The Arch and the Keystone: Unified Government, Divided Government, and the Economy

On November 3 the country votes, and the most probable outcome, on the current arithmetic, is that one chamber changes hands and the other does not. That is the forecast this site published when the midterms were ten weeks out, and if it holds, the next two years will be conducted under divided government. So the question worth asking now is not who wins. It is what divided government actually accomplishes.

The folk theory runs in two directions and both are confident. One holds that a single party controlling the House, the Senate, and the White House is the condition for competence: unified government acts, divided government quarrels. The other holds the opposite, that gridlock is good, because a government that cannot pass anything cannot do harm. Neither theory is well supported, and the reason is instructive, because the thing divided government reliably changes is not the economy. It is the statute book.

I. Two words, and what they cover

A government is unified when one party holds the presidency and majorities in both chambers. Divided government is everything else, which includes six combinations: a president of one party facing either chamber of the other, or both.

Since the Second World War, unified government has been the less common condition. Of the forty Congresses convened since 1947, seventeen opened under a trifecta and twenty-three did not. The modern examples are familiar: Truman’s elected term, Kennedy and Johnson, Carter, the first two years of Clinton, the 107th through 109th Congresses under George W. Bush, Obama’s first two years, Trump’s first two, Biden’s first two, and the current one.

That list matters for the argument, because it is short and its members do not resemble one another. It contains the New Frontier and the Great Society, and it also contains the 107th Congress, which held a trifecta for a few months before a senator changed parties. Any claim about unified government has to survive the fact that the category contains both the Great Society and the Jeffords switch.

II. The claim that gridlock is good

The idea has a respectable lineage, and it is worth stating at its strongest.

Alberto Alesina and Howard Rosenthal’s Partisan Politics, Divided Government, and the Economy (1995) is the canonical treatment. Their model is not about gridlock as a brake. It is about balance as a signal, and their claim is that middle-of-the-road voters deliberately split their tickets. “Because Democrats and Republicans usually maintain polarised preferences on policy,” they write, “middle-of-the-road voters seek to balance the President by reinforcing in Congress the party not holding the White House. This balancing leads, always, to relatively moderate policies and, frequently, to divided government” (Alesina and Rosenthal, 1995).

Read that carefully, because it is not the claim people attribute to it. The claim is that divided government produces moderation, not paralysis, and that the mechanism is the voter rather than the constitution. A divided government, on this account, is the electorate’s way of installing a median outcome it could not get from either party alone.

There is a second, blunter version, popular on the right. William Niskanen’s 2003 essay for the Cato Institute tabulated real federal spending growth by administration and concluded that “the rate of growth of real (inflation-adjusted) federal spending is usually lower with divided government” (Niskanen, 2003). That essay is advocacy, not peer-reviewed work, and the tabulation is a list of administrations rather than a regression, which means it cannot separate divided government from the personalities and eras involved. It is worth citing only as the strongest form the claim takes, and it is thin.

III. What the growth data shows

So let me compute it rather than quote it. Taking each year from 1930 to 2025, assigning it to whatever government was in session, and comparing real GDP growth gives this.

Growth figures are the annual change in real GDP, from the Bureau of Economic Analysis series (FRED, 2026).

The full sample looks like a real effect, and at t = 2.14 it clears the conventional bar. Then the sample splits, and the effect evaporates. Across the postwar period the gap is 0.72 points and not significant. Since 1981 it is negative eighteen hundredths of a point, which is to say nothing at all.

The reason is visible in the era split. Between 1930 and 1946, the two divided-government years in the sample are 1930 and 1931, which sit inside the deepest contraction in American history, while the unified years that follow include the wartime mobilization. A comparison that spans that boundary is measuring the Depression and the Second World War and calling it a partisan effect.

The other outcomes behave the same way. Unemployment averaged 5.34 percent under unified government and 5.91 under divided across 1948 to 2025, a difference that is not significant. The federal deficit averaged 2.66 percent of GDP under unified government and 2.69 percent under divided, which is to say the two are indistinguishable. Federal outlays ran at 19.2 percent of GDP under trifectas and 19.6 percent without them. On none of these measures does the composition of government separate the outcomes.

IV. Three states, not two

The comparison above is the one everyone runs, and it is the wrong cut. There are three states, not two: unified under a Democratic president, unified under a Republican president, and divided. Split that way, the same years say something the two-way comparison cannot see. Regimes are assigned per the House Historian’s own unified-government table (U.S. House, 2025).

The apparent advantage of unified government turns out to belong to one of its two forms. Unified Democratic government is the outlier. Unified Republican government performs no better than divided government, and in the full sample slightly worse.

That reframes the question. The right test is not whether unified government helps but whether the composition of government matters within a party, because that is what separates a composition effect from a party effect:

A regression on the postwar years makes it explicit. With an interaction term between the two:

$$\text{growth} = 2.72 + 0.17,\text{Democrat} - 0.25,\text{Unified} + 1.38,(\text{Democrat} \times \text{Unified})$$

The independent effect of unified control is −0.25 points and indistinguishable from zero (t = −0.31). What looks like a unified-government premium is entirely an interaction: unified control pays only when the president is a Democrat, and even that coefficient (t = 1.24) does not reach significance.

Then the war. Democratic unified years split cleanly by era: 7.07 percent for 1933 to 1946, 4.02 percent thereafter. What survives the three-way split is still mostly Roosevelt and the Second World War.

And the Republican unified sample is smaller than its twelve years suggest, because one of them is 1930, when Hoover’s last full year returned −8.51 percent. That single year drags the mean down by nearly a point on its own. Remove it and unified Republican government reads 2.47 percent, which is divided government to the tenth of a point. The medians were never the problem: Republican-unified medians of 2.62 to 2.78 sit beside the divided medians of 2.74 to 2.86 throughout.

Since 1981 all three states read 2.60, 2.57, and 2.76 percent. The spread is hundredths of a point, and the ordering is not even the one either folk theory predicts.

The fiscal measures do not separate either, and with three states visible this matters more than it did: the deficit ran 2.65 percent of GDP under Democratic trifectas, 2.37 under Republican ones, and 2.66 divided, with outlays at 19.1, 19.3, and 19.7. Neither form of unified government restrains spending.

One result does run the other way, and it deserves its own weakness attached to it. Unemployment averaged 4.80 percent under unified Republican government against 5.91 divided, a significant difference (t = −2.98), and it holds in the modern window at 4.92 against 6.16. But read the years. Those are 2001 through 2006 and 2017 through 2018, the housing boom and the late expansion, while the Democratic-unified modern mean is dragged by 2009 and 2010, which were inherited. This is a handful of specific episodes rather than a governing effect, and it is the kind of finding a three-way split makes easy to generate and hard to defend.

The correction matters more than the arithmetic. The two-way comparison was absorbing a party difference and relabeling it a composition difference. Split three ways, the composition effect collapses into the party effect, and the party effect is the one the companion essay already showed to be largely luck, largely Roosevelt, and absent since 1981.

V. The strongest claim, and why it does not survive

There is a recent paper that argues the opposite, and it deserves to be taken seriously rather than waved away.

Papamichalis, Ryu, and Wilson — economists at Cambridge, Harvard, and Oxford — presented “Divided Government and the Stock Market” at the American Economic Association meetings in 2025. They report that “the value-weighted excess monthly stock market return under United governments is 10.3% per annum versus 1.6% under Divided governments,” a gap of 8.7 points with statistical significance, and that on equal-weighted returns unified government earned 16.8 percent a year against zero under divided government (Papamichalis, Ryu, and Wilson, 2025). Their growth claim is larger than the presidential gap that the last essay examined: “Throughout 121 years of U.S. history, the average GDP growth gap is more salient under the government cycle (4.51% for United vs. 2.01% for Divided; 2.5% government gap)” than under the party of the president.

They also isolate something the earlier work missed. The Republican underperformance in the stock market is not a Republican effect at all but a divided-Republican effect: unified Republican presidents earned returns comparable to Democrats, and the paper’s own name for it is the “Divided-Republican government puzzle.” Their causal identification uses close elections, where control was nearly a coin flip, and they replicate the pattern at the state level and in the United Kingdom.

It is the best statement of the “unified government is better” case, and it has three problems.

The first is that it is a conference paper, presented at ASSA, the EFA, and the FMA but not yet published in a journal. That is not disqualifying, and the work is careful, but it has not been through the review that turns a strong draft into a settled finding. The second is that its 121-year sample spans the same boundary mine does, and the authors are aware of it: their own robustness work notes that the prewar era is where growth was “extremely volatile,” with a standard deviation of quarterly growth near 12 percent, and that stripping out NBER recessions widens their government gap to 3.6 points rather than shrinking it. That is the opposite of what I compute. The third is that I could not reproduce their market result at all: using a public share-price index from the OECD, annual returns from 1959 to 2025 ran higher under divided government, 12.84 percent against 6.42 percent, a difference that is not statistically significant but points the other way.

I want to be careful about that third point rather than score it. Their series is CRSP, which is the professional standard and covers 1927 onward; mine is an index average, which is a crude proxy and annual rather than monthly. A negative result from a worse instrument is weak evidence. What I can say is that the claim does not survive an independent attempt to check it with public data, that its effect is concentrated where the data are worst, and that a finding which reverses the received wisdom should be held to a higher standard than one that confirms it.

VI. What divided government actually changes

Here the literature is clearer, and the finding is more interesting than either folk theory.

The reference work is David Mayhew’s Divided We Govern (1991), which inventoried landmark legislation from 1947 to 1990 and found nothing: “unified as opposed to divided control has not made an important difference in recent times” (Kirkland and Phillips, 2017, quoting Mayhew, 2005 [1991], 4). That null result became the standard citation for the proposition that divided government does not matter.

It was then partly overturned, in a way that cuts against both theories. Howell, Adler, Cameron, and Riemann re-examined the data and found that Mayhew’s null was an artifact of combining series that behave differently. Measured properly, “periods of divided government depress the production of landmark legislation by about 30%,” with no substantive effect on important-but-not-landmark laws, and “actually has a positive effect on the passage of trivial laws” (Howell et al., 2000).

That is the sharpest thing in this literature, and it is a finding that neither side advertises. Divided government does not stop legislating. It changes what gets legislated: fewer landmark statutes, more small ones, and a shift in composition toward the local and the uncontroversial rather than the national and the hard. Later work found the same shape, that legislators respond to gridlock by pivoting to district-specific bills, and Ansolabehere, Palmer, and Schneer’s comprehensive database of significant acts from 1789 to 2010 found unified control adds roughly one additional significant law per Congress in the nineteenth century and four in the twentieth, while cautioning that party control cannot explain the broad historical trend (Ansolabehere, Palmer, and Schneer, 2018).

Sarah Binder’s measure tells the same story with more precision. She scored gridlock as the share of salient issues left unresolved, using New York Times editorials to define salience, and found that deadlock “was higher in periods of divided, rather than unified, party control.” But the effect has been shrinking under her own measurement: in her later series it “misses standard levels of statistical significance (one-tailed test, p = .07),” and the main drivers turn out to be polarization and House–Senate divergence rather than which party holds what (Binder, 2014). Her most gridlocked Congress of the postwar era is the 112th, which was divided, at 71 percent of salient issues unresolved.

VII. The fiscal argument, tested

The claim that divided government restrains spending is the one most often made and least often tested at the federal level. The best evidence is at the state level, where balanced-budget rules make the test cleaner.

That evidence is real. Alt and Lowry found that “divided government is less able to react to revenue shocks that lead to budget deficits, particularly where different parties control each chamber of the legislature” (Alt and Lowry, 1994). A later regression-discontinuity study of state budgets found that where control was as-if random, divided government raised the probability of a late budget by 12 to 17 percent.

Read those two together and the pattern is not restraint. It is delay. A divided government facing a shortfall does not cut harder or spend less; it fails to respond on time, and the deadline slips. That is a real cost, and it is a cost measured in legislative performance rather than in the deficit. My own numbers say the same: the deficit under unified and divided government differs by three hundredths of a percentage point over seventy-eight years, which is noise.

VIII. What is actually true

Here is the honest state of it.

For the economy, the composition of government barely matters, and the apparent effect is a party effect wearing a composition effect’s clothes. Splitting unified government in two shows the advantage belongs to one of its forms only; the independent effect of unified control is minus twenty-five hundredths of a point and insignificant. The rest is an interaction with the president’s party, and that interaction does not survive the modern window: since 1981 all three states read within two-tenths of a point of one another. Deficits and spending behave the same way. This is not a claim that policy does not matter. It is a claim that which party holds which chamber is a weak instrument for policy.

The strongest contrary finding is real work but unproven. Papamichalis and his coauthors have produced the best case for unified government, with a genuine identification strategy and international replication. It is also unpublished, its effect is concentrated in the era where the data are worst, and my own check with public data did not reproduce it. Treat it as a serious finding awaiting review, not as a settled one.

What divided government reliably changes is legislation, and in a specific direction. Fewer landmark laws, more trivial ones, a composition tilted toward the local and the easy, and a budget process that runs late when a shortfall arrives. Binder’s own trend shows the effect attenuating as polarization rises, which suggests that in the current era what stops Congress is the distance between the parties rather than the fact that they are in different branches.

And the folk theories are both wrong in the same direction. “Unified governments act” is wrong because a trifecta does not repeal the filibuster, and the 107th Congress held one for months without much of a record. “Gridlock is good” is wrong because divided government does not produce stasis, it produces a different kind of law, and the trivial statutes it passes in greater numbers are not obviously the ones anyone wanted. Alesina and Rosenthal’s moderation story is the most interesting of the three, and it is a claim about what voters intend rather than about what the economy delivers.

An arch holds because the stones on either side of the crown are cut true and bear the load down through the springing. The keystone is what locks it, and an arch with a keystone slightly out of place still stands. It simply does not lock, and the light comes through the joint. The question for November is not whether the economy will notice. It almost certainly will not. It is which stones the next two years will set.

The three-way split is the part worth keeping, because it corrects how the question is usually asked. “Unified government” is two different countries, and only one of them has ever shown the economic advantage attributed to the whole category. The other has shown none at all, and since 1981 neither has shown anything distinguishable from noise. An argument for a trifecta is an argument about what a party will do with the power, not about what the economy will do with the party. Those are different claims, and only the first survives contact with the record.

That is the arch and the keystone together. The load-bearing structure is the economy, and it does not care which party cut the stones. What the keystone decides is whether the structure locks or lets the daylight through. Divided government does not fall down. It admits more light than its builders intended, which is either the defect or the point, depending on the purpose the building was meant to serve.

PRH | huffmanwrites.org | © Philip Huffman

Sources

• Alesina, A., & Rosenthal, H. (1995). Partisan Politics, Divided Government, and the Economy. Cambridge University Press.
• Alt, J. E., & Lowry, R. C. (1994). Divided Government, Fiscal Institutions, and Budget Deficits: Evidence from the States. American Political Science Review, 88(4), 811–828.
• Ansolabehere, S., Palmer, M., & Schneer, B. (2018). Divided Government and Significant Legislation: A History of Congress from 1789 to 2010. Social Science History, 42(1), 81–108.
• Binder, S. A. (2014). Polarized We Govern? Brookings Institution, Center for Effective Public Management.
• Federal Reserve Bank of St. Louis. (2026). Real Gross Domestic Product [GDPC1 and GDPCA; data from the U.S. Bureau of Economic Analysis]. Retrieved September 28, 2026.
• Howell, W., Adler, S., Cameron, C., & Riemann, C. (2000). Divided Government and the Legislative Productivity of Congress, 1945–94. Legislative Studies Quarterly, 25(2), 285–312.
• Kirkland, P. A., & Phillips, J. H. (2017). Is Divided Government a Cause of Legislative Delay? Columbia University and Vanderbilt University working paper.
• Mayhew, D. R. (2005). Divided We Govern: Party Control, Lawmaking, and Investigations, 1946–2002. Yale University Press.
• Niskanen, W. A. (2003, March). A Case for Divided Government. Cato Policy Report. (Advocacy publication, not peer-reviewed.)
• Papamichalis, T., Ryu, D., & Wilson, M. (2025). Divided Government and the Stock Market. AEA/ASSA Annual Meeting conference paper.
• U.S. House of Representatives, History, Art & Archives. (2025). Party Government Since 1857.