Step 6: Interaction analysis

Now it is time for the interaction analysis. With this analysis, we answer the fourth research question:

Does the relationship differ between men and women?

Note Remember that the analyses should only be based on individuals with a value of 1 for the pop variable.

Note Remember to use factor variables when including categorical/non-binary variables.

Here, we will investigate whether the relationship between conflicting demands at work and self-rated health looks different for men and women. For the interaction analysis, we use Approach B (interaction analysis based on comparison of model fit).

Note For extra review, please re-visit this section: Interaction analysis.

First, we perform a multiple logistic regression analysis for conflict_dich, sex, and health_dich.

logistic health_dich conflict_dich i.sex if pop==1

Then we save the estimates from the model and name the model “model1int.” 

estimates store model1int

Now, we perform a multiple logistic regression analysis for conflict_dich, sex, and health_dich, but also include the interaction term for conflict_dich and sex. We use a factor variable approach here. 

logistic health_dich i.conflict_dich##i.sex if pop==1

Then we save the estimates from the model and name the model “model2int.” 

estimates store model2int

Note We called our saved models “model1int” and “model2int”, but you can choose any name you like.

Next we perform a likelihood ratio test to compare the two models. 

lrtest model1int model2int, stats

From the likelihood-ratio test, we see that the p-value is below 0.05, which suggests that the model that contains the interaction term (model2int) fits the data significantly better than the model without the interaction term (model1int). This can also be confirmed with the AIC value, which is lower for model2int than model1int. The BIC value is not lower for model2int than model1int, but it should be noted that BIC tend to penalise complex models more than AIC does. The results suggest that there is a moderating effect of sex in the relationship between conflicting demands at work and self-reported health.

Next we produce marginal effects and generate a marginal plot for the interaction effect. 

margins i.conflict_dich#i.sex

marginsplot

Note You can use the Graph Editor (see Graph) to edit the marginsplot.

Summary Based on the interaction analysis and the marginsplot, we can see that sex moderates the relationship between conflicting demands at work and self-reported health, which means that the relationship looks different between men and women. The differences in self-reported health between the women who had a high extent of conflicting demands at work compared to the women who had a small extent of conflicting demands at work were greater than the differences in self-reported health between the men who had a high extent of conflicting demands at work compared to the men who had a small extent of conflicting demands at work. We can see that because the slope in the marginsplot is steeper for the women.

When we have included all of the relevant variables, performed our multiple linear regression model, as well as our interaction analysis, the next step is to perform model diagnostics.