Almost everything sold as election "persuasion" rests on a hopeful picture of the voter: a reasoning individual weighing issues, waiting to be convinced. The evidence says that picture is mostly wrong, that persuasion barely moves who people vote for, and that the votes campaigns can actually change are somewhere else entirely.
The first move, and the one most political talk skips, is to separate two questions that are not the same. Turnout is whether a person votes at all. Choice is who they vote for. They have different causes, respond to different tactics, and conflating them, treating a campaign as one big "win votes" problem, is the root of a lot of wasted money.
On choice, the founding finding still holds up. The American Voter argued that party identification, acquired early in life and largely stable thereafter, is the single strongest predictor of how someone votes, working like a standing lens through which candidates and issues are perceived rather than a conclusion drawn from them (Campbell et al., 1960). The rival tradition is Downs's rational, spatial voter, who picks the candidate closest to their own position in policy space (Downs, 1957), and Fiorina's retrospective voting, where voters simply reward or punish the incumbent for how things have gone (Fiorina, 1981). Almost every modern voter segmentation, base, persuadable, opposition, is some blend of these traditions.
The contemporary correction, and the heart of this article, is bracing. Achen and Bartels assembled the evidence against the comforting "folk theory" of the rational, issue-weighing voter and found it wanting: voters punish incumbents for things plainly outside anyone's control, droughts, even a spate of shark attacks, which is not reasoned retrospection but blind, mood-driven blame. What actually drives choice, they argue, is identity and group attachment far more than policy positions; people mostly arrive at issue opinions to fit the party and groups they already belong to, not the other way round (Achen and Bartels, 2016). This connects straight to the low ideological constraint covered in L3-12: the reasoning voter is more the exception than the rule.
Which sets up the single most useful empirical result for anyone who campaigns. Kalla and Broockman pooled 49 field experiments and found that the average effect of campaign contact, ads, mail, calls, canvassing, on vote choice in general elections is approximately zero. Persuasion is not entirely dead, it shows up in primaries, on specific ballot measures, and well before election day, but the image of a persuadable general-election voter waiting to be talked round is largely a myth (Kalla and Broockman, 2018). There is one striking exception on the attitude side: deep canvassing, long, unhurried, non-judgemental conversations in which the canvasser listens more than argues, can produce durable change, shown in a careful field experiment on attitudes toward transgender people (Broockman and Kalla, 2016). Notably, that finding is also a scientific-integrity landmark, it was the honest replication that stood after an earlier, celebrated result on the same question turned out to be fraudulent.
Turnout, by contrast, genuinely moves. The largest and most replicated get-out-the-vote effect comes from social pressure: mailings that tell people their voting record is public and that their neighbours will be told whether they voted lift turnout measurably, because voting is a social norm people want to be seen upholding (Gerber, Green and Larimer, 2008). You can change whether people vote far more easily than for whom.
The honest, uncomfortable synthesis is that most of what political consultants sell as "persuasion" is really base mobilisation wearing a persuasion costume. When a campaign "moves the numbers", it has usually turned out people already inclined to support it, not converted opponents or the undecided. The "large persuadable middle" that segmentation decks love is routinely inflated; the genuinely persuadable are a thin slice, and, awkwardly, they tend to be the lowest-information, lowest-interest voters, the hardest people of all to reach and hold.
None of the underlying findings are partisan, which is worth stating plainly: weak persuasion, identity-driven choice, and blind retrospection describe voters across the spectrum equally. And there are real limits. Most of this evidence is American, and turnout and party systems differ enormously across countries. Field experiments measure short-to-medium-term effects and can miss slow, cumulative persuasion over years. And "approximately zero on average" hides real variation, some messages, moments, and contests do move choice, which is precisely why honest measurement, rather than confident assertion, matters.
The single biggest caveat is that almost all of this evidence comes from one setting: the United States, with its two dominant parties, voluntary voting, and candidate-centred campaigns. Treat every finding here as a prior to calibrate against local data, not a universal law, because the levers genuinely invert across systems. In multi-party and proportional systems, where several adjacent parties compete for the same voters and coalitions are in play, real persuadable switching is likely larger than the near-zero US figure implies. Where turnout is already near-universal, or compulsory, turnout stops being the movable lever and the contest becomes almost entirely about choice, the reverse of the US pattern. The identity that anchors choice may be ethnic, regional, religious, or tied to patronage rather than a stable party attachment, which makes the identity finding if anything stronger but relocates it. And whole mechanisms that dominate elsewhere, mobilisation through community and religious authorities, local brokers, and clientelist networks, barely appear in the US field-experiment record at all. Beyond geography, field experiments isolate a tactic's short-run effect well but capture slow, diffuse cultural shifts poorly, and self-reported vote intention is a weak guide to actual behaviour.
How much genuine persuasion happens slowly and cumulatively, below the resolution of a field experiment? Do these near-zero persuasion findings hold outside the US, in multi-party and compulsory-voting systems? And can deep canvassing, which clearly works, ever scale, given how slow and labour-intensive it is?
The usable core: separate turnout from choice, treat the persuadable middle as small until proven otherwise, and remember that the same weakness of persuasion that governs votes governs most attempts to change minds, including advertising.
One caveat first, and it matters more the further you are from a US-style two-party race: everything below is a prior to calibrate against local data, not a universal truth. In multi-party, high-turnout, or compulsory-voting systems the turnout-versus-choice balance can flip entirely, and the identity that decides a vote may be ethnic, regional, or religious rather than partisan, so read these as starting hypotheses your own measurement should confirm or overturn. With that held firmly, the discipline here saves money and self-respect.
Build separate models for turnout and for choice. They are different problems with different levers, and the commonest budgeting error is to fund one while believing you are funding the other. Stop pouring persuasion money into a general-election "middle" that the field-experiment evidence says barely moves (Kalla and Broockman, 2018), and be ruthless about how small that persuadable slice really is: the honest prior sits far below any consultant's deck, and much apparent persuasion is base turnout misattributed. In practice that usually means reallocating spend away from converting a thin, hard-to-reach middle and toward getting the people already inclined to support you to actually vote.
This is not guesswork any more. The field-experiment tradition has a practitioner home in Green and Gerber's Get Out the Vote (Green and Gerber, 2019), the handbook that turned decades of randomised turnout studies into costed tactics with a measured return per contact: door-to-door canvassing tends to give the biggest reliable lift per voter, live phone calls less, and untargeted mail and robocalls least. The applied habit it teaches is to treat every tactic as testable and to spend on what the evidence, not the pitch, says works.
Where you do want to move choice, the evidence points to slow, respectful methods over blast media, and to primaries and specific policy fights over general-election conversion. Deep canvassing, the long, unhurried, non-judgemental doorstep conversation in which the canvasser listens far more than argues, is the rare method shown to shift attitudes durably (Broockman and Kalla, 2016). Its honest limitation is operational: it is slow, labour-intensive, and expensive per voter, so it scales poorly and suits high-stakes attitude change rather than mass vote-chasing.
The dual-use line is sharp, and holding to the responsible-communicator standard is what keeps a campaign honest. Social-pressure turnout tactics genuinely work, but "we will tell your neighbours whether you voted" shades from a civic nudge into coercive shaming, and the test is whether the message serves the voter's own stake in the election or merely the sender's numbers. Lighter, non-punitive versions, reminding people that voting is common and that their community turns out, capture much of the lift without the intimidation. And selling a client a large persuadable middle you cannot deliver is not strategy, it is taking their money, and it is the fastest way to burn the trust the whole practice runs on.
The transfer is direct and humbling. If campaign persuasion barely shifts something as engaged and identity-laden as a vote, be sceptical of advertising's power to convert rather than reach and reinforce. This is exactly the Ehrenberg-Bass lesson coming in L5-02: most advertising works by keeping a brand mentally available to people already open to it, not by arguing loyalists of a rival into switching. Model your "persuadable" customers with the same ruthlessness a good campaign models persuadable voters, and expect the truly convertible slice to be smaller than the plan assumes. Borrow the campaigners' other discipline too. The reason the turnout field knows what works is that it ran randomised experiments with holdout groups instead of trusting the story, and a brand that measures incremental effect the same way, testing against a genuine control, will usually find its "conversion" number is mostly people who would have bought anyway. The honest version of a marketing plan, like the honest version of a campaign plan, budgets for the reach it can prove and treats claimed conversion as a hypothesis to test, not a number to bank.
Turnout is movable by design, and the levers are documented rather than mysterious. Making voting easier and more visible, through simpler registration, convenient voting windows, and clear reminders, raises it, and nonpartisan get-out-the-vote work of this kind is a legitimate democratic goal when pursued even-handedly across all groups. The same Get Out the Vote evidence base that helps campaigns applies here, minus the partisanship: electoral commissions and neutral civic bodies can use the tested tactics to lift participation without favouring a side. The evidence also counsels humility about "information" campaigns. Telling citizens the facts rarely changes identity-anchored views, so public communication aimed at genuine attitude change, on a vaccine, say, or a contested reform, should look more like deep canvassing, with listening, time, and respect, than like broadcasting the correct answer louder. The ethical edge cuts here too: any turnout effort that leans on shaming, or that targets only favourable groups, stops being civic and becomes manipulation.
Three habits. Always ask whether a tactic is really about turnout or about choice, because campaigns and pundits blur them constantly and the honest levers differ. Treat any "large persuadable middle", of voters or customers, as a claim to be proven, not assumed, because the field evidence says it is usually thin. And carry the deeper lesson into every attempt to change a mind: direct persuasion is weak against identity, so respectful, listening-led conversation beats louder broadcasting, and testing beats asserting.
.png)