e.g. Amazon, EBay, etc.
If you’ve observed 8 ravens and they’ve all been black, how certain should you be the next raven you see will also be black?
According to the Rule of Succession, 90%.
In general, the probability is
k+1
p = ---
n+2
that the next observation will be positive, given k positive observations out
of n total
In plain English: Add two "imaginary" reviews: one positive and one negative. Now divide the new total number of positive reviews (k+1) by the new total number of reviews (n+2), and that's your adjusted rating.
Now, a walkthrough of the mathematical logic (not a proof, merely the logic).
We'll do a product which has 4.7 stars with 118 reviews.
- How "positive" is 4.7 stars out of five? If you think of 5 stars out of 5 as being 100% positive, 4.7 stars is 94% positive (4.7/5 = 0.94)
- OK, that's something to hang one's hat on; the percentage of reviews which are positive should be 94% as well (118x0.94 ~= 110)
- Add one positive review (110+1 = 111), then up the total number of reviews by 2 (1pos and 1neg) (118+2 = 120)
- 111/120 = 0.925 - this product has a 92.5% "satisfaction rating"
The interesting thing is how intuitively this works.
- 4.2 stars, 17,000 reviews
- 4.2/5 = 84% - off to a bad start :)
- 17000*0.84 = 14280
- 14281/17002 = 83.99%
More reviews really just reinforces the star rating.
If it still seems weird, remember the oversized impact that one F or (even worse) 0 has on a GPA...
Not 100% correct yet, needs fixing.Should be good now.