You are right on the edge (n=25 per condition) where it’s not as critical, but it’s something I just learned myself so I thought I’d share. Essentially, time data are skewed positively so they’re not on a normal curve (like a nice, symmetrical bell curve). Therefore a standard average, or even a median, tends to be skewed up as well.

The geometric average adds a step where it converts the data to log, averages the data, and then un-logs them. This log transform handles the skew of the data more accurately. Some bright mathematical UX folks at measuringu.com have a nice little article about it: https://measuringu.com/average-times/

Anyway, I really enjoyed your post, and your method was very thorough! Thanks for posting.