topic Re: 2 stage regression in Statistical Procedures

2 stage regression

Jahanzaib — Tue, 13 Dec 2016 03:23:57 GMT

I want to run second regression from the predicted value of first regression. I have made it on this paper in order to understand it. Bdr is the dependent variable in first regression and in the second stage the value from first regression is used as predicted bdr as regressor.

Re: 2 stage regression

PGStats — Tue, 13 Dec 2016 03:35:48 GMT

Use an OUTPUT statement in the first regression to get the predicted values into a dataset and use that dataset to run the second regression.

Re: 2 stage regression

Jahanzaib — Tue, 13 Dec 2016 04:01:15 GMT

Can you explain more how to connect output from first regression with second regression.

Re: 2 stage regression

Reeza — Tue, 13 Dec 2016 04:22:55 GMT

Regression from step 1 creates a data set called reg_results.

Pass that to your second regression.

Proc glm data=SASHELP.class;

model weight = age height;

output out=reg_results p=Predict r=Resid;

run;quit;

proc glm data=reg_results;

model = ...;

Re: 2 stage regression

Jahanzaib — Wed, 14 Dec 2016 09:08:24 GMT

I have used following regression but it gives an error as shown in Picture.

Proc glm data=want1;
model bdr= fam EBITDA_TA MTB LNTA FA_TA RD_TA stdd;
output out=results p=Predict r=Resid;
run;quit;

proc glm data=results;
model cum= predict(bdr) EBITDA_TA MTB LnTA LnTA2 FA_TA RD_TA stdd assmat;
run;

Re: 2 stage regression

Reeza — Wed, 14 Dec 2016 10:03:17 GMT

Predict(bdr) -> what is this supposed to represent?

I don't think it does does what you think it does.

Re: 2 stage regression

Jahanzaib — Wed, 14 Dec 2016 10:09:54 GMT

predict(bdr) is the predicted value of bdr which i suppose to get from first level regression. then use that in the second regression as predicted value of bdr.

Re: 2 stage regression

Reeza — Wed, 14 Dec 2016 10:25:10 GMT

As I said, it's not doing what you think.

Review your OUTPUT statement from the previous step.

P=Predict -> what do you think this portion does?

Re: 2 stage regression

Jahanzaib — Wed, 14 Dec 2016 10:27:26 GMT

i think it will predict all value of all the variables. moreover i do not understand r=Resid.

Re: 2 stage regression

Reeza — Wed, 14 Dec 2016 10:43:59 GMT

Like I mentioned, not what you think.

It it creates a predicted variable called PREDICT. The residuals (predicted - actual) are stored in a variable called Resid.

Open the dataset and examine if manually or run a proc contents and explore the variables there.

Proc contents data=reg_results;

run;