topic Re: How to find if marginal effect is significant? in Statistical Procedures

How to find if marginal effect is significant?

BobSmith — Wed, 21 Sep 2016 19:46:07 GMT

I have a model like so:

proc surveyreg data=table1;
	class year industry;
 	model estimate = var1 var2 var1*var2 var3 var4 year industry /solution ADJRSQ;
run;

My results show var1 is significantly positive and var1*var2 is significantly negative. I want to find if the net effect on var1 (var1+var1*var2) is significant. How do I do that?

Thanks!

Re: How to find if marginal effect is significant?

PGStats — Wed, 21 Sep 2016 20:59:11 GMT

You will have to decide for which value of var2 you want to estimate the marginal effect of var1.

In the following analysis, the effect of height is estimated at age=12 :

proc surveyreg data=sashelp.class;
class sex;
model weight = height height*age sex / solution;
estimate  'Marginal Height at Age=12' height 1  height*age 12;
run;

Re: How to find if marginal effect is significant?

BobSmith — Wed, 21 Sep 2016 21:41:11 GMT

@PGStats: Thank you. So if var2 is a binary variable and I'm looking for when var2 = 1, I would do something like this:

estimate 'Marginal Effect' var1 1 var1*var2 1;

Can you please explain to me why we are setting var1 (or height, in your example) to 1? Is this essentially testing the hypothesis that var1 = var1*var2?

Re: How to find if marginal effect is significant?

PGStats — Thu, 22 Sep 2016 03:08:06 GMT

Estimate 'Marginal Effect' Var1 a Var1*var2 b;

will estimate the quantity a*Var1 + b*Var1*Var2

where Var1 and Var1*Var2 are regression estimates.

The marginal effect of Var1 is the change in the dependent variable that is predicted when Var1 is changed by 1 unit and everything else stays the same. When Var2 = 1, the predicted change in estimate is Var1+Var1*Var2 (i.e. a=1, b=1).

The hypothesis that is tested is Var1 + Var1*Var2 = 0. You can change the right hand side of that hypothesis with the TESTVALUE= option.

Re: How to find if marginal effect is significant?

BobSmith — Sat, 24 Sep 2016 06:43:29 GMT

@PGStats Brilliant, thank you.