Of the 10,770 road segments rated, 7,660 (71.1%) were in urban areas and 3,110 (28.9%) in rural areas. After weighting for road segment length, this corresponded to 61.3% and 38.7%, respectively. Urban segments were generally shorter than rural segments. Half of the urban road segments were less than 134 m long (interquartile range: 92–212 m) and 5% were longer than 420 m. In contrast, in rural areas the median road segment length was 173.5 m (interquartile range: 102–316 m) and 5% were longer than 849 m.
The frequency and weighted percent (weighted for road length) for the GIS and audit items overall and by urbanicity for the entire audit sample (n = 10,770) are shown elsewhere [see Additional file 2
]. The numbering in the table corresponds to the items used from the PIN3 Neighborhood Audit instrument [see Additional file 2
]. Compared to urban road segments, rural road segments were more often cul-de-sacs, longer in length (> = 240 m), and had higher speed limits, but less often had 3- or 4-way intersections. More rural road segments (50.7%) had no visible security warning signs on the road segment as compared to urban segments (34.3%). Nineteen percent of rural road segments included agricultural land, compared to only 0.8% on urban road segments. Urban road segments had 3 times more segments with sidewalks than rural road segments (38.4% and 11.6% respectively). Any road oriented lighting along the street was much more common on urban segments (71.2%) as compared to rural segments (23.1%).
Exploratory Factor Analysis on A Priori Constructs
Several items on the PIN3 neighborhood audit were not considered in our factor analysis. We did not consider the following variables due to their nominal coding and in some cases narrow distribution: type of housing (#3), overall condition of most buildings (#16), condition of vacant land (#20), park condition (#25), type of litter (#28), bicycle parking (#43_8), and billboards (#43_16). We used a GIS-derived speed limit measure, as described earlier, rather than our audit assessed measure (#40) since many segments did not have a speed limit sign posted. We also used the item on the number of residential units (#2) only as an indicator of whether the rest of the section on residential housing was answered. Lastly, the subjective assessment (#1) was an indicator of the rater's assessment and was not intended to represent any of the a priori constructs and therefore not used.
Standardized Cronbach's alphas are presented in Table for all a priori constructs. Physical incivilities had the highest internal consistency and its consistency was similar among urban and rural segments (alpha = 0.57 and 0.59, respectively). All other constructs had lower internal consistency (range 0.11 to 0.44). Territoriality and aesthetic views had higher consistency among rural segments, whereas social spaces and destinations had higher consistency among urban segments. Safety, aesthetic streetscape, and functionality had similar internal consistency by urbanicity.
Standardized Cronbach's alpha by urbanicity and multi-group one-factor analysis§ for a priori constructs, exploratory sample (n = 6,388).
Table also presents the invariance test (e.g., same factor loadings and thresholds or not by urbanicity) for 1-factor models and their goodness-of-fit for non-invariant models. For all a priori constructs, except for aesthetic views, the invariance test was highly significant indicating that the measures were not the same between urban and rural segments. However, the variability of the aesthetic views and destinations constructs were not significantly different from zero, and hence these constructs were not useful. Further, non-invariant models had poor fit except for physical incivilities and, hence this data did not support these hypothesized models for the remaining constructs. The second-order confirmatory factor model for functionality with four elements (walking/bicycling surface, streets, traffic, permeability) did not converge due to negative factor variances for traffic and permeability, and hence we simplified functionality to one factor explaining all items simultaneously. We also explored the functionality construct analyzing the four components separately (e.g., walking/bicycling surface, streets, traffic, permeability). Internal consistencies were low and factors fit poorly (data not shown).
Table presents the standardized and unstandardized factor loadings for the partial-invariant model (some factor loadings and thresholds constrained to be the same) of physical incivilities, which was the only construct that had an improved model fit compared to the non-invariant model. In rural road segments, litter was the item that was best explained by the physical incivilities factor (R2 = 0.47) and general condition of public spaces was the least explained (R2 = 0.19). In urban road segments, the variability of physical incivilities explained by overall condition of units, grounds, and public spaces was much higher. Overall condition of most residential units and resident-kept grounds were best explained (R2 = 0.71 and 0.74, respectively). The loading for visible dogs was not significantly different from zero in urban segments. Overall, the variability of physical incivilities in urban segments was twice that in rural segments.
Unstandardized and standardized factor loadings for physical incivilities by urbanicity, using the exploratory sample (n = 6,388)
Exploratory and Confirmatory Factor Analysis on New Constructs
Given that not all a priori constructs held, we decided to conduct an EFA with all audit and GIS variables to derive, by urbanicity, orthogonal constructs suggested by the data. This was done using the exploratory sample (n = 4,553 urban and n = 1,846 rural segments). According to the Scree plot criterion, four factors emerged for rural road segments and three factors for urban road segments. However, we selected a four-factor solution for urban segments (rather than a 3-factor solution as indicated from the Scree plots), since from visual inspection of the factor loadings for decoration, the fourth factor was similar to the one identified for rural segments. Table presents the exploratory factor loadings ≥ 0.4 for the 4-factor exploratory solution for urban and rural road segments.
Exploratory and confirmatory factor loadings† for four-factor models by urbanicity, for the exploratory sample (n = 6,388)
One factor loaded very high on higher speed limit, several lanes to cross, and pavement markings. It also loaded on 3- and 4-way intersections, paved road, presence of commercial uses, and presence of a shoulder or bike lane. We called this factor "arterial or thoroughfare". Factor loadings were in the same direction for urban and rural segments, although of different magnitude. Three items (e.g., high speed, bus facilities, and pavement markings) were excluded from the EFA for rural segments due to sparse cell sizes.
A second factor loaded high on visible children and adults, presence of a neighborhood park or playground, neighborhood entrance signs, lighting oriented for pedestrians, and control devices oriented for cars. We called this factor "walkable neighborhood".
A third factor identified was very similar to the a priori physical incivilities construct. It loaded on poor condition of residential units, poor condition of grounds, abandoned units, presence of litter, presence of dogs, no trespassing signs, and absence of pedestrian-oriented lighting. This factor was called "physical incivilities".
A fourth factor only loaded on presence of porches, decoration, and border. We called this factor "decoration".
Almost all items loaded high (≥ 0.4) on just one factor, except for sidewalk buffer (#31), sidewalk condition (#30/32), pedestrian oriented lighting (#36), and presence of pedestrian yield signs (#43_5), which each loaded on two factors. There were 15 items that did not load on any factor and hence, were excluded for the CFA: short segment (GIS), steep segment (GIS), type of front yard (#6), security warning sign (#10), industrial land (#13), agricultural land (#14), home business (#18), vacant or underdeveloped land (#19), graffiti (#29), footpath (#33), trees (#35), road oriented lighting (#36), neighborhood crime watch (#43_13), beware of dog or invisible fence signs (#43_15), and signs for cars regarding bike/pedestrian (combined index from #43_2, #43_6, and #43_7)).
The Cronbach's alpha and goodness-of-fit for the 1-factor models are presented in Table . The internal validity for these a posteriori constructs was high (range 0.43 to 0.73). The fit was acceptable according to the RMSEA, but according to the CFI and TLI the fit was slightly below the recommended cut-off values for some factors, but much higher compared to the a priori constructs. The models allowed for the following correlated errors: cul-de-sac and speed, visible children and adults, and buffer and sidewalk. The 4-factor model by urbanicity had worse fit than the separate 1-factor models, but allowed testing whether the factors were correlated after constraining some of the factor loadings to zero. The walkable neighborhood factor was negatively correlated with physical incivilities for both urban and rural segments (r = -0.12 and -0.55, respectively), and with the arterial/thoroughfare factor only in rural segments (r = -0.26). The parameter estimates for the hypothesized and derived constructs using the validation sample were not significantly different from those using the exploratory sample (Table ), providing evidence of validity in our population.
Standardized Cronbach's alpha and confirmatory factor analysis for constructs derived from EFA and Pearson correlation coefficients, exploratory sample (n = 6,388)
Goodness-of-fit for multiple-group confirmatory factor analysis§: exploratory vs. validation samples†
Lastly, we explored 2-week test-retest reliability using Spearman correlation coefficients for 464 road segments (Table ). The arterial or thoroughfare factor displayed near perfect reliability in both urban and rural segments (r = 0.96 for both). Both the physical incivilities factor (r = 0.77 to 0.78) and the walkable neighborhood factor (r = 0.79 to 0.82) had substantial to near perfect test-retest reliability in both urban and rural segments. The decoration factor displayed moderate reliability in urban segments (r = 0.50; 95% CI: 0.38–0.60) and lower reliability in rural segments (r = 0.39; 95% CI: 0.25–0.52).
Two-week test-retest reliability using Spearman correlation coefficients with 95% confidence intervals (CI)