro-level questions on the relevance of specific streetscape characteristics.(2) We distinguish in our analysis between frequent and less frequent walkers in order to understand differences in characteristics, behavior, preferences, and perceptions between these two groups.This approach is based on the hypothesis that people who walk more often might be more observant and attentive while walking and might assign different priorities to built-environment characteristics compared to people who walk less.(3) We develop one single question on builtenvironment characteristics of respondents' neighborhood which should be simple but at the same time it should well differentiate neighborhoods in terms of population density and land use diversity.In addition, the question should well match with the typical building and urban structures in German cities which have not been fully covered in the mainly American studies yet.(4) We create representative evidence for residents in German cities with at least 100,000 inhabitants.With these four main contributions, this study should inform efforts in urban and transport planning as well as in public health for promoting walking, public life, and physical activity.
The remainder of this paper is organized as follows: The state of the art on determinants of walking is presented in the section "Literature" , based on a literature review at both the individual and the streetscape level.The section "Methods" describes the study design followed by section "Statistical Analysis" that introduces the methods used for statistical data analysis.Afterwards results are presented in the section "Results" including two models focusing on determinants of leisure versus utilitarian walking behavior and on the differences between frequent and less frequent walkers.The models are built up groupwise by thematic blocks, to investigate the explanatory power of individual, meso-scale and micro-scale factors on walking.The final section discusses the results, compares them with the state of the art, develops recommendations for policy making in urban and transport planning and reflects the methodological strengths and limitations of the study.

Characteristics of the built environment
For the built environment, the "5 Ds" (Density, Diversity, Design, Distance to public transport, Destination accessibility) are consistently significant in quantitative analysis at both the individual and streetscape level.Density, measured e.g. by floor area ratios or population densities, and diversity, captured by entropy measures describing the number and variety of different land use types in a given area, are of particular importance for pedestrian volumes at specific street sections (Ewing et al., 2016;Lai & Kontokosta, 2018) as well as for residents' walking behavior (Christiansen et al., 2016;Gascon et al., 2019;Liao et al., 2020;Sugiyama et al., 2012;Vos et al., 2023).Shorter distances, particularly to railbased public transport, consistently and significantly increase pedestrian volumes at streetscape level (Ewing et al., 2016;Kim et al., 2019) and also residents' walking activity (Br€ uchert et al., 2020;Gascon et al., 2019;Knuiman et al., 2014;Lam et al., 2022;Paydar et al., 2020).Design variables such as the connectivity of the street network are significant in some studies but not in others (Ewing et al., 2016;Sugiyama et al., 2012).Destination accessibility shows an overlap with diversity and is significant mainly at the individual level (Christiansen et al., 2016;Sugiyama et al., 2012;Wang et al., 2021) but less in studies on pedestrian volumes at streetscape level (Ewing et al., 2012;Ewing et al., 2016).
The Ds also apply to the streetscape level but are used at this level almost only for studies investigating correlates of pedestrian volumes.Density, measured e.g. as the total building floor area for parcels abutting the street divided by the total area of tax lots, consistently impacts significantly on pedestrian volumes (Ewing et al., 2016).Imageability, enclosure, human scale, transparency, and complexity are further relevant design variables at streetscape level (Ewing et al., 2012;Ewing & Handy, 2009).Gascon et al. (2019) investigate the impacts of built-environment variables both at the home location and at the study or work location on residents' walking behavior based on GISdata, they find higher relevance for built-environment characteristics at the home location (see also Vale & Pereira, 2016).Blue and green infrastructures are relevant particularly for recreational walking (Christiansen et al., 2016;Wang et al., 2021).Liao et al. (2020) point out in their GIS-based study that particularly areas with residential land use and inland water have positive impact on walking frequency.Coverage with and characteristics of pedestrian infrastructure hardly gets significant at the individual level for residents' walking behavior (Sugiyama et al., 2012) but positively impacts on pedestrian volumes (Kang, 2015;Kim et al., 2019).Christiansen et al. (2016), based on GAMM-models and GIS data, find a linear relationship between residents' weekly walking time and land-use mix but not for residential density.This leads to the question, whether tipping points might exist for some of the correlates of walking.
Built-environment variables are generated from GIS-data (Gascon et al., 2019) or from survey-based questions on perceived walkability (Vos et al., 2023).The NEWS-questionnaire (Saelens et al., 2003) and its variations (see e.g.Sallis et al. (2010) for PANES as a short form of NEWS) are the most frequently used instruments that operationalize built-environment determinants of residents' walking behavior based on a validated list of items to be surveyed in a standardized questionnaire (Cerin et al., 2018;Kerr et al., 2016;Sallis et al., 2020;Van Dyck et al., 2012).NEWS in its original form measures 68 perceived attributes on (1) residential density, (2) land-use mix-diversity, (3) land-use mix-access, (4) street connectivity, (5) infrastructure and safety for walking, (6) esthetics, (7) pedestrian traffic safety and (8) crime safety (Cerin et al., 2006).Walking levels in studies that investigate the impact of perceived built-environment characteristics are mostly measured based on the IPAQ-questionnaire.This questionnaire distinguishes between walking for transport and for recreation and thus generates insights on context-specificity of walking behavior.Residential density and land-use mix diversity apply for both walking purposes (Boakye et al., 2023;Cerin et al., 2014;Kerr et al., 2016) with a stronger effect for recreational walking (Lam et al., 2022).Crime and safety (Boakye et al., 2023) as well as esthetics are significant for recreational walking (Boakye et al., 2023;Van Cauwenberg et al., 2018).Hsieh and Chuang (2021) distinguish between purposive and discursive walking and find similar effects.Residential density, land-use mix diversity, and traffic safety are associated with purposive walking while esthetics and crime safety are associated with discursive walking.Land-use mix access and street connectivity are common correlates of both walking patterns.No studies could be identified that investigate the impact of built-environment characteristics on walking behavior separately for persons with low or high walking frequencies.

Person-related determinants of walking behavior
Younger and older person groups have higher walking frequencies compared to persons aged in between in some studies (Liao et al., 2020) while the age variable does not get significant in others (Gascon et al., 2019).Women tend to walk more than men (Wang et al., 2021;Wasfi et al., 2017).Van Dyck et al. (2012) investigate interaction effects with built-environment characteristics and find some significant effects.For example, men are shown to be more sensitive to land use mix-access than women in terms of walking.Gascon et al. (2019) find less walking with higher education levels but no effect of education was found in most other studies (Cerin et al., 2009;Wang et al., 2021;Wasfi et al., 2017).Income is included only in few studies with mixed findings.For example, Moudon et al. (2007) and Br€ uchert et al. (2020) find more walking with higher income whereas Wang et al. (2021) find the opposite effect.People not working walk more compared to working people (Gascon et al., 2019), people with access to a car consistently walk less (Gascon et al., 2019;Wang et al., 2021).The same holds for access to bicycles (Gascon et al., 2019).Mixed findings exist for the effect of household characteristics on walking behavior.Wasfi et al. (2017) find higher walking levels for people living alone than in couples for younger but not for older persons whereas Wang et al. (2021) find more walking for couples.People who are physically active in their leisure time walk more than people who are not (Wasfi et al., 2017) and also healthy people walk more than people who have any difficulties to walk (Moudon et al., 2007;Wasfi et al., 2017).Only few studies exist that analyze the associations between socio-psychological variables and walking behavior.Pro-walking social norm is consistently related to higher walking levels (Gascon et al., 2019;Moudon et al., 2007).Based on a principal component analysis, Gascon et al. (2019) generate composite factors and analyze their impact on weekly walking duration.They find significant increases of walking with higher importance of safe, healthy, sustainable, and private travel as well as decreases when short, flexible, and predictable travel times are important.They find positive impacts of good opinions on walking and in addition interdependencies between walking levels and the use of the alternative modes.

Study design and data structure
This study is based on a cross-sectional online-survey on active mobility conducted in 2017 in twelve German cities with at least 100,000 inhabitants.The survey aimed at collecting data that represents resident behavior for German cities of this size.The sample was therefore drawn in a twostage process.First, eight German cities with different city size (100,000-200,000 inhabitants; 200,000-500,000 inhabitants; � 500,000 inhabitants) and topography (flat and hilly) were selected randomly; four additional cities were set from the beginning because these had been chosen as living labs where measures to promote active mobility were foreseen.
The second stage included the random selection of residents � 18 years of age from the registry of residents for each of the selected cities.Finally, 4,637 persons aged between 18 and 99 from the twelve German cities, grouped by city size and topography, provided a fully completed questionnaire.The characteristics of the sample are summarized in Table 1.They are presented in comparison with the results of the household travel survey "Mobility in Cities-SrV 2018" which reliably represents the population and their travel behavior in German cities (Hubrich et al., 2019).Official population statistics could not be used for comparison as these exist only for each individual city and because relevant variables such as car availability are missing.
The two samples match well for the variables gender, number of children in the household, and also for the accessibility of public transport.The sample in this study is in average younger than in SrV 2018, this is consistent with differences in the employment status.Car ownership is with 1.12 cars per household higher in this study than in SrV 2018 with 0.9 cars per household.One reason for this might be the bias of the sample in this study toward highly educated persons, the share of respondents with high school graduation is with 74% substantially higher in this study than in SrV 2018 with 53%.There is consensus in the literature that education levels relate to income and that income levels relate to car ownership respectively car availability (Buehler et al., 2017).It is thus very likely that the bias toward higher education levels impacts on car ownership (German Environment Agency, 2020;Hubrich, 2017).The number of bicycles per household in this study is with 2.19 bicycles per household higher than in SrV 2018 with 1.6.This indicates a possible bias toward persons who are interested and engaged in active mobility as the topic of this study.Descriptive statistics for the area type and accessibility of facilities for daily needs (food/supermarket, bakery, drugstore, pharmacy, parcel counter) are also presented but could not be compared to SrV 2018.There is a balanced distribution of area types, only 12% of respondents can reach all five facilities for daily needs within a 5-min walk.

Dependent variable for logistic regression models and model selection
The dependent variable for the logistic regression models in this study is the "frequency of walking in general" as the main target and ambition in urban and transport planning with substantial synergies to public health with its focus on the duration of walking and physical activity.Respondents stated their typical walking behavior within the last twelve months in seven categories ranging from "daily or almost daily" to "never" without any limitations in terms of time or distance.This question is consistent with standard household travel surveys (Hubrich et al., 2019).For this study, the dependent variable "frequency of walking in general" (n ¼ 4,637) is binary coded (0/1) in � Persons walking less than three to four days a week (share ¼ 43%) � Persons walking daily or almost daily (share ¼ 57%) This classification is chosen for two main reasons: First, both groups of frequent and less frequent walkers are represented with almost equal shares in the sample which is beneficial particularly for setting up the statistical models.Second, our initial tests of different possible groupings of this variable showed that a more detailed classification of walking frequency does not yield significant differences between the groups.These differences seem to emerge only with the distinction between (almost) daily walking and all other walking practices.
We investigate built-environment determinants at the micro-scale for this dependent variable in the two different contexts of promenading in the city and of walking to the supermarket leading to the two different logistic regression models, each of them built-up by thematic blocks as introduced below.

Overview of independent variables for logistic regression models
The independent variables for this study as shown in Table 2 are chosen based on the literature review.They cover three levels: (1) socio-demographic/economic variables at the individual level, (2) built-environment variables at the neighborhood level (meso-scale) and (3) at streetscape level (micro-scale).Builtenvironment variables are the main focus of this study, they are classified along the 5 Ds as introduced above.For population "Density" and land use "Diversity" at the meso-level, our ambition is to formulate one question that is simpler and easier to answer than the NEWS-questionnaire but still captures these two important variables for typical built-environment contexts in German cities comprehensively.Two questions on "Destination" accessibility and "Distance" to transit are added at the meso-level.Variables for "Design" are formulated at the micro-level of specific streetscapes for the two contexts of leisure and utilitarian walking.These variables are classified into the three groups of streetscape-level walking determinants developed by Gehl (2010), this is "Protection" as the basic requirements (e.g.safety, security), "Comfort" (e.g.space, surface quality) and "Delight" (also sometimes referred to as "Amenities" , including e.g.streetscape proportions, street furniture).

Meso-scale factors at neighborhood level
The concept for describing the built environment at neighborhood level is visualized in Figure 1.First, respondents were asked in question 1.1 to characterize the building types in their neighborhood.If in question 1.1 the category "rowed apartment houses" was ticked as a particularly diverse building type that can appear as parallel buildings or as blocks with backyards, and that fits in various types of areas (mixed or residential areas), respondents were asked to specify in a second question 1.2 the predominant type of rowed apartments in their neighborhood in more detail which corresponds to the diversity of land use in the neighborhood.In question 1.1, respondents had the possibility to check off multiple response categories in order to account for possible heterogeneity of building types within one neighborhood.For a distinct classification of the neighborhood type for each respondent within one variable, the multiple answers from question 1.1 were aggregated based on the potential number of persons per housing unit in each category into the new variable "Predominant Building Type-Ranked".The category "Rowed Apartment Houses" was expected to have the highest values in terms of potential number of persons, hence it got the highest rank (¼5) whereas detached single-family houses were assigned the lowest rank (¼1).Respondents who ticked "not present" or "all buildings correspond to this building type" for all categories in question 1.1 were excluded from further analysis for reasons of implausibility (n ¼ 62).The final new variable "Area Type by Predominant Building Type" to be used for the analysis was finally built as a combination of the new variable "Predominant Building Type-Ranked (grouped)" and question 1.2 as shown in Figure 1 (see also Figure S1 and Table S1 in the Supplementary Material).

Micro-scale factors at streetscape level and model selection
Based on the literature review, we hypothesize that builtenvironment-walking factors at the meso-scale can be uniformly formulated for leisure and utilitarian walking purposes.Therefore, we use general variables at this level as introduced above.For the micro-level, for specific street sections, however, we hypothesize that the relevance of the different built-environment characteristics differs depending on the trip purpose.People might be more attentive to their environment when walking for leisure compared to utilitarian walking and particularly in cases when the boundary between walking to get to destinations and walking as an activity in its own gets blurred, when people come to streets as destinations and not to get to destinations.They might give particular relevance to esthetics and nice surroundings in these cases.We therefore distinguish in Table 2 two sets of microscale-built-environment factors, this is first "leisure walking" and second "utilitarian walking".The example of promenading in the city is used as the context for leisure walking and the weekly trip to the supermarket as the context for utilitarian walking, it seems to be most suitable in the context of local mobility, seeing that walking is a transport mode with highest sensitivity to distance and that hardly any respondent (only around 10% of our sample) walks all the way to work or to education.Micro-scale items for these two contexts are chosen as shown in Table 2 based on the following hypotheses: (1) the speed while walking for leisure is particularly low and the awareness of streetscape characteristics is higher than for utilitarian trips (Hillnh€ utter, 2021), (2) the relevance of distance is high for utilitarian walking trips but low for leisure trips, (3) different types of equipment of pedestrian facilities matter for utilitarian and leisure trips, e.g.public toilets are important for leisure but less for utilitarian walking whereas functional qualities of pedestrian facilities such as space availability or smooth surfaces might be more important for utilitarian walking.
Two models are developed based on these considerations, one for leisure walking and one for utilitarian walking.The "frequency of walking in general" is the dependent variable in both models.There is no difference between the two models in the person-related variables and the built-environment factors at the meso-scale but at the micro-scale as explained above.Micro-scale-built-environment factors are binary coded for leisure walking trips into 0¼ (somewhat) unimportant and 1¼ (somewhat) important; and for utilitarian walking trips into 0¼ I (somewhat) do not agree and 1¼ I (somewhat) agree.The full sample can be used for model 1.For model 2, only respondents are considered who state that they usually walk to the supermarket.Figure 2 gives an overview of the main characteristics of the two models.

Initial tests of data structures and model selection
Seeing that walking frequency is chosen as the binary coded dependent variable, binary logistic regression is the appropriate method for the modeling part of this study (see e.g.Hosmer et al., 2013).To legitimate the application of a "one level"-logistic regression and to obtain correct estimations for the regression coefficients, the independence of data structures and of the different levels of analysis has to be tested first (Ali et al., 2019;Austin & Merlo, 2017;Sommet & Morselli, 2017).Two levels need to be considered; this is first respondents' person and household characteristics (level 1: individual) and second their home location, by definition respondents live in either of the twelve German cities selected for this study as described above (level 2: context).
Multilevel logistic regression in the form of a Random Intercept-Only Model is applied without specific predictors (independent variables) to verify whether data structures are nested and to understand which proportion of the variance of the dependent variable can be assigned to the respondent (level 1: individual) respectively to the city (level 2: context) (Tausendpfund, 2020).The application of this model with the help of the generalized linear mixed-model package in SPSS Version 27 shows that the random intercept variance var(u 0j ) is with 0.002 not statistically significant (p > 0.05).The Intraclass Correlation Coefficient (ICC) can be calculated based on this result as shown in Equation ( 1) below.The ICC quantifies the variance of the dependent variable at the context level 2. According to the literature, in social science cluster effects occur typically for ICC-values between 0.05 and 0.25 (Snijders, 2012;Tausendpfund, 2020).
With an ICC of 0.0006 and the non-significant random intercept variance, data structures can be assumed as independent.This means, that the variance of the dependent variable can largely be explained by respondent characteristics, the context level does not need to be considered further (Tausendpfund, 2020).With this result, the first condition for running a "one-level" standard binary logistic regression is fulfilled.
Avoiding multicollinearity between independent variables is the second requirement for logistic regression analysis.It is tested with the Pearson-Product-Correlation Coefficient which is appropriate for testing the correlation of metric and dichotomous dummy-coded (0/1) variables (Bortz, 2010;Brosius, 2018).Predictors with strong correlations (r � 0.5) are excluded from logistic regression.The final correlation matrices of model 1 and model 2 are provided in the Supplementary Material (Tables S2-S3).Based on these two initial tests, a binary logistic regression analysis can be applied to estimate the impact of individual, meso-scale and micro-scale variables on the frequency of walking.

Model selection and diagnostics
Two models are computed using SPSS Version 27 as shown in Tables 3 and 4, the first one for leisure walking and the second one for utilitarian walking.Thematic blocks of independent variables are added one after the other for each of the two models in order to investigate the explanatory power of individual, meso-scale and micro-scale factors on walking.
The goodness of fit criterion R 2 Nagelkerke and the information criteria Akaike Information Criterion (AIC) as well the Bayesian Information Criterion (BIC) are used for model diagnostics and comparison.R 2 Nagelkerke as shown in Tables 3 and 4 is acceptable seeing that R 2 -statistics are consistently lower in logistic regressions compared to linear regressions (Backhaus et al., 2018).Cao (2010) find in his binary logit model a comparable pseudo-R 2 -value which he classifies as typical for disaggregated data with large sample sizes, this applies also for our study (see also Steyerberg, 2019).Higher numbers of explanatory variables also reduce pseudo-R 2 -values (Wolf & Best, 2010).Backhaus et al. (2008) state that the explanatory power of models in social sciences with values from 0.5 can be interpreted as very good.Overall, our ambition is not to fully explain the variance in walking behavior with the models but rather to identify main mechanisms and significant effects in how built-environment characteristics impact on behavior.Seeing these specificities of R 2 Nagelkerke, we apply AIC and BIC as complementary criteria in order to ensure the comprehensive evaluation and comparison of the different models (Wolf & Best, 2010).Both information criteria consider the number of explanatory variables (k) as shown in the Equations ( 2) and (3) but only the BIC takes sample size into account and does thus better fit for studies with larger sample sizes.
With the increase of pseudo R 2 statistics, both information criteria AIC and BIC should decrease so that models with the lowest values can be considered as most suitable (see e.g.Gehrke, 2019; Wolf & Best, 2010).In this study, in both models the negative tendency of AIC and BIC becomes visible with the successive addition of independent variables (see Figures 3 and 4).
Regarding the correct classification of the samples within the models, both show values above the random classification probability (see e.g.model 1.5: 63%; model 2.6: 73%).Both models are significant (p < 0.001), which is a result of the Likelihood-Ratio-Test and indicates that the models containing explanatory variables improves the fit in relative to their intercept-only models (see e.g.Crownson, 2020).
Besides parameters of goodness of fit, the stability of effect directions and Odds as well as the logical interpretability of models, which are all given in our models, also contribute to the explanatory power of the models.

Descriptive statistics for micro-scale built-environment factors
Figures 5 and 6 show the importance and relevance of micro-scale-built-environment characteristics for leisure and for utilitarian walking.Overall, the relevance of the built environment is higher for leisure walking than for utilitarian walking and higher for frequent walkers compared to less frequent walkers.For leisure walking, traffic safety and security are paramount, between 95 and 97% of respondents indicate (somewhat) importance to these factors following by good lighting with 95% which concerns both, safety and comfort.High importance for leisure walking is also assigned to trees and planting along the sidewalk, space, surface quality, low vehicle traffic, street furniture and attractive buildings with 79 to 93% of respondents indicating (somewhat) importance to these variables.First floor usage (64%), streetscape proportions (61%) and public toilets (50%) get lowest levels of agreement.The low levels of agreement for public toilets might result from differences between age groups and respondents' health conditions.The availability of public toilets in general is (somewhat) important for about 64% of all respondents in the age-group of 65 years and older.Within the younger age groups the importance decreases.For 49% of the respondents in the age group of 35 to 64 years and for 44% of the 18 to 34 years old respondents, the availability of public toilets is (somewhat) important.Differences between frequent and less frequent walkers are generally low with highest values for "Lot of Space to Walk" and "Street furniture".
The relevance of built-environment characteristics looks different for utilitarian walking.Short distances to destinations are of highest relevance, 95% of respondents rank these as (somewhat) important.All other variables are less important.Walking to the supermarket as the fastest alternative is indicated by 79% and sufficient widths of sidewalks by 77% of respondents as reasons for choosing the walking mode, followed by crime-related security (51%), green areas (39%), low traffic (38%) and attractive buildings (34%).Differences between frequent and less frequent walkers mainly exist in the variables with overall lower levels of agreement, this is "Attractive Buildings" and "Attractive Green Areas" along the walkway.Differences between frequent and less frequent walkers are higher for utilitarian walking compared to leisure walking.Both person groups seem to be similarly attentive to their environment when they walk for leisure whereas for utilitarian walking, frequent walkers seem to be more attentive than less frequent walkers.

Models for leisure and utilitarian walking
Tables 3 and 4 show the final logistic regression models for leisure and utilitarian walking with five (model 1) respectively six stages (model 2).All coefficients and odds ratios   (OR) are plausible and stable for the different model stages which is an indication for their stability within the models and strengthens their explanatory power.Both models show comparable results for the socio-demographic and socio-economic factors and also for the built-environment variables at meso-scale due to the substantial overlap of the two samples.The reduced sample size in model 2 causes only minor differences.
Car availability and employment are the two consistently significant individual factors both for leisure and for utilitarian walking.For each extra car within one household, the chance of (almost) daily walking (y ¼ 1) is reduced significantly by 1.3 times (p < 0.001) in both models.This relationship can also be found in the literature (see e.g.Cervero & Radisch, 1996;Haybatollahi et al., 2015;Kaplan et al., 2016;Wang et al., 2021).The number of bicycles within households is not significant.The effect of employment is particularly strong for utilitarian walking for non-employed and retired persons who have a 2.8 respectively 3.0 higher chance of being a frequent walker than persons in education, the ORs are with 1.9 respectively 1.3 lower for leisure walking but still significant.Employed persons aged below 35 years have an almost two times lower chance of daily leisure walking (OR ¼ 0.6 for utilitarian walking) than persons in education.Employed persons aged above 35 years show a 1.6 times lower chance for (almost) daily leisure walking and no significant differences to persons in education for utilitarian walking.These results are plausible, since employed persons tend to be busier, have longer trip distances (e.g. to their workplace) and more complex trip chains, these trip patterns are less suitable for walking as a mode of transport.In addition, non-employed or retired persons are more often subject to financial constraints and are thus, in combination with possibly more flexibility in choosing their destinations and time use, more affine to frequent walking.
Meso-scale factors include the predominant building type and accessibility to facilities of daily needs and public   transport stops.They are consistently significant with high ORs in both models.The chance of walking frequently doubles in areas with rowed apartment buildings in mixed-use areas (Model 1.5: OR ¼ 2.32, p < 0.001) compared to areas with detached single-family houses as the reference category.ORs are somewhat smaller but still significant and important for the factors "Detached Apartment Buildings" (Model 1.5: OR ¼ 1.48) and "Rowed Apartment Buildings in Residential Area" (Model 1.5: OR ¼ 1.47).ORs in model 2 are in the same range and also significant.These findings are consistent with the literature which shows consistently significant effects on walking for density, diversity and further D-variables (Cervero & Kockelman, 1997;Kang, 2015;Stead & Marshall, 2001).All accessibility factors are significant for leisure walking in model 1 and partly significant for utilitarian walking in model 2. The chance of (almost) daily walking increases significantly (Model 1.5: OR ¼ 1.52; p < 0.001) for households that reach all five considered facilities of daily needs within a five-minute walk compared to those households that reach less facilities.The result remains insignificant in the second model but still, this variable is close to statistical significance (p ¼ 0.06) and the model stages 2.2 to 2.5 show significant effects.Access to public transport within a five-minute walk significantly increases the chance of (almost) daily walking (see Model 1.5: OR ¼ 1.54; p < 0.001).The significance of these two accessibility factors shows their complementary character to the predominant building type and confirms that all three meso-scale factors should be considered for further studies.
For interpreting model results at the micro-scale, it is important to remember that questions at this level address the importance of streetscape factors and not their existence or shape like this is the case at the meso-scale.Space to walk, crime-related security, and street furniture are the three significant micro-scale variables for leisure walking in model 1 with positive coefficients and ORs between 1.41 and 1.68 (p < 0.05).Public toilets are also significant (OR ¼ 0.86; p < 0.05), their negative effect on walking prevalence is attributed to the relationship between age, walking frequency and the importance of public toilets.Public toilets are more relevant for people with lower health status levels, these persons tend to walk less.Different results were obtained for utilitarian walking.Attractive green areas along the walkway (OR ¼ 1.49; p < 0.01), attractive buildings (OR ¼ 1.40; p < 0.5), and low traffic (1/OR ¼ 1.37; p < 0.05) are the three significant micro-scale factors in model 2. This means that frequent walkers assign higher relevance to greenery and buildings next to the street but less relevance to volumes of motorized traffic compared to less frequent walkers.

Discussion
The descriptive and model-based results confirm findings in the literature as well as the initial hypotheses for this study.Walking is a basic mode of transport, 57% of respondents in our sample walk on a daily or almost daily basis.People walk across all socio-demographic and socio-economic person groups, only car availability and employment status are found to significantly impact on walking prevalence.The distinction of employed persons by age below respectively above 35 years proved to be suitable, younger employed persons are significantly less likely to walk on a (almost) daily basis than employed persons aged above 35 years.Further variables such as gender, education or household type show no significant effects in the models, this is consistent with the literature (see e.g.Gascon et al., 2019;Wang et al., 2021;Wasfi et al., 2017).All three meso-scale built-environment factors are of high relevance, they include the predominant building type in the neighborhood as well as accessibility to facilities of daily needs and to public transport stops.The Ds and particularly density and diversity matter also in this study, this is consistent with the literature (see e.  the variable on the predominant building type has been suitably developed to classify the different area types.Findings at these two levels of households respectively persons and mesoscale built-environment factors are similar for leisure and for utilitarian walking due to the overlap of the two samples. Differences between leisure and utilitarian walking as well as between frequent and less frequent walkers were identified for the micro-scale built-environment factors.Interestingly, the importance of streetscape factors in the descriptive results is far higher for leisure walking compared to utilitarian walking.Respondents' answers match with the criteria as suggested by Gehl (2010).Almost all respondents rank "protection" very high in terms of safety and security, followed by "comfort" with slightly lower importance for the variables space and surface quality, and by "delight" ("amenities") with variables such as attractive buildings, greenery or first floor usages.The results look very different for utilitarian walking, distance is key here with 95% of respondents ranking this variable as relevant.Further variables follow with substantially lower levels of agreement.The logistic regression models show the differences between frequent and less frequent walkers in their assessment of the importance of all the different built-environment factors at the micro-scale.Significant differences between these two groups were identified mainly for the "Delight"-variables, this is attractive buildings and greenery for utilitarian walking in model 2, and space availability and street furniture for leisure walking in model 1.
The findings confirm the dominance of urban planning and of the D-variables at the meso-scale for utilitarian walking.Short distances to relevant destinations are the most important factor in our sample, all other variables are assessed substantially less important in the descriptive statistics.At the same time, the findings confirm the volatility of leisure walking.Respondents are far more demanding for these trips.More built-environment variables are assessed to be important and importance levels overall are substantially higher than for utilitarian walking.We need to invite people; we need to provide not only safe but also comfortable and attractive facilities before they come for leisure walking and street activities.The initial hypothesis that frequent walkers are more sensitive to their environment can also be maintained, they consistently assign higher importance to the "delight"-factors compared to less frequent walkers.This group should be our yardstick for urban street design with the final goal to increase the number of frequent walkers and to provide facilities that meet the needs of this most demanding and sensitive person group in our sample.The variables on public toilets (for leisure walking) and low traffic (for utilitarian walking) are also significant in some models with negative coefficients.This somewhat non-intuitive finding can be explained by dependencies with age for public toilets and with possibly actual lower sensitivities of frequent walkers for traffic volumes.In addition, 30% of those who agree on low traffic on the way of the supermarket are living in areas with lower density and less diversity.It is common sense, that in these types of areas traffic is low because of the design of the street network (e.g.cul-de-sac, no grid pattern), and because of (more likely) mono-functional land use (residential area).

Conclusions and outlook to further research
This study confirms and complements findings from the literature; it provides more detailed results at streetscape level than previous resident surveys on built-environment walking determinants; it is less detailed than previous street-level investigations with variables such as Transparency of adjacent usages or Imageability of streetscape (see e.g.Ewing et al., 2016).This study contributes to the discussion on differences in the relevance of built-environment factors for specific travel purposes and contexts.Existing studies that are based on the IPAQ-questionnaire on physical activity already distinguish between walking for transport and for recreation but they do not specifically ask respondents to report their preferences separately for the different contexts.Our context-specific questions at the micro-level of streetscapes allow to disentangle the different mechanisms that apply for utilitarian versus leisure walking with the latter being far more volatile and responsive to streetscape characteristics.In addition, this study distinguishes for the first time between frequent and less frequent walkers.It demonstrates differences in perceptions and preferences between these two groups which should be considered in future research and also in street design practice which might preferably focus on the most demanding person group of frequent walkers.The newly developed questions for characterizing the built environment at the meso-scale and for investigating the relevance of micro-scale factors proved to be suitable, they can be used in future surveys as the basis for prioritizing efforts from promoting walking but also for monitoring purposes.
The representative sample for German cities with at least 100,000 residents is a strength and limitation of this study, future similar investigations in other countries are encouraged for demonstrating the transferability of our findings.Similar representative investigations with the focus on smaller cities could also be insightful for urban and transport planners.The questions on the importance of microscale streetscape factors for leisure or utilitarian walking might also contain respondents' perceptions on whether these factors are actually present on their walking trips.The separation of the perceived existence and the perceived importance in future studies might help to better disentangle these two effects.

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

Figure 1 .
Figure 1.Procedure and composition of the new variable "area type by predominant building type".

Figure 2 .
Figure 2. Comparison of characteristics of the two models.

Figure 5 .
Figure 5. Importance of built-environment characteristics for leisure walking.

Figure 6 .
Figure 6.Motivational built-environment factors for utilitarian walking to the supermarket.

Table 1 .
Sample characteristics in this study in comparison to "Mobility in Cities -SrV 2018".
Accessibility Accessibility of 5 out of 5 Facilities of Daily Needs � in a 5-Minute Walk 12 % No information Less Than 5 Facilities of Daily Needs in a 5-Minute Walk Accessible 88 % No information Accessibility of at Least One Public Transport Mode (Bus or Tramway) in a 5-Minute Walk 81 % 77 % No Public Transport Mode Accessible (Bus or Tramway) in a 5-Minute Walk 19 % 23 % � food/supermarket, bakery, drugstore, pharmacy, parcel counter.

Table 2 .
Structure of independent variables and reference to D-variables.

Table 3 .
Model 1 -determinants of walking frequency for leisure walking.

Table 4 .
Model 2 -determinants of walking frequency for utilitarian walking.