This is a poor way of representing the data. It distorts the data based on the overall population of the area. A better approach would be to look at all the job postings in a given area and measure the percentage that sound like startups.

Remember, that this data representation was only a 'smell test' as they defined it - and for most purposes here that's all that's needed ;)

Here's the original article with much more info:

http://radar.oreilly.com/archives/2006/06/startup_centers.html