Hello, I’m using IPUMS- USA 2018 ACS 1-year and I’m struggling to recreate (with in the margin of error) the estimated civilian non institutionalized population between the ages 18-64 for the United States (197,888,546; ME 26,117) as found in table B18120.
I’m using the following filter in R and applying the person weight. This gives me an estimate of 198,010,297
filter(empstatd != 14 & empstatd != 15,
gq != 3,
age %in% (18:64))
I’m wondering if there is something I’m missing with one of the variables, or if there is any additional insight about recreating the reported ACS result.
Thanks!
In general you shouldn’t expect results from the public-use microdata to exactly replicate the official statistics from the Census Bureau, because there is some data editing and also a reduction in the sample size that occurs before release of the microdata. You can read more about this here. I suspect this is what is happening in your case, because the numbers are very close. Your code looks right to me.
Mattew, thank you, I appreciate your response. I suppose I falsely believed that it should fall within the margin of error reported by Census, but I take it that in some cases, such as this, it will still fall outside the margin of error.
Right. First of all, the margin of error in the published statistics is still a confidence interval - sometimes the real value will still fall outside this range. But more important for comparison with estimates from microdata is the margin of error for your microdata estimate, which will be a bit larger. I suspect if you did a t-test for the difference between these two estimates, the difference won’t be statistically significant.
1 Like