I might be a cheat

One of the more ridiculous arguments when accusing some of of computer cheating is that X% of their moves matched Fritz/Rybka/Fruit etc This is often presented as a bare statistic, with the assumption that the greater the percentage, the greater the likelihood of computer assistance. Such assumptions are risky however, as often they aren't statistically valid. They might seem intuitively appealing, but as examples of Bayesian Inferences have shown, the 'obvious' assumption may turn out to be the wrong one.
For example here is a game I played (as White) last week. I ran it through Fritz 9 and asked it to flag any moves that it considered better than the ones I played, even if the difference in evaluation was as small as 1 centi-pawn. Out of the 34 moves it suggested 8 improvements, although only 4 of those had a significantly better evaluation. Over the final 14 moves I played Fritz's suggestion 13 times.
Of course I did not use computer assistance, merely playing moves that seemed sensible. Granted it wasn't an overly complex game, and I didn't have to calculate too many tactics, but the 'facts' are the 'facts'. So until someone develops a test that eliminates 'false positives' from the data set, colour me unconvinced.

1.Nc3 Nf6 2.e4 Nc6 3.d4 d6 4.Nf3 e5 5.dxe5 dxe5 6.Qxd8+ Kxd8 7.Bc4 Be6 8.Bxe6 fxe6 9.Be3 Bb4 10.0-0-0+ Ke7 11.Bg5 Bxc3 12.bxc3 h6 13.Bxf6+ gxf6 14.Nh4 Rhd8 15.Rd3 Rxd3 16.cxd3 Rd8 17.Kc2 Kd6 18.Re1 Rf8 19.Re3 f5 20.Rf3 f4 21.Rh3 Rf6 22.Nf3 b5 23.Rh5 b4(D)
24.d4 Rg6 25.dxe5+ Kc5 26.Nh4 Rg5 27.Rxh6 bxc3 28.Kxc3 Rxe5 29.f3 Nd4 30.Ng6 Rg5 31.Nxf4 e5 32.Rg6 Ne2+ 33.Kd2 Rxg6 34.Nxg6 1-0