Overview
Can a matched internet survey stand in for the ANES?
Opening
Ansolabehere and Rivers ask a practical question about the 2008 election. The Cooperative Congressional Election Study (CCES) is an internet survey of volunteers. YouGov, the survey firm that runs it, matches each volunteer interview to a Census profile of U.S. citizens. After that matching, does the CCES give the same presidential result as the American National Election Studies (ANES), a face-to-face survey? And does it give the same margins as the official vote?
The two samples
The ANES sent interviewers to people’s homes. About 2,100 people finished the interview after the election. That is enough to describe the whole country. It is too few to describe one state, one House district, or a small group.
The CCES is run online by YouGov, using volunteers. Matching kept 32,800 interviews. Many research teams share this one file. Weighting counts some interviews more than others so the sample looks like the country. The article’s bet is that matching plus weighting make this large file usable. Then it can answer questions a 2,000-person survey cannot.
What the group is checking
The outcome is the 2008 presidential vote. It is coded +1 for Obama, −1 for McCain, and 0 for anyone else. Party runs from 1, strong Democrat, to 7, strong Republican. The party coefficient is how much the vote score changes for each one-step move on that party scale. A party coefficient near −0.31 means one step toward the Republicans lowers the predicted vote score by about 0.31 points. That score runs from +1 for Obama to −1 for McCain.
The article reports a party slope, another name for this coefficient, of −0.313 in the CCES and −0.312 in the ANES. Our shared model, which uses only traits both surveys contain, gets −0.317 in both. That match is the main result. The state check, which compares the CCES vote in each state with the official result, is weaker. We are about 1 point too Democratic on average. The typical state error is 3.4 points. The article reports 0.27 and 2.43.