It would also be interesting to look at the inverse relationship - how many average stars does a "licensed" repository have vs an "unlicensed" one. I suspect that you're right that repositories with licenses will have more stars, and thus more widely used.

Another interesting metric would be lines of code (licensed vs. unlicensed). This would probably make your argument more obvious... more useful repositories will have many more lines of code than unuseful ones; and thus also more likely to have a license.

I believe that the author may have just re-discovered the concept of the long tail, reproduced in source code repositories.