Hello again! I’m attempting to generate standard errors for an ATUS-X export using the RWT06 series of replicate weights. Using R, I’ve carefully reproduced the formula documented in the RWT06 description page (and in the ATUS reference documentation), however, my computed standard errors do not match those published by the BLS here:
Should they match? If so, are you aware of a reference implementation I could compare to? If not, are there any published, bullet-proofed results from the ATUS-X data that I could try to replicate?
I was able to replicate the BLS table “Avg hrs per day - Watching TV - TUU10101AA01014236” using the rwt06 replicate weights from IPUMS ATUS and the BLS time use variable BLS_LEIS_TV divided by 60.
Is there a specific table you are unable to match?
As it happens, that was exactly the series I was trying to replicate. I must just have a subtle error in my code. I’m going to take another crack at it. If you happen to have any reference code I could take a look at, that would be much appreciated.
Okay, you’re right that works perfectly. I didn’t think I could do it that way because of this note in the RWT06 docs:
“The successive difference replication approach (SDR) is different from other methods for creating replicate weights such as balanced repeated replication (BRR) and jackknife estimation.”
The R survey library doesn’t support SDR, so I thought I had to compute these from scratch. However, after seeing your code I did some more Googling and found this sentence in the IPUMS-CPS replicate weights docs:
“Correspondence with StataCorp statisticians and IPUMS testing revealed that successive difference replicate weights can be treated as jackknife replicate weights if the options are specified correctly.”
This still doesn’t explain why my (really pretty simple) code doesn’t give the correct answers, but I’m not going to worry about that for the moment. I might suggest adding a mention of that to the ATUS replicate weights docs.
Not an answer but instead, much more basic questions …
I am using STATA 13 and creating a simple cross-tabulation of vetstat (veteran status) and empstat (employment status). I want the weighted observation counts of veterans to approximate 20 million using CPS/ASEC(March) for 2015.
a) What weight variable(s) should I use (repwtp, wtsupp, or repwtp1-repwtp160)?
b) If I want to compute standard errors and confidence intervals, what weight variable(s) should I use?
b) Exactly how should I specify the tabulate command?
I am very much a STATA rookie. So, any detailed help that you can provide will be much appreciated.
For generating standard deviations using Stata and replicate weights, I would recommend starting with this documentation from the IPUMS USA website. While that page discusses using the IPUMS USA data, the syntax is basically the same You can then use the svy command to correctly apply the weights. Not all commands are available with svy in Stata. See this instructional page for an applied example of svy usage.