They based this on an interestingly small amount of data. The 5,318 Google employees, for example, is a pretty small fraction of the 49,829 employees they had in Q1 of 2014.
Stats 101: Sample size alone describes only variance in measurement and I think even the sample size of 1000 would be pretty acceptable in this case. Whether these numbers are biased or not depends on if sample was drawn randomly (i.e. if it reflects actual distribution).
If you look at the text between Microsoft and IBM, it seems to imply that the number 5318 is just the sum of the top 5 universities that feed to Google. They are only considering the top 5 feeder universities for each company.
It doesn't really matter how many employees google has, if you just interview a few hundred-a few thousand chosen at random you will get good estimates of the fractions.
sytelus|11 years ago
skwirl|11 years ago
im3w1l|11 years ago