Pundits have predictably filled every hour since the midterm polls closed with their “analyses” of the landslide’s causes, which generally amount to blaming whatever Trump administration policies of the past year they least like. Among these factions, free traders have distinguished themselves with the most credulous efforts, insisting that the midterms were a referendum on President Trump’s tariffs, which they say have failed.
The shortcomings in this argument are two-fold, and worth examining as both parties begin to think through how they will address tariffs and trade policy in the post-Trump era. One problem is that the tariffs have not failed. While 18 months is far too little time to initiate, let alone complete, the investment necessary for reindustrialization, a wide range of indicators confirm that the manufacturing sector has begun to experience a revitalization.
The second problem is that exit polls provide little useful information about why voters voted the way they did. For the 2026 midterm, pollsters included five questions in the exit poll asking “Thinking about your vote for Congress, how important to you was…” the cost of living, the war in Iran, artificial intelligence (AI) and data centers, immigration enforcement policies, and government corruption. But not tariffs.
Instead, pundits found the question that asked “Do you think Trump’s tariff policies have helped or hurt the national economy?” and trumpeted the result that 66% had responded “hurt” while only 21% responded “helped” and 12% responded “no difference.” But the question says nothing about whether this was important to a vote, nor even whether the respondent supports or opposes tariffs. And split by party, 96% of Democrats said that tariffs hurt, compared with 27% of Republicans. In other words, the data showed that Democrats don’t like President Trump, and not much else.
Fortunately, American Compass has maintained a comprehensive dashboard of economic data related to reindustrialization since Liberation Day, and had already prepared a large survey of voting rationales, fielded immediately after Election Day in partnership with YouGov. The following data demonstrate that:
- The industrial economy has turned a corner and is performing as tariff supporters would have hoped, with little sign of the harms predicted by tariff opponents;
- Corruption and the Iran War were the primary issues that pushed voters toward the Democratic Party while tariffs had no significant effect on the results;
- Voters continue to support policies with short-term costs, including higher prices, if those policies promise better jobs and a stronger economy in the long run; and
- A conservative message emphasizing family, community, and industry is far more popular with potential voters than one emphasizing liberty, opportunity, and enterprise.
The Manufacturing Boom Underway
A wide range of measurements support the case that the industrial economy began responding quickly after imposition of broad-based tariffs and is now booming. To test in particular the hypothesis that voters must be frustrated with the failure of tariffs, the following charts compare performance in the six months prior to Election Day 2024 with the six months prior to Election Day 2026. Each comparison uses the most recent six months of data available before each election; trend lines throughout the periods are provided alongside the summary statistics.
Begin with Purchasing Managers’ Indices (PMIs), which provide a real-time snapshot of industry sentiment, conditions, and expectations. Tariff critics initially focused intensively on this measure when it did not instantly turn positive after President Trump took office, or after Liberation Day. But that should have surprised no one. Over the next year, PMIs moved steadily higher, and have now held in strongly positive territory for nearly a year. Also surprising no one, the critics don’t talk about PMIs anymore.
Once PMIs proved unhelpful, tariff critics moved on to manufacturing employment, which had been falling sharply at the end of President Biden’s term and continued its slide after President Trump took office. Of course, factories must be built before they can hire, which made the measure a strange one to emphasize in the short run, unless the goal were simple demagoguery. Nonetheless, employment losses continued to slow, then stopped, and then hiring began. Today, employment stands above its Liberation Day level and job openings are at their all-time high (excluding the post-Covid recovery).
Across the board, the manufacturing picture has brightened considerably. After sliding for more than a decade, productivity growth is back: rising at 1.6% per year, 8x the mid-2024 rate.
Industrial production had been falling 1.5% per year, now it is rising 4.4% per year.
Orders for goods manufactured domestically are surging, with growth now 7x the 2024 rate.
Encouragingly, that growth rate is especially strong for capital goods…
…stronger still for machinery…
…and through the roof for industrial machinery.
And finally, real fixed investment is starting to rise again, even after excluding the “information processing equipment” category driven by data center expansion, though it remains below its 2024 level.
Likewise, exclude the categories of goods concentrated in data center expansion—semiconductors, computer equipment, computer accessories and telecommunications equipment—and the trade deficit has fallen by nearly half.
The United States still has a long way to go toward reindustrialization, and the needed investment is still beginning to germinate, but the evidence looks overwhelmingly positive at this stage and provides no sound basis for arguing that tariffs have harmed the economy, or the manufacturing sector in particular. If the trade agenda were succeeding, this is how it would look.
What the Voters Say
American Compass surveyed 2,000 Americans who voted in the 2026 midterm election, asking for their opinions on the cost of living, key issues in the election, and the types of tradeoffs they would support in public policy. Broadly speaking, Americans perceived that all policy issues had caused prices to increase, with tariffs rated among the most significant drivers of cost, behind only the Iran war and corporate efforts to boost profits.
What did people vote on? The survey asked whether five different developments during the Trump administration had made respondents much or somewhat more likely to vote for either Democrats or Republicans: tariffs, immigration enforcement, the Iran war, the One Big Beautiful Bill (OB3), and corruption. Overall, the results show all issues pushing voters toward Democrats.
But this result says little. In general, Democrats were much more likely to give ‘much more likely’ responses and, for all voters, responses across issues are highly correlated. Indeed, more than half of respondents gave the same answer on each issue, and most of those said every issue made them much more likely to vote Democratic—in other words, that they are Democratic voters who were going to vote for the Democrat, regardless of particular issues.
These questions only capture how voters explain their votes, not what might truly have motivated them, and where the voter responds the same way on every issue there is no way to differentiate among them. Fortunately, the granularity of a survey (unlike an exit poll) allows for more robust analysis, in two steps: (1) focusing only on those voters who did differentiate among issues and voted for the Republican or Democratic house candidate, which still leaves a substantial sample of N = 847, and (2) isolating the effect of each individual issue, holding others constant.
As a first step in this direction, we can tabulate the effect of a “push” on each individual issue, holding other issues constant. For instance, to assess the effect of immigration enforcement on voters, we can begin by setting a “push” value for each voter on the other four issues combined, where “much more likely to vote for Democrats” is worth +2, “somewhat more likely to vote for Democrats” is worth +1, and “much more likely to vote for Republicans” is worth -2. A voter who says all four issues make him much more likely to vote for Republicans would have a score of -8, and can be compared to all other voters in the same -8 bucket.
Suppose we have 20 voters in the +3 bucket. Of those, 10 say immigration made them more likely to vote for Democrats and 10 say it made them more likely to vote for Republicans. Of those pushed toward Democrats, 7 voted for their Democratic House candidate. Of those pushed toward Republicans, 4 voted for their Democratic House candidate. In this +3 bucket, then, being pushed toward the Democrats by immigration increased the Democratic vote share by 30 percentage points.
Of course, one small bucket of 20 voters will not generate a statistically significant result. But apply the methodology across all voters in all buckets who differentiated across issues, create an average of the bucket results with most weight given to buckets with the most voters on both sides, repeat the process for all five issues, and a much more interesting result emerges. Immigration had by far the strongest effect, with voters reporting that it pushed them toward Democrats voting 40 points further toward Democrats than voters who say it pushed them toward Republicans. Tariffs had the least effect, opening only a 14-point gap between voters who say it pushed them Left and voters who say it pushed them Right.
Those with statistical training will recognize the methodology above as, in effect, a manually executed regression (done that way to make the process and result legible to most readers). We can do the same thing with an actual regression, setting the push values for the issues as the independent variables and the House vote (D = 1, R = 0) as the dependent variable.
But those with more statistical training are thinking: an OLS regression is the wrong model for that data. It should be a binomial logistic regression, the standard model for predicting a binary outcome (R or D, 0 or 1). True enough.
Here is the better model: It produces roughly the same ranking among issues, with tariffs and the One Big Beautiful Bill still last and no longer statistically significant.
All three approaches (matched comparison, linear model, logistic model) ranks tariffs at or near the bottom; in the logistic model, their small estimated effect cannot be distinguished from zero.
Now we have looked at how many voters reported being pushed by various issues and we have looked at how much being pushed on an issue affected a vote. We can also, most importantly, combine the two. Let’s apply the logit model to the actual answers of the 1,931 respondents who voted for a Republican or Democratic House candidate, weighted to represent the electorate. Based on how much each respondent said he was pushed by each issue, and how much such responses appeared to in fact push voters, how much can we say this issue affected the final election result, holding constant other attitudes?
These results are consistent with the earlier analyses. Corruption and the Iran war both had major effects on vote totals, with corruption the largest. Tariffs and OB3 had no discernible effect. The immigration result notably underscores the value of this modeling approach: On one hand, the issue was clearly significant in pushing voters. On the other hand, it uniquely pushed voters in both directions—in essence, a polarizing issue that netted little gain to either side. The effect magnitudes also align fairly well with the swing in the House vote observed from 2024 to 2026. (But note, tempting though it may be, the swing estimates for specific issues cannot be added together to produce an overall swing.)
Applying the logistic model to only the 847 respondents who differentiated among issues yields larger effects for corruption and the Iran war, but tariffs remain statistically insignificant.
The Wise Tariff Voter
What is going on here? Cost of living issues were widely cited as central to the election, and voters clearly perceive tariffs to have been one of the most significant drivers of higher prices. Yet by voters’ own accounts, tariffs did not move their votes significantly.
The issue is best understood by thinking in terms of policy tradeoffs rather than discrete policy actions. A pollster asking “do you like tariffs” will find low support, just as when asking about any other tax. Conversely, a pollster asking “do you like reindustrialization” will find high support, just as when asking about a balanced budget.
So after asking all of the issue-specific questions, the survey also asked about two sets of tradeoffs. First, economic policy often involves a tradeoff between keeping prices as low as possible and making sure American workers have good jobs that can support a family. Consider the following two options:
- An economy that emphasizes expanding what people can buy, offering as many choices as possible at the lowest possible price.
- An economy that emphasizes making things and creating good jobs that ensure workers can support their families and the nation can provide for itself.
Thinking about where the country is right now, which way would you like elected leaders to lean?
By more than 3-to-1, voters chose an emphasis on making things and good jobs over expanding what people can buy at the lowest possible price. The result holds across parties and classes.
Second, in addressing affordability, leaders sometimes must choose between actions that lower prices now but may hurt the economy in the long-run, and actions that raise prices now but may strengthen the economy in the long-run. Consider the following two options:
- Accept higher prices now if that makes better jobs, higher wages, and middle-class security more likely in the future.
- Bring down prices today, even if that puts better jobs, higher wages, and middle-class security at risk in the future.
Thinking about where the country is right now, which way would you like elected leaders to lean?
The results were not as lopsided when the tradeoff was framed this way, but the long-term perspective still prevailed by a sizeable margin. Only Democrats and respondents with lower levels of education and income were closely divided.
Politicians should bear in mind these results when interpreting election results, crafting their own messages, and—most importantly—making policy.
Which Way GOP?
As long as we had their attention, we decided to ask voters three questions about the direction that the Republican Party should head as it enters the post-Trump era. The first is simply with reference to Trump: should the next Republican leader keep his policies, his style, both, or neither? Respondents were given four options (keep or change policies cross with keep or change style), but here we recombine responses to show attitudes toward policies and style separately.
Second, we asked which direction the Republican Party should go if it wants to have a chance of building a durable governing majority. Respondents were asked to choose between a description characterizing the Reaganite “Old Right” and one characterizing a post-Trump “New Right.”
Finally, we asked the question in its simplest possible form: Pick a slogan.
Certainly, no consensus yet exists among Republican voters about their party’s future direction. But the old Reaganite consensus is a consensus no more: it leaves the party split nearly in half and holds little appeal for other voters the party must attract.
About the Data
The American Compass Post-Midterm Survey was conducted by YouGov between November 5 and November 11, 2026. YouGov interviewed 2,438 respondents, of whom 237 were removed for data quality reasons. The remaining sample was then matched down to a sample of 2,000 to produce the final dataset.
The respondents were matched to a sampling frame on gender, age, race, and education. The sampling frame was constructed by stratified sampling from the 2024 Current Population Survey (CPS) Voting and Registration supplements, and the 2024 AP Votecast exit poll, including demographics and 2024 presidential vote. All data were weighted to the CPS-VRS person weights on the public use file, with Hur & Achen reweighting adjustments to account for missing data within the voter registration and turnout fields.
The matched cases were then iteratively proportionally fitted (IPF) on 2024 presidential vote choice, a four-way stratification of gender, age (4-categories), race (4-categories), and education (4-categories), these demographics’ associated marginals, home ownership, and region, to produce the final weight. Chart percentages are weighted unless noted (regression models are estimated on unweighted data). The chart of differentiated responses is unweighted to show counts of respondents
In analysis of the results, “Class” is defined by education and income:
- “Lower” (N=298): less than a four-year degree and household income below $30K; or did not report household income and do not have a high school diploma.
- “Working” (N=517): less than a four-year degree and household income $30K–$80K; or did not report household income and have either a high school diploma or some college but no degree.
- “Middle” (N=811): four-year degree or more and household income $30K–$80K; or household income $80K-$150K; or did not report household income and have a two-year or four-year college degree.
- “Upper” (N=294): household income above $150K; or did not report household income and have a post-graduate degree.
Respondents with a four-year college degree or more but household income below $30K are excluded from analyses using the “Class” variable (N=80).
A full survey cross-tab with all question and option wording is available here.

