Statistical analysis of the NYT blog about font credibility

>I suppose we could keep running t-tests where the null hypothesis changes from “Baskerville is the same as Comic Sans” to “Baskerville is 0.5 points better than Comic Sans” to “Baskerville is 1 point better than Comic Sans,” and get a range of probabilities. But a probability distribution like that seems more like a Bayesian outcome; I’ve never heard of such a thing in frequentist statistics.

That would essentially be a more detailed version of a confidence interval for the actual difference - and since the distribution of the "actual value" is almost certainly normal, it wouldn't really tell the reader anything more than a regular one.

Article author & submitter here: I wrote this post as self-assigned "homework" to push my (very amateur) statistics knowledge. It responds to an article that made HN last week, saying that font choice affects the believability of a claim. I've love to learn more about how to do this kind of analysis, so feel free to tear my post apart. :-)