Step 14: Interaction analysis

To answer the third research question, we want to examine if the relationship between postpartum depression in mothers and psychiatric problems in children appears different depending on the level of social support that the mother had during the child’s first year of life. We will use interaction analysis and Approach B (interaction analysis based on comparison of model fit), to investigate this.

Note To brush up on interaction analysis and Approach B, re-visit Interaction analysis.

First, perform a multiple Cox regression with depression and support as x-variables. Save the estimates from the model and name it “m1”. We save these estimates in order to compare the multiple Cox regression model to the model with the interaction term later on.

stcox i.depression i.support, noshow

estimates store m1

Next, you perform a multiple Cox regression with depression and support as x-variables, using a factor variable approach to include the interaction between depression and support in the model. Save the estimates from the model.

stcox i.depression##i.support, noshow

estimates store m2

Use a likelihood-ratio test to compare the models. Is there a statistically significant interaction between depression and support?

lrtest m1 m2, stats

Note Here, the saved models are called “m1” and “m2”, but you can choose any name you like.

In the table above, you can see that the p-value for the likelihood ratio test is below 0.05 (0.0222), which suggests that the model that contains the interaction term (m2) fits the data better than the model without the interaction term (m1). This is also confirmed by the values for AIC, which are lower for m2. The BIC value is not lower for m2 than for m1, but it should be noted that BIC tend to penalise complex models more than AIC does. We can conclude that there is a statistical significant interaction between depression and support.

Finally, we will produce marginal effects and a marginal plot.

margins i.depression#i.support

marginsplot

The association appears to be stronger for those with low and medium levels of social support, than those with higher levels of social support.

Note For more review, re-visit Interaction analysis.

Summary We should now be able to answer the last research question: 3. Does the relationship differ depending on the level of social support that the mother had during the child’s first year of life? Yes. If one has had low levels of social support, the hazard is highest, followed by medium level of social support. With high levels of social support, the hazard is lower than those who have not had depression at all levels of support. In other words, even if the mother has had postpartum depression, her child has a lower hazard of psychiatric problems if she received high levels of social support during the child’s first year of life. High levels of social support for the new mother thus eliminates the increased risk of psychiatric problems in children.