I can't remember any specific references I'm afraid. I see it as part of the general human problem of effectively measuring outcomes. The more important results of a particular measure are, the more likely is that this measure will be successfully gamed. If school test results are used to set teachers pay/school funding, then it's more likely that childrens education will be distorted in a way that leads to higher test scores, but problems in other areas. (I was reading a recent piece on Washington's school systems problems with this). If those designing a CBT programme for depression know the sorts of questions that will be asked to asses the impact of their treamtnet, then they will inevitably focus their efforts upon changing the answers to these questions... but this can undermine the ability of these questionnaires to accurately measure what it is we think is truly important. Sorry not to have any links to solid evidence or research. When I next stumble upon some I'll try to remember to post it here.