Model building in logistic regression and likelihood ratio tests

[Nita Chan]

Hi,
I am using a forward stepwise logistic regression to build a model of independent predictors of my outcome variable. I am a little bit confused as to how I carry out the likelihood ratio tests after each variable I add to the model to decide whether I should keep it or not. I have the command lrtest but I am not sure if I am using it correctly. I am using the following command for the forward stepwise logistic model:
xi: logistic outcomevar i.variable1

Any help will be appreciated.

Kind regards,

[Maarten Buis]

You should not do stepwise to begin with, see: http://www.stata.com/support/faqs/st...sion-problems/

If you still want to do it, then just use the stepwise prefix, see help stepwise.

Also you should no longer use the xi: prefix, just use factor variables instead, see help fvvarlist.

[Nita Chan]

I have to use forward stepwise logistic regression because it is the best way to achieve what I want to achieve. I have looked at the stepwise prefix and it is still confusing to me. What I really want to know is how to carry out likelihood ratio tests and if I have the right command "lrtest" to decide what variables to keep in or exclude.

[Nita Chan]

Has anybody got any useful suggestions for me? The stepwise prefix doesn't work for me and I get an error message of "factor variables and time-series operators are not allowed". I also want to know how to carry out likelihood ratio tests afterwards which would tell me which variables to include and exclude.

Kind regards,

[Richard Williams]

Stepwise doesn't support factor variables so you either need to compute them yourself or use the otherwise horrible xi: prefix. If the categorical variable has more than 2 categories you may wish to force them to enter as a block. So something like

webuse nhanes2f, clear
xi: sw, pe(.05) lr: logit diabetes (i.race) weight

Then see

http://www.stata-journal.com/article...article=dm0032

Which shows you how to take the results from stepwise and feed them into nestreg to get the results you want.

I believe in letting people do what they want. But having said that, I would seriously consider Maarten's argument that stepwise is often/usually a bad idea.