On the Choice of Browser and Numerical Intelligence

or: Chrome Users are Smartest, then Firefox Users, then IE Users

You can download the detailed analysis here (PDF, 4 pages). Feel free to use any of the information on this page or in the paper (but do refer to the source of course).

We analyzed usage data of the www.calcudoku.org number puzzle website for the years 2010 and 2011, consisting of over 1 million solved puzzles, attempting to determine the numerical intelligence of users of Internet Explorer (IE), Firefox, and Chrome (insufficient data was available for users of other browsers).

We found that for puzzles "against the clock" (or timed puzzles):

1. Chrome users were faster than Firefox and IE users

Average solving time as a percentage of the Chrome average (so smaller is better)

2. IE users were more likely to give up on a puzzle than Firefox or Chrome users

Average percentage of abandoned timed puzzles (so smaller is better)

These differences were found to be statistically significant (p = 5%). Based on the fact that Chrome users solve Calcudoku number puzzles the fastest, and that IE users give up on solving them the most, it appears that Chrome users have the highest numerical intelligence, followed by Firefox users, then by Internet Explorer users. Note that it does not follow that using Chrome makes you smarter, for example ("correlation does not imply causation"). Also, we can only speculate about the causes of the differences: perhaps Chrome is the browser of choice for more technically inclined people, who tend to have better number skills. And maybe because IE is the default browser for Windows, people who do not choose a different browser possibly are less technically skilled.

Once sufficient data is available, we would like to include Safari users in our tests. It may also be interesting to look at differences between IE versions (6.0 through 9.0), as well as between Facebook and non-Facebook users. We are also planning to publish all usage data on which this analysis is based (after proper anonymization).

You can download the detailed analysis here (PDF, 4 pages). Feel free to use any of the information on this page or in the paper (but do refer to the source of course). Your comments and suggestions are very welcome at

March 4: after review by Slashdot

Many Slashdot readers provided helpful comments, for which thanks! In a next version of the paper I'll report the actual p-values found, the sampling error, look at effect size, and check to what extent the results are affected by the multiple comparison problem.

Patrick Min
London, February 29, 2012

Author biography:
Patrick Min is a freelance scientific programmer. He specializes in geometry software, but has worked in many other areas, such as search engine technology, acoustic modelling, and information security. He has published papers and open source software in several. He holds a Master's degree in Computer Science from Leiden University, the Netherlands, and a Ph.D. in Computer Science from Princeton University. He is also a puzzle enthusiast, devising math puzzles for his father since the age of 7. This continues to this date, with dad solving his son's Calcudoku puzzles. Patrick lives in London.