Among the 25 stores that sold at least 3 fruit items (1 supercenter, 12 supermarkets, 10 grocery and 2 convenience), apples, oranges, avocado and banana were the most commonly found (Table provides both the number of stores with price data by item along with the proportion of stores doing so). On a per weight basis, bananas were the cheapest fruit type, while berries were the most expensive. Among the 32 stores that sold at least 4 vegetable items (1 supercenter, 12 supermarkets, 11 grocery stores and 8 convenience stores), carrots, lettuce, onions, potatoes and tomatoes were the most common. Potatoes were the least expensive vegetable type, while green beans were the most expensive.
Price Availability, Conditional Mean Price and Consumption Shares of Fresh Fruit and Vegetables Types
There were 25 stores that sold at least 3 types of fruit; however, 22 stores posted the requisite information--the prices of apples, oranges and bananas--to calculate the basic variety price index (1 supercenter, 12 supermarkets, 8 grocery stores, and 1 convenience store). Similarly, of the 32 stores that sold at least four types of vegetables, 23 posted the requisite price information to calculate the basic vegetable price index (1 supercenter, 12 supermarkets, 8 grocery stores and 2 convenience stores). The mean cost of meeting the USDA recommended level of fruit consumption from a high variety basket of fruit types in our sample of stores was just over $27.50 per week (Table ). In contrast, relying on only the three most common fruits lowered the weekly expense to just under $17.25 per week, a reduction of 37.6%. The effect of moving from a high variety to a low variety basket was much less when considering vegetable consumption: a 4.3% decline from $29.23 to $27.97 per week.
Summary statistics of produce availability, store types and CBG characteristics
Although there are 101 CBG in the six counties of the rural Brazos Valley region, the overwhelming majority either does not have a food store located within their boundaries or the only food stores are convenience stores that do not sell fresh fruits or vegetables. Nevertheless, the 23 food stores with sufficient price information to be included in the subsequent analysis are located across diverse neighborhoods. The median household income at the level of the CBG was highly variable, ranging from $14,400 to $50,500. On average, African American comprised 21.2% of the CBG population, while Hispanics accounted for 16.5%. There were also several minority-majority CBG in which more than 50% of the population was either African American or Hispanic. Indeed, the CBG were these 23 food stores are located (Table ) appear slightly more diverse than the region in general.
Multivariate regression analysis of the store-level price indices on CBG characteristics (Table ) revealed that stores in CBG with higher median household incomes tended to charge more for fresh fruits and vegetables, holding racial composition equal. The coefficient on the logarithm of median household income was positive and statistically significant (P < 0.05) for both vegetable price indices. For example, every ten percent increase in the CBG median household income was associated with an increase of $1.39 (P < 0.05) in the high variety index and $1.61 (P < 0.05) in the basic index. The coefficient estimate was also positive and was statistically significant at the 10% level for the high variety fruit index, implying that a 10% increase in median income was associated with a $1.02 (P = 0.089) increase.
Coefficient estimates from OLS multivariate regression models
In addition, stores in CBG with greater proportions of Black residents tended the charge more for fresh produce, holding income equal. The coefficient estimates on the proportion of Black residents were positive for all four indices and were statistically significant at the 10% level for two. Holding everything else constant, a 10 percentage point increase in the proportion of African American residents was associated with an increase of $1.49 (P = 0.085) in the high variety fruit index and $2.30 (P = 0.062) in the basic vegetable index. In contrast, the coefficient on the proportion of Hispanic residents was typically close to zero and never approached significance (the smallest P-value is 0.621). Finally, the mean VIF was at or below 2.0 in each regression and the maximum was never larger than 2.5, indicating that collinearity between explanatory variables is not problematic.
The results using multivariate analysis differ quite substantially from the univariate analysis (Table ). For example, stores in higher income areas do not charge more for the basket of fresh fruits and vegetables when differences in racial composition are not simultaneously controlled. Similarly, stores in neighborhoods with higher proportions of African American resident do not tend to charge more for fresh fruits and vegetables. The difference between the multivariate and univariate analyses is explained by the negative relationship between the proportion of African American residents in a neighborhood and the median household income.
Coefficient estimates from OLS univariate regression models
We also repeated our multivariate regression analysis replacing price indices with the prices (imputed price if missing) of individual fruit and vegetable items. Among fruits (Table ), the price of avocado was strongly related to both income and the racial/ethnic composition of the neighborhood, whereas the remaining fruit types were very weakly associated with either. This result likely explains why the R2 for the high variety index reported in Table is much larger than that for the basic index, which does not incorporate avocado. It is worth noting that this finding cannot be explained by the imputation process, as the price of avocado did not need to be imputed at any of the 21 stores used in the analysis. Although the relationship between household income and affordability of fruit (and between racial/ethnic composition and affordability) is principally the results of one item, avocado is nonetheless an important ingredient in local Tex-Mex cuisine and accounts for a non-trivial proportion of total fruit consumption, 9.1%.
Association between neighborhood characteristics and the price of individual fresh fruit items1
Among vegetables (Table ), green beans actually exhibit a negative relationship to both income and the proportion of Black or Hispanic residents, but more common items like onions, lettuce, potatoes and tomatoes display qualitatively similar results to those presented in Table for the price indices. Although the coefficient estimates in these regressions are not statistically significant, estimating a statistically significant association between a particular neighborhood characteristic and an index of prices when the individual price relationships are not statistically significant should not be troubling. Since stores that sell high prices for one type likely charge a higher price for similar items, i.e. a positive covariance between prices, the sum of variances will be less than the variance of the sum. Contrary to being troubling, this is evidence that the intuition underlying our imputation strategy was reasonable. It also suggests that studies of affordability should report results using both aggregate and disaggregated measures of affordability, since aggregate prices can obscure important behavior among individual items, while results for individual goods may not adequately account for the positive covariance of prices within a given store.
Association between neighborhood characteristics and the price of individual fresh vegetable items1
Finally, we repeated the analysis using mean imputation (not reported). In this particular application, the choice of imputation method did not affect coefficient estimates, but it may nevertheless be influential in other settings.