Street Eats, Safe Eats: Methods
To examine differences between food trucks, carts and other types of food establishments—particularly restaurants—this report relies on inspection data collected from government agencies in Boston, Las Vegas, Los Angeles, Louisville, Miami, Seattle and Washington, D.C. The Institute requested data going back to 2008 or the first year with accessible data that included mobile vendors. Data were collected through part or all of 2012 or, in the cases of Boston and Louisville, through July 2013. In all, the Institute reviewed 263,395 inspection reports across the seven cities. During the inspections, officials count the number of food-safety violations they observe.7 For example, inspectors look for minor things like clean counters and proper labeling, bigger concerns like proper food storage and hand-washing facilities, and serious issues such as sick employees and spoiled foods.
For each city, the Institute calculated the average number of violations per establishment for each category of food service—food trucks, restaurants and so on. These raw numbers are useful, but not sufficient for determining how mobile vendors compare to brick-and-mortar establishments. Other factors, such as variations in traffic or greater frequency of inspections, could be driving any differences. Additionally, any differences in the raw numbers could be simple random chance—it just so happens that during a given period of time when a random group of establishments was inspected, one category of food service received fewer violations—instead of a genuine distinction.
To control for factors that could muddy comparisons and to determine whether the differences between mobile vendors and brick-and-mortar restaurants are genuine or mere random chance, this report relies on two types of statistical analyses. The first, fixed-effects OLS regression, provides the average number of violations for each food-service category compared to mobile vendors. In other words, the first type of analysis estimates how many more or fewer violations restaurants would receive, on average, than mobile vendors, after controlling for various factors.8 The second type of analysis, Poisson regression, provides a rate estimating how many times more or fewer violations each food-service category would receive, on average, compared to mobile vendors.9
When looking at the rate of violations, keep in mind that the average numbers of violations were low for all types of food service in all cities. Thus, some eye-popping comparisons are not as dramatic as they may appear. For example, it may be startling to see the Boston results below (Table 2) suggesting that restaurants received 385 percent more violations than food carts, but food carts averaged just one violation per cart, so 385 percent more is only about four violations per restaurant.
In some cities, the data did not make it possible to distinguish between food trucks and food carts, so they were lumped together in one “mobile vendor” category. In others, trucks and carts are separate categories, so separate analyses compared each of them to restaurants, grocery stores and so on. Further details about the analysis can be found in Appendix A, and Appendix B provides full regression results.10
7 In Las Vegas, Los Angeles, Louisville and Seattle, violations are given demerit values depending on the severity of the violation. For example, a foodborne violation may have a demerit of five whereas a business practice violation may have a demerit of one. In these cities, the sum of the demerits is the number provided by the agencies and is reported here as number of violations.
8 Analyses controlled for when an establishment was inspected—day of the week, month and year—because variations may occur with higher traffic and lower traffic days and with seasonal and yearly fluctuations in demand, weather, foods, pests and other factors. The analyses also controlled for each individual establishment because some businesses may be inspected more often or have consistent issues based on something other than the type of food establishment they are. The analyses for Seattle and Washington, D.C., also controlled for risk categories assigned by the cities. These categories are assigned based on establishments’ methods of food preparation and delivery—pre-packaged versus fresh food, ice cream versus warm lunch entrees and so forth. Analyses controlled for these categories so that an abundance of high-risk, and therefore potentially high-violation, establishments in one category would not skew results.
9 The Poisson regression is commonly used for analyzing count data, which we have here (i.e., counts of violations). However, the results of OLS regression tend to be easier to understand and are included here for ease of interpretation.
10 The full regression output for models in Boston, Miami and Washington, D.C., using the numbers of critical and non-critical violations can be supplied upon request.
|<<Return to Street Eats, Safe Eats main page||Results>>|