Unfortunately, I think these studies are a bit naive because the proprietary, data-driven, probabilistic fingerprinting models used by Facebook and LinkedIn (to name two of the most elaborate fingerprinters) are years ahead of anything a few researchers could come up with.<p>1. Get a gazillion users on your site<p>2. Require a user account tied to a real person<p>3. Log IP, host, geolocation, and as many JavaScript/browser APIs as you can (there are hundreds at this point)<p>4. Among the fields you track, find the ones that ones that are the most stable and unique over time<p>5. Assign some probabilities to these fields to eliminate false positives<p>6. Generate personas for users for when they are at home, work, one their phone, etc.