It's not a matter of accuracy in this case, it's a matter of whether the regression has any meaning at all. Any regression can fit two data points, even a parabola. Of course, there are intuitive reasons to suspect that that is not the actual form of the distribution, but you can't actual show it from data with just two points.

But I have to assume that you did more data points when you actually performed this study yourself. Up to that point the methodology is really useful, so don't take this as criticism of the entire post.

Got ya. I did do more data points myself. Thanks for the commentary.

Even with more data points, I'm not convinced that App Store ranking is a good independent variable. At best, this seems like a very poor proxy for ad store impressions or downloads. However, I would be very, very surprised if the relationship between those legit metrics and app store ranking were anywhere near linear when you're comparing the top 5 or so to the rest of the top 25.

Paid App Store rankings are determined by # of downloads, so not sure why that doesn't work as a reasonable (albeit not perfect) proxy.

And agree 100%, def not linear.