A summary(z.out) does not tell me the increase in
deviance when one of
the predictors would be removed from the model, as drop1() does, and
as is often of interest to applied researchers.
This information is different from the p-values of the seperate
parameters, because multiple parameters can be estimated for one
single term, for example when either the term itself or dependent
variable is a factor with more than 2 levels.
Jeroen
On Sun, Sep 28, 2008 at 6:54 PM, Kosuke Imai <kimai(a)princeton.edu> wrote:
even if anova() command does not work for mlogit, all necessary
informaition
(residual deviance and degrees of freedom) is available in
summary(z.out1).
Kosuke
--
Department of Politics
Princeton University
http://imai.princeton.edu
On Sat, 27 Sep 2008, Jeroen Ooms wrote:
> Thank you very much. I was assuming the as.factor() was not necessary
> when the variable already is a factor, but I should have read more
> carefully.
>
> Another question about the mlogit model: is it possible to perform
> some kind of variance analysis on the predictors of the model? For
> example, if I fitted the model <COLOR~AGE+RACE, model=mlogit> in which
> color is someones favorite color, and I wish to answer the question:
> does race significantly predict someones favorite color? So I would
> like to do an omnibus test for all parameters fitted for the term
> 'race', which are (color-1)*(race-1) parameters in the mlogit case if
> I'm not mistaken.
>
> For normal and logit model I implemented this using the SS type 3,
> using the drop1 function (which calculates the decrease in LR when the
> term would be dropped from the model). For example:
>
> data(mexico);
> options(contrasts=c("contr.sum", "contr.poly"));
>
> z.out1 <- zelig(age ~ pristr + factor(othcok) + othsocok,model =
> "normal", data = mexico)
> drop1(z.out1,attributes(z.out1$terms)$term.labels,test="Chisq");
>
> z.out2 <- zelig(as.factor(vote88) ~ pristr + factor(othcok) +
> othsocok, model = "mlogit", data = mexico)
> drop1(z.out2,attributes(z.out2$terms)$term.labels,test="Chisq");
>
> anova() and aov() give similar errors for mlogit models. Is there any
> other way to perform a per-term variance analysis, or doesn't this
> make any sense?
>
>
> Thank you!
>
> Jeroen
>
>
>
>
>
>
>
> is it possible perform some kind of variance analysis on the
> predictors of a fitted mlogit model?
>
> On Fri, Sep 26, 2008 at 5:54 PM, Ferdinand Alimadhi
> <falimadhi(a)iq.harvard.edu> wrote:
>>
>> Hi Jeroen,
>> Can you try to use :
>>
>> mymodel<-
zelig(as.factor(RACE)~AGE+REGION,data=US1991,model="mlogit")
>>
>> Zelig needs to know when the same formula is used for each level
>> (as.factor()) or different formulas for each level (id())
>>
>>
http://gking.harvard.edu/zelig/docs/mlogit.pdf
>>
>> Thanks,
>> Ferdi
>>
>>
>> Jeroen Ooms wrote:
>>
>> The mlogit model returns an error:
>>
>>
>>
>> mymodel<- zelig(RACE~AGE+REGION,data=US1991,model="mlogit")
>>
>>
>> Error in attr(tt, "depFactors")$depFactorVar :
>> $ operator is invalid for atomic vectors
>>
>> I think it is a problem with the VGAM package. According to this topic
>> <http://tolstoy.newcastle.edu.au/R/e3/help/07/12/5772.html> using $ on
>> an atomic vector raises a warning sinsce R2.5.0 and an error since R
>> 2.7.2.
>> However, the VGAM package has been updated at 2008-05-14 so I
>> dont understand why it would not be fixed. Am I doing something wrong?
>> I am using R version 2.7.2 (2008-08-25), VGAM_0.7-7, Zelig_3.3-1.
>>
>> VGAM is giving a lot of warnings when it is loaded, could there be
>> some conflict?
>>
>> Attaching package: 'VGAM'
>>
>> The following object(s) are masked from package:splines :
>> bs,
>> ns
>>
>> The following object(s) are masked from package:boot :
>> logit,
>> simplex
>>
>> The following object(s) are masked from package:stats :
>> biplot,
>> coefficients,
>> deviance,
>> df.residual,
>> effects,
>> fitted,
>> fitted.values,
>> poly,
>> predict,
>> resid,
>> residuals,
>> weights
>>
>> The following object(s) are masked from package:graphics :
>> persp
>>
>>
>> The following object(s) are masked from package:base :
>> identity,
>> scale.default
>>
>> -
>> Zelig Mailing List, served by Harvard-MIT Data Center
>> Send messages: zelig(a)lists.gking.harvard.edu
>> [un]subscribe Options:
http://lists.gking.harvard.edu/?info=zelig
>> Zelig program information:
http://gking.harvard.edu/zelig/
>>
>>
>
> -
> Zelig Mailing List, served by Harvard-MIT Data Center
> Send messages: zelig(a)lists.gking.harvard.edu
> [un]subscribe Options:
http://lists.gking.harvard.edu/?info=zelig
> Zelig program information:
http://gking.harvard.edu/zelig/
>