What the equation for the model fitted by proc glmselect

[BMI_kid]

proc glmselect data=imputed PLOTS=ALL;
*class NoEvalBus NoEvalComp;
model Responce=&cluster
/ selection=stepwise(select=sl) hierarchy=single
stats=all showpvalues orderselect stb;
score out=predStepwise predicted residual;
run;

If i run this code, how do i interpret the output. I am interested in the equation for the model, and I wonder if it fits the same model as the default model fitted by proc genmod since they both fit "Generalised linear models."

[PaigeMiller]

I wonder if it fits the same model as the default model fitted by proc genmod since they both fit "Generalised linear models."

I think it would be very unlikely (but not impossible) that the stepwise model created by GLMSELECT would match the (non-stepwise) model created by GENMOD. Nevertheless, you can look at the model coefficient estimates yourself and see if it is the same model.

If i run this code, how do i interpret the output. 

You would probably want to take a look at the examples for PROC GLMSELECT in the SAS documentation to see how to interpret the various outputs. If you have a specific question about the output, show us a screen capture of the output and ask your specific question.

[BMI_kid]

I am looking for the equation of the model maybe for instance 

I want to be able to explain the parameter estimates/coefficients and maybe give weights to the effect of the predictors.

[PaigeMiller]

---

@BMI_kid wrote:

I am looking for the equation of the model maybe for instance 

I want to be able to explain the parameter estimates/coefficients and maybe give weights to the effect of the predictors.

---

Please be specific. These beta values are slopes. What else about the parameter estimates do you need to explain? What do you mean by "give weights" to the effect of the predictors?

[sbxkoenk]

Hello,

To be clear:
You don't need that explicit equation

• to understand your model (full interpretation)
• nor to score (in the future) new data with your model.

With respect to the former point:

I'm so used to it by now ... the way SAS presents parameter estimates.

I see the equation (or equations) in front of me like this ... in my head.
It is a matter of getting used to it.

With respect to the latter point:

Some people ask this question (your question) so that, with that fully written out equation, they can build a scoring app in Excel. I (and not only me) would strongly advise against that.

>> Just do the scoring in SAS. Plenty of ways to do this and you are sure results will be entirely correct.

BR, Koen

[StatDave · Accepted solution]

GLMSELECT fits the "general linear model" that assumes that the response distribution is normal and it directly models the response mean. GENMOD fits the "generalized linear model" which allows for any response distribution in a family of distributions and it models a function (the "link" function) of the response mean. If you don't use model selection in GLMSELECT, it will fit the same model as GENMOD when you select the normal response distribution and the identity link function, though different estimation methods are used. 

The estimated parameters (some call them "weights") of the model can be found in the "Parameter Estimates" table from GLMSELECT. The model is then  mean(y)=b0+b1*x1+ ... similar to the quoted text you show.

[Rick_SAS]

It's always best to use example data that everyone can run. Let's use the numerical variables in the Sashelp.cars data set. For the response variable, I will choose MPG_City. For the independent variables, I will choose five numerical variables:

%let cluster = EngineSize Horsepower Weight Wheelbase Length;

proc glmselect data=sashelp.cars plots=none;
   model MPG_City =&cluster
      / selection=stepwise(select=sl) hierarchy=single
      stats=all showpvalues orderselect stb;
   score out=predStepwise predicted residual;
   ODS SELECT  SelectedEffects ParameterEstimates;
run

For this example, two of the five candidate effects were selected for the final model, along with the Intercept term. The OLS model that best predicts the mean response (conditional on the two selected indep variables) is 
MPG_City = 38.34 - 0.00357*Weight - 0.0256*Horsepower