I wonder what the studies considered an error. Are syntax errors counted, or did it focus on runtime bugs? I imagine that syntax errors would really skew the results if they were counted.
This is a good point. Methods matter. I tracked down the exact references:
[0] A. R. Brown and W. A. Sampson, Program debugging: the prevention and cure of program errors. Computer Monographs, 18. 1973
[1] Thomas J. Ostrand and Elaine J. Weyuker, Collecting and categorizing software error data in an industrial environment. Journal of Systems and Software, Volume 4, Issue 4. 1984
I couldn't actually get a hold of [0]; however, [1] was easy to find, and it goes into great detail about it's categorization of "faults". In fact, the abstract explicitly mentions that it catogorizes and analyzes 173 faults. On page 7 (page 295 of journal), we finally get to your question:
"Of these 171, 15 were attributed to clerical mistakes, such as incorrect typing or misunderstood handwriting."
So 15/173, or about 8.7%, of the faults they found were of the kind you mention. Interestingly, from earlier on the same page, "... 12 represented compiler or operating system faults," so only 20% less frequent than "clerical errors".
This paper also compares with a lot of other studies, so it's worth reading if you're interested.
