How do weighted scores get calculated or rather Why is the weight so weak?

Feb 11, 2015 4:05 PM

Offline

Oct 2013

Every entry in the anime database with at least 50(?) ratings gets a weighted score (in the manga database it seems to be only 25 ratings).

My question is, why is the weight so weak? Entries with barely any ratings get hit "hard" by this weight, but in case an entry is more popular (like Parasyte) the weight is not noticeable.

Shouldn't a weighted score bring all entries close to a Gauss distribution? So shouldn't you calculate a weighted score based on popularity, average of user ratings, dispersion of user ratings and all of this by considering the progress of users and convert it into a Gauss distribution with an average of 5.5? Because this way, all anime entries would have an actual weighted score and use the 1-10 scale to its full extent.
In other words similar how the rating on anidb.net works which is shown side by side with the arithmetic mean.

And another question: Why is the "scored by ... users" number slightly different to the sum of all votes in the score stats? Because right now, Parasyte has "score by 37980 user", but a sum of votes (in the stats menu) of 38018.

Feb 12, 2015 12:52 AM

Serhiyko

Offline

Oct 2009

7667

http://myanimelist.net/info.php?go=topanime

Qonell said:
And another question: Why is the "scored by ... users" number slightly different to the sum of all votes in the score stats? Because right now, Parasyte has "score by 37980 user", but a sum of votes (in the stats menu) of 38018.

Because of the 1/5 rule

Feb 12, 2015 8:18 AM

Qonell

Offline

Oct 2013

Serhiyko said:
http://myanimelist.net/info.php?go=topanime

Because of the 1/5 rule

Well, that's wrong (afaik). The 1/5-rule only applies to entries which already finished airing/publishing. Parasyte is still airing. And what does the 1/5-rule have to do with the sum of ratings anyway?

Feb 12, 2015 2:02 PM

xbobx

Offline

Jun 2010

2561

Unfortunately I can't contribute to this since I never paid much attention to the statistic behind those scores...
However, I can tell you that we found out long ago that both "Mean Score" and "Score Dev." on our list are glitched. "Score Dev" being the most apparent one, as some lists show a value of (e.g) "-11.3" which is clearly impossible. MAL also mixes some data on mean and deviation of your own stuff with the general userbase, so it's rather confusing.

It seems MAL uses a Bayesian Posterior Mean to weight the score:

But I have no idea why, and neither why is this better than a simple Gauss distribution.
Luckily engineering only goes up to statistics II... hehe.

http://fakemorisummer.wordpress.com/

Feb 12, 2015 4:46 PM

nantuko

Offline

Apr 2012

483

The /stats page is more up-to-date with its score than the summary on the left of an anime, that's why the sum and rating is different.

Sometime before when MAL had some of its many issues it happened that the summary rating on the left was stuck and not updating for some or all anime or was simply wrong and Luna or someone else mentioned it somewhere that the numbers are calculated/read differently for whatever reason.

Feb 13, 2015 1:39 AM

Serhiyko

Offline

Oct 2009

7667

Qonell said:
And what does the 1/5-rule have to do with the sum of ratings anyway?

Then it's scores from users who have it in "Plan to Watch"
Or maybe the users who voted on it with zero episodes watched? Not sure if such scores are included

Qonell said:
And what does the 1/5-rule have to do with the sum of ratings anyway?

Because it shows the sum of all scores, even the ones that are not later included in the calculation of the mean score

SerhiykoFeb 13, 2015 1:50 AM

Feb 13, 2015 9:23 AM

Qonell

Offline

Oct 2013

Thanks to xbobx and nantuko, you basically have answered my questions.

The fact that there is a small time delay between these two sums sounds reasonable.

And yep, it seems to be the Bayesian average. I'm a bit embarrassed that I totally forgot to consider this mean. It would explain how the weight works and the fact it gets so considerably weak the more popular something is. I would still prefer a Gauss distribution though (like anidb.net).

See All Discussions

Sticky: » Anime DB - Post Modification Requests Here ( 1 2 3 4 5 ... Last Page ) cyruz - May 15, 2016	8686	by JohnnyNumbuh28 »» 2 hours ago
Sticky: » Anime DB - Post Genre/Theme Requests Here ( 1 2 3 4 5 ... Last Page ) Kineta - Mar 28, 2022	681	by SurgingSoulZX »» Apr 24, 7:33 AM
» My Daemon ( 1 2 ) marcolacson - Nov 26, 2023	50	by veduHMY »» Apr 23, 3:36 PM
» will the Scott Pilgrim anime be added to the database? ( 1 2 3 4 5 ... Last Page ) GanjiTheDude - Mar 30, 2023	756	by YoshiYogurt »» Apr 20, 1:12 PM
» Could we stop having Justin Cook at the start of every anime entry's staff list? El-Melloi - Jul 29, 2023	10	by MadanielFL »» Apr 20, 11:44 AM

How do weighted scores get calculated or rather Why is the weight so weak?

More topics from this board

Sticky: » Anime DB - Post Modification Requests Here ( 1 2 3 4 5 ... Last Page )

Sticky: » Anime DB - Post Genre/Theme Requests Here ( 1 2 3 4 5 ... Last Page )

» My Daemon ( 1 2 )

» will the Scott Pilgrim anime be added to the database? ( 1 2 3 4 5 ... Last Page )

» Could we stop having Justin Cook at the start of every anime entry's staff list?

MoreTop Anime

MoreTop Airing Anime

MoreMost Popular Characters