proc corr/reg by group

[lizzy28]

Hi community,

I am trying to examine correlation between two variables by group, and there are over 100 groups. What would be a quick and efficient to do that rather than running corr/reg group by group?

For example, I have variables group, var1 and var2 in data "have". There are over 100 groups (1-150), and each group has around 50 observations. I need to get correlation between var1 and var2 by group. My dumb way is:

proc corr data=work.have;

     var var1 var2;

    where group=1;

run;

Thank you all.

Lizi

[Reeza · Accepted solution]

proc corr data=work.have;

     var var1 var2;

    BY GROUP;

run;

[lizzy28]

I guess my question should be: is it possible to format the correlation coefficient result as below:

Group	corr_coeff
1	corr1
2	corr2
3	corr3

....

up to as many groups as I have.

Thanks again!

[SteveDenham]

If you use BY group processing, then adding an ODS output statement should accomplish what you need:

proc corr data=work.have;

     var var1 var2;

     BY GROUP;

     ODS OUTPUT PearsonCorr=PearsonCorr;

run;

You can then post process the dataset pearsoncorr to contain only what you need.

Steve Denham

[lizzy28]

You guys are AWESOME!

Thank you very much!

[SteveDenham]

is awesome.  I just read the documentation really, really fast.

Steve Denham