Moderation / Interaction Effect with Split Sample Comparison

[Julia Bayer]

Dear Statalist members,

I want to examine a moderation effect of gender on the relationship between openess to experience resp. agreeableness and risk taking. Gender is a dummy called female (with 1 "female"). In my Stata seminar, we were taught to analyse an interaction effect through split sample regressions. Therefore, I conducted a regression for the whole sample, a regression for female only and a regression for male only:

*Overall sample:
reg riskabovemean ope agr, robust

*For men (female=0):
reg riskabovemean ope agr if female==0, robust

*For women (female=1):
reg riskabovemean ope agr if female==1, robust

Now I don't know how to interpret the outcomes below:

riskabovemean	overall sample		sex=0 („male“)		sex=1 („female“)
	Coef.	P>|t|	Coef.	P>|t|	Coef.	P>|t|
Openess	0.0598968	0.000	0.0698625	0.000	0.0390326	0.017
Agreeableness	0.0450676	0.000	0.0550326	0.000	0.0359563	0.025

Can I say that gender is a moderator? And if yes, does gender weaken or strengthen the relationship between openess to experience resp. agreeableness and risk taking?

Thank you so much!

[Bruce Weaver]

Hello Julia. If you include the female*ope and female*agr interactions, you'll get the comparisons you want in a single model. Try this:

regress riskabovemean c.ope##female c.agr##female, robust

The coefficients for the product terms will give you the differences between males and females.

Finally, the name of your outcome variable, riskabovemean, caught my eye. A variable name like that suggests (to me) a dichotomous (Yes/No) variable. If it is a dichotomous variable, then you should be using some other type of model--e.g., logistic regression.

HTH.

[Julia Bayer]

Thank you Bruce! Well, the method with the interaction term is also what I know from other statistical programs. I now got the following results:

riskabovemean	Coef.	P>|t|
ope	0 .0696922	0.000
1.female	-0.0633182	0.002
female#c.ope 1	-0.0267155	0.172
agr	0.0530806	0.000
female#c.agr 1	-0.0151558	0.435

So just to be sure: "female#c.ope 1" and "female'c.agr 1" give me the possible interaction effect. But as p>0.1 for both, the effects are not significant and thus gender is not a moderator?

[Bruce Weaver]

Re #3: Right.