Do I need to declare a survey design to use weights?

I’m trying to analyse the hours worked of various groups of young people from the time period 1994-2018. I’m using basic monthly data from the CPS. I think the relevant weight is ‘WTFINL’.

To use this weight, is it necessary to declare my survey design beforehand? My regressions appear to be ok when I run ‘reg ahrsworkt i.month i.year sum if age==18&ahrsworkt<70 [pweight=wtfinl]’.

But because the CPS is a stratified survey, is it necessary to declare survey design beforehand using ‘svyset’? If yes, what would be the appropriate command?

Apologies if my question is basic. I am new to this area.

I would deeply appreciate some guidance. Thank you.

A few points might be helpful.

First, it sounds like you are using the HOURWAGE variable from the outgoing rotation group. If so, you should be using the EARNWT variable. This variable will correct for selection into the outgoing rotation group.

Second, in principle you can apply sampling weights in Stata either by declaring the survey design (via svyset) or by using the in-line sampling weight capabilities of the reg command. Results should not depend on this choice. If you want to declare your data’s survey design you can use the svyset command. Here is a nice “how to” on svyset.

Hi Jeff,

Thank you very much for your insights. It’s great to hear that both weight approaches work.

Just to clarify on the first point you made. The variable that I am using is AHRSWORKT, not HOURWAGE. As AHRSWORT comes from the Monthly Basic data, I assumed that the correct weight to use is WTFINL. Is this the case? Apologies on a lack of clarity on my part.

Ah, yes. You are correct about using WTFINL with AHRSWORKT. Sorry about the confusion. I misunderstood your initial question.