The abstract from a recent talk by Gabriel Lenz of MIT:
Retrospective voting is central to theorizing about democracy. Given voters’ ignorance about politics and public policy, some argue that it is democracy's best defense. This defense, however, assumes citizens are competent evaluators of incumbent politicians' performance. Although little research has investigated this assumption, voters' retrospective assessments in a key domain, the economy, appear flawed. They overweight election-year income growth in presidential elections, ignoring cumulative growth under the incumbent. In this paper, I present evidence that this myopia arises from a more general “end bias” in retrospective assessments. Using a three-year panel survey, I show that citizens' memories of the past economy are inconsistent with their actual experience of the economy as they reported it in earlier interviews. They fail to remember the past correctly in part because the present shapes their perceptions of the past. I then show similar behavior in the lab. When participants evaluate economic and crime data, I again find that election-year performance shapes perceptions of overall performance, even under conditions where the election year should not be more informative. Finally, I search for and appear to find a cure. Presenting participants with cumulative information on performance (e.g., total income growth or total rise in murders during incumbents’ terms) cures this myopia. On one hand, these results are troubling for democracy because they confirm citizens’ incompetence at retrospection. On the other hand, they point to a remedy, one that candidates and the news media could adopt.
That's a remedy as long as the candidates and news media don't simply lie about the fact. Good luck with that one.