Hi,
I am trying to use the linking functionality of IPUMS’ CPS data, and I’m having a bit of trouble determining the correct weight to use. Essentially, I am linking observations for four consecutive months (March 2020-June 2020) in order to follow participants through each of the four months to see how their labor force status changes. As such, I have merged the monthly data using the “cpsidp” variable to generate a wide-form data set for the 1/4 of the sample that is present in each of the fourth months (i.e., the MISH1/MISH5 in March 2020) where each row represents a person and all variables are named x_1, x_2, x_3, and x_4.
However, after successfully linking the observations, I am struggling to figure out which weight to use. I am familiar with the “lnkfw1mwt” and “panlwt” options; however, two issues arise. First, when I use either of those in a collapse function, I end up with estimates that are about 1/4 of the actual numbers. For example, I get about 550,000 people as unemployed on temporary layoff in March, even though the actual number is ~1.8 million. And second, since I have multiple observations for each (e.g., lnkfw1mwt_1, lnkfw1mwt_2 and lnkfw1mwt_3), so I’m not sure which to use.
On the first problem, my hunch is that since these weights (lnkfw1mwt and panlwt) are for linking two consecutive months, they are not large enough to make the sample representative of the population (since my sample is 1/4, not 1/2, as large as a given month). I had an idea to just multiply the weights by 4, but that seemed haphazard so I thought I’d ask the forum instead.
Any help would be great!
Thanks,
Jimmy
It sounds like you are only including respondents who are present in all four months of data–please correct me if I am wrong. If so, the variables LNKFWMIS14WT will allow you to weight records in the first four months of data collection (MISH 1-4) and LNKFWMIS58WT will allow you to weight linked records in the last four months of data collection (MISH 5-8). Note that these weights are for basic monthly data only. Because these are forward linking weights, they should be equal to 0 in all months other than the first month the person is observed in the subset of linked months (MISH 1 for LNKFWMIS14WT and MISH 5 for LNKFWMIS58WT); this should correspond to the first observation (lnkfwmis14wt_1 or lnkfwmis58wt_1) in your wide file. You may also be interested in the this page of resources for linking CPS data, which includes more information on weights as well as sample code that you can modify to create your own weights. I am not sure if you are using the collapse command as part of your creation of a wide data file or for some other purpose; the Stata syntax tips on the IPUMS CPS page with linking resources include information on using reshape.
I hope this helps. Please let me know if you have further questions.
Hi,
I have a follow-up question about linking weight for observations in 4 consecutive months (eighter Mish1-4 or Mish5-8), particularly when the panel data is unbalanced (i.e., respondents are not present in all four months).
In my current study, I do time-varying hazard analysis using the person-month panel structure of CPS data. I focus on employment status changes over the four consecutive survey months, either the first or second round of rotation (i.e., before and after 8-month hiatus). If a respondent has both first and second four-month periods, I arbitrarily choose the first four months for analysis. I will conduct a sensitivity test using the second four-month period.
In constructing analytic data, I do not limit the sample to those who are present in all four consecutive months of the survey. The minimum number of observations per respondent is 2, and the maximum observation is 4. Can I still apply LNKFWMIS14WT and LNKFWMIS58WT sampling weights when analyzing unbalanced panel data?
Thank you for your support.
Warm regards,
Hyojin
LNKFWMIS14WT and LNKFWMIS58WT are forward-looking longitudinal weights for linking the CPS across all four months in the period; they are based on the population counts of people who were eligible to be linked across these time periods. If only linking between two time periods, the sample size will be larger and you will likely want to use different weights. I am linking to guidance on the IPUMS CPS longitudinal weights, which includes information and code for constructing your own weight.