Stateofdps

[Deleted]

Hi,
I'm wondering why in the official forums everbody is citing the median values of stateofdps as measurement of DPS performance.
The median of 200 parses for each boss is actually a nice number , which is statistically significant. But in stateofdps in the adjusted
totals they build the median of a median. So thats a median of twelve values, which vary between 14000-30000k dps.

Therefore it's clear, that the median can easily vary +/- 1000 or 2000 dps for each class. And from the peapple say the class performs
at 90% of the other spec? 1000-2000dps variance makes easily up for this. I don't understand, why they don't calculate the average of
the medians, which is i think a much better value (if you exclude outliers like Halfus). Sorry but taking the median of 12 values varying
so strongly is really stupid. (look for example at Retri Data and Marks Data - stateofdps adjusted totals: 21901 for marks and 213269 for
Retri...only a small gap... if you create the average of the medians it's 22000 for Marks and 20648 for Retri).

Sry for my bad English .

[Revengencer]

The median of a median is actually a good statistical indicator. It eliminates any extreme outliers. If one fight, for whatever reason, favors one particular class overwhelmingly as a matter of mechanics (think Nef and ferals), it would pull the whole mean average of all the fights way up for druids. The average and the median of the median values of each fight are both legitimate methods of analysis, honestly, and to say that either one of them is "really stupid" is an extremely disingenuous statement to make.

[Deleted]

Ok it's not stupid, but it's stupid to draw from this conclusions on dps. If you have a outlier like Halfus you can exclude it. But if you create the median of twelfe values varying between 14 and 30 (try to do this with matlab: create random numbers between 14 and 30 and create the median...) the median can vary a lot (easily 1000-2000dps). So if people claim from this site, that a certain class is 10-20% higher, their conclusion may be very inaccurate or even wrong.

A median for each boss for 200 parses on the other hand is fine.

(Just an extrema: Creating the median of 1000 1000 and 3000 is 1000 and not very fair. You can only use median for larger samples.)

[Amerika]

What I dislike about people using that website is that they don't click on the links that explains how the numbers are determined. There is no way for the guy who writes Stateofdps.com to come up with accurate numbers based on when Hotfixes go out due to how World of Logs codes the dates on it's parses. So, when a class gets nerfed via a hotfix it still looks like it is doing well even a couple of weeks after the nerf simply because the old pre-nerf numbers can not be excluded.

People then go and use state of dps to respond to others that are having issues with their classes DPS performance and are being told they are fine or learn to play because the numbers are inflated due to old parses. When classes are buffed this is a non-issue because the pre-buff parses get replaced quickly.

[Revengencer]

You have to take the entire thing with a grain of salt. The thing I hate about statistics, as much as I love them to use, is that people misinterpret the shit out of them, usually by not really interpreting them at all. When you look at the mean average of any data set, you need to accept the fact that it may be skewed by outliers. When you look at a median average of that same set, you need to realize that it won't work if the data is very clustered. You can eliminate halfus because it's mechanics have a profound effect on DPS for all classis, and are variable weekly on the normal mode.

State of DPS is nothing more than a gauge of the approximate known potential of every class. As long as someone limits the scope of their assumptions to THAT, they are doing ok.

[Amerika]

Halfus data can actually now be included since he now only gains the extra damage buff when all the other drakes are dead. Or at least that is how it works in heroic mode (haven't done normal since the change) but I assume it's similar.

[Deleted]

at Revengencer:

I agree with what you say. I just hate people in the official forums citing that median and demand nerfs or buffs because of that.
The only interesting valuable information on this site the the performance of the classes for each boss (and there the median).

Averaging this data including movement fights, aoe, multidotting and whatever only results in confusion.

For the discussion of median of median and average of median, i agree of course that both are appropiate methods, but for small sample
sizes with such clustered data a median screws your average much more than creating the average while removing the Top 1 or 2 and Bottom 1 or 2.