Statistical Methods
Prespecify Trimming for a Robust Independent-Group Comparison
Define the task
A trimmed-mean comparison changes the target from an ordinary mean to a location summary that omits prespecified fractions of each tail. State why that target is relevant before seeing results. This differs from deleting individually inconvenient values after inspection. Use independent groups and retain the raw observations, exclusion rules and chosen trimming fraction. If groups are paired, clustered or weighted, an ordinary independent-sample trimmed test does not automatically handle that design.
Carry out the check
The documented trimmed t-test uses the floor of fraction×sample size to determine how many observations to remove from each tail. Its uncertainty calculation uses winsorized variation and the retained effective sample counts, not the ordinary variance of just the remaining observations. Record the variance assumption and alternative hypothesis explicitly. Inspect how integer trimming differs across group sizes. A robust method does not justify ignoring data errors or assuming the same effect under every distribution shape.
Worked hypothetical example
For hypothetical values (0,1,2,3,100) with trimming fraction 0.2, one value is removed from each tail and the trimmed mean is 2. The ordinary mean is 21.2. Winsorized values are (1,1,2,3,3), which support a different variance calculation than simply applying an ordinary t-test to (1,2,3). This example checks preprocessing, not a p-value. Report the intended estimand, retained counts and appropriate uncertainty method together.
Method checked 18 November 2026 against Primary method documentation: ttest_ind.