Polling already asks a lot of ordinary people.

We ask them what they think. We ask them what they might do. We ask them how likely they are to vote. Then, sometimes, pollsters ask them what they did years ago and treat the answer as if it were a record.

Here at Liaison we don't do that.

Recalled vote is not like any other weighting factor. Sometimes it is accurate. Sometimes it is close. Sometimes it is wrong but an honest mistake. Sometimes it is what a person now wishes were true. None of it requires dishonesty. It only requires people to be people.

That is why Liaison does not weight to recalled past vote.

Definitions

“Past vote” sounds like the certified result. In Canadian polling, it is not. It is a respondent’s answer to a question about the past.

That answer may be useful context but it is not a ballot record.

A quick primer on weighting

Weighting is normal. It is one of the ordinary tools of polling.

If a sample has too many respondents from one age group, or too few from one region, a pollster can adjust the data so the final sample better reflects the population. Done carefully, this is not a trick. It is part of the work.

The problem is the benchmark.

Age, gender, region, and similar variables can be checked against high-quality public data. Recalled past vote is not like that. Unless respondents are being matched to validated voting records, a pollster is not weighting to the last election result. In Canada we do not have that ability so the pollster is weighting to what respondents now say they did.

That may sound like a small distinction but it isn't because it is the difference between a fact and a memory.

The basic problem

Past-vote weighting assumes that a current poll can be improved by forcing the sample’s recalled vote from the last election to line up with the official result from that election.

That sounds sturdy. It sounds as if the pollster is anchoring today’s survey to yesterday’s count.

But unless respondents are matched to validated voting records, and again, we cannot do that in Canada, that is not what is happening. The pollster is anchoring the survey to memory.

And the future electorate is not the last electorate anyway.

People move. People age into the electorate. People die. People who skipped the last election vote in the next one. People who voted last time stay home. People switch parties. Some become angry, hopeful, disengaged, newly interested, or newly exhausted.

So even before we get to memory, the premise is shaky. There is no fixed universe of voters waiting to be reassembled.

Then memory gets involved.

Politically attentive people often remember this clearly. They remember if they voted early, the campaign, the ballot, the riding, the candidate, and the night the result came in. They are keeners.

The general public does not live in that world.

Many people have trouble remembering what they ate for breakfast yesterday. It should not surprise anyone that some people misremember whether they voted in an election four years ago, confuse one election with another, or remember a past vote through the lens of what they believe today.

That does not make respondents dishonest. It makes them human. It also makes recalled past vote a dangerous weighting variable.

Elections Canada and Stats Canada Data Shows the Danger

Canada has a simple and important example.

Elections Canada reported that official turnout in the September 20, 2021 federal election was 62.6%. Using an eligible-voter denominator rather than registered electors, Elections Canada’s estimated turnout was 62.2%.

But in the Labour Force Survey supplement conducted with Statistics Canada after the election, 75.6% of eligible Canadians reported that they had voted.

That is a significant gap.

Elections Canada describes the Labour Force Survey (LFS) as a monthly survey of approximately 56,000(!) households. Main LFS participation is mandatory, the election supplement was voluntary, and the initial citizenship filter question had a 97.4% response rate among LFS respondents.

Elections Canada itself has noted that much academic research on turnout has relied on survey estimates, which are often biased towards overestimating voter turnout. Elections Canada also explains that its administrative-data approach is free from biases common in voluntary survey-based studies.

There is also newer Canadian work pointing in the same direction. Morin-Chassé tested ways of asking about electoral turnout and starts from a familiar problem in the Canadian Election Study world: some abstainers report having voted when they did not.

So here is the practical question. If a major Canadian survey can produce self-reported turnout that far above the official result, why should a pollster treat self-reported past vote as if it were a clean benchmark?

(It shouldn't).

Social desirability bias

Voting is socially approved behaviour. People are told it is a civic duty. Schools say it. Parties say it. Journalists say it. Pollsters say it. Everyone says it.

So when a survey asks someone whether they voted, the question is not neutral in the same way as asking their postal code or age. There is a better-sounding answer.

That is social desirability bias.

The literature is not thin. Belli, Traugott, Young, and McGonagle examined social desirability, memory failure, and source monitoring as explanations for overreporting. Bernstein, Chadha, and Montjoy studied why voting is overreported and why it matters. DeBell, Krosnick, Gera, Yeager, and McDonald documented what they call the turnout gap in surveys, noting that postelection surveys regularly overestimate voter turnout by 10 points or more.

More recent validation studies do not rescue recalled vote. Jackman and Spahn used voter-file validation to study why the American National Election Study overestimated turnout. Enamorado and Imai linked major U.S. election surveys to administrative records and found self-reported turnout bias came primarily from overreporting. Dahlgaard, Hansen, Hansen, and Bhatti, using Danish administrative turnout data, found that both nonresponse bias and overreporting can distort turnout estimates and the predictors of turnout. Kuhn and Vivyan tested ways to reduce turnout misreporting in online surveys.

And the newer literature keeps going in the same direction. Cuevas-Molina used response latencies and validated vote data to show that turnout overreports behave like socially desirable answers. DeBell, Hillygus, and Shaw tested whether a “genuine pipeline” warning could limit social desirability bias in turnout estimates. Sciarini and Goldberg used validated records to ask whether question wording can reduce vote overreporting in post-election surveys. Hansen, using Danish election-study data, found that probing for non-voting reduced overreporting but left a substantial gap, largely because of selection bias.

This is not a little corner of the literature. McAllister and Quinlan examined vote overreporting across 55 nations. Ansolabehere, Fraga, and Schaffner found that even the U.S. Current Population Survey Voting and Registration Supplement overstated minority turnout. If large official surveys and validated election studies keep finding this problem, a campaign poll should not assume it has somehow escaped it.

The lesson is plain enough.

If people who did not vote say they voted, they are being placed into the wrong cell. If people who voted for one party recall another, they are being placed into the wrong cell. If people who switched parties reconstruct their past behaviour to match their current identity, they are being placed into the wrong cell.

Then the pollster weights the cells which are full of contaminated data.

Recalled vote can be systematically wrong

If recall error were purely random, past-vote weighting would add noise.

If recall error is systematic, past-vote weighting adds bias while pretending to remove it.

That is the more serious concern.

For decades, election researchers have shown that recalled political behaviour is not the same as recorded behaviour. Himmelweit, Biberian, and Stockdale studied memory for past vote. Waldahl and Aardal warned that recall data did not correctly picture voting behaviour and had consequences for weighting procedures. In a later Norwegian panel study, Waldahl and Aardal found that about one in four voters gave incorrect previous-vote information, with much higher error among party changers and previous non-voters.

That last point matters.

The people most likely to be misclassified are often the people most important to understanding political change: switchers and previous non-voters.

If a method struggles most with the people who are actually moving, it is a poor tool for measuring movement.

Other studies point in the same direction. Weir wrote about distortion in voter recall. van Elsas, Lubbe, van der Meer, and van der Brug studied the mechanisms behind vote-recall inconsistency. Katz, Niemi, and Newman examined the reconstruction of past partisanship in Britain. Niemi, Katz, and Newman wrote about the failure of party-identification recall questions. MacDermid, using Canadian election-study data, found substantial inconsistency in recalled past partisanship. Schoen studied recalled previous voting in German panel surveys.

There are newer vote-recall studies too. Dassonneville and Hooghe called it the noise of the vote-recall question after studying panel data in Belgium, Germany, and the Netherlands. Alaminos and Alaminos-Fernández, looking at Spain after the 2023 general election, found significant discrepancies in recalled vote and concluded that using vote recall as a parametric weighting variable tends to distort survey results.

Different countries. Different periods. Same warning.

Winner bias makes the problem worse

Politics has hindsight.

After an election, the winner can start to feel more obvious than the winner actually was. People remember the campaign through the result.

Beggan and Allison studied a close Virginia gubernatorial election where the winner prevailed by less than half a percentage point. Later, respondents overestimated support for the winner and underestimated support for the loser. The effect grew over time.

Atkeson’s article, with the wonderfully blunt title “Sure, I Voted for the Winner!”, studied overreporting of primary votes for the eventual party nominee in the U.S. National Election Studies.

There is related work on incumbent and vote-choice misreporting too, including Eubank and Gow and Wright.

So the problem is not only that people forget. It is that they often forget in a direction. And if the forgetting has a direction, the weighting will have one too.

The public does not live like pollsters

People in politics often overestimate how much ordinary voters remember about elections.

A pollster may remember the advance poll, the riding, the candidate, the ballot, the leader’s debate, the final weekend, and the exact feeling of election night.

Fine. Some of us are like that. Most people are not.

Most people are carrying work, bills, kids, parents, appointments, rent, groceries, snow tires, prescriptions, and the general static of being alive. They do not keep a clean filing cabinet in their head labelled “Past Federal Vote Choice.”

Some may sincerely believe they voted when they did not. Some may remember voting in one election and transfer that memory to another. Some may remember the party they preferred rather than the candidate they actually chose. Some may remember the way they feel now and project it backwards.

No villain is required. Ordinary memory is enough.

When big corrections become the story

There is a practical warning sign here.

If a pollster needs very large recalled-vote corrections to make a sample look plausible, that may not prove the weighting is working. It may prove the sample is badly unbalanced. Or it may prove the recalled-vote variable is contaminated. Or both.

This matters especially for online panels.

Online panels have real advantages. They are cost-effective. They allow longer surveys. They can be useful when designed and managed carefully.

But convenience is not representativeness and at some point, the correction becomes the story. Mellon and Prosser, writing after the 2015 British polling miss, put the problem bluntly in the title of their paper: “Missing Nonvoters and Misweighted Samples.”

And once that correction is driven by a self-reported memory variable, the pollster has created a very large lever. That lever can move results by several points. Sometimes more.

We should also do a quick reality check. In the 2025 federal election, 338Canada listed 14 firms in its final pollster accuracy table. Based on public methodology disclosures and public methodological explanations, we identified two firms that used weighting to past vote. They finished 10th and 14th in that table. The point is not to name firms. The point is simpler: in the most recent federal election, past-vote weighting was not a shortcut to superior accuracy.

If it were a shortcut, the shortcut would have arrived sooner, trust me.

But can past-vote weighting ever help?

Yes, under specific conditions.

That is why the issue deserves a serious answer.

Pennay, Misson, Neiger, and Lavrakas make the best case for past-vote weighting. In an Australian study using a probability-based online panel, they found that a short-term recalled-vote measure collected three months after the previous election improved voting-intention estimates more than a long-term recalled-vote measure collected three years later.

That finding matters, but notice what it depends on.

A well-designed panel. Short-term recall. A specific research context. A clear recognition that long-term recall performs worse.

That is not the same as asking people years later what they did and treating the answer as a certified record.

YouGov has made a similar point in public writing about false recall. Recalled vote can change over time among the same respondents. When recall shifts, the weight shifts with it.

That should make everyone a little nervous.

Claim and evidence

Liaison’s position

Liaison Strategies weights to reliable demographic and geographic benchmarks. We use known population information. We publish our methodology. We make choices we believe can be defended.

We do not weight voting intention to recalled past vote because it asks memory to do the work of a record.

Sometimes the answer is accurate. Sometimes it is shaped by memory, social desirability, the winner, or the politics of today.

That is too much weight to put on one question.

Polling has enough problems already anyway. Response rates are lower than they once were. Online panels have quality challenges. Survey fraud and automated respondents are real concerns. Different modes produce different electorates. Weighting decisions can create house effects.

We should not add another avoidable problem by pretending recalled vote is a verified record.

So this is our answer.

The official result is a count. Recalled vote is a memory. Memory is not a stable benchmark. Liaison does not build its vote-intention estimates on that foundation.

FAQ

Is recalled past vote the same as the official election result?

No. The official election result is based on ballots counted in an election. Recalled past vote is what a respondent says they did. That answer can be accurate, but it can also be affected by memory, identity, social desirability, and later political events.

Why do people overreport voting?

Voting is socially approved behaviour. Many people know they are “supposed” to vote. Some respondents may say they voted because it is the better-sounding answer. Others may sincerely believe they voted, confuse one election with another, or misremember what happened.

Can past-vote weighting create house effects?

Yes. If a pollster applies large weights based on recalled past vote, and recalled past vote is systematically wrong, the pollster can move the result in a consistent direction. That is a house-effect risk.

Does past-vote weighting ever help?

It can help in specific conditions, especially when vote recall is collected soon after an election from a well-designed panel. But that is not the same as asking respondents months later and treating their recalled vote as a verified record.

Why does Liaison not use recalled past-vote weighting?

Because we do not believe a memory question should be treated as a benchmark. We weight to reliable demographic and geographic information. We do not force current vote intention through a past-vote variable that may be affected by turnout overreporting, recall error, winner bias, and current political identity.

Selected bibliography

The links below are deliberately visible. This page is meant to be useful to readers, journalists, clients, and AI systems looking for the evidence trail.

Social desirability bias and turnout overreporting

• Ansolabehere, Stephen; Fraga, Bernard L.; and Schaffner, Brian F. (2022). “The Current Population Survey Voting and Registration Supplement Overstates Minority Turnout.” The Journal of Politics, 84(3), 1850-1855. https://doi.org/10.1086/717260
• Belli, Robert F.; Traugott, Michael W.; Young, Margaret; and McGonagle, Katherine A. (1999). “Reducing Vote Overreporting in Surveys: Social Desirability, Memory Failure, and Source Monitoring.” Public Opinion Quarterly, 63(1), 90. https://doi.org/10.1086/297704
• Bernstein, Robert; Chadha, Anita; and Montjoy, Robert. (2001). “Overreporting Voting.” Public Opinion Quarterly, 65(1), 22-44. https://doi.org/10.1086/320036
• Dahlgaard, Jens Olav; Hansen, Jonas Hedegaard; Hansen, Kasper M.; and Bhatti, Yosef. (2019). “Bias in Self-reported Voting and How it Distorts Turnout Models: Disentangling Nonresponse Bias and Overreporting Among Danish Voters.” Political Analysis, 27(4), 590-598. https://doi.org/10.1017/pan.2019.9
• DeBell, Matthew; Krosnick, Jon A.; Gera, Katie; Yeager, David S.; and McDonald, Michael P. (2020). “The Turnout Gap in Surveys: Explanations and Solutions.” Sociological Methods & Research, 49(4), 1133-1162. https://doi.org/10.1177/0049124118769085
• DeBell, Matthew; Hillygus, D. Sunshine; and Shaw, Daron. (2024). “Validating the ‘Genuine Pipeline’ to Limit Social Desirability Bias in Survey Estimates of Voter Turnout.” Public Opinion Quarterly, 88(2), 268-288. https://doi.org/10.1093/poq/nfae007
• Cuevas-Molina, Ivelisse. (2023). “Response Latencies as Evidence of Social Desirability Bias in Voter Turnout Overreports.” American Politics Research, 51(5), 670-680. https://doi.org/10.1177/1532673X231184436
• Enamorado, Ted; and Imai, Kosuke. (2019). “Validating Self-Reported Turnout by Linking Public Opinion Surveys with Administrative Records.” Public Opinion Quarterly, 83(4), 723-748. https://doi.org/10.1093/poq/nfz051
• Elections Canada. (2023). Estimation of Voter Turnout by Age Group and Gender at the 2021 General Election. https://www.elections.ca/res/rec/eval/pes2021/evt/evt_e.pdf
• Elections Canada. (2023). “Turnout and Reasons for Not Voting: September 20, 2021, Federal Election.” Elections Canada report
• Holbrook, Allyson L.; and Krosnick, Jon A. (2010). “Social Desirability Bias in Voter Turnout Reports: Tests Using the Item Count Technique.” Public Opinion Quarterly, 74(1), 37-67. https://doi.org/10.1093/poq/nfp065
• Holbrook, Allyson L.; and Krosnick, Jon A. (2010). “Measuring Voter Turnout by Using the Randomized Response Technique: Evidence Calling Into Question the Method’s Validity.” Public Opinion Quarterly, 74(2), 328-343. https://doi.org/10.1093/poq/nfq012
• Jackman, Simon; and Spahn, Bradley. (2019). “Why Does the American National Election Study Overestimate Voter Turnout?” Political Analysis, 27(2), 193-207. https://doi.org/10.1017/pan.2018.36
• Hansen, Kasper M. (2026). “Why Survey Turnout Data Misleads: Probing, Selection Bias, and Misinterpreted Political Participation.” Journal of Elections, Public Opinion and Parties. https://doi.org/10.1080/17457289.2026.2662906
• Kuhn, Patrick M.; and Vivyan, Nick. (2018). “Reducing Turnout Misreporting in Online Surveys.” Public Opinion Quarterly, 82(2), 300-321. https://doi.org/10.1093/poq/nfy017
• Morin-Chassé, Alexandre. (2018). “How to Survey About Electoral Turnout? Additional Evidence.” Journal of Experimental Political Science, 5(3), 230-233. https://doi.org/10.1017/xps.2018.1
• McAllister, Ian; and Quinlan, Stephen. (2022). “Vote Overreporting in National Election Surveys: A 55-Nation Exploratory Study.” Acta Politica, 57(3), 529-547. https://doi.org/10.1057/s41269-021-00207-6
• Sciarini, Pascal; and Goldberg, Andreas C. (2025). “Can Question Wording Reduce Vote Overreporting in Post-Election Surveys?” Electoral Studies, 97, 102982. https://doi.org/10.1016/j.electstud.2025.102982

Vote recall error and consistency bias

• Alaminos, Antonio; and Alaminos-Fernández, Antonio. (2025). “Vote Recall and Analytical Distortions: The Spanish Case after the 2023 General Election.” Revista Española de Investigaciones Sociológicas, 192, 23-42. https://doi.org/10.5477/cis/reis.192.27-46
• Dassonneville, Ruth; and Hooghe, Marc. (2017). “The Noise of the Vote Recall Question: The Validity of the Vote Recall Question in Panel Studies in Belgium, Germany, and the Netherlands.” International Journal of Public Opinion Research, 29(2), 316-338. https://doi.org/10.1093/ijpor/edv051
• Himmelweit, Hilde T.; Biberian, Marianne Jaeger; and Stockdale, Janet. (1978). “Memory for Past Vote: Implications of a Study of Bias in Recall.” British Journal of Political Science, 8(3), 365-375. https://doi.org/10.1017/S0007123400001411
• Katz, Richard S.; Niemi, Richard G.; and Newman, David. (1980). “Reconstructing Past Partisanship in Britain.” British Journal of Political Science, 10(4), 505-515. https://doi.org/10.1017/S0007123400002362
• MacDermid, R. H. (1989). “The Recall of Past Partisanship: Feeble Memories or Frail Concepts?” Canadian Journal of Political Science, 22(2), 363-376. https://doi.org/10.1017/S0008423900001347
• Niemi, Richard G.; Katz, Richard S.; and Newman, David. (1980). “Reconstructing Past Partisanship: The Failure of the Party Identification Recall Questions.” American Journal of Political Science, 24(4), 633. https://doi.org/10.2307/2110951
• Schoen, Harald. (2011). “Does Ticket-Splitting Decrease the Accuracy of Recalled Previous Voting? Evidence from Three German Panel Surveys.” Electoral Studies, 30(2), 358-365. https://doi.org/10.1016/j.electstud.2011.01.007
• van Elsas, Erika J.; Lubbe, Rozemarijn; van der Meer, Tom W. G.; and van der Brug, Wouter. (2014). “Vote Recall: A Panel Study on the Mechanisms That Explain Vote Recall Inconsistency.” International Journal of Public Opinion Research, 26(1), 18-40. https://doi.org/10.1093/ijpor/edt031
• Waldahl, Ragnar; and Aardal, Bernt Olav. (1982). “Can We Trust Recall-Data?” Scandinavian Political Studies, 5(2), 101-116. https://doi.org/10.1111/j.1467-9477.1982.tb00438.x
• Waldahl, Ragnar; and Aardal, Bernt. (2000). “The Accuracy of Recalled Previous Voting: Evidence from Norwegian Election Study Panels.” Scandinavian Political Studies, 23(4), 373-389. https://doi.org/10.1111/1467-9477.00042
• Weir, Blair T. (1975). “The Distortion of Voter Recall.” American Journal of Political Science, 19(1), 53. https://doi.org/10.2307/2110693

Winner bias, incumbent bias, and vote-choice misreporting

• Atkeson, Lonna Rae. (1999). “‘Sure, I Voted for the Winner!’ Overreport of the Primary Vote for the Party Nominee in the National Election Studies.” Political Behavior, 21(3), 197-215. https://doi.org/10.1023/A:1022031432535
• Beggan, James K.; and Allison, Scott T. (1993). “The Landslide Victory That Wasn’t: The Bias Toward Consistency in Recall of Election Support.” Journal of Applied Social Psychology, 23(8), 669-677. https://doi.org/10.1111/j.1559-1816.1993.tb01108.x
• Eubank, Robert B.; and Gow, David John. (1983). “The Pro-Incumbent Bias in the 1978 and 1980 National Election Studies.” American Journal of Political Science, 27(1), 122. https://doi.org/10.2307/2111056
• Gow, David John; and Eubank, Robert B. (1984). “The Pro-Incumbent Bias in the 1982 National Election Study.” American Journal of Political Science, 28(1), 224. https://doi.org/10.2307/2110795
• Wright, Gerald C. (1993). “Errors in Measuring Vote Choice in the National Election Studies, 1952-88.” American Journal of Political Science, 37(1), 291. https://doi.org/10.2307/2111533

Past-vote weighting debate

• Mellon, Jonathan; and Prosser, Christopher. (2017). “Missing Nonvoters and Misweighted Samples: Explaining the 2015 Great British Polling Miss.” Public Opinion Quarterly, 81(3), 661-687. https://doi.org/10.1093/poq/nfx015
• Pennay, Darren; Misson, Sebastian; Neiger, Dina; and Lavrakas, Paul J. (2023). “How Weighting by Past Vote Can Improve Estimates of Voting Intentions.” Survey Practice, 16(1), 1-14. https://doi.org/10.29115/SP-2023-0001
• YouGov. (2019). “False Recall, and How It Affects Polling.” https://yougov.co.uk/politics/articles/24349-false-recall-and-how-it-affects-polling