Statistics Levels statistics - Research for monthly reports

Nice talking around here

Well... I don't really know where to start...

First thing I've got in my mind is that the log2 method is exactly what I was seeking for at the begining of this thread: a universal/general equation which is able to bundle everything and give a simple result. So that's a great answer to me. Moreover, coding a such equation is so simple that everybody (almost) is able to do the same with a simple spreadsheet software (assuming you've got the data) and understand the method. It's simple and easy to understand.

But since these days I've thought it's probably not a good method to try to melt everything in only one equation. Why? Because this is trying to put in the same bucket several data that are not relative and considering they are the same while they are not. This is like saying that apples, oranges and tomatoes are the same (well... I don't have more elaborate example for now, sorry) trying to treat each one in the same manner, ignoring their differences. But as they are not the same, we search a way to make them speak in the same way. There's something not logical to me in this. That's too where views become a problem because it is more difficult to integrate them in the equation, while I believe they are a good factor of popularity, as long as they are considered with the other points (you tell me if you disagree).

With Quartiles, suddenly things seemed more respectful to each kind of data: ratings, rates, likes, views and even comments stayed in relation with themselves (but I would agree that comments doesn't seem relevant for our subject). We are NOT trying to arrange ratings with rates, likes, etc. all in the same place, or trying to help them to be friend. We are processing each kind of data separately, with respect to their own nature, in order to classify them and according to a common norm. Each data is normalized and it is that norm that we can put in the same equation. I really like this idea, that sounds so "beautifull" to me. Moreover, we could have 1, 5, 50 or 100 different kinds of data, it would be "easy" to get a result. Then we can decide of a weight for each normalized data so that the more valuable/prefered has the more influence on the result. That's almost the same consideration to me when at school you've got a notation that represents your knowledge or ability in a course; then not all the courses has the same importance so we are poundering the result of each one by a multiplier.

So... Of course Quartiles methodology is not as understandable as a general and uniq equation, but (to me) it's more meaningful in regard to the different kind of data. In this way it's seems more "universal". Coding this is also not as easy than a simple formula but it definetly worthes the work.

Now, considering the TOP 30 quartiles result, I have found from several trials that levels with a score less than 10 are not as relevant than above 10. I don't really know why... That looks like a breaking point to me.

 

@chemi : I have planed to wait a week before releasing them. So you're right: it's about time I will do it soon.

 

@chemi (and others) - Statistics from June to October has been published!