The sample consisted of restaurants in the eight EHS-Net jurisdictions. Restaurants were defined as establishments that prepare and serve food or beverages to customers but are not institutions, mobile food units, temporary food stands, supermarkets, restaurants in supermarkets, or caterers. In each EHS-Net jurisdiction, health department staff chose a geographic area in which to recruit restaurants for study participation, based on a reasonable travel distance (mean = 88.1 min, range = 30 min–4 h). Staff called or visited randomly selected restaurants to recruit for the study and scheduled a data collection visit with managers who agreed to participate. Data were collected from March 2018 through March 2019.

Each EHS-Net jurisdiction’s institutional review board approved the study protocol.

All data collectors (EHS-Net health department staff) participated in training focused on standardizing data collection methods. At each restaurant, data collectors obtained verbal informed consent from and conducted a standardized interview with, the manager (person with authority over the restaurant). They also recruited on-site food workers (people who prepare food in the restaurant) for study participation. They obtained verbal consent from participating workers and administered a written survey or provided a link to the survey online.

The worker survey included questions on worker characteristics (Table 1), including food safety knowledge (seven multiple choice items) (Box 1). The survey also included 28 items that assessed workers’ perceptions of their restaurant’s food safety culture (using a Likert-scale [1-strongly disagree to 5-strongly agree]) (Box 2). The online worker survey was available in both English and Spanish and was administered using SurveyMonkey, a user-friendly and widely used online survey platform (Momentive, San Mateo, CA, USA). See Kramer et al. (2023) for details on the development of the food safety culture measure. An identifier linked restaurant, manager, and worker data, but no data identified individual restaurants, managers, or workers. The manager interviews collected data on restaurant characteristics (e.g., seating capacity) (Tables 2 and 3).

First, we conducted an exploratory factor analysis to determine if the food safety culture survey items measured multiple factors. We retained survey items in our analysis that loaded onto unique common factors, had a primary factor loading of ≥0.4, and did not load onto another factor at ≥0.3. Sixteen items composing four unique factors were retained in the model. We assessed the reliability of these four factors using Cronbach’s alpha (alpha coefficient [α] of 0.65 or higher was considered acceptable). Study staff then reviewed the items and labeled the factors leadership: (6 items, α = 0.88); management commitment (3 items, α = 0.73); employee commitment (4 items, α = 0.87); and resource availability (3 items, α = 0.72). We then created a composite measure for each factor by averaging the item scores in the factor (negatively phrased questions were recoded so that higher scores would equate to positive agreement). These analyses were first reported by Kramer et al. (2023); additional details can be found there.

We calculated frequencies for the worker and restaurant characteristics obtained from the worker survey and manager interview. Then, we conducted hierarchical linear modeling (HLM) to determine how the four food safety culture constructs were related to worker and restaurant characteristics. We used two-level HLM to estimate the association of worker-level (e.g., years in food service) and restaurant-level (e.g., seating capacity) characteristics with the four food safety culture constructs, nesting worker-level data within restaurant-level data. This methodology allowed us to associate multiple workers within individual restaurants. We also calculated a variance partition coefficient (VPC), or intraclass correlation coefficient, to assess whether workers within the same restaurant were more likely to provide similar responses than those in different restaurants. This helped determine if our HLM analyses needed to account for correlations of worker responses within restaurants.

We developed four separate multivariable models, one for each construct. We used tests to determine whether to include each of the 23 characteristics of interest (listed in Tables 1–3) in the models. The likelihood ratio test examines the goodness of fit of two competing models: in our case, a null model and a model containing the characteristic of interest. The higher the likelihood ratio tests (LRT) (>1), the stronger the evidence that the characteristic is related to the construct. Models that contained characteristics that were statistically significant were then included in an iterative approach to determine the best combination of characteristics to include in each of the four final models based on theoretical associations and statistical fit. We developed the four final models of the best fit of statistically significant characteristics through an iterative approach. This process was guided by both theoretical considerations and statistical fit. We tested for interactions among all significant characteristics and included significant interactions in the models. We used a bootstrapping method to randomly resample our dataset 2000 times and create 2000 simulated samples (Roberts and Fan, 2004). This dataset was used to generate robust estimates of parameters and confidence intervals. We used SAS version 9.4 (SAS Institute Inc, Cary, NC, USA) to analyze the data.

We contacted 990 managers of restaurants that met the requirements for study participation; 352 agreed to participate. We received 567 food worker survey responses from 321 of those restaurants. An average of 1.8 workers responded per restaurant (standard deviation [SD] = 1.4). Denominators vary for some analyses because of missing data.

The majority of workers said they had 1–5 years of food service experience (41.7%). Most (77%) said they had received food safety training at their current restaurant. The mean worker food safety knowledge score was 70.8% (out of 100%). According to manager interview data, the majority of restaurants were independently owned (57.9%) and did not provide full service (62.3%). Tables 1–3 contain more data on worker and restaurant characteristics.

The four food safety culture construct scores were relatively high (composite scores >3 on a 1–5 scale). The resource availability score was highest (mean = 4.7, SD = 0.7, n = 557), followed by employee commitment (mean = 4.5, SD = 0.6, n = 557), leadership (mean = 4.3, SD = 0.70, n = 557), and management commitment (mean = 3.9, SD = 1.2, n = 557) (See Box 2). A detailed discussion of these data can be found in Kramer et al. (2023).

Most (66%) variation in the leadership score was in workers within restaurants; 34% was in restaurants (VPC = 0.34), indicating that worker beliefs about food safety leadership vary across workers and restaurants. LRT indicated that all the characteristics of interest were significantly related to the leadership score and were thus considered for inclusion in the multivariable model, except the worker food safety knowledge score (Table 4). The final multivariable model for leadership included the characteristics of whether the surveyed worker had received food safety training at their restaurant, employee turnover (number of food employees replaced monthly), and percentage of food safety policies that were written (Table 5). Leadership scores were higher for workers who received food safety training compared to those who did not. Leadership scores increased as the percentage of policies that were written increased. However, leadership scores decreased as employee turnover increased.

Most (65%) variation in the management commitment score was in workers within restaurants; 35% was in restaurants (VPC = 0.35), indicating that worker beliefs about management commitment to food safety vary across workers and restaurants. LRT indicated that all characteristics were significantly related to the management commitment score; thus, they were considered for inclusion in the multivariable model (Table 4). The final multivariable model for management commitment included the characteristics of worker food safety knowledge score, number of certified food protection managers in the restaurant, number of methods used for employee training on policies, and number of methods used for verifying that employees follow policies (Table 5). Management commitment scores increased as worker food safety knowledge scores increased and as the number of methods for training and verifying policies increased. Management commitment scores decreased as the number of certified food protection managers increased.

The final model included two interactions. The first interaction, between the number of certified food protection managers and worker food safety knowledge score, indicated that management commitment scores were highest when there were fewer certified food protection managers and workers who were more knowledgeable in food safety. The second interaction, between the number of methods for training employees on food safety policies and the number of methods for verifying that employees follow those policies, indicated that management commitment increased as the number of employee training methods increased and, independently, as the number of methods for verifying that policies were followed increased. However, management commitment decreased when a high number of both types of methods were used simultaneously.

Most (78%) variation in the employee commitment score was in workers within restaurants; 22% was in restaurants (VPC = 0.22), indicating that worker beliefs about employee commitment to food safety vary across workers and restaurants. LRT indicated that all characteristics were significantly related to the employee commitment score and were thus considered for inclusion in the multivariable model, except the worker food safety knowledge score (Table 4). The final multivariable model for employee commitment included the characteristics of whether the surveyed worker received food safety training at their restaurant, whether manager pay was affected by the restaurant’s food safety performance (e.g., inspection scores), and the interaction of these two characteristics (Table 5). Employee commitment scores were higher when the surveyed worker had received food safety training and lower when manager pay was affected by restaurant food safety performance. The interaction indicated that employee commitment scores were lowest when manager pay was affected by food safety performance and when workers did not receive training at their restaurant. Employee commitment scores were highest when manager pay was not affected by food safety performance and when workers received training at their restaurant.

Most (86%) variation in the resource availability score was in workers within restaurants; 14% was in restaurants (VPC = 0.14), indicating that worker beliefs about food safety resource availability vary across workers and restaurants. LRT indicated that all characteristics were significantly related to the resource availability score; thus, they were considered for inclusion in the multivariable model (Table 4). The final multivariable model for resource availability included the worker-level characteristics of food safety training at the restaurant, having a high school degree, and food safety knowledge scores, and the restaurant-level characteristics of the percentage of food safety policies that were written and employee turnover ratio (number of food employees replaced monthly divided by number of food employees). (Table 5). The final model also included the interaction of worker food safety training and food safety knowledge scores. Resource availability scores were higher if the surveyed worker had received food safety training at their restaurant and had a high school degree. Resource availability scores increased with worker food safety knowledge scores, the percentage of policies that were written, and the employee turnover ratio. The interaction indicated that although resource availability scores were higher when workers received training at their restaurant and, independently, as their food safety knowledge scores increased, they were highest when workers received training at their restaurant and were knowledgeable in food safety.