Grouping 2 variables into a single variable

[Azriel Alfajri]

Hi Everyone,

I have 2 binary variables that have the same context but not exactly the same. I want to group these 2 binary variables (i.e. married/not & pursuing kids/not) into 1 variable (i.e. family goals), and maybe this will produce a categorical variable that has 4 values (not married-not pursuing; not married-pursuing; married-not pursuing; married-pursuing).

My question is, is it statistically allowed?

Thank you in advance

[Clyde Schechter]

Yes, it is statistically "allowed." The question is whether it makes sense in the context of your specific research goals. It might help if you explain what your reason is for wanting to do this, and also your reason for doubting it. How would you use the two variables if you kept them separate? And how would you use the combined variable if you combine them? How do those uses fit into answering your research question(s)?

[Azriel Alfajri]

Hi Clyde, thank you for your answer.

The reason why I'm doing this is, those 2 binary variables basically indicators/sub-variables of 'family goals' (i'm measuring the impact of life domains on happiness), and i need only one variable to answer my hypothesis

[Clyde Schechter]

I suppose it's a question of how you might structure that variable. You listed the combinations in the following order: not married-not pursuing; not married-pursuing; married-not pursuing; married-pursuing. If you were to code those combinations, respectively, as 1, 2, 3, and 4, you would implicitly be stipulating that married and not pursuing children is a greater commitment to family goals than pursuing children and not married. Maybe that's OK, but it's not obvious to me. I can't say it's really wrong, but I also couldn't say that ordering those two in the other direction is wrong either. I suppose that depends on how you think of family goals as a construct. For my part, given that I don't see a clear ordering between those two, if I were going to combine them, I probably would do so by just adding up the number of yes's among the two, so it would be 0 for not married not pursuing, 1 for either married not pursuing or not married pursuing, and 2 for married pursuing.

My point is that it isn't obvious, at least not to me, how these variable really combine in the real world, and that might depend on your detailed conceptualization of the family goals construct. So you need to think about that.

I'll also add that when it isn't clear to me whether/how to combine variables, I generally prefer to just keep them as separate variables, perhaps with an interaction term. That would enable you to look at all four combinations without imposing a constraint about their ordering. In fact, the results of this kind of analysis might even inform you as to whether they should be combined into a single ordered variable or not, and, if so, in what order.