Rstudio version 4

All statistical analyses were conducted in Graph Pad Prism Version 9 (RRID:SCR_002798) or R studio Version 4.0.2 (RRID:SCR_001905). Data were assessed for normal distribution using a Shapiro–Wilk test. Data were considered normally distributed if the P-value was > 0.05. If the data were not normally distributed, a Kruskal–Wallis test with a Dunn's multiple comparison test (between group) was conducted for multiple groups, or a Mann–Whitney test comparing tanks between two groups. The level of significance was set to P < 0.05.

Demographic characteristics of survey participants were analyzed using RStudio version 4.0.2 (RStudio, PBC, Boston, MA). Viral testing results obtained throughout the semester were provided by university health systems or from WEDSS, and index values from serology analyses were provided by WSLH. Semester start and semester end serology results and viral testing data were analyzed using the ggplot2 and ggalluvial packages in RStudio. Serology status, trends from semester start to semester end, and the receiver operating characteristic (ROC) curve were analyzed using GraphPad Prism v 9.0 (GraphPad Software, San Diego, CA). Welch’s unpaired T tests were used to analyze differences in seropositive students at semester start and end. Mann Whitney U tests were used to compare the number of days between positive viral test date and end of semester serology collection date. Fisher’s exact tests were used to evaluate differences in symptom reporting between symptomatic and asymptomatic students with recorded positive viral tests. p-values < 0.05 were considered significant.

Descriptive statistics (mean and median levels with 95% confidence intervals) were calculated by food categories and menu labelling regulatory status for serving size, calories, caloric density and nutrients of public health concern (i.e., saturated fat, sodium and sugar) per serving and per 100 g where data were available. To account for the differences between restaurant venue types, restaurants were characterized as fast-food restaurants (FFR, n = 95) if table service was not available and sit-down restaurants (SDR, n = 46) if table service was available. Quantile regression was used to calculate the estimated difference in median serving size, calories, caloric density and nutrients of public health concern per serving and per 100 g between regulated and unregulated restaurants, adjusting for restaurant type. Bonferroni corrections were used to correct the family-wise error rate and the significant difference was set at p < 0.05. Analyses were performed using R Studio, version 4.0.2 (Boston, MA, USA).

The scores obtained for each visual field (frontal left, frontal right, lateral left and lateral right) were compared between scoring methods (Visual Field Analysis vs manual coding) using correlation tests (Pearson’s test) overall and per individual. An estimation of the number of frames needed to achieve a significant power for the correlation analysis was run prior to the study, assuming an r value of 0.9 (given the high accuracy of the tracking methodologies). According to this estimation, it would be sufficient to code only seven frames to achieve a statistically significant correlation. However, we deemed this to be insufficient to provide a reliable validation of our software. Thus, in order to provide the most precise estimation possible of the accuracy of our method, we decided to code all the frames available, greatly surpassing the minimum number required to achieve sufficient statistical power. As some videos were better tracked by DeepLabCut than others, we report the reliability of our program in relation to the tracking accuracy (measured as the percentage of frames that received identical scoring in the manual and automated scoring). The statistical tests were performed using RStudio version 4.0.2 (RStudio Team, 2020 ).