” Response options were strongly disagree (1) to strongly agree (4). For comparability to previous studies, these items were also retained in the original subscales. Self-reported weight in kilograms and height in meters were used to calculate BMI = weight/height2. Region (Seattle/King County or Maryland/Washington, DC region), gender, age, education level, ethnicity, marital status, and number of vehicles per adult in

the household were included as covariates. SPSS version 17.0 was used for analyses. Because the study design involved recruitment of participants clustered within 32 neighborhoods pre-selected to fall within the quadrants representing high/low-walkability PD332991 by high/low-income, intraclass correlations (ICCs) reflecting any covariation among participants clustered within the same neighborhoods were computed for the bicycling frequency measures. The ICCs were very near or equal

to zero: current biking frequency, ICC = 0.011; find more biking frequency if safer from cars, ICC = 0.000; and difference score (i.e., difference between current biking frequency and frequency if safer from cars), ICC = 0.009. Because the ICCs were zero or almost zero, negligible random clustering effects were expected, and traditional regression procedures were used. All variables were treated as continuous/ordinal except bicycle ownership (yes/no) and five demographic variables: region, sex, ethnicity (White non-Hispanic, vs. others), education (at least a college degree, vs. less than a college degree), and marital status (married or cohabiting vs. other). The

first Methisazone group of analyses examined all environmental and demographic variables by bike ownership. Binary logistic regression was used to identify significant associations with bike ownership in separate models for each potential correlate. The second set of analyses used linear regression procedures to examine bivariate correlates of the bicycling frequency outcomes: (a) frequency of biking (bike owners only) and (b) self-projected change (difference score) in bicycling frequency if participants thought riding was safe from cars. Although these outcome variables were somewhat skewed (+ 2.0 and + 1.0, respectively), these skewness values fall within ranges of commonly used rules of thumb, especially when using ANOVA/regression procedures that are considered robust to non-normality (van Belle, 2002, p. 10). Thus, it was judged preferable to retain the original units (e.g., 5-point ordinal categories) rather than transform the ordinal categories to log-units. Each environmental and demographic correlate was examined in separate analyses. The third group of analyses investigated whether variables significant (p < .10) in bivariate analyses remained significant (p ≤ .05) in multivariable regression models.