Monday, April 19, 2010

"Unadjusted model." Why? [methods rant]

performing yet another manuscript review provokes a methodology rant:

technical note: why report an unadjusted analysis? if your study requires adjustment for confounders / covariates, then it simply makes no sense to report the results of your study BEFORE you have gone and adjusted for the fact that you know it has biases?


i guess in an elliptical way, you could argue that the difference between the unadjusted and the adjusted analyses tells you the degree of prediction due to the adjusters / the covariates.
but if this is important, then just include them in your results table. report their weight, odds ratio, etc.

and give a "p" level if you feel like you need to.
if covariates, as a block, are interesting, then step them in on their own step in a hierarchical model.
step in demographic measures in one step.
step in comorbid conditions in the next.
step in psychosocial in another.
there. done. you have your info. done the proper way.

but reporting "unadjusted," and "adjusted," as if we are seeing things in various ways to round out our view? i don't get it. is it like when you get your eyes examined? which is better. this? ....or this (flip). this...(flip) or this?
we know which is better: the adjusted model. you would not run an adjusted model, otherwise. right? is it just me?

No comments: