Frequently Asked Questions About Carney and Poilievre Polling

Canadian federal polls typically have margins of error between ±2% and ±3% at the 95% confidence level for samples of 1,200-2,000 respondents. However, historical accuracy varies significantly. In the 2021 federal election, major pollsters were generally within 2-3 points on Liberal and Conservative vote share, though they underestimated NDP support in several regions. The 2019 election saw larger misses, with some firms overestimating Conservative support by 4-5 points nationally. Polling accuracy for Carney specifically faces additional challenges because he lacks an established voting record in federal politics, making voter intentions more volatile. Abacus Data's post-election analysis of the 2021 campaign found that polls conducted within the final week had an average error of 2.1 points per party, while polls 3-4 weeks out had average errors of 3.8 points. Given that a potential Carney-Poilievre election likely wouldn't occur until late 2025, current polls should be viewed as snapshots of current sentiment rather than predictions of electoral outcomes.

The six largest Canadian polling firms employ different methodologies that can affect results. Abacus Data uses online panels recruited through multiple sources, with quotas set for age, gender, region, and education to match census data. They typically survey 2,000-2,500 respondents and apply weighting adjustments. Leger uses a hybrid approach combining their online LEO panel (over 400,000 Canadians) with telephone surveys for quality control, usually sampling 1,500-2,000 people. Nanos Research employs live-caller telephone surveys using both landlines and cell phones, with smaller sample sizes (1,000-1,200) but longer interview times averaging 8-12 minutes. Mainstreet Research uses automated interactive voice response (IVR) technology combined with online surveys, often achieving larger samples (3,000-5,000) at lower cost. Ipsos relies primarily on their online panel with rigorous quality controls and typically surveys 1,800-2,000 respondents. Angus Reid Institute uses their proprietary online panel of approximately 100,000 Canadians, sampling 1,500-2,000 with detailed demographic weighting. Each methodology has strengths and weaknesses—live-caller surveys may reach more representative samples but have lower response rates (often below 10%), while online panels allow larger samples but may underrepresent older Canadians and those in rural areas despite weighting adjustments.

Poll-to-poll variation stems from multiple factors beyond random sampling error. First, different firms use different methodologies as described above, which can systematically affect results by 2-4 points. Online panels tend to show slightly higher support for parties with younger demographics (NDP, Greens), while telephone surveys may oversample older, more politically engaged voters who lean Conservative or Liberal. Second, question wording and order significantly impact responses. Polls that ask about leadership approval before vote intention may prime respondents differently than those that lead with policy questions. Third, sample composition varies—some firms oversample key battleground regions and then weight results, while others maintain strict proportional sampling. Fourth, the timing within a polling period matters; a poll fielded Monday-Wednesday may capture different sentiment than one conducted Thursday-Sunday due to news cycles. Fifth, weighting decisions differ substantially between firms. Some pollsters weight only on basic demographics (age, gender, region), while others include education, income, past vote recall, and even media consumption patterns. Analysis by Philippe J. Fournier of 338Canada found that systematic differences between polling firms remained consistent across multiple elections, suggesting house effects of 2-3 points. When evaluating polls, examining aggregates or averages of multiple firms provides more reliable insights than any single poll.

Undecided voters typically represent 8-15% of poll respondents in Canadian federal surveys, though this varies by timing and political context. Pollsters handle undecided voters through several approaches. Most commonly, firms report results both with and without undecided voters—'decided voters only' numbers show higher percentages for each party, while 'all respondents' numbers include undecided as a separate category. Some pollsters use 'leaning' questions, asking undecided voters which party they're leaning toward even slightly, which can reduce the undecided pool to 5-8%. Research by Nanos and others shows that approximately 60-70% of voters who claim to be undecided 3-4 weeks before an election ultimately vote for the party they supported in the previous election, suggesting many 'undecided' voters are actually disengaged partisans. However, genuinely undecided swing voters, though smaller in number (typically 4-6% of the electorate), disproportionately determine election outcomes. In the current Carney-Poilievre polling, undecided voters skew younger (18-34 age group) and are more likely to be women, according to Environics data. Historical patterns suggest these voters break toward the perceived frontrunner in the final week of campaigns, though economic conditions and debate performances can shift this significantly. The 2019 election saw undecided voters break 38% Liberal, 31% Conservative, and 22% NDP in the final week, contributing to a closer result than late polls suggested.

The margin of error (MOE) represents the statistical uncertainty in poll results due to sampling a subset of the population rather than surveying everyone. A poll with ±2.5% MOE at the 95% confidence level means that if the same poll were repeated 100 times with different random samples, approximately 95 of those polls would show results within 2.5 percentage points of the true population value. However, several critical nuances are often misunderstood. First, the MOE applies to each party's result independently—if Conservatives poll at 40% (±2.5%) and Liberals at 32% (±2.5%), the actual Conservative lead could range from 3 to 13 points because the errors compound when comparing parties. Second, the stated MOE only accounts for random sampling error, not systematic biases from non-response, question wording, or weighting decisions. Third, MOE is larger for subgroups—while a 1,500-person national poll might have ±2.5% MOE overall, results for Ontario alone (perhaps 500 respondents) would have ±4.4% MOE, and results for 18-34 year-olds in Ontario might have ±8-10% MOE. Fourth, the MOE assumes a truly random sample, which is increasingly difficult to achieve as response rates decline. Mainstreet Research reports response rates of 5-8% for telephone surveys, meaning that the 92-95% who don't respond may differ systematically from those who do. The Pew Research Center has documented that politically engaged voters are 3-4 times more likely to complete political surveys, potentially skewing results. When interpreting polls, consider the MOE as a minimum bound of uncertainty, with actual uncertainty likely 1.5-2 times larger when accounting for non-sampling errors.

Regional breakdowns from national polls should be interpreted with significant caution due to smaller sample sizes and increased margins of error. A typical national poll of 1,500 respondents might include only 200-250 respondents from British Columbia, yielding a margin of error of ±6-7%, and perhaps only 50-75 respondents from Atlantic Canada, with MOE exceeding ±10%. These large error margins mean that apparent regional swings may simply reflect statistical noise rather than genuine shifts in voter sentiment. More reliable regional data comes from province-specific polls commissioned by local media or conducted by firms like Mainstreet Research, which frequently conducts provincial polls with 800-1,200 respondents, providing MOE of ±3-4%. For the Carney-Poilievre matchup, provincial polls in Ontario, Quebec, and British Columbia offer substantially more reliable insights than national poll breakdowns. Research Co. and Leger regularly conduct Quebec-specific polls with sample sizes of 1,000+, providing much clearer pictures of the complex three-way competition between Liberals, Conservatives, and Bloc Québécois. Similarly, Innovative Research Group and Research Co. conduct BC-specific tracking with sufficient samples to analyze sub-regional patterns in Metro Vancouver versus Vancouver Island versus the Interior. When examining regional data, prioritize dedicated provincial polls over national poll breakdowns, and always check the actual sample size for the region in question. The about section of our site provides additional context on how regional variations affect overall electoral math and seat projections.