Steve, I had hoped to provoke a reaction, not an abreaction from you.
My hyperbole was a failed experiment in rhetoric that had the unanticipated and unwanted effect of provoking your ire.
I did not rubbish multiple imputation or the statisticians that use it. And I certainly had no intention of implying that use of multiple imputation is tantamount to fraud.
I did rubbish a paper that used muliple imputation without explaining why there was so much missing data, what imputation methods they used, and what effect this had on the results. I was simply warning about "garbage in, garbage out" combined with blind faith in computer printouts.
Even cleverer people than me can make a mistake in analysis or be blind to obviously bizarre results caused by a mistake in analysis or corruption in data. This does not prove that they are stupid. It proves that they are human. And, that it can happen again, to anyone, at any time. This is why we should be suspicious of studies that use mathematical models without allowing people to inspect the model, its assumptions, and the data fed to it.
I have been picked out a few times in this thread for not mentioning stuff I took for granted or left out in the interests of brevity. The most serious allegation is that I would not similarly criticise a similar trial of a real medical intervention. This is not true. I do not want to mention examples from work, but will recount a recent experience I had as a patient.
The surgeons who offered me spinal injections for sciatica were offended when I asked them why they were offering me a risky treatment despite the body of evidence suggesting that it is not effective. My view is that patients accepting such treatment would be getting nothing more than social grooming with some horrible risks. The ideal would be for doctors (and naturopaths) to get truly informed consent for participation in a trial whenever they offer treatment that is without good evidence to support it. But, I guess this won't happen in my lifetime.
Iain Chalmers and Paul Glasiou were moved a few years ago to plead for less waste in the production and use of research evidence - they classified the problems, but did not quantify the waste.
http://www.ncbi.nlm.nih.gov/pubmed/19935040
This and the more recent multiple exposés of industry's control of what gets published show that wasteful medical research is more common and more problematic than research into complementary and alternative medicine.
I trust (without irony or hyperbole) that this makes you feel better.
Michael
|