EARNWT if analytic sample is not limited to those in the earner study

CPS
1

Hi friends,

I will use variables in the earner study, but my population of interest is not limited to those in the earner study. My sample will include all household heads and their spouse/partners (if they do have a partner/spouse at the time of survey) and I will pool all the 12 monthly surveys within the same year together from 1982 till now. I will still use variables in the earner study because some of the statistics that I need to calculate will be based on the wage levels of earners in the household. Should I still use EARNWT or some kind of combination of EARNWT and the WTFINL? Thanks!

2

For any calculation using the variables from the earner study, you should use EARNWT, which will give zero weight to those not in ORG’s. For calculations involving only non-earner study variables, you should use WTFINL to make use of the larger sample available for these variables. This is assuming that you are using variables from the basic monthly CPS. For supplement topics, use the weight specific to that supplement.

If you are interested in taking the earner study variables and applying them to the other observations of a given individual to use in subsetting the data, that is also possible. To do so you can make use of the unique longitudinal identifier CPSIDP to link those observations. In that case you can use WTFINL for the analysis. More resources on using the CPS data longitudinally are available here.

4

Thank you! Sorry, I wasn’t clear in the previous post. It seems that my research does not fall into any of the three situations mentioned by you. My intention is like this: I want to calculate the proportion of different types of households in each year (by pooling the 12 basic monthly CPS within the same year and constructing repeated yearly cross-sectional CPS). Type of household is defined as follows:

  1. households headed by married couples in which the wife’s EARNWEEK is more than a given number and neither partner is self-employed;
  2. households headed by married couples in which the wife’s EARNWEEK is less than a given number and neither partner is self-employed;
  3. households headed by married couples in which at least one partner is self-employed;
  4. households headed by unmarried persons.

In this case, I do use variables from the earner study to classify households into different types and calculate the distribution of household types in each year, but I do not feel that I should use EARNWT. I am thinking about this:

  1. Limit my sample to households for which MISH equals 4 or 8 (i.e. referred to as 4/8 households below);
  2. Use a transformed version of HWTFINL in the analysis. Specifically, the weight of household i in month m of year y is its HWTFINL multiplied by (total number of all 4/8 households in month m / total number of all 4/8 households in year y)

Does that sound reasonable to you?

5

Your approach seems reasonable to me. I might restrict the sample to those households with earner study variables in each month, and rescale HWTFINL so the weights sum to the correct number of households in each month. Then divide the weights in each month by 12.

I’m not the authority on this, so you may want to consult a statistician about this, or ask on stats.stackexchange.com.