Need a Part-Time Job? Love writing?

Interested in politics and like writing, then check this out!

Researching Young Voters Using Census Data

Hi, I'm the new research analyst for Rock the Vote, now on my second month of employment. I thought I would begin my blog activity with a topic we can all appreciate: researching young voters using Census data. For software I prefer to use Stata, but the commands can be translated to SPSS, SAS, or other programs. If you are a college student you should be able to purchase an inexpensive educational version of Stata from your campus software store.

The Census
To briefly review, the Census is conducted on a decennial basis to collect population counts, employment data, etc. What is less well-known is that these data are supplemented with more frequent surveys, such as the American Community Survey (ACS) and the Current Population Survey (CPS). Researchers looking at voting or elections will be most interested in the CPS sample conducted in November of even-numbered years because it includes a variety of questions on the fall elections for that year.

Getting The Data
CPS data can be downloaded directly from the Census website but it requires a lot of work to format and label everything. Luckily the National Bureau of Economic Research (NBER) does this for us and makes the cleaned data available for Stata, SPSS, and SAS. First, download the data file for the particular month (say, November 2006 [.zip, 14.3mb]) on NBER's CPS Supplements page. Then go over to the data definition files page to download the files for Stata (cpsnov06.do and cpsnov06.dct).

Formatting The Data
Once you have the data you will need to run Stata and edit the cpsnov06.do file (hit ctrl-8 to load the do-file editor), which will format and label all the variables. You may want to tweak the memory command on the second line if you have less memory (I change it from 1000m to 300m), and you will need to check that the file paths are correct for the .dat, .dta, and .dct files: change them to the directory where you saved the files (for example, line 20 for me is local dat_name "C:\data\cpsnov06.dat"). Save your changes and then click the "Do" button to have Stata run the file - it will create a file called cpsnov06.dta with the cleaned data so that you don't have to run the .do file each time.

Analyzing the Data
From there you can look at whatever you like; the election-related questions are at the bottom of the variables list. pes1 is voter turnout, pes2 is voter registration (for those who didn't vote), pes6 is election-day vs. early voting, etc. Look at the technical documentation [pdf, 1.7mb] for the full list of variables. For youth voting research you will generally want to delete observations where the age is outside of the 18-29 range (drop if peage < 18 | peage > 29) and where they are not citizens (drop if prcitshp == 5). This will leave you with 18,110 observations for the November 2006 data.

Examples
Young voter turnout by gender: tab pes1 pesex, column
-> Results: 24.4% for young males, 27.6% for young females.

Young early voting by race: tab ptdtrace pes6 if pes6 > 0, row
-> Results: 15.8% of white youth voted early, compared to 7.4% of black youth and 28.1% of asian youth.

Top reasons for not voting: tab pes4 if pes4 > 0
-> Results: 1) Too busy - 35%, 2) Not interested - 15%, 3) Out of town - 13%, 4) Forgot - 7%.

Weighting
For the most accurate results, you need to weight the analysis to correct for small differences between the CPS sample and the overall US population. In the CPS dataset the variable pwsswgt contains a probability weighting for each observation. To use it run the command "svyset [pweight=pwsswgt]", then prefix your commands with "svy:", such as "svy: tab pes1".

Conclusion
I hope this was clear enough without being too lengthy. If you have any questions or problems feel free to leave a comment or email me at chrisk [at] rockthevote.com. I'd be interested to hear what people are looking at. Finally, check out CIRCLE's research for a variety of youth voting papers and analysis using Census data.