Statistical analysis

We characterized access measures, sociodemographic and health characteristics, and park use data using descriptive statistics for the total sample and by park use. Due to sample size, we did not make comparisons between neighborhoods. We performed bivariate analyses using a chi-square test to determine whether each measure of access varied according to whether a participant reported any use of Druid Hill Park over the last 30 days versus none. For multivariable analysis, we used a negative binomial regression model to test whether each measure of park access was associated with the number of times participants reported using the park over this period. We used negative binomial regression because the primary outcome (number of park visits over the past 30 days) was a count variable and to account for over-dispersion in our data. Because park use may vary according to time of year, all models adjusted for month of survey administration. We also adjusted for neighborhood. Separate models were run for each access measure. We then ran a single model that tested the association between access measures and park use. This model simultaneously included all access measures that were statistically significantly associated with park use at the 0.05 level in the preceding model and additionally included sociodemographic characteristics (age, education, car-ownership, children in the household, and physical activity) that were significantly associated with park use. All statistical analyses were performed using Stata 15.1 (Statacorp LP, College Station, TX).