Stumpokapow
listen to the mad man
Liabe Brave said:If you want to show the overall software strength of a system, just use total sales. This is the closest to a pure measurement you can get.
The problem is that a developer or publisher who is looking into making a new title doesn't necessarily care about the number of units of software shifted; they care about their chances to shift a unit of software. In this case, the general health of the platform in terms of an average measurement is going to be more useful.
Median isn't. Multimodal distributions would result in distorted medians, and the exclusion of bombs and hits isn't a benefit for analysis as you imply, it's a loss. It's like overzealous JPEG compression. The one thing I think median would be good for is as an abstraction of the shape of sales distribution. Showing medians as well as total number of titles released for the period would indicate whether everything was selling well, or there were hits and bombs but no consistency, etc.
Well, yes and no. I mean, certainly console sales tend to be multi-modal in that we have several sort of distinct categories of sellers (ultra-bombs, low-profile, medium profile, high profile), but I've never seen anything approximately a pure bi-modal or multi-modal distribution.
The sort of multi-modal distribution we DO see is a bi-modal distribution with peaks at the very very bottom and the very very top, which are nicely smoothed out by the median. Also, our data is not complete and is more complete for our top end, so the median could be considered a more accurate measurement since it provides some "benefit of the doubt" to the lower end given that the relative incompleteness of the lower end of data versus the upper end is compensated for. Something like a bayesian weighting algorithm could probably be applied to smooth out all data points, but I think that's probably diminishing returns at that point.
For real interest, though, I think you'd want to forego the abstraction and head straight for the true data: a histogram of sales performance of all games for a system. The bonus here would be that total sales would also be included, as the area of the graph.
Again, total sales is pretty irrelevant, but I do think a histogram year-by-year would be interesting if only to see the growth in sales over the life of the system. I'd be happy to do one as well when I get the time. A histogram would show you a pretty even distribution for most consoles, though.
Since GAF by and large has no statistical background, readability and simplicity in the presentation of statistical concepts is key. I remember the old claim that Sony was "closing the gap" when they were closing the gap in weekly sales. The poor fellow that tried to teach integration and differentiation conceptually to a thread full of randoms who just refused to get it... oh, the horror!
Edit: removed a largely tangential discussion on the merits of a year-by-year histogram.