Lecture 14: Final Review
Today has four parts.
| Session | The question it answers | |
|---|---|---|
| 8 | How corruption is measured | if the act is hidden, what are we counting? |
| 9 | Machines | why pay for a vote you cannot see? |
| 10 | China and Hong Kong | can corruption and growth go together — and how did one city get clean? |
| 11 | Kleptocracy | how does public money leave the country, and why does it not come back? |
| 12 | Organized crime | what changes when the buyer can also shoot? |
| 13 | Technology | if we can detect it, does it stop? |
Thursday’s exam is Sessions 8–13 only.
The paper asks nothing about Sessions 1–7. Two things from the first half are still tools you need on Thursday:
You will not be asked to name Klitgaard or define a confounder. You will be asked to use them on Sessions 8–13 material.
The midterm is behind you. Thursday is the last third.
| When | Thursday 3 December, in class |
| Where | In Canvas, on your own screen |
| How long | 30 minutes |
| Covers | Sessions 8–13 |
| Worth | 33% of your grade |
Closed book. No notes, no books, no browser tabs except Canvas.
Seven short parts, thirty-two items. Twenty-three of them need only a label or a letter.
| Part | What it asks | Answer looks like |
|---|---|---|
| A | Name the measure | one name from a printed list |
| B | Name the exchange | one name from a printed list |
| C | Name the concept | one name from a printed list |
| D | What made the violence rise? | one label + one sentence |
| E | Explain the pattern | one phrase + a few words |
| F | Read the study | two answers, 2–3 sentences each |
| G | Match the reading | one letter |
Each list has one option per question: you use each option exactly once. The exception is Part G, which has one extra option.
Today’s exercises are the practice paper, and they work the same way. So the last answer in each part is the one to be careful about — if you have guessed your way to it, you have not checked it.
Twenty-three items are a single word. They are not where the paper is decided.
Almost every mark lost on the midterm was lost in the sentence, not the label.
Nobody counts the act. Every number counts a trace the act left behind, and the four families observe four different traces.
| What it observes | Where it is blind | |
|---|---|---|
| Expert perceptions (CPI) | what country experts and executives assess | nobody in the data was asked whether they paid |
| Citizen and firm experience | what happened to this person in this transaction | grand corruption: contracts, capture, procurement |
| Audits and missing money | what was built, against what was billed | needs engineers, and only where there is a physical thing |
| Market-based inference | a price that is wrong, in public | tells you a rent was created, not who took it |
Measure: an observable trace we agree to treat as a stand-in for something we cannot see directly.
A falling CPI score shows that assessments changed. It does not yet show that behaviour did.
Different referent, different denominator, different people answering. Both numbers are right.
Physical audit: engineers compared what was built with what was billed in 477 village road projects in Java.
Perceptions and audits tell us different things. Suspicion is not the same as theft.
Everyone in the village knows what a cubic metre of gravel costs. Nobody can look at a heap and say how many arrived.
So the theft moved into the dimension nobody could check — which is why the audit needed engineers, not just eyes. Detection shapes the crime, not only our measurement.
Photo: Greg Willis, Wikimedia Commons, CC BY-SA 2.0.
Clientelism: politicians provide targeted benefits in exchange for political support.
Hicken (2011) identifies four key characteristics:
Remember:
| Term | Who receives | What is exchanged | The test |
|---|---|---|---|
| Programmatic | anyone who qualifies | a policy rule | you get it even if you voted against |
| Patronage | party insiders | a public job | paid for party work, in a post |
| Vote buying | ordinary voters | cash, food, goods | the debt is settled at the ballot box |
| Turnout buying | the party’s own passive supporters | a ride, a favour, a small payment | nobody asks which box, only that you went |
| Clientelist service | named residents | access to the state | it comes back next time, and someone is watching |
Two questions do most of the work: who gets it, and does political support decide whether they get it.
A benefit written into a rule. The test: you receive it whether or not you voted for her — so nothing about your politics needs to be known at all.
The currency here is a public job, and it is paid for party work rather than for a vote. That is what separates it from the three below.
One payment, one election. The test: the debt is settled at the ballot box and nothing carries forward — so there is no next time to withhold.
Her own supporters, and a ride rather than a payment. The test: nobody asks which box you ticked, only that you went — and that is a thing she can actually check.
This is Nichter’s answer to the secret ballot: stop buying the thing you cannot verify, and buy the thing you can.
The flour is not what makes this clientelism. The clues are the banner — same people, same support, every election, with the years ticked off — and the clipboard.
Ask what happens after the election. If the deal ends there it was vote buying. If it comes back at the next one, and somebody is keeping a list, it is clientelism.
Before. Parties printed their own ballots — distinctive paper, visible in the hand.
After. One official ballot, marked behind a screen: el voto es secreto.
Massachusetts 1888; Argentina’s Sáenz Peña Law 1912. Compliance stops being visible — and every machine built on compliance now has a problem.
Left: Cornell University Library. Right: ProtoplasmaKid, CC BY-SA 4.0. Both via Wikimedia Commons.
If the ballot is secret, why pay for a vote you cannot verify?
| What is bought | What makes the voter keep their side | |
|---|---|---|
| Stokes | the vote | the broker may work out how you voted, and cut off the flow next time |
| Nichter | the turnout | she can see whether you showed up; your vote stays secret |
| Finan & Schechter | the vote | you feel you owe her; nobody has to watch at all |
Nichter changes what is bought. Finan and Schechter change who does the enforcing — you do.
The puzzle: Why can’t voters take the money and then vote for another party?
Perverse accountability: instead of voters holding politicians accountable, politicians hold voters accountable.
Evidence from Argentina: Who was more likely to receive benefits?
37% of respondents believed that parties could find out how people voted.
Ang distinguishes between types of corruption based on two questions:
| Taking | Trading | |
|---|---|---|
| Non-elites | Petty theft: a clerk steals public money | Speed money: paying a bribe to get a permit faster |
| Elites | Grand theft: a minister embezzles millions | Access money: bribing a politician to obtain a lucrative contract |
Ang compares corruption to drugs:
Different types of corruption have different economic consequences.
Between 1978 and 2010, China’s economy grew roughly 9% per year, despite rising corruption (Wedeman’s double paradox).
Ang’s explanation: the composition of corruption changed, with access money becoming more important.
What is the problem?
Access money can coexist with economic growth. This does not mean that corruption causes growth.
The ICAC operates independently of the police. The police cannot stop investigations into their own corruption.
ICAC headquarters, North Point, Hong Kong. Photo: Citobun, CC BY-SA 4.0, via Wikimedia Commons.
Corruption equilibrium: people engage in corruption because they expect everyone else to do the same.
The ICAC adopted three strategies:
| Strategy | Objective |
|---|---|
| Deterrence | Increase the risk of getting caught |
| Prevention | Reduce opportunities for corruption |
| Education | Change what citizens consider acceptable |
Why was the ICAC credible?
Fighting systemic corruption requires changing both incentives and expectations.
Two governments both raise $100 billion. A collects it from millions of taxpayers. B sells oil to foreign buyers.
Rent — the gap between what something costs to produce and what it sells for. Oil costs $20 a barrel to lift and sells for $80. The $20 pays for the work; the $60 is there because the oil was already in the ground.
Aid is a windfall too. In the 50 most aid-dependent countries in Svensson’s data it financed 53.8% of central government spending.
Imagine that the World Bank gives a government an additional $100 million to build hospitals.
Rent-seeking: spending resources to obtain existing wealth rather than creating new wealth.
More foreign aid can result in fewer public goods. Credible conditions on aid can reduce this problem.
The World Bank provides $10 million to build a road.
| Where the money goes | Amount |
|---|---|
| Road construction | $6 million |
| Government official | $2.5 million |
| Company’s extra profit | $1.5 million |
| Total | $10 million |
Kickback: a payment to an official in return for awarding a contract or providing another favor.
The accounts may look correct even when corruption occurs.
The government official wants to hide his $2.5 million.
Why is recovering stolen money difficult?
Shell companies and offshore accounts are not inherently illegal, but they can be used to conceal stolen wealth.
Ugland House, Cayman Islands: the registered address of thousands of companies.
Photo: Coolcaesar, Wikimedia Commons, CC BY 4.0.
Research question: Do political elites steal foreign aid and deposit the money in offshore bank accounts?
Prediction: if elites use offshore havens to hide stolen aid, deposits in havens should increase when foreign aid arrives.
Diego: bribery
Diego pays the customs officer to let his truck through.
Cartel: bribery + threats
The cartel offers less money but threatens the officer if he refuses.
Why can the cartel obtain the same favor for less money?
Imagine that a police officer requires $100 to compensate for the risk of engaging in corruption.
\[\underbrace{b}_{\text{bribe}} + \underbrace{p}_{\text{threat}} = \underbrace{c}_{\text{cost}}\]
Three implications:
Bribery and intimidation are substitutes.
| Government policy | Criminal organizations may respond with |
|---|---|
| Make bribery more difficult | More threats, fewer bribes |
| Make intimidation more difficult | More bribes, fewer threats |
How can governments address this problem?
Fewer bribes do not necessarily mean less criminal influence.
Democratization disrupted criminal protection networks without necessarily establishing effective state control.
Research question: Do government crackdowns reduce drug-related violence?
Why? Weakening a cartel creates an opportunity for rival cartels to capture its territory.
Weakening a criminal organization is not the same as eliminating criminal competition.
To evaluate any new anticorruption technology, ask:
Mechanism: Does it increase competition, reduce discretion, or increase accountability?
Enforcement: Can anyone act when corruption is detected?
Manipulation: Can officials manipulate or misuse the technology?
Effectiveness: Does corruption actually decrease?
Detecting corruption is not the same as reducing corruption.
Traditional red flags include single bidding, short deadlines, and unusually high prices.
The problem: honest contracts can look suspicious, and corrupt contracts can look legitimate.
Machine learning can identify suspicious patterns that are difficult to detect by examining individual contracts.
Still only a list of suspects. And it learned from past sanctions, which follow past audits — so it partly learns where investigators already look, and inherits their blind spots.
Imagine a speed camera that only works at lunchtime.
8 November 2021
A field during a fire.
13 November 2021
The same field after the fire.
Dasgupta uses burn scars to detect fires that the official monitoring system missed.
Sentinel-2, 10 m per pixel. Contains modified Copernicus Sentinel data (2021).
Dasgupta compares Punjab with similar agricultural areas across the border that were subject to a different policy.
Goodhart’s law: when a measure becomes a target, people may improve the measure without improving what it is supposed to measure.
One name each. Each is used exactly once. Choose from: expert perception index, direct experience survey, physical audit, market-based inference.
| The case | Answer | Why | |
|---|---|---|---|
| 1 | 40,000 households, denominator is those who had contact | direct experience survey | what happened to them, in their own transaction |
| 2 | eleven assessments rescaled into one score | expert perception index | nobody in the data is asked about a payment |
| 3 | engineers measure what was built and price it locally | physical audit | somebody measured the thing, not what people say |
| 4 | official price against the private resale price | market-based inference | the price is public and wrong, so the rent is visible |
One name each. Each is used exactly once. Choose from: programmatic politics, patronage, vote buying, turnout buying, clientelist service.
| The case | Answer | Why | |
|---|---|---|---|
| 1 | minibuses for her own villages, nothing asked about the ballot | turnout buying | her own supporters, and only attendance is checked |
| 2 | forty parks posts for the people who ran his campaign | patronage | the currency is a public job, paid for party work |
| 3 | a subsidy for every household under two hectares, at any bank | programmatic politics | a rule: you get it however you voted |
| 4 | cash two days before, then the agent never returns | vote buying | the debt is settled at the ballot box |
| 5 | paperwork, a training place, and the family at party events | clientelist service | access to the state, and it comes back next time |
One name each. Each is used exactly once.
1–4: petty theft, grand theft, speed money, access money.
5–8: windfall, rent, kickback, beneficial owner.
| The case | Answer | Why | |
|---|---|---|---|
| 1 | invents a fee and charges it on every application | petty theft | non-elite, and one side simply loses |
| 2 | $60 million into an account he controls | grand theft | elite, taking — C1 at a different altitude |
| 3 | pays so the certificate comes this week | speed money | non-elite trade: the certificate was his anyway |
| 4 | finances the stadium, gets the development rights | access money | elite trade: a privilege bought, both expect to gain |
| 5 | two-thirds of revenue from gas, not from taxpayers | windfall | revenue nobody at home had to be asked for |
| 6 | $2 to extract, $8 to sell | rent | the gap that exists from ownership, not effort |
| 7 | cheaper materials, and part of the saving goes back | kickback | the return payment for the award |
| 8 | Jersey owns it, Belize owns Jersey | beneficial owner | the one name the layering is built to withhold |
A label and one sentence. Choose from: breakdown of political protection; removal of the group that held the territory; intimidation in place of bribery. The sentence must name what changed and what stayed the same. 1 mark for the label, 2 for the sentence.
| The case | Answer | Why | |
|---|---|---|---|
| 1 | the group is removed, the port is as valuable as ever | removal of the group that held the territory | the prize is still there, so rivals move in |
| 2 | the two officials paid to look away are replaced | breakdown of political protection | neither group knows if the other is still protected |
| 3 | bribery gets risky, officers unprotected, threats rise | intimidation in place of bribery | the instruments substitute; only the currency changed |
A phrase and a few words. Choose from: direct causation (the first causes the second); reverse causation (the second causes the first); confounding (a third factor causes both); measurement gaming (the record moved, the thing measured did not). Name the third factor if you say confounding.
| The case | Answer | Why | |
|---|---|---|---|
| 1 | record falls 71%, random inspections find the waste | measurement gaming | the target was the record, not the dumping |
| 2 | narrow win against narrow loss, groups alike before | direct causation | at that margin the only difference is who took office |
| 3 | worst scores get the most assistance, programmes come after | reverse causation | the score came first and is why the aid was sent |
| 4 | bigger transfers, more emigration | confounding: the province’s poverty | poverty is why the transfers went there and why people leave |
A study published in 2024 asks whether local news coverage deters theft from local government.
The author measures corruption in each of 180 municipalities as the share of audited projects in which the national audit office recorded irregularities. All the figures come from the same year. He relates that to whether the municipality’s local newspaper runs a website.
Municipalities whose newspaper is online show fewer recorded irregularities. The author has already taken each municipality’s income per head into account.
He concludes that online local news deters theft from municipal budgets.
Because income is already accounted for, “they are richer” on its own earns nothing. What still counts is saying what that control does not remove.
(a) Why might the proportion of audited projects with recorded irregularities be a misleading measure of theft? Give one reason and explain it.
(b) What else might explain why municipalities with an online newspaper have fewer recorded irregularities? Say what evidence would help the author test your alternative.
(a) It counts recorded irregularities — what the audit office looked for, found, and wrote down. That is detection, not theft.
Why it matters: how many are recorded depends on how thoroughly a municipality was audited and how well it keeps its books, so the same stealing can produce very different numbers. A lower count may mean less looking, or tidier files.
(b) The rival is administrative capacity. A professional administration keeps cleaner records and runs cleaner procurement — and is also the kind of place that can sustain an online newspaper.
What would test it: the same municipalities before and after their newspaper went online, against municipalities that had not yet gone online. Not proof, but it stops the comparison resting on one year’s snapshot.
One letter each. One option is left over.
1 b · 2 d · 3 f · 4 g · 5 e · 6 a
| 1 Olken | (b) belief against what was taken | suspicion barely tracked the missing money |
| 2 Stokes | (d) enforcing a bargain it cannot see | the broker guesses, then withdraws the flow |
| 3 Manion | (f) an agency outside the police | credibility by design, not harsher penalties |
| 4 Andersen et al. | (g) aid into offshore accounts | disbursements against haven deposits, quarter by quarter |
| 5 Trejo and Ley | (e) the end of one-party rule | alternation replaced the officials the pact was made with |
| 6 Dasgupta | (a) officials and their own indicator | the fires left the record, not the fields |
Left over: (c) — Medina-Hernández, Kertész and Fazekas on machine learning in Mexican procurement. On the reading list; not one of the six here.
Thursday 3 December, in class, in Canvas. Thirty minutes, closed book, Sessions 8–13.
Popescu (TEC) Politics of Corruption Lecture 14: Final Review