I want to combine ACS data 2000-2015 but also use replicate weights, which don't exist b4 2005. What should I do?

I want to combine ACS data from 2000 to 2015 but I want to account for the complex survey design using replicate weights. These weights do not exist before 2005. How do I therefore account for the complex survey design for years before 2005 when combining data from 2000 to 2015?



Unfortunately, I’m not aware of an easy or perfect solution to this problem. You have at least a couple of options. First, you could try to replicate the method used to generate replicate weights for the pre-2015 ACS samples. Details about this methodology are explained here. (Note that this document discusses ASEC data from the CPS dataset, but the process is identical for ACS data.) Second, since this procedure can be quite labor intensive, you could use replicate weights for the post-2005 samples and make notes about how much use of replicate weights change the results of your analysis (e.g. how much standard errors or confidence intervals change). These notes will then inform how you discuss and interpret the results on the pre-2005 samples. Third, you may find the STRATA and CLUSTER variables useful. These variables offer a consistent way to calculate an estimate of standard errors and variance over time using ACS data. The USA Variance Estimation page details how these variables can be used.

I hope this helps.

I second Jeff – if you have strata and cluster variables (or whatever they are called), you should be using these in your software. The estimated variances would not be as accurate as with replicate weights (which account for all of sampling variability, nonresponse, and sample calibration), but you’d have to live with that in this case… usability trumps accuracy :-\