American voters went to the polls on Tuesday in an election that was widely expected to punish President Joe Biden and the Democratic Party. Inflation remains unusually high, petrol and food prices have placed substantial pressure on household budgets, Biden’s approval ratings are weak, and the president’s party almost always loses congressional seats during a midterm election. Republicans therefore entered the final months of the campaign expecting that dissatisfaction with the economy might produce a substantial “red wave”.
That wave has not materialised.
Republicans may still gain control of the House of Representatives, although the final outcome remains unresolved as I write this. In the Senate, however, Democrats have now secured enough seats to retain control after Catherine Cortez Masto won re-election in Nevada, following Mark Kelly’s victory in Arizona. Georgia’s Senate contest will proceed to a December runoff, but Democrats no longer need Georgia simply to preserve control of the chamber.
The result is striking because several conventional indicators pointed towards a much worse election for the governing party. The question is therefore not simply why Republicans underperformed. It is what their underperformance tells us about how voters actually make political decisions.
Economists and political scientists have long studied economic voting, the idea that citizens reward governments when economic conditions are good and punish them when conditions deteriorate. The mechanism makes intuitive sense. Most voters do not have time to examine every government programme individually, so economic performance becomes a relatively accessible indicator of competence. Are wages rising? Can I find work? Are prices stable? Does my household feel more secure than it did previously?
If the answers are positive, incumbents benefit. If they are negative, incumbents suffer.
By that logic, 2022 should have been extremely difficult for Democrats. American inflation reached levels not experienced for four decades. The Federal Reserve has been raising interest rates aggressively. Housing affordability has deteriorated. Although unemployment remains low and the labour market relatively strong, the most politically visible economic indicator has been the price level, and voters experience inflation repeatedly every time they purchase groceries, petrol or other necessities.
Yet elections are not referendums on one variable.
That may sound obvious, but political forecasting frequently behaves as though a small number of macroeconomic indicators can explain most voter behaviour. The 2022 midterms are a useful reminder that voters possess multidimensional utility functions, even if nobody entering a polling station describes their preferences that way.
A voter can dislike inflation while caring more intensely about abortion.
Another can disapprove of Biden while considering the Republican candidate in their state unacceptable.
Another can prefer Republican economic policy while worrying about candidates who refuse to accept legitimate election results.
Another may vote almost entirely on immigration, guns, climate change or education.
The final ballot aggregates those priorities.
This is similar to the political bundling problem I discussed after Australia’s election. Parties offer packages. Voters cannot select Democratic abortion policy, Republican taxation, Democratic climate policy and Republican immigration policy separately. They generally have to choose between candidates representing bundles of positions.
Candidate quality therefore matters because the package includes the person.
This appears to have hurt Republicans in several important races.
Donald Trump endorsed a number of candidates who performed strongly in Republican primaries but proved less competitive in general elections. In Pennsylvania, Republican Senate candidate Mehmet Oz lost to Democrat John Fetterman. In New Hampshire, Republican challenger Don Bolduc lost to Democratic Senator Maggie Hassan. In several states, candidates closely associated with Trump’s false claims about the 2020 election also struggled.
This creates a classic principal-agent problem inside political parties. The voters determining primary elections are not identical to the voters determining general elections. Primary electorates tend to be smaller and often more ideologically engaged. A candidate can therefore optimise for winning the party nomination while making themselves less attractive to the broader electorate needed afterwards.
The party wants a candidate who can win both stages.
Primary voters may prioritise ideological loyalty more heavily.
The result can be adverse selection.
A candidate demonstrates commitment to the party’s most intense voters by adopting positions that later become liabilities among independents and moderates.
Trump’s influence intensifies this mechanism because loyalty to him has become an important signal inside parts of the Republican Party. Candidates seeking his endorsement or support may repeat claims about the 2020 election that appeal to Republican primary voters while alienating voters who accept Biden’s victory.
This means Trump’s political influence can simultaneously strengthen his control over the party and weaken the party’s general-election performance.
Those are not contradictory outcomes.
A politician can become more dominant internally while making the organisation less electorally competitive externally.
Political parties therefore face a trade-off between purity and coalition size.
The more precisely a party defines ideological membership, the easier it becomes to mobilise highly committed supporters. The narrower definition can simultaneously exclude voters required for a national majority.
Successful parties usually contain internal contradictions because large coalitions require them.
The Republican Party currently has to decide how much contradiction it can tolerate between Trump’s base and the suburban or independent voters needed to win competitive states.
Democrats face equivalent tensions between progressive activists and moderates, although this week’s results make those tensions less immediately visible.
Abortion appears to have changed the electoral environment substantially. In June, the Supreme Court overturned Roe v. Wade through its decision in Dobbs v. Jackson Women’s Health Organization, ending the federal constitutional protection for abortion that had existed for nearly fifty years and returning much greater regulatory authority to individual states.
That decision changed abortion from a relatively stable constitutional baseline into an immediate policy question.
This matters because political mobilisation often depends on loss aversion.
People can become more politically motivated by losing an existing right or benefit than by the possibility of gaining something equivalent.
Before Dobbs, voters who supported abortion rights could reasonably assume that Roe provided a federal constitutional floor despite Republican electoral victories. After Dobbs, state elections and legislative control suddenly determined whether abortion would remain legal under much broader circumstances.
The stakes became more immediate.
Republicans had spent decades mobilising voters around the objective of overturning Roe. Achieving that objective changed the incentive structure.
Success mobilised the other side.
This is an important political mechanism. Movements are sometimes strongest while fighting against a policy because opposition creates a clear common objective. Once they win, the consequences of victory become visible to voters who previously treated the issue as hypothetical.
Policy success therefore changes the coalition.
The anti-abortion movement achieved one of its most important legal objectives.
The electoral response may now impose costs on the party most closely associated with that achievement.
This does not mean abortion alone explains the midterms. Elections with more than one hundred million voters never have one cause. But referendum results provide additional evidence that abortion rights remain electorally powerful even in states where Republicans otherwise perform well.
The broader lesson is that institutions can alter electoral salience. The Supreme Court’s decision changed what elections control, which changed what voters had reasons to care about.
Courts therefore influence politics even when judges themselves are unelected.
This is another example of interaction between institutions rather than simple separation of powers.
A judicial decision changes policy.
Policy changes voter incentives.
Voters change legislatures.
Legislatures may later change laws or judicial appointments.
Institutions respond to one another dynamically.
The democracy issue may also have mattered. Biden spent parts of the campaign warning that election denial and political extremism threatened American democratic institutions. Republicans frequently criticised this strategy as an attempt to distract from inflation and other material concerns.
That criticism contained a plausible assumption: voters care more about immediate economic conditions than abstract institutional questions.
Tuesday suggests the assumption may have been too strong.
Institutions become less abstract when citizens believe they are genuinely at risk.
A voter may normally care far more about petrol prices than the mechanics of electoral certification. After January 6 and two years of arguments over the legitimacy of the 2020 result, some voters appear to have treated candidate commitment to electoral rules as a relevant qualification for office.
That is politically encouraging, although it should not be exaggerated into a claim that Americans voted overwhelmingly “for democracy”. Republicans won millions of votes and many Republican candidates who accepted the 2020 result performed perfectly well.
The more precise conclusion is that election denial appears capable of imposing an electoral penalty in competitive constituencies.
If that is true, democratic institutions have acquired an important self-correcting mechanism.
Politicians respond to incentives.
If denying legitimate election results helps candidates win primaries but repeatedly causes them to lose general elections, parties eventually face pressure to select different candidates.
Voters can therefore enforce democratic norms electorally without requiring every violation to become a legal question.
This is healthier than relying exclusively on courts.
Not every anti-democratic behaviour is illegal.
Some behaviours are restrained because politicians expect voters to punish them.
Norms operate partly through electoral incentives.
This is why candidate selection matters so much. A political party is not simply an ideological organisation. It is a mechanism for filtering who becomes eligible to exercise public authority.
When that filter rewards loyalty to an individual over acceptance of institutional rules, democratic risk increases.
When general-election voters punish the result, the party receives feedback.
Whether it learns from that feedback is another question.
Trump has already begun attacking some Republicans who performed better than his preferred candidates and continues exercising enormous influence over the party. The Republican response to the midterms will therefore reveal whether electoral performance can overcome internal loyalty structures.
This resembles the lesson from Chile’s constitutional referendum in September: defeat contains information only if political movements are willing to update.
A party can always construct an explanation protecting its existing beliefs. The media was biased. Voters were misinformed. Candidates were insufficiently loyal. Election procedures were unfair. Money was distributed badly.
Some of those explanations may contain truth.
But if every defeat is explained through external factors, political learning stops.
Democracy becomes an information system only when politicians allow votes to update their assumptions.
The Democrats should apply the same caution to their unexpectedly strong performance. Avoiding a red wave does not mean voters enthusiastically endorsed Biden’s entire programme. Exit polling and the closeness of many races suggest an electorate dissatisfied with both parties.
Democrats benefited partly because elections are comparative rather than absolute.
A voter does not answer “Is Joe Biden doing a good job?” and then automatically vote according to the response. They compare the available candidates.
This is why approval ratings cannot be translated mechanically into election results.
A voter can disapprove of Biden and still prefer the Democratic Senate candidate in their state.
Likewise, a voter can dislike Trump and vote for a Republican governor.
American politics is nationalised, but not completely.
This distinction matters for interpreting mandates. If Democrats retain the Senate while Republicans eventually win the House, neither side can plausibly claim the country delivered a unified ideological instruction.
The result would instead represent another divided electorate producing divided institutions.
That may frustrate people who want elections to provide decisive national direction.
It can also constrain political overreach.
The American constitutional system deliberately distributes authority across institutions elected at different times and through different constituencies. The president, House and Senate can therefore reflect different electoral coalitions simultaneously.
This creates gridlock.
It also makes sudden policy transformation difficult.
Again, efficiency and constraint trade off.
The midterms may therefore produce exactly the kind of institutional friction the American system was designed to generate.
There is another economic lesson here. Inflation clearly mattered to voters, but its political effect was mediated by attribution. How much inflation do citizens blame on Biden? How much on the pandemic, global supply chains, Russia’s invasion of Ukraine, energy markets or the Federal Reserve?
Economic voting depends not simply on outcomes but on who voters believe caused them.
This is why governments spend so much effort constructing causal narratives.
Biden describes inflation partly through global shocks and Putin’s war.
Republicans describe it through Democratic fiscal spending and energy policy.
Both narratives contain pieces of the explanation.
Voters then decide which causal story they find more convincing.
Economics becomes political through attribution.
This is similar to Pakistan’s floods. A disaster can contain global causes and domestic policy failures simultaneously. Political actors emphasise whichever level of explanation reduces their own responsibility.
The electorate has to process competing causal models with limited information.
That is an extraordinarily demanding expectation.
Democracy assumes ordinary citizens can make political choices without becoming professional economists, epidemiologists or constitutional lawyers.
The system works partly because voters use heuristics: party identification, candidate character, economic conditions, trusted institutions and salient issues.
Those heuristics are imperfect.
They are necessary.
This is another reason candidate quality matters. When voters cannot evaluate every policy claim, they evaluate whether the person making the claim appears competent, trustworthy or extreme.
Parties ignore that at their own risk.
The result also complicates the idea that polarisation inevitably rewards extremism. Highly polarised primary electorates can reward ideologically intense candidates. General elections still create incentives towards broader coalition-building in competitive states.
American institutions therefore contain competing pressures.
Primaries pull candidates towards party activists.
General elections pull them towards median voters.
Safe districts weaken the second mechanism because the primary effectively determines the winner.
Competitive states strengthen it.
This may explain why political extremism can become concentrated in constituencies where electoral competition is weak.
The design of electoral districts therefore affects not only which party wins but what kind of candidates parties have incentives to nominate.
Institutional design again shapes political behaviour.
By tonight, one fact is already clear: predictions of a Republican landslide were wrong. Democrats have retained the Senate, and several candidates strongly associated with election denial have lost important races.
The House remains unresolved, so it would be premature to declare the midterms an overall Democratic victory.
The better conclusion is that American voters refused to behave as though one issue determined the election.
Inflation mattered.
Abortion mattered.
Candidate quality mattered.
Partisanship mattered.
Democratic institutions mattered to at least some voters.
Different voters weighted them differently.
That is what multidimensional politics looks like.
Political analysis prefers one explanatory variable because one variable produces a clean story.
Voters rarely cooperate.
Perhaps that is healthy.
A democracy in which citizens cared only about current income could tolerate severe institutional abuse during periods of economic prosperity.
A democracy in which citizens cared only about institutional norms could ignore material suffering.
Political accountability needs both.
Governments should know that economic failure can cost votes.
Politicians should also know that attacking electoral legitimacy can cost votes even when the government they oppose is unpopular.
The 2022 midterms may have delivered both signals simultaneously.
That makes the result difficult to summarise as red or blue.
It may be better understood as voters refusing the question they were expected to answer.
The campaign asked whether Americans were satisfied with Biden’s economy.
Millions answered a different question: compared with what alternative?
Elections are always comparative.
That is why political dissatisfaction does not automatically produce political replacement.
An opposition has to become credible enough to receive the dissatisfaction.
Republicans entered this election with almost every historical advantage.
The president was unpopular.
Inflation was high.
Midterms normally punish the governing party.
Yet enough voters looked at the available Republican candidates and decided those conditions were not sufficient.
That should probably concern Republicans more than losing any individual Senate race.
A political party does not win merely because the government fails.
It has to persuade voters that it represents a better alternative.
Opposition is not an entitlement to power.
It is another application for it.
