Summary
- NPR’s 11-person Swing Shift panel shows that reported unhappiness with Trump’s second term coexisted with divergent 2026 choices, varying by candidate, office and issue rather than producing uniform movement away from the president.
- NPR reported that nine of its 11 panelists were not happy with Trump’s second term, while two were; the panel’s rough alignment with national polls concerned presidential assessments, not a shared partisan destination.
- Gerald and Wally’s Georgia choices ran in opposite directions, Theresa and Stephen split their Pennsylvania tickets, and Lee and Quintero moved in opposite directions in Nevada.
- Evidence from NPR’s panel supports candidate-centered adaptation and broader monitoring for a messaging lead, not a wholesale partisan-realignment campaign.
NPR’s long-term reporting on 11 voters provides a bounded picture of a midterm electorate that can evaluate a president negatively without moving as a bloc. Only two panelists said they were happy with Trump’s second term, yet reported choices varied substantially: Gerald supported Georgia Republicans, Wally backed Georgia Democrats, three of four Pennsylvania panelists said they planned to support Josh Shapiro, and Nevada’s Lee and Quintero moved in opposing directions. The reporting therefore supports an explanation centered on candidate, office and issue-level sorting rather than a demonstrated partisan realignment. For campaign decision makers, candidate-centered adaptation alongside broader monitoring is more defensible than attempting to move voters wholesale. NPR combined completed early votes with provisional intentions in a nonrepresentative longitudinal panel, so the accounts can illuminate reported reasoning and possible messaging approaches without establishing statewide or national vote shares or probabilities for any political outcome.
What the panel establishes
NPR checked regularly with the 11 voters in its Swing Shift project, providing longitudinal accounts rather than a single cross-sectional measure. Every project voter had cast a presidential ballot for Trump over the previous decade, and most had backed him in 2024. Theresa and Quintero had voted for Biden in 2020 before voting for Trump, so the panel was not composed exclusively of consistent Republicans, although every member had voted for Trump at least once during that period. NPR’s reporting did not establish whether this voting history was a formal recruitment criterion or an observed feature of the panel’s development.
The panel’s structure made it better positioned to show departures from a recent Trump alignment than arrivals from voters without a comparable history. The accounts could display mechanisms of movement, but they could not establish the scale of that movement among the broader electorate. The midterms came nearly two years into Trump’s second term, giving panelists an occasion to weigh the country’s condition and the president’s governing record.
Only two of the 11 panelists said they were happy with Trump’s second term; nine did not. NPR described most of the group as unhappy or underwhelmed and often attributed those assessments to the war in Iran and high energy prices. The two-of-11 finding was not an approval measurement. NPR said the panelists’ presidential assessments roughly aligned with national polls in the week before Election Day, but that alignment did not produce a shared partisan destination.
The reporting also mixed different measurement states. Gerald and Wally had cast early votes, while other panelists described intentions, including Theresa’s continuing uncertainty. NPR attributed concern about the war and energy prices to most unhappy or underwhelmed panelists, not to every member. The panel therefore combined finalized choices, provisional intentions and issue-specific judgments rather than providing one settled measure of opinion.
Choices diverged by candidate, office and issue
The reported cases gave neither party a uniform electoral benefit. Republican-aligned choices included party-line support, incumbent retention and a refusal to support an unopposed Democrat. Democratic-aligned choices included incumbent familiarity and movement for change. Dissatisfaction with Trump was not equivalent to approval of the opposing party, and opposition to the opposing party was not equivalent to a settled partisan identity.
Georgia
Gerald and Wally both voted for Trump in 2024, helping him in Georgia, but their 2026 choices ran in opposite directions. Gerald gave Trump an A+++ and cast early votes for Republican Senate candidate Mike Collins and Republican gubernatorial candidate Rick Jackson. “Red all the way down the ballot,” Gerald said.
In a separate race, Gerald explained why he had refused to support the only listed candidate, a Democrat running unopposed: “There was one instance where there was nothing but a Democrat running unopposed and I preferred to write in my own name in, because I refuse to give a Democrat my vote.”
Wally had consistently given Trump an F since March. He voted early for Democratic Senate incumbent Jon Ossoff and Democratic gubernatorial candidate Keisha Lance Bottoms. Inflation was his leading issue, although he said he had little hope that any one politician or election could fix it. He gave both Democrats and Republicans in Congress an F and said, “we need wholesale change in this midterm, and we also need a big change for the next presidential election.” NPR described Wally’s ballot as a vote against Trump and MAGA as much as anything else.
Pennsylvania
Pennsylvania’s choices were also divided. Three of the four panelists said they planned to vote for Democratic Gov. Josh Shapiro, who was seeking re-election. NPR reported that three of the four had backed Trump in 2024 and that Pennsylvania swing voters said Trump had not delivered on his promise to lower prices.
Theresa, who had voted for Biden in 2020 and Trump in 2024, gave Trump a C-. She criticized the duration of the Iran war: “Because he said the war was only going to be a few weeks. And we knew that was not the case,” Theresa said. She said, “I might get in there and change my mind again.” Familiarity with Shapiro also mattered: Theresa said she was moving toward “the politician that we’re familiar with, which is Shapiro.”
Stephen, who lives in Pennsylvania’s competitive 1st Congressional District, said he would re-elect Republican Rep. Brian Fitzpatrick while supporting Shapiro for governor. “I would still re-elect Brian Fitzpatrick but everything else, I would say maybe Democratic or…maybe do a write in,” Stephen said.
John was the only Philadelphia voter reporting a vote for the Republican candidate for governor. He followed the news closely and said the country felt like it was “teetering.” After a possible Democratic midterm victory, he expected “a lot of investigations, a lot of impeachment talk and a lot of gridlock in Washington.” John said he did not know whether the country could afford that outcome and used “anxious” to describe the state of the country. His stated reasoning could not be reduced to party loyalty.
Nevada
The Nevada accounts moved in opposite directions. Lee, one of the two panelists who said they were happy with Trump, gave his performance an A-. “Because the war in Iran is taking kind of long,” Lee said. Lee had previously been out of work but had since found a new job and said his economic outlook had improved. He also said he had become more Republican and more suspicious of Democrats.
Quintero, a voter who had moved from Biden to Trump, told NPR he was swinging back toward Democrats in the 2026 midterms. The two accounts demonstrated why satisfaction with Trump did not necessarily produce one partisan destination: Lee could approve of the president while remaining concerned about war, and Quintero could reverse his earlier choice while Lee moved further toward the Republican Party.
Asked for one word or phrase to describe their feelings, panelists used “messy,” “divided,” “disarray” and “scary.” Those descriptions captured the reported national mood more clearly than they revealed a common ballot choice.
The choices point to candidate-and-performance sorting
The most consistent reading of the accounts is candidate-and-performance sorting rather than demonstrated partisan realignment. Panelists invoked combinations of presidential performance, inflation, energy prices, the war in Iran, party identity and individual candidates. That mixture was consistent with issue- and candidate-level sorting, but the reporting did not establish the mechanism that produced the choices.
Incumbent familiarity was the clearest reported electoral advantage. Theresa supported a familiar incumbent governor, and Stephen supported a familiar incumbent representative while choosing a different party for governor. The panel did not establish whether challengers could build comparable familiarity. Trump’s absence from the ballot also meant the panel did not test how a Republican-aligned message might travel through another candidate.
A generic change-centered message would not directly answer the incumbent-familiarity preference reported in Theresa’s and Stephen’s cases. Broad party framing could obscure the seat-specific reasons behind their choices. NPR’s reporting did not contain a comparative test of those approaches.
How the accounts frame the election
NPR described the election as “a referendum on the president and his policies,” although Trump was not on the midterm ballot. That frame made presidential performance relevant to House, Senate and gubernatorial decisions, but the reported choices complicated a uniform reading. Gerald’s refusal was the clearest limit on the referendum frame: because no Republican candidate was listed in that race, candidate competition could not resolve his refusal. The referendum analogy anticipated a broader verdict, while Gerald’s case recorded a narrower refusal.
The reported grades provide a scorecard-like reading of the presidency. A+++, A-, C- and F treated presidential performance as a matter under review, with the election serving as an occasion for the score. That frame explained why voters evaluated Trump through grades, but it did not fully explain the mechanics of their seat-level choices.
An office-by-office reading draws support from Theresa’s preference for a familiar politician and Stephen’s decision to re-elect Fitzpatrick while supporting Shapiro. This frame makes split-ticket choices intelligible because voters can evaluate particular offices and officeholders separately. It does not show how thousands of individual local judgments aggregate into governing power.
A country-as-condition reading treats “teetering,” “messy,” “disarray,” “scary” and “anxious” as descriptions of a national condition rather than an assessment of a government’s record. John’s account defined that condition partly through possible investigations, impeachment talk and gridlock. This frame helps explain why a voter dissatisfied with a president might nevertheless retain an incumbent party.
A party-as-movement reading treats Gerald’s refusal and Wally’s description of his ballot as rejection of Trump and MAGA as an affirmation or rejection of a governing project. That frame makes the referendum impulse visible, but it does not explain the cross-office choices.
A moral reading asks whether voters understood politics through recurring family or moral structures. Gerald’s preference for a uniformly Republican ticket and refusal to vote for an unopposed Democrat came closest to a party-as-family interpretation. Wally’s criticism of congressional parties and call for wholesale change pointed in another direction. The interviews did not contain enough common language to establish a shared moral hierarchy.
A policy reading considers whether the central problem was presidential performance, party direction, prices, war or the local representativeness of candidates. Wally emphasized inflation; Theresa criticized the war’s duration; Pennsylvania voters raised price concerns; and Stephen and John described dissatisfaction with candidates and party choices. The accounts permitted several definitions of the problem at once, consistent with the absence of a uniform remedy.
A collective-action reading treats the election as a contest over the direction of Trump and MAGA. Wally’s Democratic choice and call for change fit that interpretation, as did Gerald’s Republican vote as support for a continuing governing project and Lee’s movement toward the Republicans. The reading did not explain the split-ticket cases or John’s simultaneous dissatisfaction with both parties.
A narrative reading treats the shared descriptions of national mood as the central story. Those descriptions captured uncertainty but explained mood more readily than ballot choice. A distributive-justice reading could foreground how economic and security burdens were distributed, but the article supplied too little direct evidence about differences in exposure to establish that frame.
The best-supported working interpretation was therefore a hybrid in which voters appeared to sort issues, offices and candidates separately rather than apply one comprehensive ideology. The interviews did not provide enough common language to formalize that hybrid with confidence, and NPR’s reporting did not adjudicate among the frames. The article’s presentation reinforced the ambiguity by placing individual accounts and evocative descriptions of national mood near the center while giving less attention to the missing electorate and aggregate benchmarks. Its referendum analogy was useful, but the reported choices did not constitute a single, uniform test.
What the evidence implies for messaging decisions
The decision owner identified by the evidence is the messaging lead: whether to organize outreach around a broad judgment of the president or around candidate-specific attributes. The panel supplies grounds to test both approaches but does not support treating either as universal. Four possible responses follow from the reported evidence; none is a documented party plan.
Continue a referendum-centered message. Treating the midterms as a presidential referendum and building a uniform party signal could preserve familiar partisan identities. Automatic assumptions about loyalty would be hazardous, however, because voters could support a Republican congressional candidate while rejecting a statewide candidate, or support a Democratic governor while retaining a Republican representative.
Restore party discipline. Making party unity dominant could reduce split-ticket voting and provide a simpler national narrative. It could also conflict with the candidate-specific reasoning visible in Georgia and Pennsylvania.
Defer a major strategic shift. Waiting for broader polling and election returns could prevent overreaction to a small panel. It would also postpone responses to the price, energy, war and governing-performance concerns described by voters.
Adapt around candidate-centered concerns. Organizing outreach around prices, energy, war and governing competence while preserving separate pathways for voters willing to split tickets directly reflected the reported reasoning. Republicans would need to connect presidential performance to particular candidates and offices rather than treating approval of Trump as a substitute for those choices. Democrats would need to distinguish opposition to Trump’s record from endorsement of every element of their platform.
On the present evidence, candidate-centered adaptation paired with broader monitoring was more defensible than a wholesale attempt to realign voters. The recommendation applied symmetrically to both parties.
The decision should be revisited at the next scheduled voter-contact cycle and again before Election Day. Recontact should cover the same voters to preserve the longitudinal comparison and, where field conditions permit, continue weekly through the final week before Election Day. The recontact should be paired with adjacent battleground polling, ballot information and access data, and available turnout data.
A broad governing message would have a stronger evidential basis only if repeated evidence showed voters changing choices across multiple offices in the same partisan direction. One dissatisfied comment or one cross-office deviation would not establish that pattern.
Scenario ordering and update rule
No defensible numerical probabilities could be calculated from the panel. The scenarios were ordinal evidence bands rather than estimates of national vote shares.
Fragmented and candidate-specific voting represented the highest band. This extended the pattern most directly visible in the reported choices: voters continued separating presidential evaluations from House, Senate and gubernatorial decisions. Relevant indicators would include candidate-specific polling among former Trump voters and the frequency of down-ballot voting across conventional party lines.
Performance-driven sorting represented a material but secondary band. Inflation, energy prices and the war in Iran could alter evaluations of Trump and the parties, but their effects would not necessarily be uniform. Lee’s satisfaction alongside his concern about the war, and Pennsylvania voters’ price concerns alongside support for a Democratic governor, showed why issue movements did not necessarily produce one partisan destination. The panel did not establish how often those issues changed votes outside the group.
Reversal, consolidation or uniform directional realignment occupied a lower band. Disaffected Trump voters could return to Republican candidates, or Democrats could convert soft support into durable party loyalty. Lee’s improved personal circumstances accompanied his movement further toward the Republican Party, but that individual case did not establish that the pattern would recur nationally. Reported movement ran in both directions, so uniform directional realignment was not the most directly visible pattern; it could not be ruled out because the panel supplied no base rate and lacked voters without a prior Trump vote. Relevant indicators would include party identification and the cohesion of congressional voting blocs.
Discontinuity occupied the tail band. An unexpected development in the war, energy markets or the economy could reorder the priorities structuring the reported choices. The panel supplied no basis for assigning a numerical weight to that possibility, but it also meant no trend projection could be treated as insulated from sudden changes in political conditions.
The scenario ordering should change if a majority of recontacted panelists moved in the same partisan direction across two or more offices at two successive contacts, or if independent polling repeatedly showed that pattern in more than one battleground. If neither threshold was met, seat-level divergence would remain the pattern most directly visible in the panel rather than being converted into a national forecast.
A clean partisan outcome after Election Day would not by itself disprove the seat-level mechanism. Separate local choices could aggregate into a seemingly uniform national result, and the panel could not determine which explanation was correct.
Movement away from Trump could not be equated with movement toward Democrats when voters cited issue-specific or candidate-specific reasons. Stated intentions also could not be treated as final choices alongside Theresa’s uncertainty and the completed votes by other panelists.
A national claim about uniform movement would require evidence the reporting did not contain: repeated same-direction shifts across offices and states, together with evidence of movement into the Trump coalition among voters not represented in a panel whose members had all previously voted for him. The most useful missing measurement would be a representative measure of movement from Democratic to Republican presidential identification, alongside comparable battleground polling, turnout information and split-ticket results.
A backcast from a desired governing outcome could not be constructed because no desired endpoint had been specified. That was a gap in the scenario design rather than an implied preference.
Evidence boundaries and unresolved questions
The 11 voters were longitudinal cases, not a representative probability sample or a sentiment survey. NPR’s alignment note with national polls was qualitative, and the article did not claim that the panel represented the electorate. The accounts could document heterogeneous preferences but could not establish statewide or national vote shares.
The reporting did not establish the views of undecided voters, nonvoters, voters outside the states and districts discussed, or Trump voters outside the project. It also did not establish whether an interviewee’s willingness to split tickets would transfer to other candidates.
NPR’s article supplied no base rate, numerical poll figures, reference series for comparable voter panels, turnout information, state-level contest data or comparable split-ticket results from which to calculate national probabilities or identify a modal national scenario. It also contained no accounts of judicial or local races, a notable omission under the referendum framing it supplied.
The article did not separately identify John’s ballot, although its statement about every project voter placed him within the panel’s Trump-vote history. It also did not date each assessment relative to early voting or Election Day, so no chronology for the panel’s changing sentiment could be inferred.
A source-review question remained: the cited NPR item was dated Nov. 3, 2026, while its title described the reporting as ahead of Election Day. The supplied package did not resolve whether that reflected a late revision or a publication convention, a distinction relevant to anyone citing it as a pre-election snapshot.
Analytical techniques used in this piece
This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.
- Decision Clarity
- Articulates the real stakes, stakeholders, and interests behind a decision facing a third party.
- Frame Comparison
- Sets two or more competing frames side by side to see what each reveals and hides.
- Wicked Futures
- Explores a long-horizon, deeply entangled future with no clean resolution.