<- file stat 97zeroes.html -> - If there are "severity" scores which are usually Zero for normal, and never zero for patients, then one has to lean *heavily* on assumptions to assume that all differences are not due to severity, if severity relates much to <whatever-it-is>

Normals vs patients: "severity"?

=======================Rich Ulrich, 12 Mar 1997==========ssc Subject: Re: regression question Message-ID: <5g6jqa$r2o@usenet.srv.cis.pitt.edu> <<: rhouts@kent.edu >> Renate Houts (rhouts@kent.kent.edu) wrote: : Hi, : I'm working with a faculty member who would like to compare regression : coefficints computed for a patient group and a control group. The : protential problem is that the patient group has one more control : variable (severity of illness) than the control group. Given the : difference in the number of control variables, is it still acceptable : to compare (via the Chow or similar test) the regressions from the two : groups. How else could this be handled? -- I think you can have a pretty good estimate for 'severity of illness' in the control group by setting its value at zero (that is, whatever is the minimum score). Using zero puts a lot of weight onto the assumption of 'linearity' in the Severity score -- especially if NONE of your patients score at the minimum. The problem is a logical one, as well as a statistical one. That is, I am saying there is no easy or automatic solution, and you will have to argue from your own particular case. IF 'severity' makes no difference to the regression, then of course you could drop that variable (and its interactions? - were they being considered?). *--------

Controlling for severity?

=======================Rich Ulrich, 20 Mar 1997==========ssc From: wpilib+@pitt.edu (Richard F Ulrich) Subject: Re: regression question Message-ID: <5gs17u$ap4@usenet.srv.cis.pitt.edu> Renate Houts (rhouts@kent.kent.edu) wrote: << concerning this previous answer by me, >> : : Using zero puts a lot of weight onto the assumption of 'linearity' : : in the Severity score -- especially if NONE of your patients : : score at the minimum. The problem is a logical one, as well as a : : statistical one. That is, I am saying there is no easy or automatic : : solution, and you will have to argue from your own particular case. << snip >> : Can you elaborate about possible consequences of putting so much : weight on assumption of linearity in the sererity scores? : Thanks, : Renate -- "elaborate about ... consequences ..." ? not much. If you are "controlling for" severity, and there is a big difference between the groups as to severity, then you will control for SOME big difference between the groups, and possibly wipe out a difference that ought to be blamed on Group-membership. Or, if the severities don't overlap, then you can't really tell anything about how the two samples ought to be overlaid onto one picture for comparison -- so you could also UNDERCONTROL for the effect of severity, and conclude that there are differences between Group, despite the fact that better control would have wiped it out. consequences == wrong conclusions, in either direction. * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Document by Rich Ulrich. E-mail to wpilib+@pitt.edu

FAQ top.

Ulrich home page.

Ulrich FAQ. http://www.pitt.edu/~wpilib/stats99.html