Ethics

AWB has started an interesting conversation about work and research and personal and professional ethics, and I wanted to take the luxury of my own blog to expand a little on something I said in comments over there. (so go on over and get caught up. we'll wait. back? Ok, good.) One of the things I love about my profession is our set of professional ethical guidelines. (that's a reference to the American Statistical Association; although public health has its own professional society, they have no official ethical guidelines, which I find weird and troubling and is the topic for a whole 'nother post) And right there in section II.A.2. (" Guard against the possibility that a predisposition by investigators or data providers might predetermine the analytic result.") is the reason I take it so personally when someone jokes about 'lies, damned lies, and statistics.' Yes, as another commenter pointed out, this is a crazy ideal to aspire to (I kept trying to re-write that in my head as to which to aspire but it just sounds weird) but for me it's this great black and white rule to fall back on when reporting results to superiors. Some MD asks me to go back and try something else, look at the analyses this way, tweak the data that way (sometimes in ways that feel obviously manipulative, sometimes in more nefarious ways that just seem like legitimate data digging) and I can play the sorry-m'am-just-following-orders card. Numbers are numbers and they either say something or they don't. Yes, they can be interpreted five gazillion different ways, but the numbers themselves never lie, only people do. The Truth of the numbers sets the perfect bar for me to reach for in my interpretations. When I am being the best statistician that I can be I am presenting my analyses and my results and my interpretations in such a transparent way as to be both repeatable by any other researcher and essentially without fault. This especially means avoiding the impression of causality when discussing results when association is all that's present. (this is, of course, easier said than done, especially when the statistician is not the PI and does not have the authority to control all the text of a paper) Sure, my colleagues may question why I chose to categorize a variable a certain way or why I'm using a significance level of 0.1 rather than 0.05, but these are matters of professional opinion. If I find a difference in the average length of hospital stay for black patients versus white patients after comparable illnesses, my goal is for that conclusion to be uncontestable. Certainly, one could suggest controlling for other variables or using a larger dataset, but as far as my specific data shows, that relationship is The Truth. And this is where reporting statistics gets so frustrating. Of course we want to generalize our findings and apply them to other problems. But the only things we can really know about are the things about which we have tangible data. Anything, and I mean anything, beyond that is guesswork. It may be educated guesswork, but guesswork nonetheless. And hence the infuriating hems and haws you get from scientists and the sense the general public often gets that we're never in agreement and we're just making shit up. In reality we're (most of the time, I hope) just trying to be true to our ethics.