We calculated means and frequencies for descriptive statistics. We qualitatively examined differences between regions and respondent types. We examined associations between the independent variable of farmers’ market use and dependent variables of (1) fruit/vegetable consumption and (2) BMI, in two separate linear regression models. All analyses were adjusted for age, gender, race, and education level and were stratified by region (KY and NC) and by type of survey (farmers’ market customer intercept survey versus RDD telephone survey). Race was not included in KY farmers’ market customer analyses as nearly all customers (99%) were white. Final models for the two dependent variables (fruit and vegetable consumption and BMI) were based upon maximizing the number of observations included in the models, and maximizing R2. In the adjusted model with fruit and vegetable consumption as the dependent variable, farmers’ market use was examined as the independent variable. In models with BMI as the dependent variable, farmers’ market use was examined as the independent variable of interest, along with fruit and vegetable consumption. Because few respondents in each of the four groups reported their perceived miles to reach the closest farmers’ market, this variable was not included in any of the regression models.
The Pitt County, NC RDD telephone survey data were weighted by the inverse of the probability of selection at the phone number level, and a post-stratification weighting factor that was developed to sequentially adjusted for landline/wireless coverage in NC, and at the county-level household income, and the race and education of the heads of households. The same weighting method was used for the KY RDD analyses, except that no adjustments were made for landline/wireless coverage (since no cellular telephones were sampled) or for race (since there were very few non-white respondents). A pooled regression analysis of the NC and KY RDD data were conducted to assess the differences between the two states. For this purpose, the weights were further adjusted according to the population sizes of the two states. All analyses were conducted using Statistical Analysis Software (version 9.2, SAS Institute Inc, Cary, North Carolina). Data from the RDD telephone surveys were analyzed using survey-specific procedures in SAS.