Voter Turnout: Why Who Shows Up Changes Everything

Every pre-election poll attempts to answer a deceptively simple question: if the vote were held now, who would win? But embedded in that question is a harder one that most readers never think about: who, exactly, is going to show up? Turnout is the gap between what people tell a pollster they intend to do and what they actually do on election day — and modelling it is one of the most consequential decisions a polling firm makes.

The problem in plain language

In most democracies, not everybody who is eligible to vote actually votes. Turnout at federal elections ranges from around fifty percent in some countries to over ninety percent in others (particularly where voting is compulsory). The people who stay home are not randomly distributed: they tend to be younger, poorer, less politically engaged, and less firmly attached to a party. If a pollster talks to a representative sample of the whole adult population, they are including millions of people who will not ultimately vote — and those people have systematically different preferences from those who will.

The result is that a poll of "all adults" can look significantly different from a poll of "likely voters." In the United States, the gap between registered-voter and likely-voter samples regularly shifts the headline number by two to four points, usually in favour of the more conservative party (because conservative voters tend to be older and more habitual in their turnout). In other countries the direction varies, but the principle is the same: who shows up determines who wins, and polls that do not account for differential turnout are measuring the wrong thing.

How pollsters model turnout

There is no single agreed method. Different firms use different approaches, and the choice can move headline numbers substantially. The main families of turnout model are:

• Self-reported likelihood: Respondents are asked how likely they are to vote, usually on a scale from one to ten. Those below a certain threshold are excluded or down-weighted. Simple but unreliable, because people consistently overstate their likelihood of voting — social desirability bias makes it embarrassing to admit apathy.
• Past behaviour filters: Respondents are asked whether they voted at the last election. Those who say no are excluded or weighted down. More robust than self-reported intention, but still depends on honest recall, and misses first-time voters or people mobilised by new issues.
• Voter-file matching: In countries with public voter rolls (notably the United States), pollsters can match their sample against official records of who actually voted previously. This is the gold standard for individual turnout history, but it is only available in a few countries and does not predict future behaviour with certainty.
• Demographic modelling: Rather than asking individuals about their plans, the pollster builds a demographic model of what the electorate will look like — for example, assuming that 75 percent of over-65s will vote but only 45 percent of 18–24-year-olds. Respondents are then weighted according to these assumed turnout rates. The accuracy of the model depends entirely on whether the assumptions are correct.
• Hybrid approaches: Most sophisticated modern polls combine several of these signals — self-reported likelihood, reported past behaviour, registration status, and demographic propensity — into a composite score. The exact formula is proprietary and varies by firm.

Why different turnout models produce different numbers

Because each model makes different assumptions about who will show up, two polls fielded on the same day with the same sample size can disagree by several points purely because they model turnout differently. This is not a flaw; it is a consequence of the fundamental uncertainty involved. Neither model is "right" until election day reveals who actually voted.

Consider a hypothetical election where Party A leads by six points among all adults, but Party B's supporters are substantially more likely to actually vote. A poll using a strict likely-voter screen might show the race as essentially tied, while a poll of registered voters shows a comfortable Party A lead. Both polls are measuring real things — they are just measuring different questions. The careful reader must always check which population a poll reports: all adults, registered voters, or likely voters.

When turnout models fail

Some of the biggest polling misses in democratic history have been turnout-model failures rather than opinion-measurement failures. The polls correctly identified what people thought — they just got wrong which of those people would turn up.

This tends to happen in elections with unusual mobilisation patterns: a candidate who brings new voters into the electorate, a differential enthusiasm gap between parties, a voter suppression campaign, or a sudden surge of late-deciding low-propensity voters. In each case, the pollster's turnout model — built on historical patterns — fails to capture an electorate that differs structurally from the past.

Compulsory-voting countries (such as Australia and Belgium) largely sidestep this problem because turnout is near-universal. But even there, the informal vote rate (people who show up but deliberately spoil their ballot) and the small percentage who pay the fine rather than vote can shift tight races.

What to look for as a reader

When evaluating any pre-election poll, check the following:

• Which population does the headline number report? "All adults," "registered voters," and "likely voters" are different things. A poll that does not specify is hiding important information.
• What likely-voter screen is being used? Reputable pollsters publish their methodology, including how they filter or weight for likely turnout. If this is not disclosed, treat the numbers with extra caution.
• Does the turnout model match the expected electorate? In elections where new voters or unusual demographics are expected to play a large role, a turnout model calibrated on the previous election may undercount them.
• How does the poll compare to others using different screens? If likely-voter polls and registered-voter polls are telling substantially different stories, the election probably hinges on turnout — and that means higher uncertainty.

Turnout in proportional systems

In proportional-representation systems, the turnout problem manifests differently. Because seat allocation is proportional to vote share nationally (or within large multi-member districts), the geographic distribution of turnout matters less than in constituency-based systems. However, differential turnout by age, class, or political engagement still shifts the relative share of parties. If young left-leaning voters stay home at higher rates, left-wing parties will underperform their registered-voter polling numbers even in a proportional system.

Some PR systems also have turnout-related threshold effects. If a small party's support is concentrated among low-turnout demographics, it may poll above the electoral threshold (say, five percent) among registered voters but fall below it on election day among actual voters — a catastrophic outcome that eliminates the party from parliament entirely.

The honest conclusion

Turnout modelling is the single largest source of legitimate disagreement between reputable pollsters measuring the same race at the same time. It is also the component most likely to cause a large-scale polling miss, because it depends on assumptions about future human behaviour that are fundamentally uncertain. The reader who understands this — who knows to ask "likely voters according to what model?" — is already reading polls more carefully than most journalists.

For more on how polls handle uncertainty, see our guide to margin of error, and for a broader overview of why polls can miss, read why polls sometimes miss elections by a mile.