Darren L Dahly PhD oMG FFs jFcICYMI:
Out of balance
An essay on covariate adjustment in randomized controlled trials in medicine.
Bill GardnerWell done.
Also, much respect for your credentials: "PhD OMG FFS JFC".
I have the first, but to whom do I apply to certify the last three?
Luisa N. Borrell@statsepi Love your 0.02!
Peter Tennant, PhDYou've lost me. Isn't balancing the outcome propensity asymptotically identical to balancing the determinants of the outcome?
Following that, in a finite world, can't you get a better approximate of perfect balance in the outcome propensities by conditioning on the determinants of the outcome?
Maybe I'm being moronic. It certainly seems like we disagree on the utility of conceptualising confounding error as the random analog of confounding bias.
Matt Hodgkinson@statsepi "So if I put my finger on the scale and systematically steer the heavy smokers into one of my trial arms for a study of lung function, then that’s a bias. However, if I randomize, and I happen to get a disproportionate number of them one arm, then that’s just dumb luck — an error whose nature will change from one replication to another."
OK, but how can someone other than the triallists tell if the imbalance arose through bias or chance? #ClinicalTrials #Statistics
@mattjhodgkinson It's why we should put a great deal of effort to ensure foolproof allocation concealment.
@statsepi If trials claim good allocation concealment but that's false due to error or deliberate bias, ignoring imbalance because we assume good faith puts us at the mercy of the incompetent and the malicious when analysing those trials.
@mattjhodgkinson This is true for a gazillion other parts of the research process, so i'm not really sure what argument is being made here.
@statsepi If there's a significant baseline imbalance between arms in a factor known to affect the outcome, that's a potential indicator of a serious problem with the trial. Yet the advice from statisticians is... shut up and calculate?
@mattjhodgkinson No, who said that?
@statsepi Does significant baseline imbalance in a factor known to affect the outcome matter?
If yes, how does it affect the analysis?
If no, why not when it is consistent with serious error or fraud?