Pareto charts often include a P-P (Percent-Percent) plot such as Jon’s red curves. This represents a cumulative distribution of effect (# of apps) against the cumulative ordered distribution of categories (for Pareto).

It’s not at all ridiculous to show that such a curve passes through the (0,0) origin as well as the (100%,100%) point. In fact, at the risk of showing the obvious, that’s how I’d prefer to draw it. For example, QIMacros draws its Pareto chart in Excel that way, although it doesn’t explicitly put cumulative % on the category count as Jon finally did under duress.

If you check Wikipedia, you’ll find a distinction between P-P and Q-Q plots, and it is P-P plots that go through (0,0) and (1,1) at their edges, while Quantile-Quantile plots may go to infinity (for continuous variables) at their hypothetical vanishing edges (not generally plotted!).

I’ve downloaded Wilkinson’s article and applied its techniques to this data set. Once I understood the algorithms and ideas, making the chart was straightforward. I wish I had more time to keep up with the discussion, but I expect to post a follow-up article next week.

I’ve always had issues with how the first and last points are considered on this kind of scale. In an Excel histogram (cough!) the first bin always has one point, unless multiple points have the minimum of the data set. In fact, some of the different definitions of percentile (or in general, N-tile) came about because of different ways to do the math. (i-1/2)/N for example, which is like measuring to the middle of each bar in a bar chart, where the bars are one unit apart along X. But other methodologies use different algorithms; I remember using something like (i-j)/(N-k) in a Weibull package we used for reliability determination, which gave very similar positioning along X.

Using cumulatives allows me to skip out on the whole percentile/bin definition thing ;-)