Markus Prior, Gaurav Sood, and Kabir Khanna write:
On factual survey questions about economic conditions, opponents of the president often report significantly worse economic performance than supporters of the president. Scholars have so far interpreted this finding to mean that partisan respondents cannot even agree on matters of fact. We test an alternative interpretation: Partisans give the partisan congenial answer even when they have, or could have inferred, information less flattering to their partisan identification. To test this hypothesis, we experimentally manipulated respondents’ motivation to be accurate by either offering them a monetary incentive, or by emphasizing the importance of an accurate answer to researchers in two nationally representative surveys. Both treatments reduce partisan differences in reports of economic conditions by about half. Many partisans interpret knowledge questions about economic conditions as opinion questions, unless motivated otherwise. Typical survey conditions thus reveal a mix of what partisans know about the economy and what they would like to be true about it.
This sounds very reasonable to me. It’s related to the idea that a survey response is like a vote, and respondents want to support their side in the debate. I’m reminded of Joe Bafumi’s phrase, “The stubborn American voter.”
P.S. Unfortunately I wasn’t able to make it to Prior’s talk at Columbia. I took a look at the paper, and I’m glad to see it has some graphs. They could be presented much better, though. Figures 1 through 3 should be combined into a crisp array of line plots. As it is, there are dots all over the place and it’s hard to follow. Figure 4 would be better as a 3×2 grid of graphs, with 2 lines on each graph labeled directly, no codes and legends required. And Tables 1-3 should be replaced by coefplots.