You cannot manage, or evaluate the fight against, what you cannot measure — and corruption is hidden by its nature. The credibility revolution in economics produced ingenious ways to measure it anyway, moving from what people perceive to what can be objectively detected. This module is the methodological heart of the course, and it pairs directly with the Impact Evaluation course.
Three ways to measure corruption
- Perception — ask experts, firms, or citizens how corrupt they think things are (the WGI, the Corruption Perceptions Index). Cheap and broad, but measures beliefs, not acts; subject to halo effects, news cycles, and the de-jure/de-facto confusion of module 1.
- Experience — ask people about their own direct experience (Afrobarometer: did you pay a bribe for a service in the past year?). Closer to reality than perception, but subject to recall error, reluctance to admit, and coverage only of petty corruption people personally encounter.
- Objective / forensic — infer corruption from physical or financial evidence: compare what was paid for with what was delivered, track money from source to destination, or read corruption off market prices. The gold standard, pioneered in the 2000s.
Olken's missing expenditure
Measuring theft directly (Olken, 2007)
Studying village road projects in Indonesia, Benjamin Olken estimated corruption by comparison: he had engineers dig core samples and measure the materials actually used in each road, costed them at local prices, and compared the result with what the project reported spending. The gap — materials and labour paid for but not present in the road — is a direct, physical estimate of theft (the 'missing expenditure'), averaging around 24% of project cost. This is forensic measurement: it does not ask anyone's opinion; it measures the road.
Olken then embedded this in a randomised experiment (linking to the Impact Evaluation course): some villages were told their project would be externally audited with near-certainty, others not. The audit threat cut missing expenditure substantially, while a bottom-up community-monitoring treatment had little effect on the materials that were easy to steal — an early, influential finding that top-down audits can beat community monitoring for certain kinds of theft.
Public expenditure tracking
PETS: following the money (Reinikka & Svensson, 2004)
A Public Expenditure Tracking Survey follows funds from the centre to the frontline and measures how much arrives. Ritva Reinikka and Jakob Svensson tracked Uganda's per-pupil capitation grant and found that, in the mid-1990s, on average only about 13% of the money the central government released for schools actually reached them — the rest captured en route by officials and local elites. The leakage was invisible to perception surveys; it took following the money to see it. A subsequent newspaper campaign publishing the grant amounts (so schools and parents knew what to expect) sharply increased the share that arrived — measurement that itself became a remedy.
Reading corruption off prices and markets
A third forensic family infers corruption from market data. Raymond Fisman (2001) estimated the value of political connections in Suharto's Indonesia by measuring how the share prices of firms tied to Suharto fell on news of his ill health, relative to unconnected firms — the connection's value, capitalised in the stock. Fisman and Wei (2004) measured tariff evasion by comparing partner countries' reported exports to a country with that country's reported imports of the same goods (the 'missing imports' gap rises with the tariff, revealing smuggling and misclassification). These methods turn ordinary market data into corruption estimates.
The limits
No method is complete. Forensic methods are precise but narrow — they measure one project, one grant, one trade flow, and may not generalise. Experience surveys miss grand corruption (no one personally bribes for a Eurobond kickback). Perception indices are broad but soft. List experiments and randomised-response techniques (which let respondents admit sensitive behaviour without revealing it on any individual answer) help with the reporting problem but add noise. The craft is triangulation: combine a forensic anchor where you can get one with broader experience and perception data, and be explicit about what each does and does not capture.
Exercise
A county is suspected of large-scale corruption in a school-construction programme: inflated contracts, materials short-delivered, and funds that don't reach schools. You are asked to measure the corruption credibly, not just assert it. (1) Why would a citizen perception survey be insufficient here? (2) Design a measurement strategy combining at least three methods from this module, specifying what each would capture. (3) Explain how you could embed an evaluation (linking to impact evaluation) so the measurement also tests whether a remedy works. (4) State two limitations of your strategy and how you would acknowledge them.