Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
Power calculations were conducted as follows. Simulations were performed employing real sales data from January 2014 to August 2019. In these simulations, we assumed that extended opening hours were implemented by trade districts from September 1st 2018, December 1st 2018, and March 1st 2019, thus resembling the dates of the planned experiment, only two years before. For each outlet, the predicted increase in sales was set to be a uniformly distributed rate over the interval (0, 0.3) of mean hourly sales on Saturday afternoons in 2018; and this effect was further weighted by a relative measure of increasing sales on Saturday afternoons as compared to sales in the morning hours. The simulated increase in sales on Saturdays, given a one hour increase in trading hours, was then added to monthly sales data on which we ran regressions with standard errors adjusted for clustering. In the regression models, we also accounted for seasonal variation. With 100 simulations of different block randomisations and random effect sizes, we were unable to produce parameter estimates for wine (mean β=0.030) and spirits (mean β=0.025) which were not statistically significant at p< 0.05. However, the effect on beer sales (mean β=0.018) was mostly statistically insignificant.