Simulation study S75 · Findings
Were earlier US policy eras steadier?
Four eras of US economic policy since the Second World War, measured two ways: what the published record shows, and what the Compassionism Simulation finds when today's households live through each era's settings, with the programme and without it.
The record is mixed. On no measure was 2005-2025 clearly less stable than all three earlier postwar eras, and 1973-1983 was the unstable era. In the model, today's households under an earlier era's settings see smaller falls in wealth but larger price swings and more basket poverty. Under every era's settings, the programme's effect keeps its direction, with one unfavourable exception: in 1973-1983 the C2.2 design slows recovery from an income shock slightly.
Corrected (Nov 16, 2026). The model figures on this page were first run without two of the simulation's release-wide settings, which the frozen main row includes. They have been rerun with them on the same eras, seeds, designs and settings (preregistration Amendment 5, written before the rerun; results s75-eras-e5649ef). One finding changed: in 1973-1983 the C2.2 design now slows recovery from an income shock slightly (by 0.07-0.17 years), where the first run had it speeding recovery, so the programme's effect no longer keeps its direction in every cell. The figures that changed: cells kept 227 of 240 (was 226), with 1 showing no measurable difference (was 8) and 12 reversals, 6 favourable and 6 not (was 6, all favourable); C2.2's cut in basket poverty 40-51 points (was 41-52), the release design's 22-30 (was 23-32); the first-11-years falls in real wealth $22,682, $7,646 and $18,726 smaller (was $21,329, $6,243 and $17,680); the rule's effect on the largest fall in 1973-2025 $13,000-20,000 (was $12,000-19,000); today's rule in 1947-1972, a $273,000 larger fall than in 2008-2025 (was $261,000) and $286,000 more than the era's path alone (was $272,000); basket poverty without the programme in 1984-2007 48.7% (was 48.6%); the cells in the settings table and the charts' figures, a little. The 8 · 6 · 1 tally and every comparison of an earlier era's path with 2008-2025 are unchanged. The published record (the descriptive panel) does not use the model and is not affected.
The question
Duke Johnson's view, put to the test: 2005-2025 was less stable than the decades before it, and earlier eras had policy settings worth comparing, among them far higher top tax rates in the 1950s, and health care financed mostly through non-profit hospitals and plans before managed care grew after the HMO Act of 1973.
The test was written down before any figure was computed (preregistration). The eras were fixed in advance:
2005-2025, the window of Duke's view, is shown beside them. 1913-1946 is shown as a reading only. "More stable" or "less stable" is said only where a difference passes the preregistered test: a block-bootstrap 95% interval that excludes zero after Holm's correction across the measures compared. The eras are short (11 to 26 years), so most differences do not pass, and "no clear difference" is often the honest reading.
Eighty years at a glance
The yearly path of each era, as the simulation reads it. The high inflation and unemployment of 1973-1983 stand out. So does the zero policy rate of 2009-2015 and 2020-2021, and the home price fall of 2007-2011.
Inflation, unemployment, the policy rate, and real home and share prices, 1946-2025
One panel per measure, each on its own scale. Shaded columns are years with at least one NBER recession month. Hover or tap for a year's values.
Basis: external sources (opened): BLS, FRED, Shiller, NBER, as read into sources/s75_hx.json.
Show the numbers
Was 2005-2025 less stable? The published record
Fourteen measures compared, with no simulation involved. On 5 of them the record goes against the view: 2005-2025 had lower inflation and lower food, shelter and medical inflation than 1973-1983, steadier shelter prices, and slower medical inflation than 1984-2007 too. On the other 9, no comparison passes the test either way. None supports it.
Three measures lean Duke's way without passing the test: unemployment swung more from year to year in 2005-2025 than in 1947-1972 or 1984-2007, real home prices swung more than in 1947-1972, and food prices swung more than in 1984-2007. Each interval excludes zero before the correction (marked † below), but none does after it.
2005-2025 against each earlier era, measure by measure
Each cell reads "2005-2025 was ... than that era". The figures show the difference (2005-2025 minus the earlier era, in points) and its 95% interval.
Basis: derived calculation (file): sources/s75_eras.py → s75_eras.json; results folder s75-panel-54e88bc.
How much each measure swung from year to year, by era
The standard deviation of the yearly values, in points. Longer bars mean bigger swings. 2005-2025 is highlighted.
Basis: derived calculation (file): sources/s75_eras.json (results s75-panel-54e88bc).
Show the numbers
Who paid, who owned, how rates were set
Taxes actually paid were far more progressive in the 1950s
The 91-92% top statutory rate of the 1950s applied to very few incomes. The rates below are what each group actually paid in all taxes as a share of its pre-tax income, from Piketty, Saez and Zucman's distributional accounts. Over 1950-1959 the top 0.1% paid 57.3% and the top 1% 45.5%; by 2008-2019 they paid 34.8% and 32.9%, while the bottom half paid more (about 21.6% in 1947-1972, 25.2% in 2008-2019).
Average tax rate, all taxes, by income group and era (% of pre-tax income)
Basis: external source (opened): Piketty, Saez and Zucman (2018), 2022 appendix, Table G1; derived in sources/s75_eras.json.
Show the numbers
Distribution moved the other way from stability. The top 1%'s share of pre-tax income fell in 1947-1972 (14.9% to 11.0%), held flat in 1973-1983, rose in 1984-2007 (12.2% to 18.4%) and stayed high (19.1% in 2019). Poverty fell in three of the four eras and rose in 1973-1983.
Basis: external sources (opened): PSZ (income and wealth shares), Census (official poverty), Columbia's anchored SPM, CMS national health expenditure; derived in sources/s75_eras.json.
Who owned the hospitals
For-profit hospitals existed before 1973, but they were small: in 1946-1950 about one US hospital in five was proprietary, holding under 4% of beds. For-profit care grew after 1973. Its share of beds roughly doubled in 1973-1983 (4.8% to 8.2%), and its share of community hospital admissions doubled from 1975 to 2015. Non-profits carried about 71-74% of admissions throughout. The rise follows the HMO Act (1973), which promoted HMOs and opened the way for for-profit ones; it did not make health profit legal, and this panel does not test what caused the rise.
Community hospital admissions by ownership, 1975-2015 (% of admissions)
Basis: external source (opened): Health, United States 2017, Table 82; Historical Statistics of the United States, Bd144-157 (beds, in the table); Social Security Bulletin, November 1973.
Show the numbers
How the central bank set its rate
A simple fitted rule for each era: how many points the policy rate rose per point of inflation. Before 1984 the response was below one for one, which lets the real rate fall as inflation rises. In 1984-2007 it was 1.3. In 2008-2025 it was about zero: the rate sat at the floor through low inflation and rose only in 2022, after inflation had already climbed. These are descriptive fits, not the rules the Federal Reserve followed.
The policy rate's response to one point of inflation, by era (with 95% interval)
Basis: derived calculation (file): sources/s75_eras.json, B5 (FRED, CBO).
Show the numbers
Today's households under each era's settings
The second part asks a model question: take today's households (the simulation's 500 adults, based on 2022) and let them live through each era's measured settings. This is a model counterfactual, not a reconstruction of the era. The population, its jobs and homes, the prices of the basket of essentials and today's institutions stay as they are, apart from the settings changed:
- The era's path: its inflation, real home and share price changes, interest rates, recession years and the fall in family income in them.
- The era's central-bank rule as fitted above, or today's rule in every era.
- The era's tax schedule: each income group's tax rate actually paid in that era minus 2019's, charged on earnings and Social Security, with an equal amount per adult returned so the size of the state does not change.
- The era's health cost growth: the basket's medical prices rise faster than the rest at the era's pace.
Five stability outcomes, lower is steadier on all five: P1 the swings in the basket's price; P2 the swings in yearly real income; P3 the largest fall in real wealth (the median adult's); P4 the years to recover from an income shock; P5 the share of adults pushed below the basket line in a recession year. Each run uses 100 seeds; intervals are Holm-adjusted across the five.
Without the programme, today's households are not steadier in every way under an earlier era's path. The earlier eras give smaller falls in real wealth and smaller swings in yearly income. But the basket's price swings more in 1947-1972 and 1973-1983, recovery from an income shock is slower in 1947-1972 and 1984-2007, and more people are pushed below the basket line in the recession years of 1973-1983 and 1984-2007. Of the 15 cells, 8 are steadier, 6 less steady and 1 shows no clear difference. Mixed, as the record is.
Each earlier era's path against 2008-2025, without the programme
The era minus 2008-2025, with its Holm-adjusted 95% interval. Left of zero: the earlier era is steadier. Right of zero: less steady.
Basis: simulation release and command: engine commit e5649ef, node dev/tools/s75_eras.js e1…e4 100; results folder s75-eras-e5649ef.
Show the numbers
Which setting moves it, inside the model
- The central bank's rule moves the largest falls in real wealth most, by $13,000-20,000 in 1973-2025: easing raises asset prices in a recession, and the later tightening lowers them.
- The era's tax schedule barely moves stability, and lowers basket poverty a little in every era (0.1-0.4 points). Once the equal amount per adult is returned, each era's change takes a little more from higher earnings than from lower ones.
- The era's health cost growth raises basket poverty (0.1-1.1 points), and in 1947-2007 the share pushed below the line in recession years.
- No setting moves the basket's price swings (P1): the rule has no channel to essential prices.
No programme; each setting minus the era's path alone, in the same era. Only cells whose interval excludes zero are named. P3 in dollars, poverty in points. Basis: simulation release and command (as above); every cell is in tables.md.
Today's rule put into 1947 is not a fair test. Set against the Federal Reserve's pegged rates after the war, today's rule tightens by up to 28 points in 1947, reading 1946's 18% inflation at the end of price controls. The engine's links from the rate to asset prices were estimated for moves of a point or two, and stretched this far they produce a $273,000 larger fall in real wealth than in 2008-2025. That cell says more about the model's reach than about the era.
Basket poverty without the programme is higher under every earlier era's path: 47.2% of person-years in 1947-1972, 57.7% in 1973-1983 and 48.7% in 1984-2007, against 44.4% in 2008-2025. This comes from the model's paths (their recession years, the fall in family income in them, and inflation applied to today's basket). It is not a measure of how people lived then.
The programme's effect under every era's settings
Both designs were run against each era: C2.2, which pays every adult enough to reach the cost of the basket of essentials, and the release design, the simulation's first released design, which adults join. Each is paired with a no-programme run on the same seeds.
Both designs cut basket poverty in every era, shrink the largest fall in real wealth, and push fewer people below the line in recession years. Both also make the basket's price and incomes in dollars swing more, as in the simulation's study of the central bank (S74). The release design lengthens recovery from an income shock in every cell; C2.2 shortens it, except in 1973-1983, where it lengthens it slightly.
Basket poverty under each era's path: no programme, release design, C2.2
% of person-years with income below the cost of the basket of essentials, means over 100 seeds.
Basis: simulation release and command: engine commit e5649ef; results s75-eras-e5649ef.
Show the numbers
Does the programme's effect keep its direction? 120 cells per design
Every era (4) × setting (6) × outcome (5), each against the direction found in S74. "Kept" means the same direction, good or bad.
What this does not show
- The panel describes; it does not attribute. The eras differ in many ways at once: the US's postwar position, union membership, demography, fixed exchange rates to 1971, wars, the size of government, the rules of finance. No comparison here says which policy caused which outcome.
- The tax setting covers work and pensions only. It is charged on earnings and Social Security, not on the capital income wealth pays. Much of the top 1%'s higher tax in 1947-1972 fell on capital (corporate, estate and property taxes), and that part is not in the model.
- The engine has no expectations channel. A rule that responds to inflation less than one for one (1947-1983) or not at all (2008-2025) cannot show the instability it is thought to cause.
- The rate's links to asset prices are stretched wherever a rule parts from the actual rate by more than about 5 points (today's rule in 1947-1983; the era's own rule in 1947-1972). Those links were estimated for moves of a point or two.
- The model does not represent war, military spending or corruption, so no figure on this page is about them.
- Era samples are short. The simulation uses one population and 100 seeds; its intervals measure the model's randomness, not uncertainty about history. Series that do not reach back are replaced by named substitutes, each a modelling assumption listed in
sources/s75_hx.json. - The model has not yet been reviewed by an independent economist.
The aim of the design
2005-2025 was marked by war and corruption. Compassionism is meant as a way past war, with stability and prosperity as goals instead of exploitation and extraction.
Sources, files and how to check them
- Preregistration, with Amendments 1-5: prereg/2026-11-s75-deep-history.md
- Full report: dev/reports/s75-deep-history.md · the panel's note: dev/reports/s75-panel.md
- Every cell: panel tables · engine tables (the first run's, superseded: tables)
- Results, with run.txt and SHA256SUMS: s75-panel-54e88bc · s75-eras-e5649ef (supersedes s75-eras-77a85d6)
- Data sources opened: BLS CPI-U and its food, shelter and medical indexes; FRED (federal funds rate, Treasury bill, mortgage rate, unemployment, CBO's output gap and noncyclical rate); Shiller's home and share prices; NBER recession dates; Piketty, Saez and Zucman's distributional accounts; the Historical SCF (Kuhn, Schularick and Steins); Jordà, Schularick and Taylor's rates of return; CMS national health expenditure; Census poverty tables; Historical Statistics of the United States; Health, United States 2017.
- The data behind this page, in one file: data/s75-us-eras.json, built by
tools/build_s75_eras.pyfrom the simulation's files. - The simulation itself: Compassionism Simulation · its findings page
The math and code of this simulation were engineered by Claude, an AI model made by Anthropic, from the concepts in Duke Johnson's book Better To Best and his related vision for eradicating extreme poverty while enriching cultures and supporting human flourishing. The model has not yet been reviewed by an independent economist; the full code is open for anyone to check, and expert collaborators are welcome.