Anonymous survey, this is an interesting argument but is generally unfounded. In anonymous surveying for studies a margin of error of 5% has been found to be accurate even with questions that are controversial and would damage someone socially should they admit to them. That's not a huge difference, I understand the argument though.

I should point out that the term cheat in this case means to sleep with another person, not to do the various other things that most people also constitute as cheating.

Interestingly the figure is around 5% higher when only looking at the married couples. People that get married are more likely to cheat? Not by much more, but it's an interesting statistic.

I say 22%, that's from the most recent I could find, looking at studies on the whole it looks like it ranges from around 14% to 35%. 22% still sits in the middle though. I'm more inclined to say that the average of all tests is probably closest to the correct figure however I generally always look at the most recent studies as figures like this can change very rapidly, if there had been studies 50 years ago I'd bet the figures would be much lower.