Retrospective Voting

POLSCI 240 / PSY 225: Political Psychology

March 31, 2025

Topics for this week

Retrospective voting as an alternative to spatial theory and issue-based voting

  • What do we mean by retrospective voting?
  • Why is retrospective voting an attractive alternative?
  • Can retrospective voting be justified from a rational choice perspective?
  • How might retrospective voting fail to live up to expectations?

Today, we will consider the “classic” perspective on retrospective voting; next time we will develop critiques of this literature based on research in political psychology

Position vs valence issues

Distinguish two different kinds of issues:

  • Position issues: people differ about the right goal to pursue (e.g., more redistribution, legal abortion)
  • Valence issues: everyone wants the same thing (e.g., economic growth, low unemployment)

When people vote on position issues, we are in the world of spatial theory, but spatial theory doesn’t work for valence issues

  • Everyone has the same position: we want more (or less) of the thing!

Retrospective voting

Retrospective voting is a decision rule in which voters choose to either:

  1. Retain the incumbent candidates or parties, when they have performed well in achieving shared goals
  2. “Kick the bums out”, when they have performed poorly in achieving shared goals

Retrospective voting as attractive alternative

There are two different streams of argument for why retrospective voting is an attractive alternative to spatial theory:

  • Political psychology

    • Research suggests people don’t know much about issues, parties, and candidates
    • Retrospective voting is far less demanding: only need reasonably accurate perceptions of whether important goals are being achieved, or not

  • Rational choice theory

    • A rationally ignorant voter can still induce good outcomes by allocating reward and punishment in a systematic way
    • Select competent representatives, and incentivize good behavior

Political psychology arguments

Retrospective voting is less demanding than spatial voting

  • The necessary information is more objective, and easier to obtain (e.g., unemployment stats)
  • People can often use their own experiences as data (e.g., “did I lose my job?”)
  • The decision process is greatly simplified: no need to calculate spatial distance, or consider utility functions


Rational choice arguments

There are two primary, rational choice based justifications for retrospective voting

  • Selection of the most competent leaders using past performance as a guide

  • Sanctioning (vs rewarding) incumbent candidates for poor (vs good) economic outcomes during their term as a means of incentivizing effort on behalf of citizens’ interests

Rational choice arguments: Selection

Citizens want to elect leaders that are competent and will produce good outcomes in the future - what information do they have?

  • For incumbents, they have their performance while in office

  • If competence is relatively stable, citizens can use past performance to predict future performance

    • aggregate various performance metrics over the period of their time in office as a “best guess”

  • They don’t have the same info for most challengers, so need some “baseline” or prior expectation for comparison (e.g., randomly drawn from distribution)

Set up

  • Voters observe economic outcomes for T periods of the incumbent’s term in office

  • They know that outcomes are a function of:

    • the competence of the incumbent
    • random forces unrelated to the incumbent
    • they know nothing about the challenger, who is expected to be average

  • Use a Bayesian update to compare incumbent to challenger, and choose the one with higher expected value

Illustration

# set seed
set.seed(27474)

# number of observation periods
t <- 4

# incumbent's competence is drawn from 
# a normal distribution with mean 0 and standard deviation sigma_i
sigma_i <- 1
incumb <- rnorm(1, mean = 0, sd = sigma_i)

# economic outcomes are drawn from normal distribution w/
# mean = incumbent's competence and standard deviation sigma_e
sigma_e <- 1
econ <- rnorm(t, mean = incumb, sd = sigma_e)

# voter infers competence of incumbent based on economic outcomes
# using Bayesian updating
sigma2_post <- 1 / (1/sigma_i^2 + t/sigma_e^2)
mu_post <- sigma2_post * (0/sigma_i^2 + sum(econ)/sigma_e^2)

# results
round(cbind(incumb = incumb, mean_econ = mean(econ), post_mean = mu_post, post_sd = sqrt(sigma2_post)), 3)
     incumb mean_econ post_mean post_sd
[1,]  0.345     0.532     0.425   0.447

Illustration

If you use more periods of observation

If there is less randomness in economic outcomes

If there is MORE randomness in economic outcomes

Rational choice arguments: Sanctioning

Citizens want to incentivize elected officials to exert effort toward bringing about good outcomes

  • Elected officials can divide their time between self-interested and public-interested activities

  • There is some, non-negligible link between effort and outcomes

  • If voters punish bad outcomes, elected officials will increase their effort at the margin

    • Voters cannot be too “easy” or too “hard” on politicians - either will incentivize shirking

Basic model

Instead of elected officials having a pre-determined competence, they choose a level of effort

  • Voters update their beliefs based on outcomes
  • They re-elect incumbents only if they can be confident that the official is above some minimum effort level

Illustration

  • Let’s set the voter’s effort requirement at 0 for simplicity
  • Voter sets a confidence threshold, e.g., they want to be 80% sure the elected official is above 0

Illustration

Illustration

From the perspective of the elected official, we need to ask: “what is the expected value of exerting some particular amount of (costly) effort?” Several things matter:

  • Personal “perks” of office (e.g., rents, corruption)

  • Citizens’ confidence threshold

    • If very low, they almost always get re-elected no matter what
    • If very high, they almost never get re-elected no matter what

  • Variability of econ outcomes: how random are outcomes vis-à-vis effort?

Empirical work on retrospective voting

Work on estimating “V-P” (“vote and popularity”) functions provides strong evidence for retrospective voting across countries and time periods

  • Outcome variable: some measure of incumbent candidate or party popularity; most often vote share, but could also be approval

  • Wide variety of possible independent variables, but bulk of research on key macro-economic variables: unemployment, inflation, GDP and real income growth

  • Most of work is focused on national level (Congress, Parliament, President), but there is a recent literature looking at local level retrospective voting (e.g., municipalities)

Key empirical findings: ego- or socio-tropic voters?

  • Egotropic voting uses personal (household) economic conditions as the relevant criteria
  • Sociotropic voting uses national economic conditions
  • The bulk of research suggests that voters are more sociotropic than egotropic

Key empirical findings: what variables matter?

A primary role for GDP (perhaps real disposable income) growth and change in unemployment, with a significant (but perhaps lesser) role for inflation

Key empirical findings: institutions and context matter

Retrospective voting is a mechanism for holding elected officials accountable, so clarity of responsibility for outcomes matters

  • Controllable vs uncontrollable economic forces

    • Exposure to globalization and free trade
    • Crises (e.g., COVID, 2008 financial crisis)

Key empirical findings: institutions and context matter

Retrospective voting is a mechanism for holding elected officials accountable, so clarity of responsibility for outcomes matters

  • Local vs national govt, or national vs supra-national (e.g., EU)

    • Evidence that national is primary locus of concern

Key empirical findings: institutions and context matter

Retrospective voting is a mechanism for holding elected officials accountable, so clarity of responsibility for outcomes matters

  • Executive vs legislature (e.g., Congress vs President)

    • Evidence that Executive is locus of concern
    • But possibly reduced impact under divided govt

Key empirical findings: institutions and context matter

Retrospective voting is a mechanism for holding elected officials accountable, so clarity of responsibility for outcomes matters

  • One party government vs multi-party governing coalitions

    • Weaker effects in multi-party coalitions
    • But typically hold leader’s party accountable (e.g., Prime Minister, German Chancellor)

Psychological biases and retrospective voting

The success of retrospective voting in fostering policy in the public interest requires citizens to be (relatively, for the most part):

  • focused on the right information
  • accurate in attributing responsibility
  • non-partisan in their assessments (e.g., of economic changes)

As we will see, there are reasons to be concerned on all these fronts…

Myopia

In this context, “myopia” means “too focused on recent economic performance”

  • Voters seem to give more weight to the most recent economic changes
  • Increases randomness, decreases incumbent incentives to work on behalf of public
  • Can create political business cycles: incumbents overheat the economy just prior to elections

Achen and Bartels

Kicking the dog

Sanctioning only creates good incentives if elected officials are punished for things over which they have some control

  • The more voters punish for uncontrollable outcomes, the weaker the incentives to work for the public good

  • e.g., droughts, floods, shark attacks, tornadoes

  • However, there is debate about the rationality of such punishment

    • Officials cannot control weather, but can prepare for such events to minimize consequences

Evidence is mixed: some evidence voters punish for disasters rationally - e.g., for amount of damage caused by tornadoes, not number of lives lost - but also reason to be skeptical

Preparation vs mitigation

A general finding in decision making is what Kahneman calls WYSIATI: what you see is all there is

  • This means that people are too focused on what is currently salient, and don’t think about the unseen when making decisions

This can lead to perverse incentives for government to focus on highly visible efforts at mitigating the impact of disasters, rather than the less visible (but more cost-effective) efforts at preventing them

Spending on relief or preparedness

Healy and Malhotra (2009)

Irrelevant events

Healy et al. (2010)

Dividing responsibility

Voters may not do a great job dividing responsibility for outcomes approriately across levels and institutions of government, e.g.,

  • Strong focus on the executive and party of the executive (President, Prime Minister, Chancellor, etc.)
  • “Nationalization” of local elections

Though important not to overstate: there is responsiveness to relevant factors, but perhaps not enough

Partisan bias

  • There tends to be a substantial gap in partisans’ perceptions of the state of the economy
  • Responsibility attributions are often motivated: goal is, in part, to preserve a positive image of a favored party or candidate
  • Partisan bias reduces govt accountability: citizens need to be willing to punish the in-party too!

Healy and Malhotra (2013)