Interaction effect: how to set baseline?

[Grek Lai]

Hello everyone.

I am doing a longitudinal analysis using fixed effect. I am interested in how child's gender will influence parental economic outcome.

So here is how I do it. I categorized the age of parent's child into several categories. Say, 1 for 0-2, 2 for 3-7, 3 for 8-18. Then if they are not parents yet, I codify their child's age into 0. Since my question is whether child's gender matters then I would like to do an interaction effect between child's gender and child's age. Then my question is, since I am using fixed effect, I am comparing difference within the same individual. I want the reference category to be the group that are not parents yet and that's why I codify them into 0 in child's age.

My question is that, while I am doing the interaction effect, the reference category is the group with child age form 0 to 2 rather than those without a child, because the information of the child's gender of the reference group (those without child yet) is missing. Pairwise deletion then wipes out the group which I expect to be the reference category.

I know stata is doing it job. But is there something I have done wrong. My intention is to compare the outcome difference between when individual are parents yet and when they have child of difference ages, and whether this relationship will differ by child's gender.

Your help will be much appreciated.

Thank you.

Grek

[Maarten Buis]

In that case you can create an indicator (dummy) variable for whether or not the respondent has a child. Than replace the variable gender to be the value for males. If you add now both the child indicator variable and the gender variable, the effect of child is how much different a non-parent respondent is compared to a parent with a male child. You can do the same for age. After that you can include the interaction.

The challenge would how to deal with parents with more than one child. Solving that is up to you.

[Grek Lai]

In that case you can create an indicator (dummy) variable for whether or not the respondent has a child. Than replace the variable gender to be the value for males. If you add now both the child indicator variable and the gender variable, the effect of child is how much different a non-parent respondent is compared to a parent with a male child. You can do the same for age. After that you can include the interaction.

The challenge would how to deal with parents with more than one child. Solving that is up to you.

Thank you very much Dr Maarten! In my study, it is contextualized in the one child family

But maybe I didn't make myself clear enough in the post.

I only have one variable indicating whether they have child and how old he/she is. So if I tab it, it is like

0 for no child
1 for child age 0-2
2 for child age 3-6
...etc

I cannot quite follow your lead. Perhaps something wrong with my understanding. By adding a new variable indicating whether they have a child and replace later on, does it mean to, in the first place, separate family with boy from the one with girl. Then how could I include the interaction?

Could you elaborate a little more on that? It matters a lot for my research!

Thank you!