,
4 tweets,
1 min read
Read on Twitter
One of the challenging things about teaching Intro Stats and, specifically, linear regression, is trying to get across the idea that two variables need not be perfectly correlated for their relationship to be useful or meaningful.
IMO, a lot of beginning statistics students see a scatterplot with a fair amount of spread, then see a significant p-value and get confused: how can there be evidence of a "linear association" between the variables? The correlation is only 0.3!
But if one variable explains 10% of the variation in another variable, that information can be really important! It doesn't mean that I can predict Y from X by itself, but in biomedical research, that need not be the case for a relationship to be useful.
(this story doesn't really have an ending, it's just kind of a musing about teaching linear regression to beginning stat students)
