0
$\begingroup$

I am trying to identify what combination of independent variables (5 variables) explain a dependent variable and I have a few questions:

1) Is it possible to use multiple linear regression analyses to identify what independent variables predict the dependent variable and discard some variables? In this case, is it O.K to say that variables “a”, “e” and probably “d” are the variables that best explain the dependent variable?

R = 0,898 Rsqr = 0,806 Adj Rsqr = 0,709

                Coefficient Std. Error  t  P    VIF
      Constant       0,312  0,056    5,53    <0,001 
      Variable a    -0,001  0,001   -0,865    0,407 1,640
      Variable b     0,000  0,000    3,928    0,003 1,546
      Variable c     0,092  0,130    0,707    0,496 1,185
      Variable d     0,077  0,037    2,058    0,067 1,795
      Variable e     0,001  0,000    3,312    0,008 1,724

Analysis of Variance: DF SS MS F P

Regression 5 I 0,122 I 0,024 I 8,326 I 0,002

Residual 10 I 0,029 I 0,003

Total 15 I 0,151 I 0,010

2) Is it possible to identify what combination of variables are the best to predict the dependent variables? I mean, can I use the R2 values or the F-values to identify whether variables “b”, “d”, and “e” are better than variables “b” and “e” only to predict the dependent variable (based on the R2 values)? If so, what would be the best approach to do this, use the R2 values only or using any approach like AIC?

Multiple linear regression using variables b, d, and e:

R = 0,888 Rsqr = 0,788 Adj Rsqr = 0,735

            Coefficient Std. Error   t    P     VIF
Constant    0,317   0,050   6,287     <0,001    
Variable b  0,000   0,000   4,060     0,002 1,512
Variable d  0,079   0,036   2,216     0,047 1,788
Variable e  0,001   0,000   3,482     0,005 1,280

Analysis of Variance: DF SS MS F P

Regression I 3 I 0,119 I 0,040 I 14,903 I <0,001

Residual I 12 I 0,032 I 0,003

Total I 15 I 0,151 I 0,010

Multiple linear regression using variables b and e:

R = 0,838 Rsqr = 0,702 Adj Rsqr = 0,656

            Coefficient Std. Error  t     P     VIF
Constant    0,416       0,0263      15,843  <0,001  
Variable b  0,000767    0,000308    2,489   0,027   1,082
Variable d  0,000118    0,0000217   5,433   <0,001  1,082

Analysis of Variance: DF SS MS F P

Regression 2 0,106 0,0530 15,296 <0,001

Residual 13 0,0450 0,00346

Total 15 0,151 0,0101

Thanks!

$\endgroup$

1 Answer 1

0
$\begingroup$

Well the answers for question 1 and 2 kind of go together. There are several SAS options that you can use if you use the SELECTION option in your regression model. These different options can be found here: https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_reg_sect030.htm

These selection models are mostly based on the F-value and R^2.

$\endgroup$
1
  • $\begingroup$ I will check the link and see what can I do in SAS. Thank you so much! $\endgroup$
    – Billo
    Oct 13, 2017 at 0:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.