Study III · Measuring the Economy

The Same Question, Different Numbers

Four credible estimates give four answers to one question, and in every case the gap has a documented cause.

5 min read · 5 exhibits

Four credible estimates of what the current tariffs cost an American household: $800, $840, $1,000 and up to $1,500. Ask what federal regulation costs each year and you get $2.15 trillion from the Competitive Enterprise Institute and $9.6–11.9 billion from OMB (2001 dollars) — more than a hundredfold. It is tempting to read this as evidence that everyone is cooking numbers. That is almost never what is happening.

In every case examined here, the divergence has a specific, documented, methodological cause, and in most cases the institutions have written the cause down. The numbers differ because they answer different questions, use different baselines, or make one named assumption that the other rejects. The useful skill is not picking a side. It is being able to say, in one sentence, why the two figures differ.

Case one: the tariff figures measure three different things

Exhibit 1

Six published answers to “what do tariffs cost a household?”

Source. Tax Foundation, Trump Tariffs Tracker, last updated September 2, 2026; The Budget Lab at Yale, The State of U.S. Tariffs, editions of February 20, April 8 and August 24, 2026.

The Tax Foundation’s tracker, updated September 10, 2026, now gives $820.

Full size

Method. Figures as published, not adjusted to a common basis. They are not rival estimates of one quantity. The Tax Foundation figures are a tax incidence calculation — customs duties collected, distributed across households. The Budget Lab figures are a welfare calculation — the equivalent loss of income. The two Budget Lab April figures are one estimate under two policy scenarios. The February figure follows the Supreme Court’s ruling vacating the IEEPA tariffs, which cut the effective rate overnight. The Tax Foundation page does not state a dollar-year basis; the Budget Lab’s February and April estimates are in 2025 dollars.

Charted on one axis these look like a disagreement. They are closer to a category error in the reading. A duty collected and a welfare loss are different quantities, and the honest presentation labels the axis twice, or draws two charts. There is an external check available: Federal Reserve staff estimated in April 2026 that tariffs had raised core PCE prices by 0.8 percent through February — close to the Budget Lab’s price-level estimate of about 0.7 percent, and hard to reconcile with the Council of Economic Advisers’ July 2025 finding — “These findings contradict claims that tariffs or tariff-fears would lead to an acceleration of inflation” — which conceded in its own text that it did not identify a counterfactual.

Case two: they agreed on the cost and split on the growth

The 2017 tax law is the cleanest case in the file, because the two institutions scoring it agreed almost exactly on one number and diverged by a factor of more than two on the next.

Exhibit 2

The 2017 tax law: agreement, then divergence

Source. Joint Committee on Taxation, JCX-69-17, December 22, 2017; Tax Foundation, Preliminary Details and Analysis of the Tax Cuts and Jobs Act, December 18, 2017.

Full size

Method. Ten-year revenue cost in billions of dollars, as published. The conventional scores are $14bn apart on a $1.46 trillion estimate — agreement to within one percent. The dynamic scores are $623bn apart. All of the disagreement is in the growth feedback, and it has a named cause: the Tax Foundation’s model treats the United States as a small open economy (Tax Foundation, “Responding to a Critique of the Taxes and Growth Model,” December 5, 2017) in which global capital is highly elastic, so deficit-financed cuts do not meaningfully crowd out investment, while JCT weights three models (JCX-69-17) that, in the Tax Foundation’s account, incorporate crowd-out. A chart placing them adjacent without saying so would mislead.

What happened next is the part worth knowing, because it is a model of how this should go. In November 2017 Greg Leiserson at the Washington Center for Equitable Growth published a specific critique alleging two modeling errors in the Tax Foundation’s score. The Tax Foundation responded the same day, conceded “an erroneous term relating to the interaction between federal and state taxes,” corrected the model, and revised its House-bill estimate from 3.9 percent to about 3.6. On the second charge — its treatment of the estate tax — it conceded nothing and, in a post of November 21, 2017, defended the approach as a matter on which “economists can have honest differences.”

That is a correction done well: same-day, specific, partial, with the model actually changed. One thing remains unresolved, and we flag it rather than smooth it: Leiserson argued the corrections should have moved the headline from 3.9 to about 1.9; the Tax Foundation moved it to 3.6. We found no publication reconciling a two-point claim with a three-tenths correction.

Case three: a stock and a flow, reported as rivals

Exhibit 3

What does federal regulation cost per year?

Figure	What it is	Annual cost, low, $bn	Annual cost, high, $bn
$2,153bn	Competitive Enterprise Institute, Ten Thousand Commandments 2026, April 9, 2026. The accumulated burden of the entire regulatory state, costs only, benefits not netted.	2153	2153
$9.6–11.9bn	OMB, Report to Congress on the Benefits and Costs of Federal Regulations, FY2023. Annualized cost of new major rules in one fiscal year, against $30.7–49.0bn in benefits.	9.6	11.9
More than a hundredfold	The ratio between them — which is not a measure of disagreement, but of the fact that one is a stock and the other is a flow.

Source. Clyde Wayne Crews Jr., Ten Thousand Commandments 2026, Competitive Enterprise Institute, April 9, 2026; Office of Management and Budget, FY2023 benefit-cost report, January 2025 (2001 dollars).

Full size

Method. Shown as figures rather than as bars because a bar chart spanning two orders of magnitude renders the smaller value invisible and implies a comparison that is not valid. OMB annualizes the cost of new major rules in a single fiscal year, covering only the nineteen rules where agencies monetized both costs and benefits (“There were 19 such rules in FY 2023”) — 19 of the 58 FY2023 executive branch rules the report counts, or 33 percent. CEI estimates the accumulated burden of all federal regulation and does not net benefits. CEI is unusually candid about the construction, describing its total as a “conservative placeholder” intended to push Congress toward an official estimate, and stating that “total regulatory costs are immeasurable.” Each is defensible for what it measures; they are not rival answers to one question. On a like basis — OMB’s own 2022-dollar equivalent of $15–19bn a year against CEI’s $2,153bn — the ratio is about 113 to 144 (2,153 ÷ 19 and 2,153 ÷ 15).

Case four: the window decided the sign

In January 2024 the Groundwork Collaborative published a finding that corporate profits “drove 53 percent of inflation” during the second and third quarters of 2023, against 11 percent “over the 40 years prior to the pandemic.” It traveled widely. A week later Ryan Bourne at Cato published a critique with three parts, of which one is devastating and checkable.

Exhibit 4

Source. Pancotti and Owens, Inflation Revelation, Groundwork Collaborative, January 18, 2024; Ryan Bourne, “New Nonsense on Profit-Driven Inflation,” Cato Institute, January 25, 2024, reporting a calculation by Chris Conlon.

Full size

Method. Share of price growth attributed to corporate profits in a unit-cost decomposition, percent. Extending the window back three quarters flips the sign from +53 to −22, with labor costs then accounting for more than all of the price increase. Bourne’s further objection is that over the original two-quarter window prices barely moved, so the calculation measures the extent to which a very small price change is accounted for by a very small profit change. A third objection is categorical: a unit-cost decomposition is an accounting identity and cannot establish that profits drove anything.

A number that changes sign when you move the start date by three quarters is not a finding about the economy. It is a finding about the window. This is the single cheapest test any reader can apply to a decomposition, and it costs nothing: ask what happens if the period moves.

The one measure that is not a matter of opinion

Method disputes are hard to adjudicate from outside. Whether an institution lets you check its method is not. We looked at whether each publishes its models.

Exhibit 5

Who lets you check the model

Institution	Public code	What is there
Congressional Budget Office	33 public repositories	Tax, premium-growth, legal-status, budgetary-feedback and other models in Python, Stata and R.
Policy Simulation Library	44 public repositories	Tax-Calculator, OG-Core, OG-USA, Cost-of-Capital-Calculator.
The Budget Lab at Yale	29 public repositories	Tax-Simulator, Tariff-Model, tariff-rate-tracker (MIT license), and a versioned archive of chart specifications.
Tax Foundation	69 public repositories — but not the model	Datasets, publication data and interactives are public. We found no repository containing the Taxes and Growth macroeconomic model, which is the thing in dispute in Exhibit 2.
Joint Committee on Taxation	Documentation only	Publishes methodology overviews (The Joint Committee on Taxation Revenue Estimating Process, February 2019) and states its model weights in its analysis of the 2017 tax law (JCX-69-17). No public code located.
Heritage, CEI, CIS, Groundwork	None located	No public model or code repository of their own found.

Source. GitHub organization pages inspected directly, September 2026; repository counts as of October 5, 2026; Joint Committee on Taxation, The Joint Committee on Taxation Revenue Estimating Process, February 2019, and JCX-69-17, December 22, 2017.

Full size

Method. “None located” is not “none exists.” This records what a determined search found in September 2026, and any institution listed here can point to a repository we missed. Repository counts include everything in the organization, not only models — which is why the Tax Foundation row is the interesting one: a high count, and not the model. A table rather than a bar chart because the counts are not comparable quantities and the caveat cannot be drawn.

Two figures that differ by more than a hundredfold are not necessarily evidence that one party is lying. They are usually evidence that somebody asked two questions and reported one answer.

Four questions that settle most of these

What exactly is being counted? A duty collected and a welfare loss are different quantities. A stock and an annual flow are different quantities. Most of the spread in this study dissolves at this question.

Over what window, and what happens if it moves? The profit-share figure changes sign on a three-quarter shift. Any decomposition should be asked this before it is repeated.

What is the baseline? A projection under current law and a projection assuming an agenda passes and works are not competing forecasts. They are answers to different conditionals, and the gap between them is the assumption, not the analysis.

Can you see the model? The least ideological question available, and the one most rarely asked. It does not tell you who is right. It tells you who can be checked.

What we could not verify

Whether the Groundwork Collaborative responded to the window critique. We found no such publication. That is not evidence that none exists, and it is not presented as such.
The reconciliation of 3.6 and 1.9. No Tax Foundation publication was found addressing why its correction moved the estimate three-tenths of a point where the critic argued it should move two points.
The Tax Foundation’s dollar-year basis for the $1,000 and $840 household figures is not stated on the tracker page, so those figures are not deflated to match the 2025 dollars of the Budget Lab’s February and April estimates.
The Council of Economic Advisers’ 2026 position on tariffs. Chapter 3 of the 2026 Economic Report of the President covers trade policy; we could not extract its price figures and have therefore cited only the July 2025 analysis.
A comparison of immigration fiscal estimates was examined and excluded: the two studies we would have charted were published a year apart, and the later one changed several of the assumptions the earlier critique disputed. They are not a like-for-like pair and charting them as one would have been the error this study is about.

Sources

Corrections

The four tariff figures are four credible estimates from two institutions, not four credible institutions.
In Exhibit 1, $940 and $1,500 are the tops of published ranges ($760–940 and $1,200–1,500), and only two of the estimates are stated in 2025 dollars.
Both revenue scores in Exhibit 2 are ten-year; the “long run” against “ten-year window” explanation is withdrawn.
The two regulatory-cost figures differ more than a hundredfold, about 113 to 144 times on a like basis, not two hundred, and the nineteen rules are 19 of 58, or 33 percent.
OMB’s figure is $9.6–11.9 billion in 2001 dollars, as published, not “about $11 billion”.
In case two the dynamic scores differ by a factor of more than two (1,071 ÷ 448 = 2.4), not “a factor of two”.
In case four the start of the window moves three quarters, not four, and the 11 percent is “over the 40 years prior to the pandemic”.
JCT’s model weights are stated for the 2017 analysis, not for “each” analysis.

Data for every exhibit: download (CSV)

Cite this study

Higgins Strategy. “The Same Question, Different Numbers” Study III, Measuring the Economy. October 12, 2026. https://higginsstrategy.com/studies/the-same-question-different-numbers/

@techreport{higginsstrategy2026thesamequestiondifferentnumbers,
  author      = {{Higgins Strategy}},
  title       = {{The Same Question, Different Numbers}},
  institution = {Higgins Strategy},
  type        = {Study},
  number      = {III},
  series      = {Measuring the Economy},
  year        = {2026},
  url         = {https://higginsstrategy.com/studies/the-same-question-different-numbers/},
  month       = oct
}

The Same Question, Different Numbers

Case one: the tariff figures measure three different things

Case two: they agreed on the cost and split on the growth

Case three: a stock and a flow, reported as rivals

Case four: the window decided the sign

The one measure that is not a matter of opinion

Four questions that settle most of these

What we could not verify

Sources

Read next