concerning statistics

[purple]

I was thinking that statistics should be weighted in part from member ratings but mostly from points given to them from being favorited/in charts. This way, the statistics page for member "rated" stuff will be more representative of what the members actually enjoy, rather than what afew people have given them. For example, the top member rated album is an Elvis CD which only 8 people have favorited and the top meber rated song is from Arcade Fire's Funeral (which is a great album, but there are better songs). I know that there's a chart with albums based on points, perhaps this should be merged with member rated...

just some thoughts

[albummaster]

I agree, I was just commenting on another thread that people are getting a bit smart with their ratings which is really affecting the average ratngs. The algorithm does need to be tweaked a little to stop people from gaming the system. At the moment, we offer member rating charts for albums, bands, charts and tracks. Albums and bands could. in theory, be rated taking into account their appearances in member charts, but tracks and charts themselves cannot. At the moment, we use the 'mean' average rating but perhaps something else would now be more appropriate.

Regarding the overall chart, one idea which we've been pondering is to calculate the overall chart taking into account the rating that a chart has received. If a chart has been rated with 90% it would have more weighting than a chart that only has a 60% rating. Personally, I think this would lead to a much higher quality overall chart (which would have the benefit of being 'peer reviewed' as part of its compilation)

[RFNAPLES]

I am not a fan of the chart ratings. Often it appears only to be a popularity contest, or a vindictive rating system, with little thought and agreement on rating criteria. There also appears to be a bias against most "recognized" charts. Let the number/position of charts remain the criteria for ranking; forget the idea of weighting by rating.
_________________
Top 100 Greatest Music Albums by RFNAPLES
Bubbling Under The Top 100 Greatest Mus...y RFNAPLES

[purple]

I like the weighted idea, but there does need to be some way to weave out the "vindictiveness factor." Perhaps any rating could show up on a members chart page but, when calculating how to weight the chart, scores 10 or below are not counted and (this would be really tricky to program) scores below 50 which are the same as those given to the rating member are discounted (for example, many people will rate your chart a 40 if you rate their chart a 40, take a look at telefunkers page).

As a side note, because members often don't re-rate a chart when it is edited I've gotten into the habit of deleting my chart every time I edit it so members are rating my new selection of top albums. Tell me if this is messing up the system in any way or not preferred and I can stop.

thanks

[albummaster]

No, deleting the chart each time won't mess the system up and is probably the most sensible thing to do if your chart has more than a few tweaks.

I'm not sure we can do much about tit-for-tat ratings. They'll always exist and we cannot program around them. In truth, the 'mean' average is the simplest way of doing things and probably no worse than any other method (and particularly for items that have been rated a great number of times).

[RFNAPLES]

Maybe you can use an Olympic system where they throw out the highest and lowest scores.
_________________
Top 100 Greatest Music Albums by RFNAPLES
Bubbling Under The Top 100 Greatest Mus...y RFNAPLES

[albummaster]

yes, we thought about doing a truncated mean but that means discarding some votes and how many do we discard? We could pick an arbitrary value but some albums have more votes than others, so it would end up being equally as [un]valid as what we are doing now, plus it has the possibility to annoy some members if *their* vote isn't counted.

[Charicature]

[albummaster]

Yes, there's a whole host of ways we could decide to calculate this but I don't think there's a perfect answer. The more complicated we make it, the less people will understand what numbers they are looking at and the less meaningful the numbers will be. I think sometimes, keeping things simple is the best way to do things which is primarily why things have stayed as they are.

[RFNAPLES]

Yes simplicity is good. Perhaps a few other statistical measures could be given. In addition to mean, these measures might be understood--median, mode, range, standard deviation, kurtosis, and skewness.
_________________
Top 100 Greatest Music Albums by RFNAPLES
Bubbling Under The Top 100 Greatest Mus...y RFNAPLES