4 ms·

The high school stats student in me really doesn't like that regression line plotted against two datapoints.

trevmckendrick 14y ago
Fair enough. I was just doing an example with two data points. Of course the more points you add the more accurate the line.
- jamesaguilar 14y ago
  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.
  - trevmckendrick 14y ago
    Got ya. I did do more data points myself. Thanks for the commentary.
- omni 14y ago
  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.
  - trevmckendrick 14y ago
    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.
dmourati 14y ago
This. You have an implicit assumption for exponential decay or a polynomial. Why isn't this linear? Some other function?
- trevmckendrick 14y ago
  The more data you plot the more you'll find an exponential increase in sales for every increase in Gross Ranking. And if you watch the App Store rankings that makes sense. The apps at the top stay there, in the same relative order, for long periods of time. This suggests larger gaps in sales, vs lower ranking apps whose rankings fluctuate constantly. The purpose of this exercise was to make reasonable estimates, not to get it exactly right.

Fair enough. I was just doing an example with two data points. Of course the more points you add the more accurate the line.

jamesaguilar 14y ago
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.
- trevmckendrick 14y ago
  Got ya. I did do more data points myself. Thanks for the commentary.
omni 14y ago
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.
- trevmckendrick 14y ago
  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.

dmourati 14y ago

This. You have an implicit assumption for exponential decay or a polynomial. Why isn't this linear? Some other function?

trevmckendrick 14y ago
The more data you plot the more you'll find an exponential increase in sales for every increase in Gross Ranking. And if you watch the App Store rankings that makes sense. The apps at the top stay there, in the same relative order, for long periods of time. This suggests larger gaps in sales, vs lower ranking apps whose rankings fluctuate constantly. The purpose of this exercise was to make reasonable estimates, not to get it exactly right.