Median annual earnings

Hi everyone,

I am new to PUMS. I am trying to estimate the gender wage gap by median annual earnings of men and women in Pennsylvania. I am using 2017 1-year estimates. I am calculating WAGP*(ADJINC/1000000). I have made a table of RAC1P variable(race/ethnicity) and SEX variable, aggregating the person-level data using median function. 

However my estimates are coming out incorrect - the median earnings for "Black or African American alone" class is coming more for women, whereas women should be earning less(according to past historical data in american factfinder). Am I doing something wrong? 

PS- I am not using the PWGTP weights, is that the problem? 

PPS- I have tried using OIP+WAGP+SEMP instead of just WAGP, but the estimates are still in the same order.

Any help is appreciated,

Thanks.

  • Good morning -

    Here are a couple of thoughts:

    First, yes, you do need to weight cases using PWGTP.

    Second, when troubleshooting, it might make sense to first try and replicate the results in an existing table (such as number of workers in each income category in table B20005B). PUMS will not exactly match published tables because PUMS is only a sample of the entire survey sample, but PUMS results will be close. Another option is to query sub-population earnings details using IPUMS and compare your results to those (which should be an exact match).

    Third, are you sub-setting the population by any employment characteristics such as all workers, or all full-time/year-round workers? Most published tables on earnings comparisons are for workers ages 16+ who worked full-time/year-round.
  • In reply to Beth Jarosz:

    Hi

    Thank you for replying.
    Would it be correct if I simply multiplied the PWGTP weights with the wage ( calculated so, (WAGP*ADJINC)/1000000 ) ?

    I am not sub-setting the population, so I will keep that in mind.

    Thanks for the advice.

  • In reply to Shubham Kumar:

    Apologies for not seeing this sooner (spam filter issues). Yes, that is correct.