[ { "slug": "1936-literary-digest", "year": 1936, "title": "1936: millions of returns, the wrong electorate", "shortLabel": "Digest", "office": "President", "description": "Why did millions of Literary Digest responses favor Landon before Roosevelt’s landslide? Compare the poll and outcome on one denominator and examine selection and nonresponse.", "takeaway": "A huge voluntary return cannot guarantee a representative sample.", "classification": "Wrong leader in a national straw poll", "basis": "two-party", "chartRows": [ { "label": "Literary Digest", "pollMargin": -14.152334412498904, "actualMargin": 24.912887221581187 } ], "chartNote": "Both sides are normalized to Roosevelt + Landon = 100%. Third parties are excluded from both denominators. The poll uses published voluntary response counts, not a probability sample. The historical outcome compilation can differ slightly across editions.", "facts": [ [ "Compared estimate", "Literary Digest" ], [ "Election", "1936 · President" ], [ "Published / fieldwork context", "Cards mailed in late summer; cumulative returns published through October 31, 1936. Exact field cutoff not established." ], [ "Error in the selected margin", "39.1 percentage points" ], [ "Comparison basis", "Two-candidate normalized on both sides" ] ], "dialogue": [ { "question": "What did the selected poll say?", "paragraphs": [ "The published Digest returns included 972,897 Roosevelt responses and 1,293,669 Landon responses. Among those two candidates, our normalized shares are 42.9% and 57.1%. On the same two-candidate basis, the poll margin was R +14.2 and the recorded outcome margin was D +24.9: a 39.1-point difference." ] }, { "question": "What actually happened?", "paragraphs": [ "The original published Digest count favored Landon; Roosevelt won decisively. Normalizing both to two-party shares puts their comparison on the same denominator." ] }, { "question": "Why might the poll have missed?", "paragraphs": [ "Historical reanalyses identify selective participation and coverage differences. The familiar explanation that telephone and automobile ownership alone caused the miss is incomplete. Evidence status: supported, contribution uncertain." ] }, { "question": "What has not been established?", "paragraphs": [ "The familiar story that access to a telephone or automobile alone explains the failure is too simple. Coverage and the decision to return a form both matter. Historical reanalysis does not make the voluntary sample representative after the fact." ] }, { "question": "How should we use this example?", "paragraphs": [ "Gallup’s separate final estimate put Roosevelt at 55.7 and Landon at 44.3. It identified the popular-vote winner while still understating Roosevelt’s eventual two-candidate margin. That contrast demonstrates why representative selection matters and why a correct winner call is not perfect accuracy.", "Both sides are normalized to Roosevelt + Landon = 100%. Third parties are excluded from both denominators. The poll uses published voluntary response counts, not a probability sample. The historical outcome compilation can differ slightly across editions." ] } ], "sources": [ { "label": "Literary Digest, October 31, 1936: original issue scan", "url": "https://www.rossmanchance.com/chance/LD/LitDigest1936.pdf", "kind": "primary historical document", "detail": "Original magazine pages hosted by statistics educator Beth Chance. HTTPS access verified with the system trust store; no certificate-warning bypass was used." }, { "label": "Arkansas Secretary of State Historical Report 2010, printed p. 585", "url": "https://www.sos.arkansas.gov/uploads/elections/historical_report1210.pdf", "kind": "official historical compilation", "detail": "" }, { "label": "Squire 1988, Why the 1936 Literary Digest Poll Failed", "url": "https://doi.org/10.1086/269085", "kind": "original empirical research", "detail": "Publisher access may be restricted; direct automated access returned 403. The institutional teaching mirror below provides the same original research article." }, { "label": "Lusinchi 2012, President Landon and the 1936 Literary Digest poll", "url": "https://pdixon.stat.iastate.edu/stat587/resources/president-landon-and-the-1936-literary-digest-poll-were-automobile-and-telephone-owners-to-blame.pdf", "kind": "original empirical research", "detail": "" }, { "label": "Squire 1988, original article: institutional teaching mirror", "url": "https://www.michaelperess.com/teaching/sscs2022/Squire.pdf", "kind": "original empirical research, teaching mirror", "detail": "The original Public Opinion Quarterly article, pages 125–133, with publication attribution preserved." } ] }, { "slug": "1948-truman-dewey", "year": 1948, "title": "1948: the Dewey lead that disappeared", "shortLabel": "Gallup", "office": "President", "description": "Compare Gallup’s final Dewey lead with Truman’s victory. Examine quota sampling, late fieldwork and why a newspaper call is different from a poll.", "takeaway": "The sampling and timing of a poll matter more than confidence in its winner call.", "classification": "Wrong leader in final national estimate", "basis": "raw", "chartRows": [ { "label": "Gallup", "pollMargin": -5, "actualMargin": 4.399999999999999 } ], "chartNote": "This compares one Gallup final allocated estimate with source-reported election shares rounded to 0.1 point. The plotted error is approximate raw D−R margin error. This is not the polling industry’s mean error.", "facts": [ [ "Compared estimate", "Gallup" ], [ "Election", "1948 · President" ], [ "Published / fieldwork context", "Final published estimate. October 15–25 fieldwork is commonly reported in historical compilations but was not independently established from an original release here." ], [ "Error in the selected margin", "9.4 percentage points · approximate" ], [ "Comparison basis", "Raw D−R margin, with source qualifications" ] ], "dialogue": [ { "question": "What did the selected poll say?", "paragraphs": [ "Gallup’s selected estimate put Harry Truman at 44.5% and Thomas Dewey at 49.5%. On the chart’s stated basis, that was R +5.0. The corresponding recorded outcome margin was D +4.4, for a 9.4-point difference in the margin." ] }, { "question": "What actually happened?", "paragraphs": [ "The three major polling organizations used quota sampling and all picked Dewey. The selected Gallup estimate favored Dewey by 5 points; the cited rounded outcome favored Truman by 4.4." ] }, { "question": "Why might the poll have missed?", "paragraphs": [ "AAPOR's methodological review describes how interviewer choices within quotas overrepresented more educated and affluent respondents, creating bias against Truman. Evidence status: supported.", "Gallup's retrospective also attributes part of the miss to interviewing ending before late movement toward Truman. This does not establish that a late swing alone explains the error. Evidence status: institutional retrospective, not exclusive cause." ] }, { "question": "What has not been established?", "paragraphs": [ "Neither late movement nor quota sampling should be presented as the entire explanation. A contemporary dataset holding the same sampling process and turnout model fixed would be needed to isolate their separate contributions." ] }, { "question": "How should we use this example?", "paragraphs": [ "Read the actual survey and its interviewing dates, not only the famous Chicago Tribune headline. A media organization’s premature call is a separate decision from a polling organization’s estimate.", "This compares one Gallup final allocated estimate with source-reported election shares rounded to 0.1 point. The plotted error is approximate raw D−R margin error. This is not the polling industry’s mean error." ] } ], "sources": [ { "label": "Gallup final estimate and historical result table", "url": "https://www.presidency.ucsb.edu/statistics/data/election-year-presidential-preferences", "kind": "academic compilation", "detail": "" }, { "label": "AAPOR Task Force on Nonprobability Sampling, printed p. 12", "url": "https://aapor.org/wp-content/uploads/2022/11/NPS_TF_Report_Final_7_revised_FNL_6_22_13.pdf", "kind": "professional methodological review", "detail": "" }, { "label": "Gallup institutional retrospective", "url": "https://news.gallup.com/poll/3376/George-Gallup-TwentiethCentury-Pioneer.aspx", "kind": "pollster retrospective", "detail": "" }, { "label": "Truman Presidential Library 1948 election educational sheet", "url": "https://www.trumanlibrary.gov/public/1948_election.pdf", "kind": "official presidential library", "detail": "" } ] }, { "slug": "1980-reagan-carter", "year": 1980, "title": "1980: the right winner, a much larger victory", "shortLabel": "Gallup", "office": "President", "description": "Gallup put Reagan ahead in 1980 but understated his winning margin. Examine changing preferences, turnout estimation and the limits of a correct winner call.", "takeaway": "Calling the winner correctly can conceal a substantial error in the victory margin.", "classification": "Large margin underestimate; final poll leader won", "basis": "raw", "chartRows": [ { "label": "Gallup", "pollMargin": -3, "actualMargin": -9.732703566693772 } ], "chartNote": "One Gallup final allocated estimate is compared with all-vote election shares. Carter and Reagan do not sum to 100% in either set of reported shares. The raw D−R margin is used, without two-party normalization.", "facts": [ [ "Compared estimate", "Gallup" ], [ "Election", "1980 · President" ], [ "Published / fieldwork context", "Final Gallup estimate immediately before November 4, 1980; exact original field dates not verified here." ], [ "Error in the selected margin", "6.7 percentage points" ], [ "Comparison basis", "Raw D−R margin, with source qualifications" ] ], "dialogue": [ { "question": "What did the selected poll say?", "paragraphs": [ "Gallup’s selected estimate put Jimmy Carter at 44.0% and Ronald Reagan at 47.0%. On the chart’s stated basis, that was R +3.0. The corresponding recorded outcome margin was R +9.7, for a 6.7-point difference in the margin." ] }, { "question": "What actually happened?", "paragraphs": [ "Gallup's final estimate put Reagan ahead 47–44. The result was approximately 51–41, a much wider win." ] }, { "question": "Why might the poll have missed?", "paragraphs": [ "Gallup's retrospective traces movement toward Reagan after the October 28 debate and discusses the Iran hostage crisis. This is evidence of changing measured preferences, not an identified causal effect of the debate. Evidence status: pollster interpretation.", "Fenwick and colleagues studied how to allocate undecided respondents in 1980 presidential polls and validated their method with a post-election follow-up. The verified abstract does not establish why this particular final Gallup estimate missed. Evidence status: methodological context; abstract verified." ] }, { "question": "What has not been established?", "paragraphs": [ "A measured movement after a debate does not prove the debate caused every changed vote. The comparison cannot identify which respondents later changed preferences or who ultimately turned out." ] }, { "question": "How should we use this example?", "paragraphs": [ "Report the point estimate, uncertainty, timing and numerical miss together. Label this a missed margin with a correct leader, rather than turning it into a false claim that the final Gallup poll picked Carter.", "One Gallup final allocated estimate is compared with all-vote election shares. Carter and Reagan do not sum to 100% in either set of reported shares. The raw D−R margin is used, without two-party normalization." ] } ], "sources": [ { "label": "Gallup historical final-estimate table", "url": "https://www.presidency.ucsb.edu/statistics/data/election-year-presidential-preferences", "kind": "academic compilation", "detail": "" }, { "label": "Gallup 2008, Late Upsets Are Rare, but Have Happened", "url": "https://news.gallup.com/poll/111451/late-upsets-rare-happened.aspx", "kind": "pollster retrospective", "detail": "" }, { "label": "George Gallup’s November 1980 postmortem", "url": "https://www.washingtonpost.com/archive/politics/1980/11/12/what-happened-to-the-polls/a5abac1c-ccb0-4828-8902-60b99ec5d401/", "kind": "first-person contemporary pollster assessment", "detail": "Gallup attributed the miss to timing. This is his interpretation." }, { "label": "Fenwick, Wiseman, Becker and Heiman 1982, Classifying Undecided Voters in Pre-Election Polls", "url": "https://academic.oup.com/poq/article-abstract/46/3/383/1811208", "kind": "peer-reviewed original methodological research", "detail": "The publisher abstract describes allocation of undecided respondents in 1980 presidential polls and validation using a post-election follow-up. Only the abstract was verified; automated publisher access may return 403 and the full article may require access. This is methodological context, not an audit establishing this Gallup estimate’s cause." }, { "label": "Arkansas Secretary of State Historical Report 2010, printed p. 586", "url": "https://www.sos.arkansas.gov/uploads/elections/historical_report1210.pdf", "kind": "official historical compilation", "detail": "" } ] }, { "slug": "1982-california-governor", "year": 1982, "title": "1982 California: a famous label, several explanations", "shortLabel": "California", "office": "Governor", "description": "Field’s final poll favored Tom Bradley, but George Deukmejian won. Examine absentee voting, turnout and the contested explanations behind the Bradley effect.", "takeaway": "The Bradley effect is a hypothesis about mechanisms, not a complete explanation established by the miss itself.", "classification": "Wrong leader in final statewide poll", "basis": "raw", "chartRows": [ { "label": "Field Institute / California Poll", "pollMargin": 7, "actualMargin": -1.1850778079850173 } ], "chartNote": "Field’s reported Bradley 49 / Deukmejian 42 estimate is compared with all counted gubernatorial votes from the official count reproduced in Field’s audit. The raw D−R margin is used. The post-election explanation is the pollster’s assessment, not a proven causal decomposition.", "facts": [ [ "Compared estimate", "Field Institute / California Poll" ], [ "Election", "1982 · Governor" ], [ "Published / fieldwork context", "October 29–31, 1982 final Field survey, reproduced in Nieman Reports Spring 1983." ], [ "Error in the selected margin", "8.2 percentage points" ], [ "Comparison basis", "Raw D−R margin, with source qualifications" ] ], "dialogue": [ { "question": "What did the selected poll say?", "paragraphs": [ "Field Institute / California Poll’s selected estimate put Tom Bradley at 49.0% and George Deukmejian at 42.0%. On the chart’s stated basis, that was D +7.0. The corresponding recorded outcome margin was R +1.2, for a 8.2-point difference in the margin." ] }, { "question": "What actually happened?", "paragraphs": [ "Field's final survey favored Bradley 49–42; the official count reproduced in Field's audit gave Deukmejian 3,881,014 votes to Bradley 3,787,669.", "Field's audit reports Bradley ahead by 19,429 among precinct voters and Deukmejian ahead by 112,774 among absentee voters, producing the 93,345-vote statewide margin. Two county absentee splits were estimated." ] }, { "question": "Why might the poll have missed?", "paragraphs": [ "Field's February 1983 assessment discusses Republican absentee mobilization, the handgun initiative, minority turnout and racial bias. These are its post-election interpretations rather than a randomized decomposition of polling error. Evidence status: pollster interpretation; multiple causes." ] }, { "question": "What has not been established?", "paragraphs": [ "A wrong poll cannot by itself establish racial concealment. Turnout, absentee mobilization, undecided voters, response patterns and timing can interact. Calling the gap the Bradley effect does not resolve which mechanism caused it." ] }, { "question": "How should we use this example?", "paragraphs": [ "Separate the documented vote arithmetic from the explanations offered afterward. Then compare turnout assumptions, absentee participation and respondent composition using evidence collected before and after the election.", "Field’s reported Bradley 49 / Deukmejian 42 estimate is compared with all counted gubernatorial votes from the official count reproduced in Field’s audit. The raw D−R margin is used. The post-election explanation is the pollster’s assessment, not a proven causal decomposition." ] } ], "sources": [ { "label": "Field Institute original release 1201, February 1, 1983, pp 1–2", "url": "https://abcnews.go.com/images/PollingUnit/FieldPoll1982analysis.pdf", "kind": "primary pollster audit", "detail": "" }, { "label": "Nieman Reports Spring 1983, Bradley vs. Deukmejian table, source: Field Institute", "url": "https://niemanreports.org/app/uploads/2025/09/Spring-1983_150.pdf", "kind": "contemporary research using original polling data", "detail": "Poll 49/42 and October 29–31 dates verified from indexed table. Table's preliminary outcome percentages are not used." } ] }, { "slug": "2014-virginia-senate", "year": 2014, "title": "2014 Virginia: the winner was right, the margin was not", "shortLabel": "Virginia", "office": "U.S. Senate", "description": "A final PPP poll showed Mark Warner ahead by nine points; he won Virginia by less than one. See the margin error and why its cause remains unresolved.", "takeaway": "A nearly correct estimate of the winner's share can still produce a badly wrong margin.", "classification": "Large Democratic-margin overestimate; same winner", "basis": "raw", "chartRows": [ { "label": "Public Policy Polling", "pollMargin": 9, "actualMargin": 0.8114826140296287 } ], "chartNote": "One PPP likely-voter poll is compared with the current Virginia Department of Elections total, including other candidates and write-ins. Its reported margin miss is not an industry-wide polling error.", "facts": [ [ "Compared estimate", "Public Policy Polling" ], [ "Election", "2014 · U.S. Senate" ], [ "Published / fieldwork context", "October 29–30, 2014; the final PPP release located for this race. Sponsor is an advocacy organization." ], [ "Error in the selected margin", "8.2 percentage points" ], [ "Comparison basis", "Raw D−R margin, with source qualifications" ] ], "dialogue": [ { "question": "What did the selected poll say?", "paragraphs": [ "Public Policy Polling’s selected estimate put Mark Warner at 49.0% and Ed Gillespie at 40.0%. On the chart’s stated basis, that was D +9.0. The corresponding recorded outcome margin was D +0.8, for a 8.2-point difference in the margin." ] }, { "question": "What actually happened?", "paragraphs": [ "The selected PPP poll put Warner at 49 and Gillespie at 40; Warner finished near 49.15 while Gillespie finished near 48.34. The published margin missed by about 8.2 points even though the poll's leader won.", "Warner's estimated share changed little in the comparison; the numerical miss is concentrated in Gillespie's share. That does not prove that particular undecided or Sarvis respondents switched to Gillespie." ] }, { "question": "Why might the poll have missed?", "paragraphs": [ "No race-specific methodological audit establishing a single cause was verified for this case. Possible turnout, nonresponse and late-choice mechanisms remain possibilities. A numerical difference between candidate shares does not track individual voter movement." ] }, { "question": "What has not been established?", "paragraphs": [ "No race-specific methodological audit establishing a single cause was verified for this case. Possible turnout, nonresponse and late-choice mechanisms remain possibilities. A numerical difference between candidate shares does not track individual voter movement." ] }, { "question": "How should we use this example?", "paragraphs": [ "Look at multiple independently documented polls on the same dates, while retaining their uncertainty and sample definitions. This page uses one transparent example; it does not infer that every Virginia poll made the same mistake.", "One PPP likely-voter poll is compared with the current Virginia Department of Elections total, including other candidates and write-ins. Its reported margin miss is not an industry-wide polling error." ] } ], "sources": [ { "label": "PPP original questionnaire/crosstabs, October 29–30, 2014", "url": "https://www.scribd.com/document/245109946/VA-Sen-PPP-for-LCV-Oct-2014", "kind": "primary pollster release", "detail": "937 likely voters, Warner 49, Gillespie 40, Sarvis 5, undecided 6; sponsored by LCV Action Fund." }, { "label": "Virginia Department of Elections, current 2014 Senate results", "url": "https://historical.elections.virginia.gov/contest/44423", "kind": "official state results", "detail": "" }, { "label": "FEC 2014 Senate results, earlier edition", "url": "https://www.fec.gov/documents/1698/2014senate.pdf", "kind": "official federal compilation", "detail": "" } ] }, { "slug": "2016-midwest", "year": 2016, "title": "2016: small state margins, consequential polling misses", "shortLabel": "Wisconsin", "office": "President", "description": "Compare final archived polling estimates with FEC results in Wisconsin, Michigan and Pennsylvania. Examine education weighting, late deciders and the national-versus-state distinction.", "takeaway": "A reasonably close national popular-vote estimate can coexist with consequential errors in the states that decide the presidency.", "classification": "Wrong leaders in selected state aggregates", "basis": "raw", "chartRows": [ { "label": "Wisconsin", "pollMargin": 5.325556, "actualMargin": -0.764343195067453 }, { "label": "Michigan", "pollMargin": 3.191111, "actualMargin": -0.2230332691293118 }, { "label": "Pennsylvania", "pollMargin": 3.181765, "actualMargin": -0.7183871226205008 } ], "chartNote": "Original retrospective calculation: retain the latest median-fieldwork-date eligible poll per firm and race in the archive's final-14-day window, average same-date ties, and give each firm equal weight. Partisan polls retained. Dates below are selected archive fieldwork midpoints, not publication dates or final interviewing dates. These are our retrospective archive means, not contemporaneously published RCP or FiveThirtyEight polling averages. The signed errors below use independently checked FEC valid-vote totals. They are errors of a mean, not mean absolute errors of individual polls.", "facts": [ [ "Compared estimate", "Our retrospective mean, not an election-eve branded average" ], [ "States", "Wisconsin, Michigan, Pennsylvania" ], [ "Selected firms", "9 in WI · 9 in MI · 17 in PA" ], [ "Outcome source", "FEC 2016 official count compilation" ], [ "Comparison basis", "Raw state D−R margin; all valid votes in denominator" ] ], "dialogue": [ { "question": "What did the state polls suggest?", "paragraphs": [ "The selected archive estimates averaged Clinton leads of 5.3 points in Wisconsin, 3.2 in Michigan and 3.2 in Pennsylvania. Each average gives one equal-weight latest eligible estimate to a firm/race, with same-date ties averaged.", "These are retrospective calculations from a stated archive window, not a claim about any particular website’s contemporaneous polling average. The archive’s interview midpoint dates are not publication dates or final fieldwork dates." ] }, { "question": "What did the votes show?", "paragraphs": [ "The FEC totals put Trump ahead in all three states: about 0.76 point in Wisconsin, 0.22 in Michigan and 0.72 in Pennsylvania. The selected mean margins therefore overstated Democrats by roughly 6.1, 3.4 and 3.9 points, respectively.", "Clinton won the national popular vote while Trump won the presidency. This is why national popular-vote accuracy and Electoral College expectations must be evaluated separately." ] }, { "question": "Why did the state polls miss?", "paragraphs": [ "AAPOR's 2016 investigation found evidence that late-deciding voters and failure to adjust for educational differences contributed to state misses. Those general findings do not attribute a measured percentage of any specific retrospective aggregate's error to a single cause.", "Education mattered because college graduates were more likely to participate in surveys and supported Clinton more strongly. Failing to correct the imbalance could make a survey look more Democratic than the electorate. Late-deciding voters also broke toward Trump in important states. These findings concern the evidence evaluated by AAPOR, not a precise decomposition of every poll’s miss." ] }, { "question": "Does this prove voters hid support for Trump?", "paragraphs": [ "No universal concealment explanation was established by the AAPOR 2016 evaluation. The documented evidence about education adjustment and late choice is stronger than an assumption that every respondent concealed a preference.", "Errors in related states can move together. Averaging polls reduces some random noise, but a shared selection or turnout-model problem can survive the average. The three state outcomes are not independent demonstrations of three unrelated polling failures." ] }, { "question": "What should readers change?", "paragraphs": [ "Treat a small state lead as uncertain. Read the interviewing period, population, weighting disclosures and likely-voter assumptions. Distinguish a measurement of current preferences from a model assigning a probability of winning the Electoral College.", "Do not carry the same correction into the next election automatically. A later cycle can have a different response pattern, candidate coalition, turnout environment or direction of error." ] } ], "sources": [ { "label": "AAPOR evaluation of 2016 election polls", "url": "https://aapor.org/wp-content/uploads/2023/01/AAPOR-2016-Election-Polling-Report.pdf", "detail": "Evidence on late-deciding voters and education adjustment; national versus state performance." }, { "label": "FEC Federal Elections 2016", "url": "https://www.fec.gov/resources/cms-content/documents/federalelections2016.pdf", "detail": "2016 presidential electoral and popular vote state summary, printed page 6 (PDF page 11)" }, { "label": "Pinned FiveThirtyEight poll archive", "url": "https://raw.githubusercontent.com/fivethirtyeight/data/be387bc1649413be1c85f2181b2c32aacee24315/pollster-ratings/raw_polls.csv", "detail": "Original retrospective calculation: retain the latest median-fieldwork-date eligible poll per firm and race in the archive's final-14-day window, average same-date ties, and give each firm equal weight. Partisan polls retained. Dates below are selected archive fieldwork midpoints, not publication dates or final interviewing dates." } ] }, { "slug": "2024-iowa-selzer", "year": 2024, "title": "Iowa 2024: a striking miss, not a settled explanation", "shortLabel": "Iowa", "office": "President", "description": "Selzer’s final Iowa poll showed Harris ahead; Trump won by 13.2 points. Compare a 16.2-point margin miss with another Iowa poll and examine what remains unexplained.", "takeaway": "One prominent outlier can dominate the election story. Evaluate it against the result and other polls without pretending that its cause has been established.", "classification": "Wrong leader in one prominent statewide poll", "basis": "raw", "chartRows": [ { "label": "Selzer / Iowa Poll", "pollMargin": 3, "actualMargin": -13.209510515742053 }, { "label": "Emerson", "pollMargin": -10, "actualMargin": -13.209510515742053 } ], "chartNote": "Raw D−R margins. Selzer fieldwork: October 28–31; released November 2. Emerson fieldwork: November 1–2. Both are compared with the same valid presidential-vote denominator; overvotes and undervotes are excluded. These are two individual polls, not an industry aggregate.", "facts": [ [ "Selzer estimate", "Harris 47% · Trump 44%" ], [ "Sample", "808 Iowa likely voters" ], [ "Fieldwork", "October 28–31, 2024" ], [ "Release", "November 2, 2024" ], [ "Recorded result", "Harris 42.52% · Trump 55.73%" ], [ "Selected margin miss", "16.21 percentage points" ] ], "dialogue": [ { "question": "Why did this poll get so much attention?", "paragraphs": [ "The final Des Moines Register/Mediacom Iowa Poll from Selzer & Company put Harris ahead 47–44. In a closely watched presidential campaign, an unexpected Democratic lead in Iowa was a conspicuous point estimate.", "It was still one poll, with its own sample and likely-voter screen. A three-point lead was not a high-confidence claim that Harris would win the state; the reported candidate-share sampling margin was 3.4 points, and uncertainty in a candidate lead is larger than uncertainty in either share alone." ] }, { "question": "What did the final vote show?", "paragraphs": [ "Iowa’s certified canvass recorded 927,019 Trump votes and 707,278 Harris votes. Using 1,663,506 valid presidential votes gives Trump a 13.2-point margin. Selzer’s D+3 margin missed that result by 16.2 points.", "That is an error in the difference between two candidate shares. It does not mean both candidates’ vote shares were each wrong by 16.2 points. Undervotes and overvotes are not candidate votes and are excluded from this denominator." ] }, { "question": "Did every Iowa poll show the same thing?", "paragraphs": [ "No. Emerson’s November 1–2 poll put Trump at 53 and Harris at 43, a ten-point Republican lead. Against the same outcome, its margin miss was about 3.2 points.", "The comparison illustrates substantial disagreement between polls. Their interviewing periods, samples and methods differ, so it is not a controlled test establishing why one method worked better. The original unallocated Emerson 53–43 result is used rather than its separate estimate reallocating undecided voters." ] }, { "question": "Why did Selzer miss by so much?", "paragraphs": [ "Selzer said in a December 13, 2024 Iowa PBS interview that she had not established why the poll missed. She discussed the likely-voter screen, response patterns and possible late turnout changes as possibilities; none was demonstrated as the explanation in that interview.", "A large residual tells us that the estimate was far from the outcome. It does not, on its own, tell us which people failed to respond, which voters were screened incorrectly, how preferences changed afterward, or how much each mechanism contributed." ] }, { "question": "What can we conclude—and what remains open?", "paragraphs": [ "The selected poll’s margin was a major miss. The existence of another poll much closer to the result prevents treating the Selzer result as the error of every Iowa poll.", "The explanation remains an empirical question. Read the firsthand post-election interview and the original methods before adopting a single-cause theory. Neither the size of the error nor the pollster’s reputation substitutes for that evidence." ] } ], "sources": [ { "role": "original poll release", "title": "Des Moines Register: Iowa Poll, November 2, 2024", "url": "https://www.desmoinesregister.com/story/news/politics/iowa-poll/2024/11/02/iowa-poll-kamala-harris-leads-donald-trump-2024-presidential-race/75354033007/", "retrieval_note": "Original publisher URL could not be opened by the web reader; the publisher's syndicated article was indexed and read at Yahoo. No exact release clock time is asserted.", "detail": "Original publisher URL could not be opened by the web reader; the publisher's syndicated article was indexed and read at Yahoo. No exact release clock time is asserted." }, { "role": "syndicated original publisher article", "title": "Brianne Pfannenstiel, Des Moines Register, syndicated by Yahoo", "url": "https://www.yahoo.com/news/iowa-poll-kamala-harris-leapfrogs-230122850.html", "retrieval_note": "Search reader supplied the original article text and methodological details; a subsequent direct open returned HTTP 429.", "detail": "Search reader supplied the original article text and methodological details; a subsequent direct open returned HTTP 429." }, { "role": "certified outcome", "title": "Iowa Secretary of State: 2024 General Election Canvass Summary", "url": "https://sos.iowa.gov/elections/pdf/2024/general/canvsummary.pdf", "page": 15, "note": "The printed Total column includes 8,586 undervotes and 1,919 overvotes. Valid presidential votes equal 1,674,011 minus those 10,505 nonvotes: 1,663,506. Candidate and write-in totals are preserved.", "detail": "The printed Total column includes 8,586 undervotes and 1,919 overvotes. Valid presidential votes equal 1,674,011 minus those 10,505 nonvotes: 1,663,506. Candidate and write-in totals are preserved." }, { "role": "certified valid-vote total cross-check", "title": "Iowa State Board of Canvassers: Abstract of Votes", "url": "https://mccmeetingspublic.blob.core.usgovcloudapi.net/iowasttres-meet-970af300230e49f4a38af865c7c6f3e2/ITEM-Attachment-001-c3569f875a514294b58f293705523ed2.pdf", "certification_date": "2024-12-02", "detail": "certified valid-vote total cross-check" }, { "role": "first-person methodological review", "title": "Iowa PBS: J. Ann Selzer", "url": "https://www.iowapbs.org/shows/iowapress/episode/11885/j-ann-selzer", "date": "2024-12-13", "detail": "first-person methodological review" }, { "label": "Emerson College Polling: November 2024 Iowa poll", "url": "https://emersoncollegepolling.com/november-2024-iowa-poll-trump-53-harris-43/", "detail": "Use the original unallocated 53-43 result, not the alternative undecided-allocated 54-45 estimate, for this comparison. The comparison demonstrates disagreement between polls; it does not estimate the Iowa polling industry's overall error." } ] } ]