Active member
- Joined
- Mar 25, 2018
- Messages
- 269
I really like the site approach to developing new features. Thank you!
Re Bayesian ratings and rating scales: I typically rate series about one full unit below the mean rating. On the other hand, I tend not to rate series that I've dropped or that put me off in the first couple of chapters. Perhaps I'm giving the author the benefit of the doubt. After all, in theory it could get better.
That said, since the numerical ratings have verbal descriptions on the site, starting the Bayesian rating at "average" does have merit. That's even if most ratings given on the site are higher. Unfortunately, doing that really isn't in the spirit of Bayesian statistics. An uninformative prior should be the best estimate given the absence on information, i.e., the estimate which will converge to the population statistic as quickly as possible. Hence, the average of all ratings on the site is a good predictor of the final rating for a series that gets lots of ratings. So the question is, are series with few ratings unpopular or just new? What weight do you give to implicitly low ratings that were never given?
Note that using the average of all ratings, rather than all series with ratings, increases the upward bias from popular series getting more ratings.
Re Bayesian ratings and rating scales: I typically rate series about one full unit below the mean rating. On the other hand, I tend not to rate series that I've dropped or that put me off in the first couple of chapters. Perhaps I'm giving the author the benefit of the doubt. After all, in theory it could get better.
That said, since the numerical ratings have verbal descriptions on the site, starting the Bayesian rating at "average" does have merit. That's even if most ratings given on the site are higher. Unfortunately, doing that really isn't in the spirit of Bayesian statistics. An uninformative prior should be the best estimate given the absence on information, i.e., the estimate which will converge to the population statistic as quickly as possible. Hence, the average of all ratings on the site is a good predictor of the final rating for a series that gets lots of ratings. So the question is, are series with few ratings unpopular or just new? What weight do you give to implicitly low ratings that were never given?
Note that using the average of all ratings, rather than all series with ratings, increases the upward bias from popular series getting more ratings.
