Link. Nothing surprising considering previous work on the subject, but a good reminder.

A study by three scientists in the American Political Science Review finds that voters are not competent at accurately evaluating incumbent performance and are easily swayed by rhetoric, unrelated circumstances and recent events.

Gregory Huber, Seth Hill, and Gabriel Lenz constructed a 32-round game where players received payments from a computer "allocator." The goal is to maximize the value of those payments.

Halfway through, at round sixteen, the player had to decide whether to get a new allocator or to stick with the old one.

The allocators pay out over a normal distribution based on a randomly selected mean. Getting a new allocator means that a new mean is selected. This was meant to simulate an election based on performance. 

The group ran three experiments where they changed some of the rules of the game in order to find out how voters could be manipulated or confused over performance. Essentially, how good were voters at accurately analyzing the performance of the "allocator?" 

• The first experiment merely alerted the player at round twelve that they would have the chance to pick a new allocator at round sixteen. This "election in November" reminder made the player weight recent performance in rounds 12-16 over earlier performance in rounds 1-12.

• The second experiment involved a lottery held at round eight or round sixteen. The payout was either -5000, 0, or 5000 tokens. The participant was told that the lottery was totally unrelated to the current allocator, but players still rewarded or punished their current allocator based on their lottery performance.

• The third experiment primed the player with a question right before the election. The question took an adapted form of either Ronald Reagan's "Are you better off than you were four years ago?" or John F. Kennedy's "The question you have to decide on November 8 is, is it good enough? Are you satisfied?" 

The conclusion: 

Participants overweight recent performance when made aware of the choice to retain an incumbent closer to election rather than distant from it (experiment 1), allowed unrelated events that affected their welfare to influence evaluations of incumbents (experiment 2), and were influenced by rhetoric to focus less on cumulative incumbent performance (experiment 3).

If you were ever wondering why Congress has a 95% incumbency rate despite an approval rating in the high teens, this study may be worth a read.