[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Evaluation of Logistic Regression of Complex Survey Data

Subject   Re: st: Evaluation of Logistic Regression of Complex Survey Data
Date   Sun, 5 Jul 2009 10:49:57 -0400


To plot the ROC curve,

 Roger Newson's -somersd- (downloadable from SSC) will compute and
compare the areas under  ROC curves with weighted, clustered data
together.  See caveats at:

To mimic Stata's -linktest-, see:  The
original topic was about -svy: reg- . After logistic regression, make
sure that you generate the linear predictor, not the probability
with "predict yhat, xb"


On Fri, Jul 3, 2009 at 11:52 AM, Hisako Kobayashi<> wrote:
> I am analyzing a logistic regression with  svy command.     As you
> know, with “svy”  many statistics, i.e. dbeta, dx2, ddeviance, ROC,
> etc. etc.   for assessment of model are not calculated in logit
> postestimation, except “svylogitgof”.   This may be a stupid
> question.. but my question is how  I can  assess the fit of svy logit
> regression model?

> Here are two approaches that I can think of:
> 1. According to Hosmer and Lemeshow, one approach is to compare
> design-based analysis with model-based analysis.
> 2. The other approach would be to estimate those statistics only with
> cluster (without pweight) and evaluate them.
> Is there any better approach to evaluate survey logit model?  If no,
> which approach is better?

Steven Samuels
18 Cantine's Island
Saugerties NY 12477

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index