It's a log scale. The bottom axis should use tick marks to make it is obvious. A poorly marked log scale is certainly something to fix. It's not the same as fudging the data.
[edit: If you wish to say a log scale is inherently misleading in this context ... go for it. That's different then saying the data is manipulated.]
[Edit2: The area is not meaningful. The widths are meaningful if there is a total ordering and the scale is labeled. ... I do however agree I am probably way overestimating how obvious a [log(1+cumulative percentile),log(1)] mapped to the xaxis is. Also the chart does scream compare areas, and those are strictly meaningless unless comparing within the same OS.]
Can someone explain how a log scale could work for this sort of stacked graph? (Serious question, not rhetorical question.)
If the X axis is a log scale of market share, what would happen if Apple and Android both had 40% market share? Both bars would overlap.
If X is cumulative market share, the bar width would depend on order and the two hypothetical 40% companies would have different widths.
If each bar has area proportional to the log, how would that work? The logs of market share are negative unless there is an arbitrary constant in there. Also the vertical breakdown doesn't make any sense in that case, because the areas of the vertical blocks don't add up to the OS total. Also small market shares would have negative width?
So can someone explain how log scale could even theoretically work here?
At least since the appearance of data sets that lent themselves towards such visualization. For example, how better to compare the growth of various platforms, from TRS-80 days to the iPad while the entire industry grows exponentially?
[edit: If you wish to say a log scale is inherently misleading in this context ... go for it. That's different then saying the data is manipulated.]
[Edit2: The area is not meaningful. The widths are meaningful if there is a total ordering and the scale is labeled. ... I do however agree I am probably way overestimating how obvious a [log(1+cumulative percentile),log(1)] mapped to the xaxis is. Also the chart does scream compare areas, and those are strictly meaningless unless comparing within the same OS.]