100 blog posts is not even a remotely sufficiently large sample size for such an analysis, especially since metrics on from different types of blog posts can cause things to be skewed heavily. You'd need 1,000 minimum from the same type of blog post to reduce the bias.

Even worse, the article asserts that the correlation between # of text/images in an article is the cause of an article's success, which isn't statistically proven from the post.