Two likely voter polls can show different results without either poll being dishonest or careless. The difference often starts with one question: who counts as part of the electorate?
A registered-voter poll describes the views of people who say they can vote. A likely-voter poll estimates the views of people who are expected to cast ballots. Those groups overlap, but they aren't identical. Before we compare poll numbers, we need to know which group each poll is measuring.
Registered Voters and Likely Voters Are Different Groups
A registered voter is usually someone who tells a pollster they are registered to vote. Pollsters may verify that claim against a voter file, but many surveys rely on the respondent's answer. Gallup provides a useful explanation of how registered-voter and likely-voter categories differ in practice.
A likely voter is a registered voter who appears likely to participate in the upcoming election. Pollsters make that judgment using answers about voting plans, past participation, interest in the campaign, and knowledge of the voting process.
The distinction looks simple, but it changes the question the poll answers.
| Poll population | What the poll measures |
|---|---|
| Adults | Views among all adults surveyed |
| Registered voters | Views among people who say they're registered |
| Likely voters | Views among registered voters expected to cast a ballot |
A registered-voter poll may include someone who registered years ago but rarely votes. A likely-voter poll may give that person less influence, exclude them, or assign them a low turnout probability.
That doesn't make registered voters irrelevant. They can show how opinions look among the broader pool of people eligible to participate. They also help pollsters study people who may vote later, especially when the election is still months away.
Likely voter polls answer a narrower question: what might the preferences of the actual electorate look like? The word "might" matters. The electorate doesn't fully exist when the survey is taken. Pollsters are estimating who will show up on Election Day.
A registered-voter poll measures a defined group. A likely-voter poll measures a predicted group within it.
The two results can be close. They can also move apart when voters who are less likely to participate hold different views from habitual voters.
How Likely Voter Polls Estimate Turnout
Likely-voter models are not one universal formula. Each pollster chooses its own questions, scoring rules, data sources, and cutoff.
Some models begin with a broad survey and screen respondents. Others use commercial voter files that include registration records and, in some states, a history of whether a person voted in previous elections. A pollster may combine that information with survey answers about current interest and voting plans.
Common questions include:
- Whether the respondent plans to vote.
- How interested the respondent is in the election.
- Whether they know where or how to vote.
- Whether they voted in previous elections.
- How often they usually vote.
- Whether they have already voted, where early voting is available.
Gallup's traditional model has used a seven-question scale. Answers that indicate stronger participation raise a respondent's score. Nonregistered respondents receive no place in the likely-voter pool, while someone who reports already voting receives the strongest possible turnout score.
Other pollsters use different indexes. Pew Research Center has tested measures involving past voting history, campaign interest, and knowledge of voting procedures. The results can change depending on how much weight a model gives to each factor.
The AAPOR presentation on likely voters puts the central problem plainly: likely-voter models estimate a population that doesn't yet exist at the time of the poll, the future electorate.
That future group is hard to predict because turnout changes by election. Presidential elections usually attract more voters than local contests. Midterm elections often bring a smaller and different electorate. Primary elections, special elections, and ballot measures can produce still other turnout patterns.
A strict model may prioritize people with a strong voting history. An expanded model may include people who say they are certain to vote, even if their past participation is limited. Neither choice is automatically correct.
Likely-voter models are estimates, not superior samples by definition. They can improve a forecast when the turnout assumptions fit the election. They can also miss voters when participation rises among people the model treats as unlikely.
Sample Composition Is Not the Same as Weighting
Poll discussions often blur two separate steps: deciding who belongs in the sample and adjusting the sample after interviews are complete.
Sample composition describes who was included. A poll might interview registered voters across several age groups, regions, education levels, or party affiliations. A likely-voter screen then narrows that pool based on expected turnout.
Weighting adjusts the influence of people already included. If a poll interviews too many older voters compared with the target population, the pollster may give each older respondent less weight. Younger respondents may receive more weight.
Weighting cannot add people who were never interviewed. If a likely-voter screen removes too many low-propensity voters, later weighting usually doesn't restore the full group. The two processes affect the poll in different ways.
A poll can also weight for several characteristics at once. Age, race, gender, education, geography, and past vote may all enter the adjustment process. The more a poll relies on modeled targets, the more important it becomes to read the methodology rather than the headline.
Registered voters and likely voters may also come from different sampling systems. A survey of adults may ask respondents whether they are registered. A campaign poll may draw names from a voter database. Those approaches can reach different people before turnout modeling begins.
The difference isn't always dramatic. In one F&M poll comparison, the likely-voter and registered-voter estimates in a Senate race were only two points apart in August. That example matters because it shows that changing the population doesn't guarantee a large change in the result.
When the numbers do differ, we shouldn't assume the likely-voter result is automatically more realistic. We should ask what the model excluded and whether those exclusions match the election being studied.
Why Results Can Change as Election Day Gets Closer
Early in a campaign, many voters haven't formed firm plans. Some aren't paying attention. Others may become engaged after debates, major news events, candidate changes, or new information about voting rules.
That creates a timing problem. A person who looks unlikely to vote in June may become a reliable voter by November. A person who expresses strong intentions early may not follow through.
Pollsters handle this uncertainty in different ways. Some report registered-voter results until the campaign is further along. Pew has generally waited until after the nominating conventions before reporting likely-voter findings in presidential elections. The timing gives pollsters more information about interest and participation.
Turnout is also shaped by practical barriers, not only political enthusiasm. Registration deadlines, polling-place changes, work schedules, transportation, illness, and confusion about voting procedures can affect who actually casts a ballot. Our coverage of election participation obstacles shows why a person can be eligible, registered, and politically interested without successfully voting.
This is where likely-voter screens can create their largest errors. If a model favors people with a long voting history, it may undercount new voters. If it accepts stated intention too freely, it may include people who don't follow through.
The best model depends on the election. A method that performs well in a high-turnout presidential contest may not fit a low-turnout municipal election. Poll readers should treat the model as part of the evidence, not as a stamp of certainty.
How to Read Likely Voter Polls Without Overreading Them
When we encounter a poll headline, the topline number is only the beginning. We look for the population, field dates, sample size, mode of interviewing, weighting variables, and turnout screen.
The population label should come first. "Among registered voters" and "among likely voters" aren't interchangeable descriptions. A poll of adults may produce a different result again because it includes people who aren't registered.
The field dates matter because opinions and turnout intentions can change quickly. A poll conducted before a debate doesn't measure the same moment as one conducted afterward.
The pollster's likely-voter definition deserves close attention. Look for the questions used to identify likely voters, the score or cutoff, and whether the model uses verified voting history. If the methodology page doesn't explain those points, the result deserves less confidence.
Sample size matters, but a larger sample doesn't erase a flawed turnout model. The margin of sampling error describes uncertainty from surveying a sample rather than everyone in the population. It doesn't capture every source of error, including nonresponse, inaccurate voter records, question wording, or a turnout assumption that misses the electorate.
When comparing two polls, use similar populations whenever possible. Comparing a registered-voter poll with a likely-voter poll may tell us something useful, but it isn't a clean trend line. The polls may be measuring different groups with different rules.
A practical reading order looks like this:
- Identify the population being measured.
- Check when the interviews took place.
- Read how the pollster defines a likely voter.
- Review the weighting and sampling method.
- Compare results with other polls using the same population.
- Treat the margin of error as one uncertainty measure, not a full accuracy guarantee.
A poll with transparent methods can still be wrong. A poll with a small margin of error can still miss turnout. Good reporting keeps both facts in view.
What the Difference Means for Poll Readers
Registered-voter polls are often useful when the election is distant or turnout is unsettled. They show the broader opinion environment and include people who may become more active later.
Likely-voter polls can be more focused when Election Day is near and the pollster has credible evidence about who will participate. They may better approximate the voters who will decide the result, but only if the model recognizes the election's actual turnout pattern.
The difference can also affect campaign coverage. A candidate may lead among registered voters but trail among likely voters. That isn't necessarily a contradiction. It may mean the candidate's support is stronger among infrequent voters, younger voters, or people who are less engaged at that point in the campaign.
The reverse can happen too. A candidate may perform better among likely voters because habitual voters favor that candidate. In either case, the poll doesn't prove that one group is more legitimate. It describes two different electorates.
We should be cautious when a headline presents one sample as the "real" public. The public is larger than the people predicted to vote, while the electorate is smaller than the public. Both matter, but they answer different questions.
Conclusion
Likely voter polls narrow a survey to people a pollster expects to vote. Registered-voter polls include a broader group of people who say they're registered. The difference comes from turnout prediction, not from a universal rule that one sample is always better.
When we read poll results, we check the population, timing, sample composition, weighting, and likely-voter screen before comparing numbers. That discipline prevents a familiar mistake: treating an estimate of the future electorate as if it were a direct count of voters already standing in line.