Predicting Firefox Growth

Our Metrics intern, Eric Hergenrader, did some terrific work analyzing fluctuations in Firefox usage.  Using a linear regression, he found that this summer’s decline in Firefox use was largely in line with expectations; a 2-3% decline in July and August is typical.  In addition to seasonality, Eric found that three factors driving usage drops are weekends (-18%), holidays (-30%), and Firefox releases (-13.5%).

Two months later, Eric’s model does a remarkable job predicting future usage.  Below, I have plotted actual against predicted usage.  The correlation between these two lines is remarkable.  Since September, the average prediction was off by just 1.63%!

Screen shot 2009-10-12 at 2.47.16 PM

To better understand daily growth variations, I created a heat map that visualizes month over month usage growth. Each cell contains one day’s growth rate over the same day of the week, 28 days earlier.


Some highlights:

  • Strongest growth occurs during Spring and Fall
  • Periods of growth and decline are clustered
  • Growth is outperforming predictions by 2.58%

We must be careful not to overreach with our conclusions.  We are working with a limited range of data and have likely missed significant confounding variables.  That said–thanks to Eric’s work–we’re off to a great start.  Please leave your thoughts and any suggestions on how to improve our projections in the comments.

5 responses

  1. Donnie Berkholz wrote on :

    If you really want to illustrate the similarity between prediction and reality, show the difference between them as a percentage error…

    Also, as an analogy with stock markets, be careful with month-to-month growth. Try with the same month last year instead.

  2. David Naylor wrote on :

    Very interesting!

    I’m curious though, that is the Y-axis in that graph? Active Daily Users?

    I’m also baffled as to why Firefox usage goes down around Firefox releases – wouldn’t you expect the opposite?

  3. Dwayne Bailey wrote on :

    I found the heat graph interesting.

    I assume this is en-US data? Since other people have different holidays I’d be curious to see if that matches other locales. Also can you please avoid terms like Spring and Fall, they mean nothing to someone from the Southern Hemisphere.

  4. berkeley water filter wrote on :

    Maybe it doesn’t pick up the data correctly from version to the next and there’s a slight blip.

  5. Berkey Water Filter wrote on :

    Eric’s model is very close to the trend in which the visitors to my site are doing. A large portion of them are still using IE but there is an uptick in Firefox users.