topic Re: Effect Plot f or GEE Negative Binomial Rate Model in Statistical Procedures

Effect Plot f or GEE Negative Binomial Rate Model

telligent — Mon, 26 Aug 2019 21:20:54 GMT

I would like to make plots like the ones in this post:

https://blogs.sas.com/content/iml/2016/06/22/sas-effectplot-statement.html

but I have 2 problems

1) I have an offset and a binomial distribution and these charts only work for Poisson. Binomial fits better based on qic, given my robust standard errors can I just go with poisson to get my to chart output?

2) My x variable is is not really continuous, it is a categorical time variable and SAS only wants continuous variables on the x-axis.

Is there a way to make a plot like with an output dataset and sgplot.

Re: Effect Plot f or GEE Negative Binomial Rate Model

Ksharp — Tue, 27 Aug 2019 11:40:40 GMT

Since it is written by Rick. Calling @Rick_SAS

Re: Effect Plot f or GEE Negative Binomial Rate Model

Rick_SAS — Tue, 27 Aug 2019 11:49:16 GMT

1. Please post the SAS code you are using.

2. The X variable does not have to be continuous, See the documentation for the EFFECTPLOT statement for examples of plotting the response versus a CLASS variable.

Re: Effect Plot f or GEE Negative Binomial Rate Model

telligent — Tue, 27 Aug 2019 16:35:53 GMT

proc genmod data=whatever;

class time;

model y =time a b c time*a time*b time*c/link=log dist=nb offset=log_n;

effectplot fit(x=time plotby=a) /moff ilink; *doesn't work;

effectplot slicefit(x=time sliceby=b) /moff ilink; *doesn't work;

run;

Could I switch to dist=poisson just to get my charts to run even though dist=nb gives me a much lower qic, given that GEE has robust standard errors? Are they robust to choice of poisson or Negative binomial?

I could also switch my ordinal time to continuous to make my charts output, but I have no idea what that would do statistically.

Re: Effect Plot f or GEE Negative Binomial Rate Model

Rick_SAS — Tue, 27 Aug 2019 16:57:24 GMT

Look at the table on this page of the documentation.

It shows which options and suboptions support CLASS variables.Try the INTERACTION plot, which will plot the response versus levels of the CLASS variable, as shown in this example.

Re: Effect Plot f or GEE Negative Binomial Rate Model

telligent — Tue, 27 Aug 2019 17:29:34 GMT

I get this error.

I use this option because I want the y axis to be rates.

ERROR: The MOFF option is only available for Poisson regression with an offset variable and the log

link.

I get this error when I remove the moff option:

ERROR: Continuous variables are not allowed in the X= option.

These errors result from using:

effectplot interaction(x=time*a);

So, I don't think I'm asking a not reading the documentation type question. I am asking for an alternate way to do what SAS doesn't automatically do or how to modify my data to make it do what SAS automatically does.

Re: Effect Plot f or GEE Negative Binomial Rate Model

telligent — Wed, 28 Aug 2019 03:47:17 GMT

I was able to make something close using output from the lsmeans and the at command.

Re: Effect Plot f or GEE Negative Binomial Rate Model

Rick_SAS — Wed, 28 Aug 2019 09:32:00 GMT

Great! Could you please share the code you used so that future people who have this question can find the answer?

Re: Effect Plot f or GEE Negative Binomial Rate Model

telligent — Thu, 29 Aug 2019 01:43:26 GMT

ods output lsmeans=lsmeans;

lsmeans time/at a=5 exp;

lsmeans time/at a=10 exp;

lsmeans time/at a=15 exp;

etc. . . for as many values of a as needed.

Repeat for b and c.

proc sgplot data=lsmeans;

vline time/response=y group=a;

run;