27 Dec 12, 03:10PM
(27 Dec 12, 08:15AM)Lantry Wrote: I think that the net people per load isn't the best way to rank the map+modes.
Net players (joins - leaves) provides a more meaningful ranking because it shows how many people joined (or left) the server because of the map+mode.
I can't disagree more. Non normalised NET varies in sample/frame/whatever it is size, therefore the values are not correct and should not be compared. Now, when you normalise by respective number of map loads, you get a constant 'NET per map load' size, which can be compared easily.
(27 Dec 12, 09:57AM)GDM Wrote: Even with normalization by #map loads, you can't reasonably compare 1078 loads for the 3rd place with 13 loads for the 4th place.
Some of the samples are way too small (10 vs 1000), yes, and I whole heartedly agree, but if I went with something like geometric mean (~86 loads) about half the mode/map cobos would not be part of the data set. The data set would definitely have a lot more "confidence" associated to it, though.
(27 Dec 12, 09:57AM)GDM Wrote: You have made no determination of the # of map loads needed for a representative sample. I don't remember much stats, but iirc you could probably calculate a #of samples needed to achieve some confidence in the average?If only I knew how :| I am in no way a statistician, so that's why I've released a spreadsheet so people experienced in the field could mess around with the data.