Best Ever Albums

Re: I guess I don't understand how averages work here...

albummaster — Sat, 3 Nov 2012 06:14:17 GMT

Author: albummaster

Posted: 11/03/2012 10:14

Andrash wrote:

My chart has two ratings: one is 95 and the other one is 85. So the computed average is... 77. 😲 How is that possible?

Thanks!

As above, the average is a Bayesian average which helps to smooth the figures. In the case of items with a few votes, the site average rating will pull the figure down, and that's why the computed average is 77. (As the chart gets more ratings, the average will likely rise if people keep rating at those types of levels.)

Re: I guess I don't understand how averages work here...

Jabapac — Sat, 3 Nov 2012 05:49:27 GMT

Author: Jabapac

Posted: 11/03/2012 09:49

Also, don't forget about this: (from last February updates)

Quote:

The average rating algorithm has been updated to use the item average instead of the average of all ratings, and a change has also been made to reduce the impact of troll votes on an item's average rating. Previously, outlying votes could disproportionately affect an item's average rating, but now ratings will be excluded if they lie more than three standard deviations from the item average.

Re: I guess I don't understand how averages work here...

Defago — Sat, 3 Nov 2012 02:23:27 GMT

Author: Defago

Posted: 11/03/2012 06:23

Eggman/Walrus/Taxman wrote:

Average Rating = (n ÷ (n + m)) × av + (m ÷ (n + m)) × AV

I always thought it should be (n ÷ (n + m)) × av + ((z) ÷ (n + m)) where if n>m, z is 0 and if n
That way, you take into account the site mean average only while your chart's amount of ratings is below m, and after that you forget Bayes and use simple arithmetic. I'm sure there are reasons why this is not used, though...

Re: I guess I don't understand how averages work here...

Facetious — Fri, 2 Nov 2012 14:39:34 GMT

Author: Facetious

Posted: 11/02/2012 18:39

From the "How It Works" page:

"The average rating is not a simple arithmetic mean, instead it is calculated using a Bayesian average:
Average Rating = (n ÷ (n + m)) × av + (m ÷ (n + m)) × AV

where:
av = mean average rating an item has currently received.
n = number of ratings an item has currently received.
m = minimum number of ratings required for an item to appear in a 'top-rated' chart (currently 10).
AV = the site mean average rating.

A Bayesian average helps to stop items with only a few high ratings from being ranked too high in the charts. It also helps to dampen the effects of a few bad ratings on the overall average."

Hope you understand that.

I guess I don't understand how averages work here...

Andrash — Fri, 2 Nov 2012 14:15:57 GMT

Author: Andrash

Posted: 11/02/2012 18:15

My chart has two ratings: one is 95 and the other one is 85. So the computed average is... 77. 😲 How is that possible?

Thanks!