| |
Seven
Registered: Jan 2002
Posts: 202 |
Voting System: Weighted Averages?
I'm still somewhat confused about this weighted average thing, somebody willing to explain how this is calculated? =) |
|
| |
Perff
Administrator
Posts: 1680 |
I will not describe exactly how it is calculated, but I'll try to come with a brief explanation here.
The weighted average is a modified mathematical average which take into account the spread and number of votes.
This results in that the weighted average will always be smaller (or the exact same) as the mathematical average.
The higher the spread and/or the fewer votes, the lower the weighted average is compared to the mathematical average.
Noone said that this is the correct way to do it, but it's better than simply using the mathematical average.
|
| |
Steppe
Registered: Jan 2002
Posts: 1510 |
I agree. It prevents the chance of fake-votes. You know, people who try to "downvote" a release to bring up their own stuff in the ratings. Good example is www.c64.ch, it's a pity that the ratings are so unreliable there. Votings like 10 times a one for good stuff, the rest of the votes all above six, that's annoying. |
| |
Seven
Registered: Jan 2002
Posts: 202 |
Looking at e.g. the Saga ranking, I'd say it doesn't work just yet, wouldn't you agree? =)
|
| |
Perff
Administrator
Posts: 1680 |
I have a sugestion for 'improving' the weighted average (short WA from now on)
What about, when calculating the WA, you ignore say about 10% of the best & badest votes?
You can assume that there is always some that don't vote according to what they really think about it, but rater to raise/lower it's position in the ranks, and theese votes can then be filtered out by this method.
Idea? |
| |
Seven
Registered: Jan 2002
Posts: 202 |
10% of 5 votes is what? =)
Since you still didn't share the actual calculation of the WA so far, I wouldn't know if ignoring some of the votes would help.
I think as soon as some more people get their a$$es here and vote some, this will pretty much regulate itself with time, and a normal vote average would be the way to go then... |
| |
Perff
Administrator
Posts: 1680 |
that is 0.5.. ;)
Well. I think you are right. As long as there are only 5-10 votes on everything it very difficult to give a real ranking. That is also why the ratings are so low - the few votes degrades the WA compared to the real average of the votes.
Just hope people will vote some more, and I think we'll all see the rankings form more correctly. |
| |
anonym
Registered: Jan 2002
Posts: 267 |
Hey, and while we are talking about this, sometimes I see "Top Coders", sometimes I don't -> And shouldn't there be more people than just three in the coder charts?
/Frank |
| |
anonym
Registered: Jan 2002
Posts: 267 |
I just noticed something else. Groups the charts that have the same amount of points are still ranked on different spots, even though they should be on the same.
/Frank |
| |
Perff
Administrator
Posts: 1680 |
To both Frank's (anonym) posts:
1: The reason that sometimes the "Top Coders" shows up is that this place is used for general scener ratings.
What happens is that a random profession is selected, and then the best 5 in this profession is listed.
That leads us to the next problem. There arn't that many votes on sceners, so there are almost never anything in this position.
VOTE PEOPLE!
2: About the equal votes.
The votes that is listed is rounded number of the real WA, so even padua and crest at present time both got 7.7 according to the "Top Groups" they don't got the same exact rating.
Pauda got: 7.7109...
Crest got: 7.7086...
I think it would be nasty to display all theese decimals, so you just have to trust the rating. :)
|
| |
Seven
Registered: Jan 2002
Posts: 202 |
...which, btw., nicely demonstrates the flaw of the WA... looks like someone seriously downvoted Crest by casting a single vote...
or am I mistaken? |
| |
Perff
Administrator
Posts: 1680 |
Quote: ...which, btw., nicely demonstrates the flaw of the WA... looks like someone seriously downvoted Crest by casting a single vote...
or am I mistaken?
Nope. :/
You are absolutly right about this.
If there are a lot of votes on something this dosn't mean anything, but this is not the case.
As I sugested earlier an option is to look away from the best/worst votes when calculating the WA.
Anyone got other ideas? |
| |
Seven
Registered: Jan 2002
Posts: 202 |
I still think a normal vote average is less influented by these kinda things... |
| |
CyberBrain
Administrator
Posts: 392 |
I think normal average would suck. If a group had 3 votes of, let's say, 7 each, and a group had 30 votes of 7 each, both groups would be on the same place with normal average, even though only 3 people voted on the first group. (so f.e. just a few "not well thoughtout"-votes would dramatically change the charts, and it wouldn't show anything about what most people would think (when 3 votes would give the same weight as 30 votes).
WA *MUCH* better represents the votes, in my oppinion.
About the WA problem: Cutting off the 10% lowest/highest is certainly better than nothing.
|
| |
Seven
Registered: Jan 2002
Posts: 202 |
why not cut off *all* the votes?
why should I care to vote at all when my votes get cut off anyway?
yes, 3 votes with an average of 7 is the same as 10 votes with an average of 7... people don't like this - cast a vote yourself... but at least this way a single vote does not affect a rank _as_ much as is the case with the Crest downvote...
do your math? |
| |
CyberBrain
Administrator
Posts: 392 |
I don't understand the WA problem (and i don't know how WA is calculated, but judging from how WA should work). I would think that with WA if a single vote was far from the others, it would *not* affect the rating as much as normal average (and thereby not put Crest to the bottom of the charts just because of one vote)
Obviously this is not the case?
If not, it should be taken into consideration in the WA formula. |
| |
chatGPZ
Registered: Dec 2001
Posts: 11390 |
1) talking of something like "weighted averages" is total BULLSHIT unless you tell us how its calculated. pronto presto pretty please! :o)
2) i agree that using something like quadratic- or cubic- average is probably better for calculating "true" numbers
3) averages shouldnt be shown at all until atleast a certain amount of ppl (10 or so) have voted. with less votes, you may just as well use a dice :o) |
| |
Flex
Registered: Feb 2002
Posts: 111 |
To prevent "downvoting" etc. misbehaviour, why shouldn`t we have a some kind of a register for that.
Would be nice to see who has voted and how... |
