people reject unfairness
but normalize inequality

evidence from a large-scale experiment

mauricio bucca

puc chile

mario d. molina

nyu abu dhabi

September 3, 2025

the puzzle

the puzzle

• A large body of research shows people are deeply averse to unfairness and resistant to large economic gaps.

• Yet inequality keeps rising worldwide —often tolerated, rarely contested.

Why do people end up accepting the very inequalities they claim to oppose?

prevailing account

• inequality is tolerated when it originates from a fair process (Starman 2017, NHB)

equal opportunity

unequal outcomes

prevailing account

• inequality is tolerated when it originates from a fair process (Starman 2017, NHB)

unequal opportunity

unequal outcomes

• yet, opportunities are not equally distributes so, why is still inequality tolerated?

prevailing account

• limited awareness of systemic inequities

  • misinformation (McCall 2017, PNAS; Sands 2019, Nature)
  • biased belief systems reinforced by inequality itself—for example, through segregation (Newman 2015, AJPS; Mijs 2021, SR)

core assumption: people would reject inequality if they were fully aware of unfair origins.

a preview of what we find

evidence points to a dual-track process of fairness evaluation:

• People rely on moral judgment to assess procedural fairness.

• Their standards for acceptable disparities are anchored in the level of inequality they experience.

study design

challenges

• unfeasible to manipulate societal inequality in observational data.

• experiments often manipulate arbitrary inequalities, triggering perceptions of unfairness (Starman 2017, NHB).

• ➡️ we need controlled experiment that resembles a real social environment:

  • individual are endowed with (un)equal opportunity
  • outcomes plausibly depend on individual choice
  • (!) yet, the actual role of skill or effort must be controlled.

experiment overview

• pre-registered, interactive online card game (\(N = 3,335\))

• conducted on cloudresearch connect (US sample)

• fully crossed 2×2×2 design:

  • opportunity inequality: equal vs. opportunities details
  • outcome inequality: low vs. high payoff gap identification
  • social position: winner vs. loser

• several game features suggest skill matters, but outcomes are driven almost entirely by chance or rules tested

sample

scheme	inequality of opportunity	inequality of outcomes	social position	payment	n
1	equal	high ($5 gap)	loser	$5	210
1	equal	high ($5 gap)	winner	$10	199
2	equal	high ($5 gap)	loser	$2.5	258
2	equal	high ($5 gap)	winner	$7.5	200
3	unequal	high ($5 gap)	loser	$5	208
3	unequal	high ($5 gap)	winner	$10	203
4	unequal	high ($5 gap)	loser	$2.5	205
4	unequal	high ($5 gap)	winner	$7.5	202
5	equal	low ($2.5 gap)	loser	$5	222
5	equal	low ($2.5 gap)	winner	$7.5	210
6	equal	low ($2.5 gap)	loser	$7.5	198
6	equal	low ($2.5 gap)	winner	$10	200
7	unequal	low ($2.5 gap)	loser	$5	214
7	unequal	low ($2.5 gap)	winner	$7.5	200
8	unequal	low ($2.5 gap)	loser	$7.5	203
8	unequal	low ($2.5 gap)	winner	$10	203

experimental conditions and payment schemes

experimental platform

two-player, 5-round game

• each round, players receive 7 random tokens

• exchange rules before each round:

  • random exchange = equal opportunity
  • regressive exchange = unequal opportunity

• overall winner/loser = most rounds won

• payoffs: payoff gaps of $2.5, (low inequality) or $5 (high inequality).

dependent variables

• Fairness (1–7 slider)
  1 = completely unfair, 7 = completely fair

• attributions: ,ost/lear important factor
  • luck, talent, rules

• Fair Distribution
  “Keeping the total reward constant, what would be a fair distribution between winner and loser?”
  Slider initialized at actual distribution

  • Rank Preservation: did participants preserve winner > loser?
  • Fair Inequality: winner’s share in proposed distribution.
  • Jasso’s-like Justice: log ratio of observed vs. fair inequality.

findings

rejection of unfairness

• unequal opportunities produce a substantial drop in perceived fairness

• the size of the outcome gap has no detectable effect

• across all conditions, winners judge outcomes as fairer than losers

normalization of inequality

rank preservation

• across all conditions, a clear majority favor unequal reward distributions that preserve the original winner–loser ranking

fair inequality

• perceived “fair” inequality mirrors participants’ exposure

• those in high-inequality conditions endorse larger gaps than those in low-inequality conditions

• holds regardless of opportunity distribution or personal outcome

• winners see more inequality as fair

experienced vs fair inequality (Jasso’s justice)

• losers saw actual inequality as slighlty greater than what they considered fair

• winner in high-inequality viewed observed gaps as fair

• winners in low-inequality felt underpaid and favored larger winner shares

causes of inequality

• winners credited talent — even when game was rigged.

• losers, especially under high inequality, less likely to credit talent.

• RA more often cited luck; RE more often cited rules.

other findings

• selective effects on broader societal beliefs (GSS questions, importance of family wealth and talent to get ahead) details

• no detectable effect on redistributive behavior (dictator game and ultimatum game) details.

main takeaway

Why do people accept (unfair) outcome inequalities if they reject inequality of opportunity ?

• No need to assume limited awareness of unfairness

• We document a dual-track process where:

• Rejection of procedural unfairness: Unequal opportunities are seen as less fair than equal opportunity, but the size of outcome inequality doesn’t affect fairness evaluations.

• Normalization of inequality: People anchor their evaluation of outcome inequality in what they experience; opportunity distribution doesn’t affect tolerance for disparities.

• Personal consequences amplify effects: Winners see more fairness, accept larger gaps, and credit talent—regardless of rules or results.

discussion

• provides microfoundations for the macro puzzle of persistent inequality.
  
• unifies cognition, fairness judgments, and stability of unequal arrangements.
  

so …

THEY DON’T! THEY JUST GET USED TO IT.

thank you

mauricio bucca — puc chile · mebucca@uc.cl

mario d. molina — nyu abu dhabi · mdmolina@nyu.edu

supplementary materials

outcome treatment

• \(Y_i\): outcome for individual \(i\)
  
• \(\text{Position}_i \in \{\text{winner}, \text{loser}\}\): social position
  
• \(\mathbb{I}^{\text{out}} \in \{\text{high}, \text{low}\}\): outcome inequality level
  
• \(P_i \in \{\$10,\ \$7.5,\ \$5,\ \$2.5\}\): payment received by individual \(i\)

causal effect of outcome inequality: \(\mathbb{E}[Y \mid \mathbb{I}^{\text{out}}=\text{high},\ \text{Position}_i=p,\ P_i=p^*] - \mathbb{E}[Y \mid \mathbb{I}^{\text{out}}=\text{low},\ \text{Position}_i=p,\ P_i=p^*]\)

• compares same position (winner/loser),
  
• with same payoff \(P_i = p^*\),
  
• but different inequality levels (high vs. low).

earnings are fixed ⇒ differences reflect exposure to inequality only

opportunity treatment

RA/RE induces equal/unequal opportunity

limited role of talent/effort

Skill minimized; success ≈ hand strengts

external validity: GSS items

redistributive behavior

power analysis

• \(N = 3,335\), 2×2×2 factorial.
  
• Detectable effects:
  • Main: 7–8 p.p. differences
    
  • 2-way: ~10 p.p.
    
  • 3-way: ~14–16 p.p.
    
• Bonferroni corrected α = 0.0083.

sample composition

• Sample size: 3,333 participants

• Average age: 39 years (range: 18–80)

• Education: Nearly half hold a college degree; 17% have graduate/professional degrees

• Gender: 53% female, 45% male, 1.5% other

• Income: Broadly distributed, with most between $25,000 and $99,999

• Political ideology: Average score 6 (on a 1–9 scale)

• Political party: Majority identify as Democrat or Democrat-leaning; 18% Republican or Republican-leaning; 14% Independent

• Race: 70% White, 12% Black, 9% Asian, 8% Latino/Hispanic, 3% Other

• Region: 30% West, 27% South, 23% Midwest, 20% Northeast

• Religion: 44% report no religion; 22% Protestant/Lutheran; 16% Catholic; 14% Other; small percentages Jewish and Muslim