# Stacking 3 Individual Years of Data vs. Waiting for the 3-year Sample

I am working on a project for the Dept. of Edu that describes English Learners (ELs) in the U.S. using 2011-13 ACS data. I have a strong preference to use the IPUMS data over the PUMS data, but the 2011-13 IPUMS data are not available yet. I have two quick questions to help me choose which data set to use:

1. Does anyone know when the 2011-13 IPUMS data will be released? Can you say whether it will be released by 2/1/15?

2. Another researcher suggested combining the 2011, 2012, and 2013 individual year samples. The researcher said the main adjustments I would need to make are dividing the sample weights by 3, and adjusting the dollar variables to constant dollars.

Is adjusting the sample weights for the three year sample as easy as dividing the weights by 3?

The project involves calculating a series of population averages (for ex., the % of ELs by state, the % of ELs enrolled in college, the % of ELs who are single parents).Would these population averages be bias if I divided the sample weights for each year by 3 and stacked the samples into one data set?

The IPUMS-USA Project is currently working on improving their data creation process.This will ultimately improve the IPUMS-USA project as a whole, but with the unfortunate side-effect of delaying the release of the 2013 multi-year ACS data files. While there is no current estimated release date for these samples, I do not believe they will become available through IPUMS-USA before February of 2015.

However, your proposed method of “stacking” the thee single-year ACS files and adjusting the weights and dollar variables should be a sufficient approximation of the Census Bureau’s 3-year 2013 ACS file. While the Census Bureau does use more complex methods for generating multi-year weights (see Weighting Methodology from the 2012 3-year Accuracy Statement), simply dividing by three is the primary adjustment and provides extremely similar values to the Census Bureau calculated weights. You can verify this by trying to match the 2011-2013 PUMS Estimates for User Verification. If you are still uncertain about the “stacking” method, I would recommend replicating your analysis using the stacked 2010 to 2012 ACS samples and compare your results with the actual 2012 3-year ACS sample.

I hope this helps.