RE: st: regression using generalized linear model

[Joe Canner <jcanner1@jhmi.edu>]

I think you mean "independent" variables. (The dependent variable in an ANOVA is usually continuous.)

I'm not sure what it means to have a non-integer classification variable.  I suspect -glm- will be a better choice, but I defer to others with more experience in this area.

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Trung (Aidan) Nguyen
Sent: Tuesday, March 11, 2014 3:06 PM
To: statalist
Subject: Re: st: regression using generalized linear model

Thanks Joe and Nick,

I used the encode command to turn the class variable 'a' into a
numeric value. However, for anova command to work, all the dependent
variables must be of integer value. If I assigned integer values for
all the dependent variables, then it would not lead me to the same
results I get from the GLM command in SAS. Yes, I did use the # for
the interaction effects in my STATA code. Would you suggest the glm
command in STATA?

x is the dependent var
a and b are the classification effects

Thanks.



On Tue, Mar 11, 2014 at 12:51 PM, Joe Canner <jcanner1@jhmi.edu> wrote:
>
> Actually, Aidan provides us with most of the answer in his post.  He indicates that variable -a- is a "string character" and that -u- "contains non integer and negative values".
>
> Nick has suggested a fix for string variables (-encode-). -destring- may also work.  I don't remember enough about PROC GLM to know how u*u should be translated into Stata.  If the asterisk is an interaction, perhaps a hash(#) would work better in Stata.
>
> Incidentally, how did Aidan get the asterisks to work at all in Stata?  Is that new in version 13?
>
> Joe
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nick Cox
> Sent: Tuesday, March 11, 2014 2:40 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: regression using generalized linear model
>
> I've never used SAS but I have used Stata [NB].
>
> But the message does explain the problem, which may be why no-one
> answered this first time this was asked.
>
> At least one of your variables -v y u t a b- contains something other
> than integers. So,
>
> . codebook v y u t a b
>
> to see which variable(s) don't come up to scratch. Perhaps one is
> string. Even if SAS permits string predictors (I don't know its
> terminology), Stata expects you to -encode- string variables before
> using them in -anova-.
>
> (Please note our policy on repeating posts. It's best to wonder why a
> post didn't get answered, and revise it, and not to seem too
> impatient.)
>
> Nick
> njcoxstata@gmail.com
>
> On 11 March 2014 18:28, Aidan Trung Nguyen <aidant@mit.edu> wrote:
>
> > I'm trying to write the STATA equivalent of the followingSAS statement:
> >
> > proc glm data=datasetclass a b model x = v*y*y    u*u    t    a*b
> > weight w
> > output out = outputdataset
> >
> >
> > a is a string character
> > b, x, v, y, u, t are numeric
> > u contains non integer and negative values
> >
> > I tried the below command but without success:anova x  v*y*y   u*u  t   a*b
> >
> > error message : factor variables may not contain non integervalues Any guidance is much appreciated.
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/