topic Re: proc logistic:class parametrization in Statistical Procedures

proc logistic:class parametrization

kgb — Tue, 09 Jan 2018 12:09:11 GMT

Hello!

I would like to have the odds ratio spcified not according to a reference category of a variable but according to the "overall" risk.

I explain myself with an example:

explanatory variable name: hospital

explanatory variable categories:

hospitalA
hospitalB
hospitalC
hospitalD

response variable:

0=no readmission
1=readmission

N total of observations= 12000;

Pct Readmissions= 3,5%

I would like to obtain the Odds ratios like these:

hospital A Vs overall avarage

hospital B Vs overall avarage

hospital C Vs overall avarage

hospital D Vs overall avarage

...so with a different parametrization from the classic one similar to the following:

hospital A Vs hospital D

hospital B Vs hospital D

hospital C Vs hospital D

if I use the following code I don't obtain results that I want, how should I change it?

proc logistic data=dataset-name PLOTS=all;
class hospital (param=effect );
model response variable(event='1')= hospital
/noint;
run;

thank you very much in advance!

Re: proc logistic:class parametrization

PeterClemmensen — Tue, 09 Jan 2018 12:23:28 GMT

"if I use the following code I don't obtain results that I want, how should I change it?"

What results do you obtain then?

param=effect is the default of PROC LOGISTIC

Re: proc logistic:class parametrization

kgb — Tue, 09 Jan 2018 12:40:12 GMT

like this: OR IC
hospital A Vs hospital D ..... ...... ......
hospital B Vs hospital D ..... ...... ......
hospital C Vs hospital D ..... ...... ......
but
maybe the problem is that I don't change the class statement, but include a contrast or estimate statement for obtaining OR on
hospital A Vs overall avarage
hospital B Vs overall avarage
hospital C Vs overall avarage
hospital D Vs overall avarage?

Re: proc logistic:class parametrization

Reeza — Tue, 09 Jan 2018 16:10:08 GMT

I think, you're referring to the GLM parameterization, you can use PARAM=GLM, but it's also the default I believe.

Then you can take your estimates and exp(estimate) to get the odds ratio.

Or you can use the ODDSRATIO or ESTIMATE statements to build your hypothesis.

Re: proc logistic:class parametrization

StatDave — Wed, 10 Jan 2018 22:57:06 GMT

See this note which compares the log odds for each group to the average log odds.

Re: proc logistic:class parametrization

kgb — Mon, 15 Jan 2018 13:49:57 GMT

thank you very much for your helpful indications!bye!