Very unsurprising for me. You are measuring your ability to solve algorithm puzzles. Most engineers don't actually do many algorithm puzzles in day-to-day work, especially the types of algorithms that interviews tend to focus on like sorting and dynamic programming. So "years of experience" is not measuring experience in what you're actually being tested on. On the other hand, you do exactly those types of things in many CS classes, and in Coursera classes, algorithms are exactly what you practice. So it makes sense it correlates.

Top company is a predictor for the obvious reason - it's selection bias for people who already passed those interviews at the company. You're not good at the interview because you worked at the company, you work for the company because you're good at the interview.

Master's degrees seem like largely for international students needing visas, career switchers, etc so not surprised they are not a strong predictor. And if anything the course material moves past the intro data structures stuff the whiteboard interviews tend to test.

The only huge surprise for me here is that Coursera is a stronger predictor than top company and top school. I would have predicted top company > top school > Coursera.

The post that I would be much more interested in is correlating performance reviews to interview performance. That gets suggested as a possible future post.