Atlas › The world in numbers

Case study 42 of 198

Trolley problems in 42 countries: does Jev know who pushes?

For the Switch, Loop and Footbridge dilemmas answered by 70,000 people in 42 countries, does Jev know how many people in each country would sacrifice one to save five, and how does its own answer compare?

result129 questions

For three classic trolley dilemmas answered in 42 countries, Jev knows the universal pattern (people everywhere accept switching the track more than pushing a man) but not the variation: its country estimates barely track the real ones (rank correlation 0.30 for the switch, -0.30 for the push). It underestimates how many people would push the man by 28 points, and asked itself, it would push with 20% probability, where 55% of the visitors said they would.

0255075100

France · SwitchSwitch

Czech Republic · SwitchSwitch

Portugal · SwitchSwitch

Hungary · SwitchSwitch

Argentina · SwitchSwitch

Mexico · SwitchSwitch

Belgium · SwitchSwitch

Israel · SwitchSwitch

Colombia · SwitchSwitch

Ireland · SwitchSwitch

Romania · SwitchSwitch

Italy · SwitchSwitch

Spain · SwitchSwitch

United States · SwitchSwitch

Finland · SwitchSwitch

Turkey · SwitchSwitch

Brazil · SwitchSwitch

Canada · SwitchSwitch

Sweden · SwitchSwitch

Netherlands · SwitchSwitch

Germany · SwitchSwitch

United Kingdom · SwitchSwitch

Viet Nam · SwitchSwitch

Norway · SwitchSwitch

India · SwitchSwitch

Ukraine · SwitchSwitch

Slovakia · SwitchSwitch

Poland · SwitchSwitch

Denmark · SwitchSwitch

Australia · SwitchSwitch

Russian Federation · SwitchSwitch

Greece · SwitchSwitch

Switzerland · SwitchSwitch

Austria · SwitchSwitch

Singapore · SwitchSwitch

New Zealand · SwitchSwitch

Belarus · SwitchSwitch

Hong Kong · SwitchSwitch

Japan · SwitchSwitch

Taiwan, Province of China · SwitchSwitch

Korea, Republic of · SwitchSwitch

China · SwitchSwitch

Hungary · LoopLoop

France · LoopLoop

Czech Republic · LoopLoop

Israel · LoopLoop

United States · LoopLoop

Belgium · LoopLoop

Spain · LoopLoop

Mexico · LoopLoop

Singapore · LoopLoop

Colombia · LoopLoop

Portugal · LoopLoop

Ireland · LoopLoop

Canada · LoopLoop

Australia · LoopLoop

United Kingdom · LoopLoop

Netherlands · LoopLoop

Germany · LoopLoop

India · LoopLoop

Romania · LoopLoop

Argentina · LoopLoop

Sweden · LoopLoop

Greece · LoopLoop

Switzerland · LoopLoop

New Zealand · LoopLoop

Russian Federation · LoopLoop

Ukraine · LoopLoop

Finland · LoopLoop

Norway · LoopLoop

Italy · LoopLoop

Austria · LoopLoop

Viet Nam · LoopLoop

Denmark · LoopLoop

Poland · LoopLoop

Brazil · LoopLoop

Slovakia · LoopLoop

Turkey · LoopLoop

Korea, Republic of · LoopLoop

Hong Kong · LoopLoop

Belarus · LoopLoop

Japan · LoopLoop

Taiwan, Province of China · LoopLoop

China · LoopLoop

Viet Nam · FootbridgeFootbridge

United States · FootbridgeFootbridge

Israel · FootbridgeFootbridge

United Kingdom · FootbridgeFootbridge

Brazil · FootbridgeFootbridge

Australia · FootbridgeFootbridge

Turkey · FootbridgeFootbridge

Czech Republic · FootbridgeFootbridge

Slovakia · FootbridgeFootbridge

Canada · FootbridgeFootbridge

Mexico · FootbridgeFootbridge

Portugal · FootbridgeFootbridge

Italy · FootbridgeFootbridge

Ireland · FootbridgeFootbridge

Sweden · FootbridgeFootbridge

Argentina · FootbridgeFootbridge

New Zealand · FootbridgeFootbridge

Hungary · FootbridgeFootbridge

Belgium · FootbridgeFootbridge

Romania · FootbridgeFootbridge

France · FootbridgeFootbridge

Poland · FootbridgeFootbridge

Spain · FootbridgeFootbridge

Norway · FootbridgeFootbridge

India · FootbridgeFootbridge

Finland · FootbridgeFootbridge

Colombia · FootbridgeFootbridge

Netherlands · FootbridgeFootbridge

Singapore · FootbridgeFootbridge

Germany · FootbridgeFootbridge

Denmark · FootbridgeFootbridge

Switzerland · FootbridgeFootbridge

Ukraine · FootbridgeFootbridge

Russian Federation · FootbridgeFootbridge

Greece · FootbridgeFootbridge

Hong Kong · FootbridgeFootbridge

Austria · FootbridgeFootbridge

Belarus · FootbridgeFootbridge

Korea, Republic of · FootbridgeFootbridge

Japan · FootbridgeFootbridge

Taiwan, Province of China · FootbridgeFootbridge

China · FootbridgeFootbridge

Jevpeopleshare who would sacrifice one (magenta Jev, ink observed)

How to read this: One row per country and dilemma, the switch first, then the loop, then the footbridge, each sorted by the share of visitors who would sacrifice the one (ink diamond), with Jev's estimate beside it (magenta square). In the footbridge rows Jev's squares sit far to the left of the diamonds.

126 country-dilemma estimates; mean error by dilemma: Switch -8 points, Loop -13 points, Footbridge -28 points.

In short

Jev knows the textbook trolley pattern, switching the track beats pushing a man, in all 42 countries, but not how countries differ.
It badly underrates the pushers: it guessed about 13% of American visitors would push the man, where 61% said they would.
Asked itself, Jev would push the man far less often than the visitors said they would, so its picture of moral intuition is more squeamish than theirs.

What the data shows

Drake No/Yes meme: push the man off the footbridge to stop the trolley; pull a lever so it hits one person instead

The order, yes: in every country, Jev's estimates put the switch above the loop above the push, as the real answers do.
The countries, no: its estimates barely follow the real differences: 0.30 for the switch, 0.20 for the loop, and -0.30 for the push, where it gets the ordering of countries slightly backwards.
It thinks people won't push: it underestimates the footbridge share by 28 points on average. In the United States 61% of visitors said they would push; Jev guessed about 13%.
And it wouldn't: asked itself, Jev would push with 20% probability; 55% of visitors said yes. On the switch and the loop it matches them (81% vs 82%, 76% vs 74%).

What it means, and what it doesn't

Jev knows the trolley problem the way a textbook does: the switch is fine, the push isn't. It doesn't know how people in different countries actually answer, and it assumes the push is far more taboo than it is for the people who took this test.

It doesn't mean those visitors represent their countries. It does mean Jev's picture of human moral intuition is more squeamish, and more uniform, than the answers of 70,000 real people.

Caveats

Not national samples. The answers come from self-selected visitors to the Moral Machine website, an English-first site, not national samples. The study's authors say so too; "the share in Japan" means the share of Japanese visitors to that site.
The wording. The dilemmas were written for this project, paraphrasing the website's classic trolley pages; visitors saw the site's own text and pictures.
Small spread between countries. Countries differ by roughly 25 to 35 points from lowest to highest on each dilemma, and a correlation across 42 countries is still noisy.
A textbook result. The switch-versus-push contrast is one of the most taught results in moral psychology, so getting the order right is expected knowledge, not a sign of insight into people.

Jev on this experiment

Would a person find it interesting to read?: Yes76%
Does it describe you?: No58%
Would you have predicted it?: Yes58%
How fair is the comparison?: The comparison is shaky
How much should a reader rely on it?: A little
Which caveat matters most?: Not national samples61%

Why ask this

The trolley problem comes in versions. Would you pull a lever to send a runaway trolley onto a side track, killing one worker instead of five? Most people say yes. Would you push a large man off a footbridge to stop it? In the classic lab studies most said no, though the arithmetic is the same.

In a study published in 2020, a team led by Edmond Awad reported how 70,000 people in 42 countries answered all three classic versions (the switch, a looped track, and the footbridge) on the Moral Machine website. The order was the same everywhere, but the levels differed: visitors from East Asian countries were less willing to sacrifice anyone in every version. Knowing the order is textbook knowledge; knowing the variation is knowing people.

How this was done

The people and the data

About 70,000 visitors to the Moral Machine website, an English-first site, answered the three dilemmas as part of the study. Nobody was sampled: people came to the site on their own, so "Japan" means Japanese visitors to that site, not Japan. The study's public data gives 230,475 answers; the 42 countries here are those with at least 200 answers to each dilemma, and each country's share who would sacrifice the one is the comparison for Jev.

What Jev was asked

Two kinds of questions. Per country and dilemma, the share of visitors who would sacrifice the one (in 5% steps). And the three dilemmas put to Jev itself:

A runaway trolley is heading down the track toward five workers, who will be killed if it goes on. You can pull a lever to switch the trolley onto a side track, where it will kill one worker instead of the five. Would you pull the lever?
Yes, I would pull the lever · No, I would not pull the lever

Jev said yes with 81%; of 72,982 visitors, 82% did. That's 129 questions in all, each asked with the options shuffled and averaged, and the three personal dilemmas also asked for what Jev thinks most people would say.

How it was measured

For each dilemma, how well Jev's country estimates order the countries (rank correlation: 1 same order, 0 no relation, negative means reversed) and how far off they are on average; whether Jev keeps the universal order (switch above loop above push) in every country; and Jev's own answer against the pooled share.

Where these questions live

129 questions across 2 topics of the map. Each opens on the map with every question in it.

Countries

Trolley Dilemmas by Country126

Values & Morality

Moral Foundations3

Every question

All 129 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 129