Hacker News Reader

Sometimes I wonder about the actual utility of the T-test compared to just looking at a pair of boxplots, with jittered points (or some other indication of the number of data points).

If it isn't plainly evident from the boxplots (assuming you've got "enough points") do T-tests alone ever make a truly compelling argument?

View on HN

That would not be exactly scientific. T-test can be calculated independently and verified.

But you still need to choose a significance threshold, which is just an opinion. There's nothing "scientific" about p=0.05; surely God loves p=0.06 almost as much.

What I mean is, if the plots don't clearly make the case, I would not trust p-values alone.

Reply on news.ycombinator.com