Unmatched conditional logistic regression, possible?

[Tom Hsiung]

I don't have matched pairs.

. clogit firsttherapeuticinrreached i.gendercode i.ageforttest i.af i.hypertension i.nonaaornona
> nona i.chf i.potentialamiodaroneddi i.ttestindication i.ttesttargetinr i.afandwarfarinhistory
group() required
r(198);

[Mike Lacy]

The whole point of a conditional logit model is that the observations are grouped (by design or otherwise), and you wish to control for the explanatory variable that defines the groups. If you don't have matching of some kind, then -clogit- would not be applicable. If you explain your situation a bit, and what makes you think you need to use -clogit-, you'll likely get better advice.

[Tom Hsiung]

Some people say conditional logistic is preferred than unconditional when number of independent variables is relative large to number of observations. In addition, under some situations, unconditional logistic regression results in a much greater OR than conditional logistic regression.

[Phil Bromiley]

You don't seem to have taken Stata's response and Mike's comment sufficiently seriously. Without a meaningful grouping variable (usually something like panel effects where you want to control for characteristics that are equal within groups), you can't even estimate a conditional logit. So, if you had a pile of variables that don't vary within groups (panels) and what you do care about does vary within panels, then I guess a conditional logit would be a way to avoid including all those non-varying variables. However, it does it at the cost of all the parameters for groups, so I'm not sure it has an efficiency gain. [It does control for variables that don't vary within panels that you haven't measured which can be an advantage.]