I have compared Compete's numbers to actual data from about 150 sites.
On that sample the numbers were never higher than actual and on average the Compete numbers were 2-3x lower than the actual numbers.
So I have used it as a reasonable estimate for half of the traffic a site sees. Although I may be placing my foot in my mouth because the stats for HN don't quite seem accurate: http://siteanalytics.compete.com/news.ycombinator.com/
Are they sites with less than 20,000 uniques per month? I've found those to be the outliers in my data set. They're wildly off on "low" traffic sites but have been a decent yardstick for sites getting more than 20k uniques in my limited experience.
On that sample the numbers were never higher than actual and on average the Compete numbers were 2-3x lower than the actual numbers.
So I have used it as a reasonable estimate for half of the traffic a site sees. Although I may be placing my foot in my mouth because the stats for HN don't quite seem accurate: http://siteanalytics.compete.com/news.ycombinator.com/