Column 1 + 2: rows represent the results of a different effect size (Cohen’s d) scenario, indicated left. The level of attrition in either treatment group is stated (left to right) as “8 + 8” without attrition and “7 + 8” to “6 + 7” with the total number of missing animals increasing from one to three. Column 1 + 2: probability of positive trials after random loss (first column) or non-random attrition of extremes that are not in favor of the effect (second column) for different effect sizes (rows 1–3). Colors represent the proportion of trials out of 10,000 simulations that are significant (1) independent of attrition (orange) or significant (2) only in the case of attrition (red), non-significant (1) independent of attrition (cyan), or non-significant (2) only in the case of attrition (dark blue). Column 3: ratio of type 1 error rates (falsely accepting the alternative (H₁) hypothesis if there is no true effect, first row) or type 2 error rates (falsely failing to reject the null hypothesis if there is a true effect, second and third row), respectively, for different levels of attrition relative to the rates acquired with the full sample (“8 + 8”). Ratios for random attrition are colored in black, and ratios for non-random attrition are colored in red, in arbitrary units (a.u.). Fourth column: effect size estimated from positive trials only. Mean estimated effect sizes are displayed in black (+) for random attrition and in red (×) for non-random attrition.

Column 1 + 2: rows represent the results of a different effect size (d) scenario, indicated left. The level of attrition in either treatment group is stated as “8 + 8” without attrition and “7 + 8” and “7 + 7” with the number of missing animals increasing from left to right. Probability of positive trials before and after attrition of outliers in the samples that are not in favor (first column) or in favor (second column) of the expected effect. Colors represent the proportion of trials out of 10,000 simulations that are significant (1) independent of attrition (orange) or significant (2) only in the case of attrition (red), non-significant (1) independent of attrition (cyan), or non-significant (2) only in the case of attrition (dark blue). Column 3: ratio of type 1 error rates (falsely accepting the H₁ hypothesis if there is no true effect, first row) or type 2 error rates (falsely failing to reject the null hypothesis if there is a true effect, second and third row), respectively, with increasing attrition relative to the rates acquired with the full sample (“8 + 8”). Ratios for attrition of outliers that are in favor of the effect are colored in black, and ratios for attrition of outlier that are not in favor of the effect are colored in red, in arbitrary units (a.u.).

All experiments were compared using the numbers of animals reported in the methods and results section. If this comparison was impossible, experiments were coded as “Unclear.” If these numbers were identical, experiments were coded as “Matched.” If there was a discrepancy between these numbers, then experiments were coded as “Attrition,” which could either be explained (via information in text or figure legends of the paper) or unexplained.

Boxes represent second to third quartiles, and red lines in the middle are the median. Whiskers represent first and fourth quartiles.