Forward stepwise logistic regression using NHANES 2009-2010 data

[Dianne Devine]

Hello

Can you you help please.

I am trying to set up forward stepwise logistic regression (LR) using NHANES 2009-2010 data. My dependent variable is (Y) is anyohpv (Any oral hpv). I have 20 independent variables (X)

I have reviewed the NHANES website but can't find any information on setting up forward stepwise LR model other than the the simple logistic regression which continues to add each independent variables (see below example). Is this method sufficient to use?

Also I would like to use good-of-fit test but understand I cannot use Hosmer-Lemeshow due to survey sampling design. Can you please offer any help with command/s I could use after fitting the forward stepwise LR model please. Any help is greatly appreciated! Thank you Dianne

Commands used in Stata for forward stepwise LR model

xi: svy, vce(linearized): logistic anyohpv

xi: svy, vce(linearized): logistic anyohpv i.gender2

xi: svy, vce(linearized): logistic anyohpv i.gender2 i.ridreth1

xi: svy, vce(linearized): logistic anyohpv i.gender2 i.ridreth1 i.education

[Richard Williams]

First off, unless you are using an ancient version of stata, don't use xi. Just use factor variables. Type -help fvvarlist- or better yet read section 11.4.3 of the User Guide.

To be clear, you don't want mindless stepwise selection, right, where the computer picks the variables for you? svy and sw don't get along. I would describe your approach as hierarchical.

You could add commands like

testparm i.education

if you want to test whether the education dummies, as a block, are significant.

Not everybody like Hosmer-Lemeshow anyway so not being able to do it may be no great loss: http://www.statisticalhorizons.com/hosmer-lemeshow

Do you really need all these intermediate models? If you want to, say, see how the effect of gender changes as you add more variables, there are potential problems with doing that in logistic regression. See http://www3.nd.edu/~rwilliam/xsoc73994/Nested02.pdf

[Dianne Devine]

Thank you so much Richard for all your help! Kind regards Dianne

[Dianne Devine]

Hello again Richard, I have used below command but receive error message option table not allowed

estat gof, group(8) table

I'm not sure how to sort this as I would like table.

Any help would be greatly appreciated thanks! Dianne

[Richard Williams]

http://www.stata.com/manuals13/restatgof.pdf. It says estat gof is not appropriate with the svy prefix. If you were using svy then the error message is not as clear as it could have been.

[Jeff Pitblado (StataCorp)]

That is a documentation oversight on our part.

We'll need to update that manual entry to point to a corresponding entry in the
SVY manual instead.

If you type

help estat gof

in Stata you should be directed to a table of choices. The one Richard references is
for estat gof after logistic which was not designed to be used with svy
estimation results. The one Dianne is using is documented in the SVY manual.

In Stata type

help svy estat

and search for estat gof. This estat gof available after svy does not
support the table option, the test statistic is computed differently that in the
non-survey situation.

[Dianne Devine]

Thank you very much Jeff for your help! Kind regards Dianne