These figures are derived from univariate Linear Probability Models (LPMs) (https://bit.ly/3aKaCeG
) and multivariate LPMs, where we control for users’ gender and age (including a 2nd degree polynomial for age). Results are robust (in terms of sign, magnitude and statistical significance) to using logistic regression (https://bit.ly/3aHN0HI
). LPM results are preferred for ease of interpretation
These data use information provided by over 13,000 Evergreen Life users. As there are more responding users in some areas than in others, the map information is a stronger reflection of the true percentages in some areas than others. The percentage reported is the mean of the corresponding Beta distribution, with the prior pseudo-counts both set to equal 1 (https://bit.ly/3ax2n5I
). The confidence intervals for each proportion are extracted as the upper and lower bounds of the regularised incomplete beta function (https://bit.ly/3byoERT
) , where it takes the value of 0.025 for the lower CI and 0.975 for the upper.
National trend line disclaimer:
These graphs use information provided by over 13,600 Evergreen Life users. It represents the national trend in the number of users that each day have downloaded and used the App and have most recently reported that their drinking habits. Percentages are displayed with an exact, Clopper-Pearson 95% confidence interval (https://bit.ly/2XmFGOP