their own welfare. If jobs have been lost in a recession, something is wrong. If sons
have died in foreign rice paddies, something is wrong. If polluters foul food, water
or air, something is wrong. And to the extent that citizens vote on the basis of such
judgments, elections do not signal the direction in which society should move so much
as they convey an evaluation of where society has been” (Fiorina 1981, p.5-6; emphasis
in the original).
This is a plausible idea, but like many verbal theories it is somewhat vague and incomplete.
In particular, how do voters evaluate governmental performance? How do they decide that an
incumbent has “performed poorly or well”? And what are the effects of retrospective voting,
either microscopic (e.g., the voting trajectories of individual citizens) or macroscopic (e.g., electoral
outcomes)? We try to address these and related questions by developing a deductive model of
retrospective voting. Formalizing Key’s verbal theory has the usual benefits: it not only clarifies
central notions, such as “good” governmental performance, it also allows us to extract testable
implications.
Our model is designed to stay close to the general sense of Key’s ideas. Since he clearly wanted
to be realistic about voters and their capabilities, our model assumes that voters are not perfectly
rational.
2
But in line with his famous remark that “voters are not fools” (1966, p.7), our voters
are adaptively rational: they use sensible heuristics.
3
The rest of the paper is organized as follows. Section II lays out the general ideas and defines
2
There is a series of retrospective voting papers, pioneered by Ferejohn (1986), which assume completely rational
voters. However, this line of work focuses on a different issue: not the capabilities of voters but whether politicians
could credibly commit to implementing platforms advocated in campaigns. (Standard Downsian models assume that
they could; in, e.g., Ferejohn (1986) words mean nothing; only the incumbent’s actions matter.) Thus, Ferejohn-
type models are explicitly principal-agent models, precisely as that term is used in economics. We completely agree
with the assertion of Achen and Bartels that “rational retrospective voting is harder than it seems” (2004, p.36; see
Wolfers [2002] for evidence on this point), and we suspect that Key would find the completely rational citizenry of
these principal-agent models to be quite different from the American voters he studied empirically.
3
For clarity and simplicity we examine voting heuristics that are purely retrospective: the citizens’ votes are based
totally on politicians’ past performances. Fortunately, one can easily prove that most of our results are robust: if
electoral choice is a weighted average of retrospective and prospective voting then they continue to hold if “most”
(but not all) of the weight is on the past.
3
Convention