Comparing ACS Table 18120 with PUMS (2019 1 year estimates)

Hello,

I'd like to seek help with a couple of questions about comparing ACS Table and PUMS estimates.

My project is related to estimate the number of people with vision difficulty and then calculate the employment-population ratio. I used two ways. 

(1) Based on the ACS Table 18120 (2019 ACS 1 year estimates from data.census.gov), for people with vision difficulty (age 18-64), the numbers were: employed  (1,734,139), unemployed (161,777), and not in the labor force (1,859,756). Then, I calculated the employment-population ratio as 1,734,139 / (1,734,139 + 161,777 + 1,859,756) = 46.2%

(2) I also used the PUMS person data (2019 ACS PUMS 1 year) to make the calculation. I first selected people with vision difficulty (age 18-64) by using DEYE = "1". Then, I utilized PROC SURVEYMEANS in SAS. I set negative replicate weights as zero before running the following code. 

proc surveyfreq data=pums19_1864w varmethod=jackknife;
weight pwgtp;
repweights pwgtp1-pwgtp80 / jkcoefs=0.05;
table Deye*ESR/row chisq;
run;

Based on the SAS output, for people with vision difficulty (age 18-64), the weighted numbers were:

1 Civilian employed, at work 1,691,122
2 Civilian employed, with a job but not at work 52,231
3 Unemployed 164,666
4 Armed forces, at work 6,587
5 Armed forces, with a job but not at work 289
6 Not in labor force  2,039,589

Then, I calculated the employment-population ratio as (1,691,122 + 52,231 + 6,587 + 289) / (1,691,122 + 52,231 + 164,666 + 6,587 + 289 + 2,039,589) = 44.3%

My questions are: 

(1) Is there anything wrong with my calculation about the employment-population ratio using ACS Table or PUMS? Should I remove "Armed forces" records from my calculation? 

If I remove "Armed forces", I consider the employment-population ratio as (1,691,122 + 52,231) / (1,691,122 + 52,231 + 164,666  + 2,039,589) = 44.2%

(2) Why is there a difference between the ACS Table estimates and the PUMS estimates? Does 46.2% vs 44.3% (or 44.2%) look normal? 

 

I look forward to your insights. Thank you. 

Carol

Parents
  • Table B18020 uses the civilian non institutionalized population for the universe, so you should exclude ESR = 4,5 and RELP=16. So, no Armed Forces. The removal of institutional GQ will reduce the denominator (most are NILF), which is likely to bring the estimate up.

    That said, the PUMS will not produce the exact same estimate as the tables. This is because the PUMS is a subsample of the internal micro data file set to a ~1% US sample or about 3.2 million person records, compared to almost 5 million observations on the internal file, used for the tables.

     - Matt

Reply
  • Table B18020 uses the civilian non institutionalized population for the universe, so you should exclude ESR = 4,5 and RELP=16. So, no Armed Forces. The removal of institutional GQ will reduce the denominator (most are NILF), which is likely to bring the estimate up.

    That said, the PUMS will not produce the exact same estimate as the tables. This is because the PUMS is a subsample of the internal micro data file set to a ~1% US sample or about 3.2 million person records, compared to almost 5 million observations on the internal file, used for the tables.

     - Matt

Children
No Data