The primary concern from a research standpoint is that you don't know whether other subjects would revise their previous answers if asked. Second, of course, is the concern that retrospective ratings usually are less reliable. 

In practical terms, in this case I'd probably change his first answer and add a note to the data you eventually publish, but leaving the original answer and adding a note to the data would perhaps be more defensible.

I suppose in the long run you will want to decide how you will deal with such post-hoc revelations, and include it in your lab's Standard Operating Procedures (http://www.columbia.edu/~wl2513/sop-safety-net.pdf). 

http://www.columbia.edu/~wl2513/sop-safety-net.pdf

Hi Shan, I agree with Kevin - always difficult to know what all persons might otherwise have indicated. As this is an observer rated scale, it could also be suggested that the interviewer was not skilled enough to probe. Personally, I would not change the data, if you have a large enough sample size variations such as this should equal out... you can make a note in the discussion, as often happens when we have clinical samples we do not sample people at 'their worst', as often they are too ill to be interviewed/answer questionnaires at that time. Best of luck with it. Kate

I agree with Kate here. I work in clinical trials and the retrospective change of any psychometric scores is ethically fraught. It should be the responsibility of the rater to adequately probe, clarify, follow-up etc. so that such minimization would be detected in the first instance. I agree that this should be noted in the discussion, that the score not be changed and that if the study is adequately powered this should not make a significant difference. I might also add that if you believe that this subject was potentially overall a poor historian and this can be documented, that their data be excluded.