Star ratings

[Smashed]

What's with the star ratings on manga? It's never anywhere near close to an average of what everyone rate it, what you'd assume it'd be. Baka updates has a weird way of weighing their average but I have seen examples here that would be impossible to generate even with that (a series having nothing but 1 star ratings archiving a 7.9 overall).
To add to that the lowest score I've ever seen something get here is a 6.5 and the highest a 9.3, this squashes everything down into a 2.8 star range instead of the advertised 10 star range and that's misleading at best. I get not using the top end of the scale because nothing is perfect but why are you manipulating the statistics to avoid giving anything a "fine" or less rating?

 

[Roler]

MangaDex mainly uses Bayesian average, so the initial rating is based on the average rating of all titles on the site.
If a few people rate a title 1 or 10, it won't automatically become 1 or 10, but it'll increase or decrease the more people rate it.

 

[Smashed]

Is their assumed mean the average bayesian average of all other manga on the site and not 5, the number that they claim is an average rating?

 

[Roler]

Yes, it's basically the average rating.

https://forums.mangadex.org/threads/totally-un-biased-ratings-on-mangadex.1215736/post-17165887

also https://api.mangadex.org/docs/03-manga/statistics/#the-bayesian-formula

 

[RogueKitsune]

Is their assumed mean the average bayesian average of all other manga on the site and not 5, the number that they claim is an average rating?

The bias value is the arithmetic average of all ratings across all titles, and yes its shockingly high at first glance (somewhere around 8 iirc). However, if you think about the behavior of "most" users of the site it starts to make sense.

People don't read things they don't like (or if they don't think they will like it). This in turn has the side effect of people not rating things poorly because they simply are not reading many things that would cause them to give a low rating (also people have trash tastes).

Personally, I don't rate titles until I have completed it or dropped it after reading more than just a few chapters (i.e. if I drop it during the 3 chapter test I usually don't give a rating).

Bayesian Average Oversimplified:
Its a fancy weighted average. When there are only a few ratings the bias value has a very large impact on the title's rating. The more ratings a title has the less impact that bias has. This means eventually with enough ratings (good or bad) the Bayesian Average and Arithmetic Average will be the same.

Examples:
Redo of Healer: 6.43(B) 6.42(A) well over 4K votes
Shinmai Shimai no Futari Gohan: 8.92(B) 9.02(A) ~1K votes
Sousai Inferno: 8.06(B) 6.00(A) 1 vote

 

[Smashed]

Ok then this is the issue

don't lable 5 as average if it isn't the average. Either change these words to match reality, apply some normalization at any step of the way, or atleast use 5 as the assumed mean.

 

[Grogzilla]

Ok then this is the issue
don't lable 5 as average if it isn't the average. Either change these words to match reality, apply some normalization at any step of the way, or atleast use 5 as the assumed mean.

What do you mean by this? 5 IS average between 1-10.

8 is the calculated average between all titles. It is assumed a title will receive a score of 8 until ratings start coming in. Once more information has been gathered (more people rate a mango), the rating then begins to reflect the Arithmetic average score (see RogueKitsune’s comment again).

If you think 8 is “average” by any stretch of the imagination, your standards are too high.

 

[RogueKitsune]

Ok then this is the issue

don't lable 5 as average if it isn't the average. Either change these words to match reality, apply some normalization at any step of the way, or atleast use 5 as the assumed mean.

I am not sure if you are trolling at this point or not.... But let us assume you are not.
The wording next to the numbers in that list are just a description to help guide a user to pick a value that aligns with how they might feel about a given title. In this case "(5) average" should be taken as the title was neither exceptional or awful, just "middle of the road"/"meh"/"average".

 

[Smashed]

If you think 8 is “average” by any stretch of the imagination, your standards are too high.

I don't but mangadex does.

 

[Smashed]

In this case "(5) average" should be taken as the title was neither exceptional or awful, just "middle of the road"/"meh"/"average"

But in practice it isn't used this way. I have not seen a single series rated below a 6. So you are telling everything that exists is above average.
Defining average as

being about midway between extremes

 

[T10_]

I don't but mangadex does.

No, the users do.

But in practice it isn't used this way.

There is nothing that can be done about that. Just because you don't, doesn't mean that others don't. Users are given the option, but they choose to rate titles overwhelmingly positively
with the main cause being:

People don't read things they don't like (or if they don't think they will like it). This in turn has the side effect of people not rating things poorly because they simply are not reading many things that would cause them to give a low rating (also people have trash tastes).

Personally, I don't rate titles until I have completed it or dropped it after reading more than just a few chapters (i.e. if I drop it during the 3 chapter test I usually don't give a rating).

Everyone here (obviously) isn't the first person to complain about it and so is every thread about it since the forum was reopened. I can only say "users were a mistake".
The only solutions (I can think of) are:
1. to only show the score after it has enough votes, like MAL.
2. to show a different score than the bayesian average. And I can assure you, if it's even on the list, the priority would be somewhere at the bottom.

I have not seen a single series rated below a 6.

https://mangadex.org/titles?order=rating.asc

 

[Roler]

I don't but mangadex does.

You seem to be confusing two different averages:

• the "average" when you (one user) rate a title, is there to guide you to pick 5 if you think the title is average.
• while the "average" used for the Bayesian calculation is the average of all titles on the site, meant to represent all users.

The dev explained why that "average" is used for Bayesian, but you, as a single user, should still use 5 (or lower) if you think a title is average, even if other people on the site rate titles higher than they should (or don't rate them lower).

 

[Manko-sensei]

If you're bothered by the fact that it doesn't show an objective result then you're barking at the wrong tree, you won't get objective results in any platform as few people take rating seriously. When emitting a vote, you shouldn't care at all if it's used Bayesian or whatnot, and the description next to the number helps people deciding how to rate things. You should just check the other thread they added for you to avoid repeating the same discussion, and the reason for using this method is to avoid bomber rating titles because they have a certain tag or belong to a certain author. So read, think and if you still don't understand something then ask.

 

[RogueKitsune]

But in practice it isn't used this way. I have not seen a single series rated below a 6.

But it is used that way, your problem is that other users don't tend to read things they don't like and thus don't give things low ratings.

I am kind of guilty of this as well, while I have read a few things that I have rated below a 5 I would say a majority of things I have rated I have given a 6 or 7 because I tend to read things I like.

And trust me when I say I have seen so much BL rape omegaverse crap that I would probably give a 1 if I actually read them, but I don't hate myself that much.

 

[Smashed]

https://mangadex.org/titles?order=rating.asc

this us exactly my point. There are less than 20 "below average" manga on the entire site. Doesn't that seem oxymoronical to you?

2. to show a different score than the bayesian average. And I can assure you, if it's even on the list, the priority would be somewhere at the bottom.

You could still use bayesian and just use 5 as your assumed average instead of the average bayesian average. That would draw the average bayesian average closer to the score you list as average

 

[T10_]

this us exactly my point. There are less than 20 "below average" manga on the entire site. Doesn't that seem oxymoronical to you?

⬇️

Everyone here (obviously) isn't the first person to complain about it and so is every thread about it since the forum was reopened.

me included before I became a mod
I remember ranting about literally the same thing in 2022 to @tristan9

You could still use bayesian and just use 5 as your assumed average instead of the average bayesian average. That would draw the average bayesian average closer to the score you list as average

then it wouldn't be a bayesian average anymore, since it relies on data and manipulating that data would invalidate that. Which brings me back to ⬇️

to show a different score than the bayesian average. And I can assure you, if it's even on the list, the priority would be somewhere at the bottom.

 

[Roler]

this us exactly my point. There are less than 20 "below average" manga on the entire site. Doesn't that seem oxymoronical to you?

You could still use bayesian and just use 5 as your assumed average instead of the average bayesian average. That would draw the average bayesian average closer to the score you list as average

You seem to not understand why Bayesian average is used in the first place.

Bayesian average has little to no impact on titles that are rated enough, and a lot of titles there are rated enough (their averages and Bayesian averages are really close to each other).

Changing it to 5 would not affect them much, but it would drastically affect titles that are not rated enough (the very thing Bayesian average is trying to prevent) because titles rated 5 would still be considered very bad on MangaDex like they are now and new titles (with not enough ratings) would be considered equally as bad as them (instead of what is considered average on MangaDex).

 

[tristan9]

There are less than 20 "below average" manga on the entire site. Doesn't that seem oxymoronical to you?

Maybe, but that's not something we can do anything about. Go and try to convince ~4 million people to rate things 5 when they find it average, instead of the 8 they do. 🤷‍♂️

Also in general no it's not contradictory like you think if you give it more than 2 seconds of thought.
For the reasons people have mentioned to you above and which you ignored:
1. People don't read a random selection of manga, they read things that they think they will like from the get go
2. So on average they like what they read more than what they would find by selecting things truly randomly
3. Plenty of us won't give a 0 to a title if we try it, dislike chapter 1, and move on. This also skews averages up.

All of this has nothing to do with MangaDex and everything to do with the fact that if people have 100,000 titles to choose from, with precise filters to read (almost) only what they will like.

You can go and read the thousands of threads about the same "issue" on every single entertainment platform that has user ratings.

The only alternative is normalizing ratings sitewide, which would be possible but would end up causing a 9.5/10-rated title to become something like an 8/10. Is that more desirable? I don't think so.

 

[tristan9]

to show a different score than the bayesian average. And I can assure you, if it's even on the list, the priority would be somewhere at the bottom.

We already do, by the way. If you hover the rating text on desktop, it shows the non-Bayesian average rating too in a tooltip.

 

[Teasday]

I told them way back in the day to just have ratings of 1-2, maybe 1-3 but I was overruled! smh