Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: checking model fit with svy estimation on

From   Steven Samuels <[email protected]>
To   [email protected]
Subject   Re: st: checking model fit with svy estimation on
Date   Thu, 17 Feb 2011 23:08:58 -0500

See: and  Also 
-estat gof- works after svy: logistic.  


To understand the difference 

On Feb 17, 2011, at 6:20 PM, Cheney wrote:

I am having some troubles checking if my model is good (goodness of fit, no
influential outliers etc) and comparing model 1 to model 2 (with other iv's
or w iv's expanded), because of the limitations in stata for post est with
(eg.   svy: logistic outcome iv1 iv2 iv3 iv4 iv5 iv1*iv3) My outcome is
binary and my IV's vary between categorical and binary (gender). I also know
you can test the model to see if you should have the iv's expanded into
dummy variables, I think you run one model, then another (with vars
expanded) then lrtest (I think- i haven't done it for a while) but you can't
do this with svy. I know you can do something with a wald test- but I don't
fully understand how to interpret that. Anyone have any suggestions? I need
a goodness of fit test... 

What I have done so far:
checked for interactions- found then added to the model.
checked for collinearity- found, centered vars and re ran model with
centered vars.
checked for outliers- (plotted pearson resid's) but I couldn't estimate
deviance residuals (because of svy) so I'm not sure if just one plot is
enough? I had only one outlier.
specification error- i did a linktest which gave me this output:
       _hat |   1.145419    .274488     4.17   0.000      .607336   
     _hatsq |   .0298467   .0544568     0.55   0.584    -.0769059   
which might mean that i don't have a problem with specification error. but I
wanted to do the 'boxtid' to check for non linear relationships, however
this is not supported with pweights... so I just ran the regression without
the svy etstimation and ran the boxtid. I ended up with non linear reasults
coming up, but none of them significant (in the example online they are
significant of course) so I didn't know if this meant they are non linear
but not to an extent that matters?
age      |   .4471324   .0736975      6.067   Nonlin. dev. 0.241   (P =
      p1 |    2.85128   3.722516      0.766

Any other suggestions for checking the model is sound?
thanks in advance!

View this message in context:
Sent from the Statalist mailing list archive at
*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index