Re: Covariate Models Using Weight

Date: Tue, 23 Nov 1999 09:14:00 +0000 From: James <J.G.Wright@ncl.ac.uk> Subject: Re: Covariate Models Using Weight Dear nmusers, I think everyone in this (extremely long) thread believes in using prior knowledge in some way. The problem is that this is inevitably a subjective process and having done some work with renal function in cancer patients I can safely say that none of the existing formula work terribly well in this population. As such, I don't want to use them (because of population specificity, problems with the assay method etc.). If you happen to be working in a population you consider exchangeable with that used in the original study then maybe you could use their information, but the way to do this may well be with prior that acknowledges uncertainty. However I can here the frequentists calling if you have a big sample the prior doesn't matter and if you have a little sample you will get biased results - it depends on the quality of your prior information. To return to Steve's original question - if both fit the model equally, then this probably isn't a big issue (not that you'd believe it). Personally I would stick to the covariates which come from a clearly defined source, rather than putting age & weight in a predefined formula and then interpreting this to mean their effects on renal function are completely accounted for. James PS Is this the longest thread ever on the nmuser list?