Beyond the Horse Race: How Consensus Choice Polling Can Transform Elections and Journalism
by Carah Ong Whaley
Elections aren’t the only institutions overdue for reform—public opinion polling and journalism must also embrace better methods to understand and report on what voters actually prefer.
In addition to use in actual elections, Consensus Choice can also be employed as a polling method to better understand the preferences of voters before ballots are cast on election day. Most pre-election polling focuses on the “horse race," with simple survey questions designed to gauge who is “winning” or “losing” the election at a particular point in time. Typically respondents, whom pollsters filter for “likely voters,” are asked who they would vote for if the election were held today. A growing body of research has shown that horse race coverage can be harmful to both voters and candidates by focusing on who is in the lead and who is behind because it detracts from coverage of the real policy issues candidates are being elected to address and can advantage or disadvantage certain candidates. Such coverage can have deleterious consequences for democracy by increasing political cynicism and decreasing trust in politicians and news media, as well as by contributing to a confused and demobilized electorate.
By focusing on who voters prefer with either ranked or pairwise comparisons instead of only whom they plan to vote for, Consensus Choice polling can encourage respondents to express their true likes and dislikes, minimizing strategic answers and revealing a deeper understanding of voter preferences. It also can highlight candidates who may not receive attention in traditional election polls and increase coverage of them, rather than just a couple of dominant candidates. Reporters can use data from Consensus Choice polls to focus on which candidate is more broadly preferred by the electorate and why, rather than just who is winning or losing, thereby giving more attention to the real preferences of voters and which candidates are better suited to serve the common good.
More Data = More Insight
Instead of a standard horse race poll (which simply asks who is leading), the two Consensus Choice poll designs illustrated below ask voters to rank or compare candidates, revealing preferences that typical election polls often miss. It allows the media to open up discussion about more than just the dominant candidates and introduce voters to the full range of options in a contest. Using a Consensus Choice poll would be particularly advantageous for pollsters, media and voters in fluid and dynamic multicandidate primary elections. Media outlets can use data from a Consensus Choice poll to report more than just “Candidate X leads”; they can report how each candidate compares to all of the others and say who is the most broadly acceptable candidate in an electorate focusing on compromise and problem-solving, rather than division.
In addition to better understanding the preferences of the broader electorate, Consensus Choice polling would also reveal deeper preferences of different demographic groups and Independents who are crucial in close elections, but often aren’t well-represented in top-choice-only polls. Instead of forcing respondents to pick just one, Consensus Choice polling reveals their full order of preferences—so journalists, candidates and campaigns can analyze whether different types of voters consistently prefer moderates, mavericks, or anyone except the major party nominees.
Consensus Choice polling offers deeper insights into the preferences of different demographic groups compared to the overall electorate. For instance, young voters might rank progressive candidates more highly, making their Consensus Choice very different from that of older voters. This approach can reveal which candidates are broadly acceptable to all, and which appeal mainly to specific groups. By examining pairwise strengths, Consensus Choice polling uncovers “bridge” candidates—those who may not be anyone’s first pick, but are a popular second choice across groups. Moreover, candidates often overlooked in traditional polls can emerge with significant cross-group support through this richer, more nuanced method.
How Consensus Choice Polling Works: Two Designs
Graphic of ranking preferences verses paired candidate comparisons.
Design 1: Paired-Comparison Poll
Pollsters are already likely familiar with polls that ask primary electorates to express preferences between candidates before a field has winnowed. A paired comparisons poll expands on this kind of polling.
Format: This design presents respondents with a series of specific one-on-one candidate matchups and asks them to choose whom they prefer in each pair. Instead of ranking all four at once, voters consider candidates two at a time. Each pair is a separate question, typically phrased as “If the choice came down to Candidate X versus Candidate Y, whom would you prefer?”
With four candidates, there are 6 unique pairings (since each pair of names can be matched up once: A vs B, A vs C, A vs D, B vs C, B vs D, C vs D). In this design, each respondent will expresses their preference in all six head-to-head contests. This yields a complete pairwise preference matrix for each respondent (and for the electorate as a whole), which is exactly what’s needed to determine the Consensus Choice. In essence, it’s like doing a round-robin tournament where every candidate faces every other candidate, and we see who wins the most matchups overall.
Hypothetical Example with Four Candidates:
Instructions for Respondents: “You will be asked about all possible pairs of candidates. There are six matchups in total. For each pair, choose the candidate you personally prefer. If you truly have no preference between the two, you may say so, but if you do have even a slight preference, please indicate it. By answering all pairwise questions, you help us understand your complete ordering of the candidates.”
Question Format: The survey lists each pair of candidates one by one (typically in random order to avoid any bias from a fixed sequence). Each question is a binary choice like “Candidate X vs Candidate Y: Who do you prefer?” with options [X / Y] (and optionally [No preference]). For example, the six questions (in no particular order) could be:
Ann vs Bob? – [Ann / Bob / No preference]
Ann vs Carol? – [Ann / Carol / No preference]
Ann vs David? – [Ann / David / No preference]
Bob vs Carol? – [Bob / Carol / No preference]
Bob vs David? – [Bob / David / No preference]
Carol vs David? – [Carol / David / No preference]
Each of these explicitly asks the respondent to pick who they like more out of that pair.
Example Response: Let’s say one respondent answers as follows:
Ann vs Bob → Ann (respondent prefers Ann over Bob)
Ann vs Carol → Carol (respondent prefers Carol over Ann)
Ann vs David → Ann (respondent prefers Ann over David)
Bob vs Carol → Carol (respondent prefers Carol over Bob)
Bob vs David → Bob (respondent prefers Bob over David)
Carol vs David → Carol (respondent prefers Carol over David)
From this we can construct their personal ranking: Carol is this respondent’s most preferred candidate (Carol was preferred over Ann, Bob, and David in the above responses), so Carol is this respondent’s top choice. Between Ann and Bob, they chose Ann, and Ann over David, so Ann is ahead of both Bob and David for them, making Anna their second choice. Bob beats David in their response, so Bob is third, David fourth. Indeed, the pairwise responses contain enough information to derive that ordering: Carol > Ann > Bob > David. If any inconsistencies were present (like if their choices implied a cycle, e.g. A > B, B > C, C > A), that means they have indifference or a very close call – but allowing a “No preference” option allows the respondent to express ties instead and can prevent a cycle. The key takeaway is that the respondent has made an explicit comparison in every possible matchup.
Interpreting the Results: For the individual, it’s straightforward: their six answers can be tallied to confirm a consistent rank order (or identify ties). For the overall poll results, each pair question is aggregated to provide percentage results for each of the six pairings among all respondents, e.g., “Ann vs Bob: Ann preferred by 58% of respondents, Bob by 42%.” We can compile these into a table or matrix. Here’s a hypothetical example:
Table of hypothetical matchups with percentages by candidate.
From such a table, pollsters and analysts can see who wins each one-on-one comparison. In this example, Carol wins against everyone (Carol is preferred to Ann 55-45, Bob 60-40, David 70-30), so Carol is the Consensus Choice. Ann is preferred over Bob and David but not over Carol, placing her likely as overall second choice. Bob is preferred over David but over Ann and Carol, so Bob is third. David is not preferred over any other candidate so is the fourth choice.
The advantage of an all-pairs poll is that the data to identify this ranking and the Consensus Choice is open and transparent, and can be more easily understood by journalists and the public. Journalists can even present results using analogies of round robin sports tournaments, with which many in the public are already familiar. However, journalists will need to be careful to not just cherry pick findings and be transparent about the results of every matchup. While weighted results should always be reported, journalists should also be transparent about whether results change after weighting for partisanship and other demographics.
In the unlikely event that the pairwise results produced a cycle (A beats B, B beats C, C beats A), journalists simplify by saying “voters are split and no candidate is the Consensus Choice: different candidates win against different opponents, showing how divided preferences are, and no one candidate has majority backing against all others.”
Design 2: Ranked Poll
Format: Each respondent ranks all candidates in order of preference, with ties allowed if the person has equal feelings about two or more candidates. This is similar to a ranked voting ballot. Respondents assign a “1” to their top choice, “2” to their next choice, and so on until all candidates are ranked (if they truly have no preference between certain candidates, they can give them the same rank or not rank a candidate). A ranked poll can be used with three or more candidates and will perform best with 3-5 of the strongest candidates and avoid overwhelming respondents.
Hypothetical Example with Four Candidates:
Instructions for Respondents: “Rank the following four candidates in order of preference, with “1” for your favorite. You may assign the same rank to candidates if you feel equally about them. Rank as many as you would like. You do not have to rank all of the candidates, but the more candidates you rank, the more say you have in comparisons of candidates.”
Question Format: The poll presents a list of all four names – for example, Ann, Bob, Carol, and David – each with a space or dropdown to input the rank. The question might read: “Please rank the candidates below from 1 (best) to 4 (worst) according to your personal preference.”
Example Response: One respondent might answer: 1 – Ann, 2 – Bob, 3 – Carol, 4 – David. Another respondent who is indifferent between two candidates could respond: 1 – Ann, 2 – Bob and Carol (tied), 4 – David. This indicates Ann is the favorite; Bob and Charlie are equally liked next; David is the least preferred. Another respondent may choose to only rank one candidate: 1 – Bob. This indicates that Bob is the favorite and the respondent has chosen not to express a preference for the other candidates.
Interpreting the Results: Pollsters can derive pairwise outcomes from the overall respondent rankings – e.g. how often each candidate was ranked above each other candidate – to identify the Consensus Choice. Ties are accounted for by treating tied candidates as having the same rank. The Consensus Choice is the candidate who was ranked higher than each of the other candidates on a majority of responses.
This format collects a wealth of information in one question. It lets people express nuanced opinions – for example, someone can show that a candidate is their second choice even if not their first. It also encourages honesty: a voter can safely rank a long-shot candidate first and a more likely winner second, knowing their full feelings are recorded (in an actual election they might hesitate to “waste” a vote, but in a poll they can be candid). Because all candidates are ranked, the poll captures broad support and not just plurality support. For example, a news report might highlight that “while Ann isn’t the top choice of most voters, she is the best compromise candidate – the rankings show she beats every other candidate head-to-head, indicating broad appeal.”
Conclusion
Consensus Choice polling is an innovative opportunity to improve both election polling and journalism. By providing a more honest and nuanced picture of what voters want, it can shift media coverage from focusing on who’s ahead or behind to highlighting which candidates are best equipped to serve the common good. Moving beyond the limitations of traditional horse race polling, Consensus Choice methods capture a fuller range of voter preferences, enabling the public and decision-makers to better understand which candidates unite, divide, or quietly appeal across partisan and demographic lines. As our democracy navigates complex challenges and evolving coalitions, more transparent and insightful polling can help ensure every voter matters to every candidate and contribute to more representative outcomes in elections and governance. Consensus Choice polling is a vital tool in making this future possible.
Try it out and create a Consensus Choice poll of your own here and let us know what you find by emailing info@betterchoices.vote.
