Suggestion for d20reviews ratings, the true Bayesian estimate

[Psionicist]

I'd suggest you to use the true Bayesian estimate when calculating the rating of a specific book, if it qualifies for a list or not.

true Bayesian estimate
weighted rank (WR) = (v / (v+m)) * R + (m / (v+m)) * C

where:
R = average for the book (mean) = (Rating)
v = number of votes for the book = (votes)
m = minimum votes required to be listed in the toplist
C = the mean vote across the whole report

For example:

R= 3.8
v= 23
m = 10
C = 4.2

(23 / (23+10)) * 3.8 + (10 / (23+10)) * 4.2 = 3.92

 

[Morrus]

Why?

 

[Psionicist]

Because it's a very, very good formula for that sort of things.

 

[Morrus]

Why?

 

[hong]

Bayesian estimates make the baby Hong cry.

 

[pennywiz]

Why?

 

[Darkness]

Why?

Click here to find out - maybe.

 

[CRGreathouse]

The formula Psionicist presented is designed to "normalize" rankings - if something is below average it raises it; if above average, it lowers it.

It has two major problems I see: I don't want to water down scores like that, and it doesn't take care of extreme scores like the other formula does.

The reason we tried to move away from the mean (average) ranking system is that it can be skewed by a reviewer who gives it a well-chosen score (1 to a product with a 4.5). Using the mode (most common) score reduces this problem, but sometimes causes odd results, like scores "jumping" from 5 to 3 on a single review, even when the product has many people rank it. Using the median (middle) score comepletely takes away the possibility of score tampering, but (like the mode), has a limited range of output without fractions. As a result, I suggested Sqr(mean * median).