Why There Are Only Traces

Why can nobody just count the act itself?

• The act is illegal, so both sides hide it
• No victim files a report when both parties gained
• Definitions differ: gift, fee, commission, or bribe?
• Researchers can sometimes watch a bribe; usually they cannot
• So a measure settles for what corruption leaves behind

On direct observation of bribe payments in the field, see Sequeira (2012).

Family 1: Expert Perceptions

The Corruption Perceptions Index, published since 1995.

• Combines at least 3 of 13 existing surveys and assessments
• Respondents are country experts and business executives
• Scale runs 0 (highly corrupt) to 100 (very clean)
• Nobody in the data is asked “did you pay a bribe?”
• Covers almost every country every year – its great advantage

Source: Transparency International, How CPI scores are calculated.

What the Experts Are Actually Asked

The CPI itself asks nothing; it aggregates questions like these.

1. Imports and exports
2. Public utilities
3. Annual tax payments
4. Awarding of public contracts and licences
5. Obtaining favourable judicial decisions

From Thirteen Sources to One Number

The CPI runs no survey of its own; it averages thirteen other people’s.

%%{init: {"theme":"base","width":1150,"height":520,
  "themeVariables":{"fontSize":"20px","fontFamily":"Helvetica, Arial, sans-serif",
    "primaryColor":"#f4f6f8","primaryTextColor":"#1e293b",
    "primaryBorderColor":"#64748b","lineColor":"#334155","textColor":"#1e293b"},
  "flowchart":{"useMaxWidth":true,"htmlLabels":true,"nodeSpacing":26,"rankSpacing":55,
    "curve":"basis","padding":30}} }%%
flowchart LR
  A["13 sources<br/>at least 3 per country"] --> B["Rescale each<br/>onto 0-100"]
  B --> C["Simple<br/>average"]
  C --> D["A score,<br/>plus a range"]
  classDef inp fill:#fdf2ef,stroke:#b44527,stroke-width:2px,color:#1e293b;
  classDef step fill:#f4f6f8,stroke:#64748b,stroke-width:2px,color:#1e293b;
  classDef out fill:#eef3f1,stroke:#4a7c6f,stroke-width:2px,color:#1e293b;
  class A inp;
  class B,C step;
  class D out;

Source: Transparency International (2026), CPI Technical Methodology Note. Rescaling is a z-score against a baseline year, set to mean 45 and standard deviation 20. Three of the thirteen sources run the opposite way and are reversed first.

Why the Sources Disagree

Mexico’s eight sources span 19 to 37. Five reasons, none of them error.

• They score different things: bribery, nepotism, state capture
• Some score the machinery instead: prosecutions, disclosure rules, whistleblower protection
• They ask different people: executives, risk analysts, academic experts
• They carry different vintages: Mexico’s run from 2024 to 2026
• Each had its own scale before being rescaled onto 0–100

Sources: Transparency International (2026), CPI Technical Methodology Note and CPI 2025 Full Source Description.

What a Falling Score Means

Four years at the same score, then a five-point drop in 2024.

Family 2: Citizen and Firm Experience

Family 3: Audits and Missing Money

A physical audit, not an opinion.

Sources: Olken (2007, 2009); Reinikka and Svensson (2004).

Why Quantities Are Invisible

Family 4: Market-Based Inference

No one writes the bribe down, but the price is public.

• After Iraq invaded Kuwait, the UN banned Iraqi oil exports
• Oil for Food (1997–2003) let it sell, for humanitarian goods
• Every dollar went to a UN-controlled account, not to Baghdad
• Iraq still chose who was allowed to buy the oil
• Until 2001 it also proposed the price, for UN approval
• It proposed a price below the world market

Source: Hsieh and Moretti (2006).

Why Sell It Too Cheap?

Because the oil was cheap, but the right to buy it was not.

%%{init: {"theme":"base","width":1150,"height":360,
  "themeVariables":{"fontSize":"19px","fontFamily":"Helvetica, Arial, sans-serif",
    "primaryColor":"#f4f6f8","primaryTextColor":"#1e293b",
    "primaryBorderColor":"#64748b","lineColor":"#334155","textColor":"#1e293b"},
  "flowchart":{"useMaxWidth":true,"htmlLabels":true,"nodeSpacing":28,"rankSpacing":58,
    "curve":"basis","padding":26}} }%%
flowchart LR
  A["Oil money locked<br/>in a UN account"] --> B["Sell the oil<br/>below the<br/>market price"]
  B --> C["The discount is<br/>a gift Iraq<br/>hands out"]
  C --> D["Buyer pays cash back<br/>to Baghdad, outside<br/>that account"]
  classDef block fill:#eef3f1,stroke:#4a7c6f,stroke-width:2px,color:#1e293b;
  classDef gift fill:#fdf2ef,stroke:#b44527,stroke-width:2px,color:#1e293b;
  class A,B block;
  class C,D gift;

Like pricing a 150-peso ticket at 100, choosing who may buy one, and pocketing 30 on the side.

Only Cheap While Iraq Set the Price

Iraqi and Saudi crude are chemically very similar. They normally cost the same.

Same oil, same buyers, same market. Only the pricing rule changed.

Why the Two Diverge

• Different referent: the state’s corruption versus my transaction
• Coverage: a scandal raises perception without raising bribery
• Salience: people recall vivid cases, not their own year
• Denominator: experience counts only those who had contact
• Who answers: experts assess countries, residents report events
• Richer, more democratic countries score better than experience alone implies

Source on the last point: Donchev and Ujhelyi (2014).

Do Villagers Know?

Olken asked villagers how much corruption there was in their own road.

• They were partly right: more missing money, more suspicion
• But barely: 10% more missing moves suspicion from 36.0% to 36.3%
• And who you are matters: education and gender shape the answer
• Villages with more ethnic diversity report more corruption
• Those same villages lost less money

Source: Olken (2009), 3,691 respondents in 477 villages. Education and gender still shape answers when one village is compared only with itself.

How Suspicion Can Lower Corruption

One chain makes both of the last two findings true at once.

%%{init: {"theme":"base","width":1150,"height":360,
  "themeVariables":{"fontSize":"19px","fontFamily":"Helvetica, Arial, sans-serif",
    "primaryColor":"#f4f6f8","primaryTextColor":"#1e293b",
    "primaryBorderColor":"#64748b","lineColor":"#334155","textColor":"#1e293b"},
  "flowchart":{"useMaxWidth":true,"htmlLabels":true,"nodeSpacing":28,"rankSpacing":58,
    "curve":"basis","padding":26}} }%%
flowchart LR
  A["Diverse village,<br/>less trust<br/>between people"] --> B["More suspicion<br/>of the<br/>officials"]
  B --> C["More of them<br/>turn up and<br/>watch"]
  C --> D["Less money<br/>actually goes<br/>missing"]
  classDef belief fill:#fdf2ef,stroke:#b44527,stroke-width:2px,color:#1e293b;
  classDef doing fill:#eef3f1,stroke:#4a7c6f,stroke-width:2px,color:#1e293b;
  class A,B belief;
  class C,D doing;

Author’s illustration, after Olken (2009).

Exercise: Interrogate a Headline

What Each Family Can Carry

Author’s synthesis, drawing on Sequeira (2012) and Olken (2009). Not exhaustive: court records, procurement red flags and direct observation are further families.

So: Better or Worse?

Back to the question the newsroom could not settle.

• Four families: perceptions, experience, audits, market inference
• CPI score fell from 29 to 27 since 2017
• Perceived frequency fell from 91.1% in 2017 to 84.1%
• Experienced corruption rose from 14.6% in 2017 to 15.6%
• Different traces, different trends, different questions
• “It depends what you counted” is the disciplined answer

Conclusion

Five things to carry out of this session.

• Most measures count a trace, not the act itself
• Mexico 2025: 84% think it frequent, 16% experienced it
• Eight expert sources scored Mexico anywhere from 19 to 37
• In Java the theft hid in quantities, where nobody could check
• More suspicion and less stolen money can go together

References

References (continued)