Comparative Politics

Skocpol's second list

Theda Skocpol published States and Social Revolutions in 1979. Everyone remembers the three revolutions in it: France from 1787, Russia in 1917, China from 1911 to 1949. Far fewer remember the second list, which is where the book's argument actually lives. Skocpol also works through England in the 1640s, Prussia after the defeat at Jena in 1806, and Japan after 1868. None of those produced a social revolution. That is the point of including them.

Here is why. France, Russia and China in their revolutionary years all had agrarian economies, landed upper classes, and states just humiliated by a militarily superior rival. You could write a book saying revolutions come from agrarian backwardness plus foreign defeat. It would sound convincing and be worthless, because you looked only at revolutions. Prussia in 1806 was agrarian, had a landed Junker class, and had just been destroyed by Napoleon at Jena; it produced the Stein-Hardenberg reforms. Japan after Perry's squadron arrived in 1853 was agrarian, had a hereditary military aristocracy, and faced overwhelming foreign force; it produced the Meiji Restoration, run by members of the old elite.

The negative cases turn a list of similarities into a claim about causes. They force Skocpol to say what the three revolutions had that Prussia and Japan lacked: a state administration that could not be repaired from inside, and a peasantry with the local organisation to act on its own. Strip out the negatives and you cannot reach that sentence, because you would not know that backwardness plus defeat is compatible with no revolution at all.

Mill's two methods, and Mill's own objection to using them

John Stuart Mill set out five canons of inductive inference in A System of Logic in 1843. Two of them are the backbone of every small-N comparison ever published.

The method of agreement takes cases that share an outcome and differ in everything else you can think of, and looks for the one condition they still share. If revolutions occurred in France, Russia and China, and those three societies differ in religion, language, century, population and technology but all had a state that had just lost a war, then defeat is a candidate cause because it survived the differences.

The method of difference takes two cases as alike as you can find them, one with the outcome and one without, and looks for the condition present in one and absent in the other. This is the stronger of the two, because it approximates an experiment: hold everything constant, vary one thing, see what happens.

Now the part that gets skipped. Mill did not think either method could be used on whole societies. In Book VI of the Logic he argues that the method of difference is unavailable in politics because you cannot obtain two nations agreeing in every particular but one, and that the method of agreement is defeated by the plurality of causes: the same effect arises in different countries by entirely different routes. He wanted deduction checked against history instead. Anyone teaching Mill's methods as the foundation of comparative politics is teaching a tool over the objection of the man who made it.

The point: the methods are not a licence. They are a statement of what perfect data would let you conclude, and the whole craft of comparison consists of getting as close to those conditions as an imperfect world allows, then being explicit about the remaining distance.

Two designs, and the different jobs they do

Adam Przeworski and Henry Teune gave the two workable approximations their modern names in 1970. Both start from the recognition that you will never hold everything constant, so you choose which uncontrolled variation you are willing to live with.

Read the third row slowly. Neither design confirms a cause; both eliminate candidates. Pairing Costa Rica with Nicaragua cannot show that abolishing the army made Costa Rica democratic. It can show that Spanish colonial inheritance, coffee exports, Catholicism and small population do not explain the difference, because both countries have all four.

Lijphart's arithmetic, and four ways to live with it

In 1971 Arend Lijphart published a short article in the American Political Science Review that stated the discipline's central practical problem in one phrase: many variables, small number of cases. There are about 190 sovereign states, and if you are studying democratic breakdown in interwar Europe you have perhaps twenty. The plausible causes number in the dozens: class structure, electoral system, party fragmentation, army loyalty, external threat, depression, colonial legacy, religion, the constitution, the president's temperament. You cannot estimate thirty effects from twenty observations, and no statistical technique manufactures the missing information.

Lijphart proposed four responses, and they remain the standard menu.

1. Increase the number of cases. Extend backwards in time, or down to subnational units: India's 28 states, Germany's 16 Laender.
2. Reduce the property space. Combine variables that travel together, so eight correlated measures of state strength become one.
3. Compare comparable cases. Restrict the sample to countries alike on the background, controlling by design rather than by statistics.
4. Focus on the key variables. Use theory to name in advance the two or three that matter, and accept the risk that the theory is wrong.

Every one of these is a trade. Indian states share a national constitution, so they buy you observations and cost you any leverage on constitutional design. There is no free move, and an author who does not name the trade has hidden the weakest part of the argument.

Choosing cases on the outcome, and what it costs

The single most common error in this field has a name: selection on the dependent variable. It means picking your cases by looking at the outcome you want to explain, then studying only the ones where it happened.

Barbara Geddes made the cost concrete in 1990. Through the 1980s a well-known argument held that the East Asian growth miracle rested on labour repression: South Korea, Taiwan, Singapore and Hong Kong had all suppressed independent unions and all grew fast. The four cases were selected because they grew fast. Geddes reselected them on the proposed cause instead, gathering authoritarian regimes that repressed labour whether or not they grew, and the relationship collapsed. Plenty of governments crushed unions and presided over stagnation. Labour repression turned out to be close to universal among the dictatorships of the period, which means it cannot explain why a handful of them grew at seven percent a year while others managed two.

The formal version matters. An effect is a difference between what happens with a cause and what happens without it, so a sample with no variation on the outcome cannot yield one. Select partially, keeping high-outcome cases more often, and estimates are pulled toward zero. King, Keohane and Verba worked this out in Designing Social Inquiry in 1994; note the direction, because it is counterintuitive. The result to distrust most from a selected sample is not a strong one. It is a null one.

What matters here: ask of any comparative argument, including the ones later in this course, how the cases got into the study. If the answer is that they are the ones where the interesting thing happened, the argument may still be true, but it has not been tested.

Working one design: why Costa Rica and not Nicaragua

Set up a most-similar comparison properly and watch what it delivers. The question: why has Costa Rica held competitive elections continuously since 1953 while Nicaragua has not?

The two share a great deal. Both were governed from the Captaincy General of Guatemala, both became independent in 1821, both joined and left the Federal Republic of Central America between 1823 and 1841, both built nineteenth-century export economies on coffee, both are overwhelmingly Catholic and Spanish-speaking, both are small: about 5.2 million people in Costa Rica against about 6.9 million in Nicaragua. Both saw repeated United States intervention. On the standard background variables they are as close as two countries get.

They differ on the outcome, and the design points at the differences that remain. Two stand out. The first is what happened to the army. After losing a disputed election result to a congressional annulment, Costa Rica fought a six-week civil war in 1948; the winner, Jose Figueres Ferrer, abolished the standing army on 1 December 1948, and the 1949 constitution wrote the prohibition into Article 12. Nicaragua's Somoza family took the opposite path: Anastasio Somoza Garcia used his command of the National Guard, an institution built during the United States occupation of 1927 to 1933, to seize the presidency in 1936 and hold the country until 1979. The second difference is landholding. Costa Rica's coffee frontier in the central valley was farmed largely by smallholders; Nicaragua's export agriculture concentrated land in far fewer hands.

The design has done real work. It has removed colonial inheritance, religion, language, crop and size from the list of things that explain the difference, because both countries have them. It has left two candidates. It has not chosen between them, and it cannot: with an N of two and two surviving differences, the comparison is exactly as underdetermined as Lijphart said it would be.

What would overturn the army argument

Take the abolition claim seriously enough to say what would kill it. If abolishing the army protects democracy, other states that dismantled their armed forces should show the pattern. Panama abolished its military in 1990 after the United States invasion and has held competitive elections since, which fits. Haiti demobilised its armed forces in 1995 and did not become a stable democracy, which does not. So the Costa Rica inference is a hypothesis a design generated, not a result. Test it by collecting every case of abolition or radical demobilisation since 1945 and coding what followed; if the outcomes look like coin flips, the mechanism is not doing the work.

Common misconceptions

• Case studies cannot establish causes, only generate hypotheses. Too strong. A most-similar pair can eliminate a proposed necessary condition decisively: one country with the condition and without the outcome refutes the claim that the condition is sufficient. Elimination is causal work.
• More cases are always better. Only if the added cases belong to the same population. Adding Denmark to a study of Central American regime survival does not increase your information about Central America; it changes the question you are answering.
• Selecting on the dependent variable inflates your findings. Usually the opposite. Truncating the outcome compresses its variance and biases estimated effects toward zero, which is why a null result from a selected sample tells you almost nothing.

Where this leaves us

Four questions to put to any comparison, including your own. How did these cases get into the study, and were they chosen by looking at the outcome? Is the design most-similar or most-different, and does it match the question? How many candidate explanations survive it, and does the author admit the number? What evidence would overturn the conclusion, and has anyone looked?

Worth holding on to: Skocpol's book is remembered for France, Russia and China, but it is an argument because of England, Prussia and Japan. The cases where nothing happened are not background. They are the comparison.

Sources

1. Mill, J. S. (1843). A System of Logic, Ratiocinative and Inductive. John W. Parker. Project Gutenberg
2. Macleod, C. (2024). John Stuart Mill. In The Stanford Encyclopedia of Philosophy. plato.stanford.edu
3. Lijphart, A. (1971). Comparative politics and the comparative method. American Political Science Review, 65(3), 682-693.
4. Geddes, B. (1990). How the cases you choose affect the answers you get: Selection bias in comparative politics. Political Analysis, 2, 131-150.
5. King, G., Keohane, R. O., & Verba, S. (1994). Designing Social Inquiry: Scientific Inference in Qualitative Research. Princeton University Press.

Twenty-two survivors out of sixty-two

In 1997 Xavier Sala-i-Martin published a paper in the American Economic Review called I Just Ran Two Million Regressions. The title was literal. Economists had been arguing about which of dozens of national characteristics cause growth, and every author reported a specification in which their favourite variable was significant. Sala-i-Martin assembled 62 candidate variables, fixed three of them in every model, and ran the growth regression with every combination of three more drawn from the remaining 59. Twenty-two variables kept their sign and significance across the whole distribution of specifications. The rest did not.

That is a useful shock to hold onto for the rest of this course. A published table with three stars beside a coefficient is one draw from a very large space of tables the author could have shown you. Cross-national statistics are indispensable, because there is no other way to find out whether a pattern you noticed in two countries holds in 150. They are also the easiest place in comparative politics to fool yourself, and the ways they fail are specific enough to memorise.

Building the regression you would actually run

Take the oldest question in the field: are rich countries more likely to be democracies? Write the simplest model that could answer it. For each country i, let D be a democracy score between 0 and 1 from V-Dem, and let Y be gross domestic product per capita. You do not regress D on Y directly, because a thousand dollars means something entirely different to Malawi than to Denmark. You use the natural logarithm of income, so that the coefficient describes what happens when income changes by a constant percentage rather than a constant number of dollars.

The estimated model is D = a + b log(Y) + e. Suppose b comes out at 0.11. That number has a precise meaning and a much vaguer one, and the gap between them is where most misreadings live. Precisely: comparing two countries whose incomes differ by a factor of e, roughly 2.72, the richer one scores about 0.11 higher on a scale running from 0 to 1. Doubling income, which is a smaller step, corresponds to about 0.11 times log(2), or 0.076. Vaguely, and usually wrongly: if a country got richer, its democracy score would rise by that amount. The regression compares countries. It does not follow any country through time, so it cannot describe what happens when one changes.

In short: a cross-sectional coefficient answers the question how do rich countries differ from poor ones. It does not answer the question what happens to a poor country that becomes rich, and those are different questions with, as we will see, different answers.

Five ways this goes wrong, in order of how often

Measurement. D is not observed; it is coded. Polity assigns a country-year a score from -10 to 10 built out of judgements about executive recruitment and constraints. V-Dem's electoral democracy index aggregates ratings from multiple country experts across dozens of indicators. Freedom House scores civil liberties and political rights on separate scales. The three disagree, and not randomly: cases with elections that are competitive but unfair, such as Turkey after 2014 or Hungary after 2010, are where they diverge most. Your coefficient inherits every coding decision.

Omitted variables. Oil raises income and, on most accounts, lowers democracy. Leave oil out and part of its negative effect on D is attributed to income, dragging b downward. The direction is knowable in advance: an omitted variable that is positively correlated with your predictor and negatively with your outcome biases the coefficient toward zero. Work out the sign before you look at the result, so that you cannot be talked into the version that suits you.

Reverse causation. Democracies may become rich rather than the reverse. Secure property rights, contested budgets and a press that reports on procurement fraud are all plausible causes of growth. Nothing in the regression distinguishes the two directions, because both produce the same cloud of points.

Non-independence. Regression treats each country as a separate draw. Countries are not separate draws. Between 1974 and 1990, roughly thirty countries moved from authoritarian rule to democracy in what Samuel Huntington called the third wave, and they did so in geographic clusters: southern Europe first, Latin America next, eastern Europe in a single eighteen-month burst after 1989. Poland did not decide independently of Hungary. Francis Galton raised this objection to Edward Tylor's cross-cultural statistics in 1889, and it still carries his name: Galton's problem. If your 150 countries are really 30 independent episodes, your standard errors are too small by a factor you have not calculated.

Specification search. This is Sala-i-Martin's point turned into a warning. With 59 optional controls there are more than 32,000 ways to choose three, and somewhere in there is a table supporting almost anything. The defence is to report the distribution of estimates across specifications rather than the one you like.

The second run, where the same data reverse the finding

Now the worked example that matters, because it uses one dataset and changes only how the outcome is defined.

Adam Przeworski and his collaborators assembled regime and income data for 135 countries from 1950 to 1990. Regressed the ordinary way, being a democracy is strongly associated with being rich, exactly as the modernisation literature had claimed since 1959. Then they split the outcome in two. A country-year can produce a transition from dictatorship to democracy, or a country-year under democracy can produce a breakdown back to dictatorship. Estimate those separately and the picture changes completely.

Their headline figure gives the third row its force: in the 1950 to 1990 window, no democracy with a per capita income above 6,055 international dollars of 1985, the level of Argentina in 1975, had ever fallen. Poor democracies died often. Rich ones did not die at all. The strong cross-sectional association between wealth and democracy is therefore consistent with wealth having no effect whatsoever on becoming a democracy, and a very large effect on staying one. Same countries, same years, same income series. One modelling decision, and the causal story inverts.

Lesson 5 takes up the fight this started, including Carles Boix's response that extending the sample back to 1800 restores the effect on transitions. For now, notice the mechanism: the finding did not change because someone found new data. It changed because someone asked what the dependent variable actually was.

What regression genuinely delivers

None of this is an argument for abandoning the method, and the case studies in Module 5 would be far weaker without it. Four things a cross-national regression does well.

• It refutes universal claims cheaply. If a theory says no Muslim-majority country can sustain competitive elections, Indonesia and Senegal are enough, and the regression finds them for you.
• It sizes an effect. Knowing that an effect exists is much less useful than knowing it is worth 0.08 on a scale of 1.
• It disciplines case selection. Fit the model, then study the countries the model gets most wrong: Botswana and India as democratic outliers, Singapore and Qatar as wealthy non-democracies. Residuals are a principled way to choose cases, and unlike selecting on the outcome it is defensible.
• It makes the sample visible. A regression forces you to say which country-years you used, which is more than most narrative comparisons do.

What would overturn the argument of this lesson

The claim here is that cross-national coefficients are fragile to specification. Test it directly. If someone took the twenty or so most-cited findings in comparative politics, re-ran each across the full space of reasonable specifications, and found that the estimates barely moved, this lesson would be wrong and cross-national work would deserve far more confidence than it gets here. Such exercises exist in economics and the results are mixed rather than reassuring, but the political science version is thin. If you want a dissertation topic, that is one.

Common misconceptions

• Adding more control variables makes an estimate more trustworthy. Controls that are themselves consequences of your predictor absorb the very effect you are trying to measure. Controlling for education when studying income and democracy may remove the main channel through which income works.
• Country fixed effects solve omitted variable bias. They absorb everything that does not change within a country, which is a real gain, and they simultaneously make it impossible to estimate the effect of anything constant, such as colonial origin, legal tradition or latitude.
• A large sample means a reliable result. Not when the observations are not independent. One hundred and fifty countries clustered into a handful of diffusion waves carries far less information than the sample size suggests.

The short version

A cross-national regression compares countries as they are. Read its coefficient as a difference between countries, never as a prediction about change within one, unless the design specifically supports that. Before believing any such table, ask how the outcome was coded, what is omitted and in which direction it biases the estimate, whether causation could run the other way, whether the countries are independent, and how many other tables the author could have run.

Bottom line: Przeworski's split of one association into transition and survival is the model to imitate. The most valuable thing you can do to a cross-national finding is not to add controls. It is to ask whether the dependent variable is really one thing.

Sources

1. Sala-i-Martin, X. X. (1997). I just ran two million regressions. American Economic Review, 87(2), 178-183.
2. Przeworski, A., Alvarez, M. E., Cheibub, J. A., & Limongi, F. (2000). Democracy and Development: Political Institutions and Well-Being in the World, 1950-1990. Cambridge University Press.
3. Wikipedia contributors. (n.d.). Galton's problem. Wikipedia
4. Herre, B., Ortiz-Ospina, E., & Roser, M. (n.d.). Democracy. Our World in Data. ourworldindata.org
5. Angrist, J. D., & Pischke, J.-S. (2009). Mostly Harmless Econometrics: An Empiricist's Companion. Princeton University Press.

Four hundred and seventy-five states that no longer exist

Around 1500 the map of Europe held roughly five hundred independent political units: duchies, bishoprics, free cities, leagues, principalities, a few kingdoms. By 1900 it held about twenty-five. The missing four hundred and seventy-five were conquered, absorbed, partitioned, bankrupted or simply outbid. Charles Tilly built a theory of state formation out of that attrition, and it remains the most useful thing in comparative politics for explaining why the units we compare look the way they do.

The story runs through money, not ideas. But before the story, the definition, because the whole argument turns on it.

Weber's definition, and the three words that do the work

Max Weber gave a lecture in Munich in January 1919, published as Politics as a Vocation, in which he defined the state as a human community that successfully claims the monopoly of the legitimate use of physical force within a given territory. Every word in that sentence is load-bearing.

Monopoly is a matter of degree, not a switch. Anywhere private actors can use force without the state's leave, the monopoly is partial. Legitimate does not mean morally justified; it means that the claim is generally accepted, so that a police officer using force is doing something different in kind from a gunman doing the same thing. Territory means the claim is bounded by lines on a map rather than by who your subjects are, which is what separates a modern state from an empire of persons or a nomadic confederation. And successfully claims is the escape hatch: Weber is describing an assertion that is made to stick, not a fact of nature.

Two cases in the Horn of Africa show what happens when you take the definition seriously. Somalia holds a United Nations seat and is recognised by every state on earth, and for long stretches since 1991 its government has not controlled large parts of its territory. Somaliland, which declared independence in 1991, has its own currency, army, passports and a sequence of competitive elections, and is recognised by nobody. On Weber's definition Somaliland is closer to a state than Somalia. On the definition used by the international system, only one of them exists. Keep both definitions in mind, because indices measure one and theories usually mean the other.

Tilly's engine

Tilly's mechanism has four steps and one condition. The condition is that war is expensive and losing it is fatal.

1. A ruler faces a military rival and needs an army bigger than last time. Gunpowder, fortification and standing infantry all made the minimum viable army larger through the sixteenth and seventeenth centuries.
2. Armies must be paid, so the ruler needs revenue on a scale that plunder and crown lands cannot supply.
3. Extracting revenue from a population requires an apparatus: assessors, collectors, courts to enforce payment, records to know who owes what. Building that apparatus is state-building, and nobody undertakes it for its own sake.
4. Taxpayers with anything worth taxing resist, so the ruler bargains. What they get in exchange for money is representation, legal protection of property, and eventually a claim on how the money is spent.

Tilly put the provocation in a 1985 essay: war making and state making are the largest examples of organised crime, in that a protection racket and a government differ mainly in whether the protector also manufactures the threat. He meant it analytically. The state that emerges from the process is not designed by anybody. It is the residue of a sequence of emergencies.

Why this matters: if states are built by the pressure of war, then everything downstream, including tax systems, bureaucracies and parliaments, is shaped by what rulers had to concede to get paid. That is a testable claim, and the English record tests it.

Following the money: England from 1688

England in the 1680s was a middling fiscal power. Then it acquired a Dutch king, a permanent continental war, and a set of financial institutions in quick succession: the settlement of 1688 and 1689, which gave Parliament control of supply; the founding of the Bank of England in 1694 to lend the government money; a funded national debt serviced by earmarked taxes; and an excise service that by the 1780s was among the most professional bureaucracies in Europe, with salaried inspectors, standard gauges and written procedures.

The numbers show what the machinery did. England's funded debt stood at roughly 17 million pounds in 1697 and about 850 million by 1815, more than twice national income, and it was serviced without default. That capacity, not the size of its population, is how a country of some ten million people fought France for most of a century and won. Note the order of causation Tilly predicts and the record supports: Parliament did not grow strong and then decide to fund wars. It funded wars and, in the bargaining over how, became the body that controlled the state.

Prussia ran the other version. Frederick William I, ruling about two and a half million subjects, built an army of some eighty thousand men, and the fiscal-military administration he created, the Generalkriegskommissariat, became the civil administration as well. The contemporary quip that Prussia was not a state with an army but an army with a state describes a real institutional fact: the tax service and the war ministry were the same organisation.

Two roads to a state, and the one that won

Tilly's 1990 book generalises this into a continuum. Rulers needed both coercive means and capital, and where they found them shaped what they built.

The third column is the answer to why the map ended with twenty-five units. City-states could buy soldiers but could not survive a long war of attrition; coercion-intensive empires could raise men but not credit. The combination beat both, which is why the national state and not the city-league is the form that spread, and why every unit in your dataset today is a descendant of the winners.

Herbst's objection: the same engine, a different geography

Jeffrey Herbst asked what Tilly's mechanism predicts where the inputs are different, and answered in States and Power in Africa in 2000. Precolonial Africa was land-abundant and people-scarce. In that setting the strategic problem is not seizing your neighbour's territory, which is worth little without people on it, but broadcasting authority outward from a centre across long distances at low population density. Rulers who lost a conflict could withdraw; subjects who disliked a ruler could leave. Neither situation generates Tilly's fiscal pressure, because there is no rival about to extinguish you.

Colonial rule then froze a particular set of boundaries. The Berlin Conference of 1884 and 1885 set the rules for partition, and the lines that resulted were drawn for European convenience. At independence, African states adopted the principle of uti possidetis, confirmed by the Organisation of African Unity in Cairo in 1964: colonial borders would stand. The consequence Herbst emphasises is that African states have been almost uniquely secure from conquest. Since 1957 the map has barely changed, and no African state has been eliminated by a neighbour. Security from extinction sounds like a gift. In Tilly's logic it removes the one pressure that forced European rulers to build tax systems and bargain with their populations.

What replaces it is a state that controls its capital and its export enclave and taxes lightly, if at all, in the hinterland. The Democratic Republic of the Congo covers 2.34 million square kilometres with a paved road network measured in the low thousands of kilometres. A state that cannot physically reach a district cannot count it, tax it or police it, and the missing roads are not an accident of poverty alone; they are the infrastructure that nobody was ever forced to build.

Measuring capacity, and what the measures miss

State capacity is the concept that has to carry all this, and it is measured badly. Here are the workhorse indicators and their failures.

The first row is worth dwelling on, because tax as a share of output is the single most used proxy in the field. Denmark's central government tax take runs above thirty percent of GDP; Nigeria's has sat in the high single digits for most of the last two decades. That gap is real and it is enormous. But part of the Nigerian figure reflects a state that could fund itself from oil rents collected at a handful of terminals, which is administratively closer to owning a well than to governing a population.

What would overturn Tilly's account

The mechanism predicts that external military threat should precede and produce fiscal capacity. Two findings would damage it. First, if careful sequencing showed that in most European cases tax administration was built before the wars it supposedly funded, the arrow reverses. Second, if the many post-1945 states facing serious external threats, such as Israel, South Korea, Taiwan and Pakistan, showed no systematic advantage in fiscal capacity over comparable states at peace, the mechanism would not travel beyond early modern Europe. Both are answerable with existing data, and the second is the more interesting, because a mechanism that only worked once is a historical description rather than a theory.

Common misconceptions

• The Peace of Westphalia created the sovereign state in 1648. The treaties settled religious and constitutional questions inside the Holy Roman Empire. The tidy story of a sovereignty system born in one year is a later invention, and Tilly's five centuries of attrition describe the actual process.
• Weak states are simply poor states. Capacity and wealth are correlated and distinct. Some poor states register births and collect taxes well; some middle-income resource exporters cannot do either, because oil paid for everything the tax office would otherwise have had to learn.
• African borders are arbitrary and that is why African states are weak. Herbst's argument is nearly the reverse. The borders are artificial and, crucially, guaranteed, and it is the guarantee against conquest rather than the shape of the lines that removes the pressure to build.

Putting it together

A state is a claim to monopolise legitimate force over a territory, and claims of that kind vary in how well they stick. European states were forged by an elimination tournament in which the price of survival was a tax system, and the price of a tax system was bargaining with the people who paid. Where that tournament did not run, because distance was the enemy rather than a neighbouring army, and later because the borders were guaranteed from outside, the resulting states look different in ways you can measure: in what share of output they collect, in whether they can find a birth in a rural district, in how much road there is.

The core of it: nobody set out to build the modern state. It is what was left standing after five centuries of rulers trying to pay for the next war, which is why its shape makes so much more sense as a residue than as a design.

Sources

1. Weber, M. (1946). Politics as a vocation. In H. H. Gerth & C. W. Mills (Eds.), From Max Weber: Essays in Sociology. Oxford University Press. (Original lecture 1919.)
2. Tilly, C. (1990). Coercion, Capital, and European States, AD 990-1992. Blackwell.
3. Herbst, J. (2000). States and Power in Africa: Comparative Lessons in Authority and Control. Princeton University Press.
4. Wikipedia contributors. (n.d.). Berlin Conference. Wikipedia
5. World Bank. (n.d.). Tax revenue (percent of GDP). World Development Indicators. data.worldbank.org

One country, three verdicts

In March 2021 the V-Dem Institute's annual report reclassified India from electoral democracy to electoral autocracy. That same season Freedom House moved India from Free to Partly Free. Polity5, whose series stops in 2018, had India at 9 on its scale of -10 to 10, a solid democracy. One country of 1.4 billion people, three teams of serious researchers, three different answers.

This is not a scandal, and it is not a reason to give up on measurement. It is a reason to know exactly what each index is measuring, because the disagreements are informative and they are not random. This lesson builds the definitions first, then the instruments, then reads the places where the instruments diverge.

Two definitions, and what each throws away

Joseph Schumpeter wrote the minimal definition in 1942: the democratic method is the institutional arrangement in which individuals acquire the power to decide by means of a competitive struggle for the people's vote. That is it. No requirement about equality, deliberation, or good outcomes. Schumpeter was deliberately draining the word of its aspirations, and the reason is methodological: if you define democracy to include good government, you cannot then investigate whether democracy produces good government, because you have made the answer true by construction.

Robert Dahl thought the minimal definition too thin, because a competitive struggle for votes can be staged in a country where the press is owned by the incumbent, opposition parties cannot register, and half the adults cannot vote. In Polyarchy in 1971 he set out two dimensions rather than one: contestation, how freely opposition can compete, and inclusiveness, how much of the adult population may take part. A regime high on both he called a polyarchy, using a deliberately awkward word to keep the ideal of democracy separate from the messy real systems that approximate it.

Dahl's institutional list has six items, and it is worth learning because every serious index is a way of operationalising it: elected officials control government decisions; elections are free, fair and frequent; citizens may express themselves without danger; alternative sources of information exist and are legally protected; citizens may form independent associations including parties; and citizenship is inclusive, extending these rights to essentially all adults.

Key idea: the two dimensions are separate, and the separation earns its keep historically. Britain in 1900 scored high on contestation and low on inclusiveness, since parties competed genuinely while most adults could not vote. Several one-party states of the 1970s had near-universal suffrage and no contestation at all. A single number hides which of the two a country is failing.

Four instruments, side by side

The Democracy-Dictatorship measure deserves a note because its rule is so unusual. Przeworski and colleagues classify a regime as democratic only if offices are filled by contested elections and, crucially, if the incumbent has actually lost and left at some point. The alternation rule solves a real problem, that a party can win fair elections repeatedly and you cannot tell whether it would have accepted defeat. It also creates one: Botswana and Japan spent decades misclassified because a dominant party kept winning votes it may well have surrendered had it lost them.

Where the instruments diverge, and why

Disagreements cluster in one place: regimes that hold real elections and simultaneously dismantle the conditions that make elections meaningful. Three kinds of divergence explain nearly all of it.

Conceptualisation. Polity asks about the executive: how is the chief executive recruited, how constrained is the office, how open is political competition. It does not directly score whether the government owns the television stations. So a government that leaves elections formally competitive while capturing the media and the courts can hollow out a democracy without moving Polity much. Hungary is the case that made this famous: Polity5 recorded Hungary at 10, its maximum, right through to the end of the series in 2018, while V-Dem's electoral democracy index for Hungary fell steeply across the 2010s and Freedom House moved Hungary out of the Free category in 2019, the first European Union member so classified.

Aggregation. Suppose a country scores well on suffrage and badly on press freedom. Add the components and the good score compensates for the bad one. Multiply them and a near-zero on any necessary component drags the whole index down. The choice is not technical: it encodes whether you think the components are substitutes or necessary conditions. Munck and Verkuilen showed in 2002 that indices rarely justify the choice, and that rankings move when it changes.

Coding and thresholds. V-Dem asks multiple experts per indicator per country-year and uses a measurement model that estimates coder reliability and returns a credible interval. Those intervals are wide enough that many published rankings are distinguishing between countries whose intervals overlap completely. Users who report the point estimate and drop the interval are throwing away the most honest part of the dataset.

Remember: when two indices disagree about a country, do not average them. Find out which of the three problems is producing the gap, because the answer tells you something substantive about the regime, usually that it is competitive in form and captured in substance.

Continuous or dichotomous

One more dispute, because it will recur every time this course uses a democracy variable. Should democracy be a threshold or a scale?

The case for a dichotomy: many theories are about a change of kind rather than degree. If your question is what causes transitions, you need a moment at which a country crosses over, and a continuous score gives you no event to explain. The case for a scale: countries clearly differ in how democratic they are, and forcing Denmark and Mexico into one category discards information you paid for. Collier and Adcock argued in 1999 that the choice should follow the research question rather than a general position, which sounds like a compromise and is in fact the strongest available answer. Studying breakdown, use a threshold; studying quality of representation, use a scale.

What would overturn the reliance on expert-coded indices

The central worry about V-Dem-style measurement is that experts read the newspapers, so their judgements might follow international coverage rather than events on the ground. Two tests would settle it. First, code a set of country-years from documentary records alone, blind to the period and the country's reputation, and see whether the blind codings match the expert ones. Second, check whether index movements systematically precede or follow major foreign press attention: if the score moves after the coverage rather than with the underlying event, the instrument is measuring attention. V-Dem publishes coder-level data, so the first test is possible for anyone with the patience, and if it failed badly this lesson would need rewriting.

Common misconceptions

• A country either is or is not a democracy, and indices just quantify the obvious. The hard cases are numerous, not marginal. Regimes with real elections and captured referees are the growth area of modern politics, and they are exactly where the instruments split.
• V-Dem is more objective than Freedom House because it uses more coders. More coders reduce idiosyncratic error and do not remove shared bias. What V-Dem adds is a model that estimates how much coders disagree and reports it, which is a different virtue: honesty about uncertainty, not the elimination of it.
• Defining democracy to include social equality makes the concept richer. It makes it unusable for the questions comparativists ask. Build outcomes into the definition and you cannot test whether democracy produces those outcomes.

What to remember

Democracy is defined minimally by competition for votes, and more usefully by Dahl's two dimensions of contestation and inclusiveness with their six institutional guarantees. Four instruments operationalise those ideas differently: Polity through executive recruitment and constraints, Freedom House through rights as experienced, V-Dem through several thousand expert-coded indicators with explicit uncertainty, and Democracy-Dictatorship through a binary rule that requires an observed alternation. Their disagreements concentrate on regimes that keep the elections and capture everything around them, and each disagreement can be traced to conceptualisation, aggregation or coding.

Worth holding on to: before using any democracy score in an argument, open its codebook and find out what would have to change in a country for the number to move. If the answer does not include the thing you are studying, you have the wrong instrument.

Sources

1. Dahl, R. A. (1971). Polyarchy: Participation and Opposition. Yale University Press.
2. Schumpeter, J. A. (1942). Capitalism, Socialism and Democracy. Harper & Brothers.
3. Munck, G. L., & Verkuilen, J. (2002). Conceptualizing and measuring democracy: Evaluating alternative indices. Comparative Political Studies, 35(1), 5-34.
4. V-Dem Institute. (n.d.). The V-Dem Dataset. University of Gothenburg. v-dem.net
5. Wikipedia contributors. (n.d.). Polity data series. Wikipedia

The richest democracy ever to fall

On 24 March 1976 the Argentine armed forces removed President Isabel Peron and began seven years of dictatorship that killed thousands of people. Argentina in 1975 had a per capita income of about 6,055 international dollars in 1985 prices. Two decades later, when Adam Przeworski and his collaborators assembled regime and income data for 135 countries from 1950 to 1990, that figure turned out to be the ceiling: in their entire sample, no democracy with a higher income had ever collapsed. Argentina marks the edge of the map.

Around that fact sits the longest-running argument in comparative politics. Rich countries are overwhelmingly democracies; poor countries mostly are not. Everyone agrees on the correlation. What it means has been fought over since 1959, and the fight is worth following in detail, because it is the clearest example in the field of a dispute that turned on how the question was specified rather than on new data.

Lipset's claim, and the evidence he actually had

Seymour Martin Lipset published Some Social Requisites of Democracy in the American Political Science Review in 1959. His claim was blunt: the more well-to-do a nation, the greater the chances that it will sustain democracy. His evidence was a set of cross-tabulations. He split European and English-speaking countries into stable democracies and unstable democracies or dictatorships, split Latin American countries into democracies and stable dictatorships, and compared the group averages on wealth, industrialisation, urbanisation and education. The stable democracies were richer on every indicator.

The mechanism he proposed was social, not economic. Wealth creates a large middle class that moderates conflict and rewards centrist parties; it raises education, which raises tolerance; it produces cross-cutting associations that stop politics dividing along one line; and it lowers the stakes of losing, since the losers of an election in a rich country do not thereby lose their livelihoods. This became modernisation theory, and in its stronger versions it implied a sequence: get rich, then become democratic.

The split that changed the question

Przeworski and Fernando Limongi asked in 1997 what a correlation between wealth and democracy could be made of. Two stories fit it equally well.

• Endogenous democratisation. Development causes dictatorships to become democracies. Wealth accumulates, a middle class forms, the regime liberalises, and the country crosses over.
• Exogenous democratisation. Countries become democracies for reasons unrelated to income, such as defeat in war, colonial withdrawal or elite splits, and once democratic, rich ones survive while poor ones die. Over time the world sorts itself into rich democracies and poor dictatorships without development having caused a single transition.

Both stories predict the correlation. They differ entirely in what a development agency should expect from raising incomes. Przeworski and Limongi separated the outcomes, as Lesson 2 described, and found that the probability of a dictatorship democratising in a given year did not rise steadily with income, while the probability of a democracy dying fell sharply and reached essentially zero at the top of the income range. That is the exogenous story, and it was a serious blow to the sequence Lipset's followers had drawn.

So what?: the policy implication reverses. If democratisation is exogenous, then getting a dictatorship rich does not make it democratise, and the case for supporting new democracies through their poor early years becomes the strongest thing you can do.

Boix and Stokes reply: look at your sample

Carles Boix and Susan Stokes answered in 2003 with an argument about which country-years were in the data. Przeworski's sample runs from 1950. Almost every rich democracy in the world democratised before 1950: Britain, France, the United States, the Nordic countries, the Netherlands, Belgium, Switzerland, Australia. If development caused those transitions, the evidence for it lies entirely outside the window being examined. Their reconstruction pushes the data back to 1800 and finds that development does raise the probability of a transition, and that the effect was strong in the nineteenth and early twentieth centuries.

They add a selection argument. In the post-1950 sample, dictatorships that got rich and were about to democratise are underrepresented, because the rich ones that would have democratised already had. Correct for that and the exogenous result weakens.

This is a good dispute because both sides are right about something specific. Przeworski is right that in the second half of the twentieth century, rich dictatorships did not reliably become democracies, and the modern cases prove it: Singapore's income per head exceeds that of most European democracies and it has not had an alternation in government since 1959. Boix is right that a sample beginning after most of the democratising happened cannot be used to say democratisation does not happen.

Acemoglu and Robinson: neither, and here is the reason

Daron Acemoglu, Simon Johnson, James Robinson and Pierre Yared went at the correlation with a different instrument in 2008. They added country fixed effects, which means asking whether a country becomes more democratic in the years when it becomes richer, rather than whether rich countries are more democratic than poor ones. The cross-country association is strong. The within-country association essentially vanishes.

Their interpretation is that both wealth and democracy are outcomes of something older. Societies passed through critical junctures, in their account largely in the early modern and colonial periods, that set them on divergent paths, and those paths produced both prosperity and representative institutions in some places and neither in others. Income does not cause democracy; a common ancestor causes both.

Their own theory of democratisation, set out in 2006, does not run on income at all. It runs on threat and commitment. Elites extend the franchise when the disenfranchised can credibly threaten revolution and when a promise of future redistribution would not be believed. Giving away the vote is a way of making the promise credible, because institutions bind future governments in a way that a ruler's word does not. Inequality enters non-monotonically: where inequality is very low the poor do not press hard, where it is very high elites will fight rather than concede, and democratisation is likeliest in between.

The four positions on one page

Read the last column, because it is what makes this a dispute rather than a debate. Each position has committed to something that could go against it.

What would settle it

Three things would move this argument more than another regression. First, better pre-1900 regime data, coded consistently, would let the Boix and Stokes claim be tested rather than asserted; V-Dem's series back to 1789 is the current best attempt and its early coverage is thin. Second, a genuine natural experiment in income, such as a large unanticipated resource discovery or a commodity price shock, offers within-country variation not confounded by institutions, and the existing studies of oil windfalls mostly point away from democratisation. Third, and most decisive, the arrival of the first case: a wealthy, non-resource-dependent, one-party state that democratises without defeat, elite split or external pressure. Singapore is the obvious candidate. Until such a case appears, Przeworski's null on transitions is holding up better than modernisation theory's supporters expected.

Common misconceptions

• Przeworski showed that development does not matter for democracy. He showed something narrower and more useful: development does not appear to cause transitions in the post-1950 window, while it strongly protects democracies once they exist. The second half is a large effect, not a null.
• Adding country fixed effects gives the true causal estimate. Fixed effects answer a different question, about change within countries, and if the real effect operates over generations rather than years, a within-country specification will find nothing even when the effect is real.
• Rich dictatorships must be oil states. Singapore has no oil. The awkwardness of that case for modernisation theory cannot be dismissed as a resource-curse exception.

The takeaway

The correlation between wealth and democracy is one of the most robust facts in the discipline, and sixty years of work has not settled what produces it. Lipset saw wealth building the social conditions for democratic politics. Przeworski and Limongi split the outcome and found wealth preserving democracies rather than creating them. Boix and Stokes showed that the sample excluded the transitions that mattered. Acemoglu and Robinson removed cross-country variation and watched the relationship disappear, then proposed a theory driven by credible threat rather than income at all.

The upshot: when a correlation survives everything and its interpretation survives nothing, the problem is almost never the data. It is that a single association is being asked to answer several different questions, and the useful move is to name which one you are asking.

Sources

1. Lipset, S. M. (1959). Some social requisites of democracy: Economic development and political legitimacy. American Political Science Review, 53(1), 69-105.
2. Przeworski, A., & Limongi, F. (1997). Modernization: Theories and facts. World Politics, 49(2), 155-183.
3. Boix, C., & Stokes, S. C. (2003). Endogenous democratization. World Politics, 55(4), 517-549.
4. Acemoglu, D., Johnson, S., Robinson, J. A., & Yared, P. (2008). Income and democracy. American Economic Review, 98(3), 808-842.
5. Wikipedia contributors. (n.d.). Modernization theory. Wikipedia
6. Roser, M. (n.d.). Economic Growth. Our World in Data. ourworldindata.org

Fifty-three percent of the vote, and a constitution

On 25 April 2010 the Hungarian alliance of Fidesz and the Christian Democrats won 52.7 percent of the party-list vote. Under the mixed electoral system then in force, that produced 263 of 386 parliamentary seats. Two-thirds of 386 is 258. They had five seats more than they needed to rewrite the constitution by themselves, and within a year they had written a new one.

No tanks moved. Nobody suspended parliament. The 2010 election was free, and international observers said so. What followed is the way democracies now end: not by seizure but by an elected government using its majority to dismantle the constraints on itself. This lesson follows that process through three countries and puts the older literature on transitions alongside it, because the two together explain why the classic warning signs no longer fire.

What a transition used to look like

The study of regime change was built in the 1980s on a different set of cases. Guillermo O'Donnell and Philippe Schmitter, writing in 1986 about southern Europe and Latin America, produced a vocabulary the field still uses.

They separate liberalisation, in which an authoritarian regime relaxes repression, permits some press freedom and releases prisoners, from democratisation, in which it accepts rules for choosing who governs. The first does not lead automatically to the second, and rulers who begin the first usually intend to stop short of it.

Their causal engine is a split inside the regime. Hardliners believe repression can hold; softliners believe some opening is needed to preserve what matters to them, often their own immunity and the army's position. In the opposition, moderates will negotiate and radicals will not. A transition becomes possible when softliners and moderates can strike a bargain that hardliners cannot veto and radicals cannot destroy. Spain's Moncloa Pacts of 1977 and Poland's Round Table talks of February to April 1989 are the model cases: negotiated, partial, and explicitly protective of some interests of the outgoing regime.

Why this matters: the transitions literature treated uncertainty as the defining feature. Nobody in the room knew who would win the founding election, and it was precisely that ignorance that made a bargain rational. When one side is confident of winning, the bargain does not form.

The old way democracies died, and the new way

Nancy Bermeo drew the contrast sharply in 2016. The classic route out of democracy was the coup: the Chilean armed forces on 11 September 1973, the Argentine junta in 1976, dozens of others. That route has become rare. What replaced it she calls executive aggrandisement, plus strategic manipulation of elections. The distinguishing features are that it is legal in form, incremental, and carried out by people who were elected.

This is the diagnostic problem. Every step is defensible in isolation. Reforming a court is a normal legislative act. Redrawing constituencies is a normal legislative act. Regulating broadcasting is a normal legislative act. Only the sequence has a direction, and by the time the direction is undeniable the means of reversing it have usually been altered too.

Hungary, step by step

Watch the sequence in one country. After April 2010, the government of Viktor Orban moved on four fronts.

1. The constitutional court. Its membership was enlarged and the appointment procedure changed so the governing majority could fill it alone; its jurisdiction over budget and tax legislation was restricted after it struck down a retroactive tax.
2. The constitution itself. A new Fundamental Law was passed in April 2011 and took effect on 1 January 2012, written and adopted by the governing majority without opposition participation.
3. Media. A media council with regulatory and licensing powers was created with members drawn from the governing party's nominees, and the public broadcasters were merged under a single news agency.
4. The electoral law. Parliament was cut from 386 seats to 199, the single-round majoritarian tier was strengthened, and constituency boundaries were redrawn.

The last step is the one to work as arithmetic, because it shows what an electoral system can be made to do. In 2014 the governing alliance won 44.9 percent of the party-list vote. Two-thirds of 199 seats is 132.67, so 133 seats are needed for a constitutional majority. Fidesz won exactly 133. A vote share that had fallen by nearly eight points produced the same constitutional power, because the rules had been rewritten in between. In 2018, on 49.3 percent, it won 133 again.

Nothing here was concealed. All of it was published in the official gazette.

Venezuela: the same sequence, a different order

Hugo Chavez won the Venezuelan presidency in December 1998 with 56.2 percent of the vote, having campaigned on replacing a party system that most Venezuelans regarded as a closed cartel. In July 1999 elections to a constituent assembly, his allies took 121 of 131 seats, a share of seats far above their share of votes because the assembly was elected by a majoritarian formula. The assembly wrote a new constitution, dissolved the sitting congress and reorganised the judiciary. In 2004 the Supreme Tribunal was enlarged from 20 to 32 justices, with the new places filled by a simple legislative majority.

Two features distinguish Venezuela from Hungary. The order is different: Chavez rebuilt the constitutional order first and won subsequent elections from inside it, whereas Orban won a supermajority under existing rules and then changed them. And the resource base is different, since oil revenue funded social programmes that sustained electoral support without the bargaining that taxation requires, which is a thread Module 6 picks up.

Poland, and the fact that the sequence can run backwards

In October 2015 Law and Justice won 37.6 percent of the Polish vote and 235 of 460 seats, an absolute majority on rather less than two-fifths of the vote, again through a seat-allocation rule that rewards the largest party. Within weeks it was in open conflict with the Constitutional Tribunal over judicial appointments; a public media law followed in December 2015; laws restructuring the ordinary courts and the National Council of the Judiciary followed in 2017 and 2018.

Then, in October 2023, turnout reached 74.4 percent, the highest in any Polish election since 1989, and the governing party lost its majority. A change of government followed. That single fact should be held against every account of democratic backsliding that treats it as one-directional. Poland shows the erosion is real, and it shows that where elections remain competitive enough to lose, they can still be lost.

What is actually being destroyed

Steven Levitsky and Daniel Ziblatt argue in How Democracies Die that written rules never sufficed, and that two unwritten norms did the real work.

Mutual toleration is accepting that your opponents are legitimate rivals rather than enemies of the nation. Institutional forbearance is declining to use the full legal powers available to you. A president may in principle pardon allies, pack a court to any size the statute allows, or govern by decree; forbearance is the practice of not doing so. Both norms are invisible while they hold, because their operation consists of things not happening.

Their second argument is about gatekeeping. Established parties, not voters, are the filter that historically kept demagogues off ballots, and the filter fails when a mainstream party calculates that an extremist is a useful ally. They give four warning signs of an authoritarian politician: rejecting the democratic rules of the game, denying the legitimacy of opponents, tolerating or encouraging violence, and showing readiness to curtail opponents' civil liberties. The signs are checkable in advance, which is the point of listing them.

In short: what fails first is not a law. It is the practice of not doing the legally permitted thing, and no constitution can require its own restrained use.

What would overturn this account

The claim is that backsliding proceeds by legal, incremental executive aggrandisement and that norms fail before institutions do. Two findings would damage it. First, if a systematic coding of recent cases showed that most democratic declines still involve overt illegality, such as annulled elections or arrested opposition leaders, then the emphasis on legality would be misplaced and the older breakdown literature would still be the right tool. Second, if a set of countries showed courts and electoral commissions being captured while mutual toleration and forbearance remained intact in surveys of elite behaviour, the ordering would be wrong. Bermeo's coding supports the first claim on the cases to date, and the norm-based ordering is much harder to test, which is the honest weakness of this literature.

Common misconceptions

• Backsliding means elections stop being held. In every case here elections continued, were contested and were reported. The manipulation happens upstream of election day, in courts, districting, broadcasting and campaign finance.
• A written constitution with strong checks prevents this. Hungary's post-1989 constitutional order was widely admired, and it contained a two-thirds amendment rule that a single election delivered to one alliance. Rules are only as strong as the coalition that will not use them.
• Once erosion begins it cannot be reversed at the ballot box. Poland in 2023 is the counterexample, and it matters because it locates the threshold: reversal is possible while elections remain competitive, which is precisely what the manipulation is aimed at ending.

Looking back

The transitions of the 1970s and 1980s were bargains struck under uncertainty between softliners inside a regime and moderates outside it. The breakdowns of the last two decades are something else: elected majorities converting temporary electoral advantage into permanent institutional advantage, one lawful step at a time. Hungary shows how an electoral law can hold constitutional power steady while support falls eight points. Venezuela shows the same destination reached by rewriting the order first. Poland shows both that the sequence runs and that it can be interrupted by a turnout of 74 percent.

Remember: ask of any democracy under strain not whether elections are held, but whether the referees are still independent, whether the rules were changed by the people they govern, and whether losing remains possible.

Sources

1. O'Donnell, G., & Schmitter, P. C. (1986). Transitions from Authoritarian Rule: Tentative Conclusions about Uncertain Democracies. Johns Hopkins University Press.
2. Levitsky, S., & Ziblatt, D. (2018). How Democracies Die. Crown.
3. Bermeo, N. (2016). On democratic backsliding. Journal of Democracy, 27(1), 5-19.
4. Wikipedia contributors. (n.d.). 2010 Hungarian parliamentary election. Wikipedia
5. V-Dem Institute. (n.d.). The V-Dem Dataset. University of Gothenburg. v-dem.net

Two votes against

On 11 March 2018 the National People's Congress of China voted to remove the two-term limit on the state presidency from the constitution. The count was 2,958 in favour, two against, three abstentions. The limit had been in the 1982 constitution, and under it China had managed two orderly handovers of leadership, in 2002 and 2012, which is more than most authoritarian regimes ever achieve.

So here is the puzzle this lesson works. Term limits and managed succession look like a solution to the deepest problem a dictatorship has. Why would a regime that had solved it undo the solution? Answering that requires knowing what the problem is, and the modern literature on authoritarian politics gives an unusually clean answer.

The problem a dictator actually has

Milan Svolik's The Politics of Authoritarian Rule begins with a statistic that reorients the whole subject. Among authoritarian leaders who lost power between 1946 and 2008 by means other than natural death or a constitutional exit, roughly two-thirds were removed by people inside their own regime: ministers, generals, party colleagues. Popular uprisings and foreign invasions together account for a much smaller share.

That fact splits authoritarian politics into two problems, and they pull in opposite directions.

The problem of authoritarian control is the relationship between the ruling elite and everyone else: how to prevent the population from organising against the regime. Its tools are repression, censorship, surveillance, and co-optation through jobs, subsidies and patronage.

The problem of authoritarian power-sharing is the relationship between the dictator and the allies who put him in office. Those allies want assurance that they will keep their share. The dictator would prefer to concentrate power, and he has an advantage in doing so gradually. Nobody can enforce a bargain between them, because there is no court above the ruling group. So the allies watch for signs of encroachment, and because they can only act while they still have the means, they are prone to strike early. This is why so many coups come from the inner circle.

Key idea: the tools that solve one problem often worsen the other. A powerful secret police controls the population and also gives the dictator a means of eliminating his own coalition, which is exactly what makes his allies nervous.

Why a dictator builds a party

Barbara Geddes, with Joseph Wright and Erica Frantz, sorted authoritarian regimes since 1946 into a small number of types according to who controls access to office and policy. The types differ systematically in how long they last and how they end.

The military row explains itself once you see it in Svolik's terms. Officers have somewhere else to go. The army as an institution has interests, a budget and a career structure that survive a return to barracks, so when holding office starts to damage the institution, withdrawal is a real option. A single-party regime has no such fallback, which is why its members fight harder and why the party invests in rules, congresses and promotion ladders that make power-sharing credible without any external enforcer.

Jennifer Gandhi extended this to legislatures. Authoritarian parliaments are not decoration; they are a place where concessions can be made visibly, in a form that opponents can verify and that the regime cannot quietly retract. That is a solution to a commitment problem, which is why the dictatorships with legislatures and parties tend to be the ones that survive.

China, and the reopening of a solved problem

Return to the puzzle. The post-1978 Chinese arrangement was close to a textbook answer to power-sharing: fixed terms, an age norm for retirement from the Politburo Standing Committee, a designated successor identified in advance, and collective leadership that limited any one figure's ability to purge the others. Two successions occurred on schedule.

The 2018 amendment removes the term limit for the state presidency, and the concentration of authority under Xi Jinping has gone considerably further than that one clause: an anti-corruption campaign that disciplined a very large number of officials, the elevation of the leader's ideology into the party constitution, and no publicly designated successor. In the Geddes framework this is movement along the axis from single-party toward personalist rule, and the framework has a prediction attached. Personalist regimes are more prone to policy error, because subordinates stop delivering bad news; they are more likely to end violently; and they have no agreed procedure for the day the leader dies. The prediction is testable, and the test will take twenty years.

Elections nobody can lose, and why they are held anyway

Most authoritarian regimes today hold elections. They are not simply theatre, and three functions explain them.

• Information. A regime without a free press does not know which regions are disaffected or which local officials are hated. An election that the ruling party wins everywhere, with varying margins, is a map of where its machine is weak.
• Co-optation. Seats, nominations and the patronage attached to them are goods that can be distributed to buy loyalty from potential rivals, and a formal contest is a way of allocating them.
• Demonstration. An enormous winning margin tells the opposition that resistance is pointless. Beatriz Magaloni's account of Mexico under the Institutional Revolutionary Party, which held the presidency from 1929 to 2000, turns on this: the party spent heavily to win by margins far larger than it needed, because the margin itself was the deterrent.

When the tilt is real but the contest is not fake

Steven Levitsky and Lucan Way named the category that swallowed much of the post-Cold-War world. In a competitive authoritarian regime, formal democratic institutions exist and are the principal route to power, but incumbents abuse the state so systematically that the playing field is not level: state media, state resources, selective prosecution, harassment of donors. The regime is not a democracy, because the contest is unfair. It is not a closed autocracy either, because the opposition genuinely contests four arenas and occasionally wins in one of them: elections, the legislature, the judiciary and the media.

The word occasionally is doing real work. Alberto Fujimori dissolved the Peruvian Congress in 1992 and left the country in 2000 amid a bribery scandal broadcast on video. Malaysia's governing coalition held power without interruption from 1957 until it lost the general election of May 2018. Cases like these are why the category is not just a polite name for dictatorship: incumbents in competitive authoritarian regimes really can lose, which is why they work so hard on the tilt.

The point: the interesting question about a modern autocracy is rarely whether it holds elections. It is which of the four arenas the opposition can still use, and what it costs the regime to close each one.

What would overturn this account

The framework rests on the claim that the main threat to a dictator comes from inside. If a recount of leader exits over the last twenty years showed that mass protest now removes more autocrats than insiders do, the emphasis would need to shift, and the wave of protest-driven exits since 2010 makes that a live possibility rather than a rhetorical one. The Geddes prediction about personalisation is likewise falsifiable: if personalist regimes over the next two decades prove no more prone to violent collapse and policy disaster than single-party ones, the typology is describing labels rather than mechanisms.

Common misconceptions

• Dictators are constrained by nothing. They are constrained by the people who keep them in power, and that constraint is enforced by the credible threat of a coup rather than by any court. It is a real constraint and an unreliable one.
• Institutions in autocracies are window dressing for foreign audiences. Parties and legislatures make concessions verifiable and distribute spoils predictably. Regimes that have them last longer, which is a strong hint that they do something.
• Personalist rule is the strongest form of dictatorship. It is the most unconstrained and among the most fragile. Removing the mechanisms that reassure your allies removes the mechanisms that would have kept them loyal after your death.

What you now know

Authoritarian politics is two games at once: controlling a population and reassuring a coalition. Most dictators fall to the second. Regime types differ in how they manage it, with parties supplying credible power-sharing and long tenure, military juntas supplying an exit route and short tenure, and personalist rule supplying neither. Elections and legislatures under autocracy gather information, distribute patronage and demonstrate futility. And a large class of contemporary regimes sits in the middle, competitive enough that the opposition sometimes wins, unfair enough that it usually does not.

In short: China removed a term limit in 2018 by 2,958 votes to two. Read that as a regime trading a solution to its power-sharing problem for a concentration of control, and you have the whole framework in one vote.

Sources

1. Svolik, M. W. (2012). The Politics of Authoritarian Rule. Cambridge University Press.
2. Geddes, B., Wright, J., & Frantz, E. (2018). How Dictatorships Work: Power, Personalization, and Collapse. Cambridge University Press.
3. Levitsky, S., & Way, L. A. (2010). Competitive Authoritarianism: Hybrid Regimes After the Cold War. Cambridge University Press.
4. Gandhi, J. (2008). Political Institutions under Dictatorship. Cambridge University Press.
5. Wikipedia contributors. (n.d.). Competitive authoritarianism. Wikipedia
6. Herre, B. (n.d.). Democracy. Our World in Data. ourworldindata.org

The same share of the vote, twice

On 7 May 2015 the United Kingdom Independence Party won 12.6 percent of the votes cast in the British general election, about 3.88 million of them. It won one seat out of 650. On 24 September 2017 the Alternative for Germany won 12.6 percent of the German second votes. It won 94 seats out of 709.

The same share of a national electorate. One seat against ninety-four. Nothing about the two parties explains the difference; the rules do all of it. This lesson works both results as arithmetic, then generalises. Bring a pen: everything here is calculable by hand, and doing it once is worth reading about it three times.

Britain in 2015, worked

Single-member plurality, sometimes called first past the post, divides the country into districts and gives each one to whoever gets the most votes there. No account is taken of anything else. Here is what that produced.

Read the last two rows against each other, because they contain the mechanism. The SNP took 4.7 percent of the national vote and 56 seats. UKIP took 12.6 percent and one. UKIP had roughly two and a half times as many votes and one fifty-sixth as many seats.

The difference is geography. The SNP's votes were concentrated: it contested only Scotland's 59 seats and won 50 percent of the Scottish vote, which under plurality delivers nearly everything. UKIP's votes were spread almost evenly across England, finishing second in well over a hundred constituencies and first in one. Under plurality, coming second everywhere is worth precisely nothing.

A single number summarises this. The Gallagher least-squares index takes the difference between each party's vote share and seat share, squares it, sums the squares, halves the total and takes the square root. For 2015 it comes out around 15, high by any democratic standard; most proportional systems sit between 1 and 5. The squaring is deliberate: it makes one enormous discrepancy count for more than several small ones, which is what a voter actually experiences.

Note the constitutional consequence, since it recurs. A party with 36.8 percent of the vote received an outright majority of seats and governed alone for five years. Majoritarian systems are defended on exactly this ground: they manufacture governing majorities.

Germany in 2017, worked

Germany's system gives each voter two ballots. The first elects a representative in one of 299 districts by plurality, exactly as in Britain. The second is cast for a party list, and it is the second vote that determines how many seats each party gets in total. District winners are seated first, and list seats are added until each party's total matches its second-vote share. This is mixed-member proportional: it looks half majoritarian and is proportional in its result.

Every seat share sits within about two points of the vote share. The Bundestag elected in 2017 had 709 members rather than its nominal 598, because the CDU and CSU won 231 of the 299 districts, far more than their second-vote share entitled them to, and levelling seats were added to every other party until proportionality was restored. Overhang is the price of grafting district representation onto a proportional result.

What matters here: the German voter casting a second vote for a party polling 12 percent gets roughly 12 percent of the parliament. The British voter doing the equivalent in 2015 got one seat in 650. That is not a difference in political culture. It is a difference in an allocation formula.

The threshold, and what it destroys

Germany's proportionality has a hard edge: a party must clear 5 percent of second votes, or win three districts outright, to receive any list seats at all. The Free Democrats scored 4.8 percent in 2013 and took zero seats. In 2017 they scored 10.7 percent and took eighty. In that 2013 election, close to a sixth of all valid second votes went to parties that elected nobody, the highest such figure in the Federal Republic's history.

Thresholds exist because pure proportionality can fragment a parliament to the point where no government can be formed, and the German rule was written with the Weimar Republic's very fragmented assemblies in mind. The trade is explicit: a threshold buys governability with disenfranchisement, and 5 percent is a choice rather than a law of nature.

Two divisor formulas, worked side by side

Within proportional systems, the formula still matters. Take one district with 10 seats and 100,000 votes: A 47,000, B 33,000, C 12,000, D 8,000.

Under D'Hondt, divide each party's votes by 1, 2, 3 and so on, and award the ten seats to the ten highest quotients. A's quotients are 47,000, 23,500, 15,667, 11,750, 9,400. B's are 33,000, 16,500, 11,000, 8,250. C's are 12,000, 6,000. D's are 8,000, 4,000. The ten largest are, in order, 47,000, 33,000, 23,500, 16,500, 15,667, 12,000, 11,750, 11,000, 9,400 and 8,250. Count them by party: A takes 5, B takes 4, C takes 1, D takes none.

Under Sainte-Lague, which Germany has used since 2009, the divisors are 1, 3, 5, 7 and so on. A's quotients become 47,000, 15,667, 9,400, 6,714, 5,222; B's 33,000, 11,000, 6,600; C's 12,000, 4,000; D's 8,000, 2,667. Taking the ten largest gives A 5, B 3, C 1 and D 1.

One seat moves from the second-largest party to the smallest. That is the whole difference between the two formulas, and it is systematic rather than random: dividing by odd numbers penalises the large parties earlier, which is why D'Hondt is preferred by governments that expect to be large.

Duverger, and why the SNP is not a counterexample

Maurice Duverger proposed in 1951 that single-member plurality tends toward two-party competition, and that proportional representation tends toward multipartism. Duverger's law runs on two effects. The mechanical effect is the arithmetic above: small parties convert votes into seats at a punitive rate. The psychological effect is what voters, donors and candidates do once they understand the mechanical effect: they abandon parties that cannot win locally, which makes those parties weaker still.

Britain in 2015 elected members from more than half a dozen parties, so is the law refuted? No, and understanding why is the point. Gary Cox showed in 1997 that the law operates in the district, not the nation. In a district electing M members, the equilibrium number of viable candidates is M plus one, so a single-member district tends toward two. Aggregating districts nationally only produces two national parties when the same two are viable everywhere. Where a regionally concentrated party is one of the two viable options in its own area, as the SNP is in Scotland, national multipartism coexists with district-level two-party competition. UKIP, viable second almost everywhere and first almost nowhere, is the law working exactly as advertised.

What the systems do to representation

Arend Lijphart's Patterns of Democracy assembled these comparisons across thirty-six democracies and concluded that consensual systems perform at least as well as majoritarian ones on governing effectiveness while doing better on representation and participation. Not everyone accepts the conclusion, and the causal direction is contested: countries that were already consensual may have chosen proportional rules rather than been shaped by them.

What would overturn the Lijphart claim

The proposition that proportional rules produce higher turnout and broader representation is a cross-national correlation of exactly the kind Lesson 2 warned about. It would be damaged by within-country evidence: cases where a country changed its electoral system and the predicted effects failed to follow. New Zealand's switch from plurality to mixed-member proportional in 1996 is the cleanest natural experiment available, and Japan's 1994 reform is a second. If turnout and the diversity of representation did not move in the predicted directions after those changes, the cross-national pattern is telling us about the countries rather than the rules.

Common misconceptions

• Proportional representation is fairer, full stop. It is more proportional, which is one criterion. It also weakens the link between a vote and a government, since coalitions are assembled after the votes are counted by parties negotiating in private.
• First past the post always produces a majority government. It usually manufactures one, and it failed in Britain in 2010 and 2017. The seat bonus depends on how the vote is distributed, not only on its size.
• Duverger's law is refuted by any country with more than two parties. The law describes district-level competition. National multipartism under plurality is what regional concentration produces, and it is consistent with the law rather than against it.

Summing up

Two parties took 12.6 percent of the vote in two neighbouring democracies and received one seat and ninety-four. Plurality rewards concentration and punishes dispersion, manufactures majorities from pluralities of the vote, and produced a Gallagher index of about 15 in Britain in 2015. Proportional systems track vote shares closely, with the deviation set by district magnitude, the divisor formula and the threshold: 4.8 percent bought the Free Democrats nothing in 2013, and 10.7 percent bought them eighty seats four years later.

The upshot: when someone tells you an electoral result shows what a country believes, ask what the same votes would have produced under a different formula. You can now calculate the answer, and it is often a different country.

Sources

1. Duverger, M. (1954). Political Parties: Their Organization and Activity in the Modern State. Methuen.
2. Cox, G. W. (1997). Making Votes Count: Strategic Coordination in the World's Electoral Systems. Cambridge University Press.
3. Lijphart, A. (2012). Patterns of Democracy: Government Forms and Performance in Thirty-Six Countries (2nd ed.). Yale University Press.
4. Die Bundeswahlleiterin. (n.d.). Bundestag election 2017: Results. bundeswahlleiterin.de
5. Wikipedia contributors. (n.d.). 2015 United Kingdom general election. Wikipedia

Forty-four percent to under five

In October 2009 the Panhellenic Socialist Movement, PASOK, won 43.9 percent of the Greek vote and an absolute majority in parliament. It had been one of the two poles of Greek politics since 1974. In January 2015 it won 4.7 percent. In five years a party that half the country had voted for lost roughly nine tenths of its support, and the collapse was permanent.

Party systems are supposed to be the stable part of politics. Governments come and go; the alignment of social groups behind parties was thought to persist for generations. This lesson explains why that was once true, what mechanism produced it, and what has been dismantling it since roughly 1970.

Two revolutions, four cleavages

Seymour Martin Lipset and Stein Rokkan published in 1967 the argument that still organises the field. European party systems, they said, were not built around whatever issues happened to be current. They were built around four durable cleavages, produced by two historical upheavals.

The national revolution, meaning the construction of centralised states, produced two. Centre against periphery set the state-building core against populations with their own language or identity: Catalonia against Madrid, Scotland against London, Flanders against Brussels. State against church set secularising governments against religious authorities over schools, marriage and burial, and it is why continental Europe has Christian Democratic parties and Britain does not.

The industrial revolution produced two more. Land against industry set agricultural interests against urban manufacturing over tariffs and grain prices, which is why Scandinavia still has agrarian parties. Owner against worker set capital against labour and produced the socialist and social democratic parties that dominated the twentieth century.

The insight is that a cleavage is not an opinion. It requires three things at once: an objective social division, a sense of collective identity attached to it, and an organisation, usually a party or a union, that mobilises it. Divisions without organisations do not become cleavages, which is why some obvious social differences never structure party competition at all.

The freezing hypothesis

Then came the claim that made the article famous. Lipset and Rokkan observed that the party systems of the 1960s reflected, with few exceptions, the cleavage structures of the 1920s. Parties formed when the franchise was extended had captured their electorates and locked them in, through unions, church associations, party newspapers and the dense social life that surrounded them. Voters inherited party loyalty from their parents along with everything else.

The mechanism matters more than the metaphor. Freezing was not inertia; it was organisational encapsulation. A Dutch Catholic in 1950 bought Catholic bread from a Catholic shop, belonged to a Catholic union, read a Catholic newspaper and voted Catholic. Dismantle the shops, the unions and the newspapers, and the vote comes loose.

Why this matters: if the freezing hypothesis is right, party systems are historical residues rather than current opinion, and comparing party competition across countries means comparing the timing of state building, secularisation and industrialisation. That is a very different research programme from asking voters what they think.

The thaw, in numbers

Everything the freezing mechanism relied on has weakened, and the decline is measurable.

• Membership. Across established European democracies, party membership as a share of the electorate has fallen from around a tenth in the 1960s to under a twentieth, and in several countries below three percent. The organisations that encapsulated voters no longer contain many of them.
• Identification. The share of voters reporting a stable attachment to a party has fallen across almost every long-running national election study.
• Volatility. The Pedersen index, which sums the absolute changes in party vote shares between two elections and halves the total, has reached postwar records across western Europe.
• Collapse. PASOK from 43.9 to 4.7 percent. The Dutch Labour Party from 24.8 percent and 38 seats in 2012 to 5.7 percent and 9 seats in 2017. The French Socialist candidate taking 6.4 percent in the first round of the 2017 presidential election, five years after a Socialist had won the presidency.

Two explanations compete. Russell Dalton's dealignment account holds that rising education and mass media gave voters the capacity to judge issues for themselves, so they stopped needing a party to think for them. Peter Mair's cartel account holds that parties themselves withdrew: as they moved from society into the state, funded by public subsidy rather than dues and converging on a narrow policy range, they stopped offering voters much to be loyal to. The two are not exclusive, and the second explains the direction of the anger better than the first.

Counting parties properly

Before comparing party systems you need a number, and the raw count is useless: a system with two large parties and eleven tiny ones is not a thirteen-party system. The standard measure is the effective number of parties: square each party's share, add the squares, and take the reciprocal.

Work it on Britain in 2015. By votes, the squares are 0.368 squared, which is 0.135, plus 0.304 squared, which is 0.092, plus 0.126 squared, which is 0.016, plus 0.079 squared, 0.047 squared and 0.038 squared, adding to about 0.254. The reciprocal is about 3.9. By seats, the shares are 0.509, 0.357, 0.086 and 0.012, whose squares sum to about 0.394, giving about 2.5.

So Britain in 2015 was a four-party system in the electorate and a two-and-a-half-party system in parliament. The gap is the electoral system from the previous lesson, expressed as a single number, and it is the cleanest way to see a rule reshaping a society's politics before anyone has voted on anything.

Giovanni Sartori added the qualitative dimension in 1976: what matters is not only how many parties there are but how far apart they stand and whether the parties at the edges have coalition potential or merely blackmail potential. Polarised pluralism, with a large centre under attack from both extremes, behaves nothing like moderate pluralism with the same party count.

What is filling the space

Into the space the old parties vacated has come populism, a word used so loosely that it needs a definition with edges. Cas Mudde supplied the standard one in 2004: populism is a thin-centred ideology holding that society is separated into two homogeneous and antagonistic groups, the pure people and the corrupt elite, and that politics should express the general will of the people.

Thin-centred is the crucial term. Populism has no economic programme of its own, which is why it attaches to a host ideology and appears on both left and right: Podemos and Syriza on one side, Fidesz and the Rassemblement National on the other. It follows that the useful questions about a populist party are what it is attached to and what its conception of the people excludes, not whether it is populist.

The definition also generates a testable claim. If populism is defined by the people-elite opposition and the rejection of intermediate institutions, then populist governments should attack referees rather than pursue any characteristic economic policy. Lesson 6 supplies the pattern that supports it.

A new cleavage, or a realignment on an old one?

The open dispute is whether what has emerged is a genuine new cleavage. One camp says yes: a division over identity, immigration, national sovereignty and cosmopolitanism, running orthogonally to the old economic axis and often labelled green-alternative-libertarian against traditional-authoritarian-nationalist, or integration against demarcation. Its evidence is that education now predicts vote choice better than income does in many western European countries, a reversal of the pattern that held for fifty years.

The other camp says this is a realignment, not a new cleavage, because the Lipset-Rokkan criteria are not met: there is identity and there is division, but the dense organisations that turn a division into a cleavage are precisely what has disappeared. On this account the new politics is volatile because it has no organisational anchor, and volatility is the evidence.

What would settle it

Watch for institutionalisation. If parties organised on the new axis build durable memberships, associated organisations and stable electorates that persist through leadership change and defeat, the cleavage claim is vindicated. If their support keeps moving between vehicles, dissolving when a leader falls and reassembling behind a new one, the realignment account is right. A decade of election studies will answer this, and the answer is not yet in.

Common misconceptions

• Populism is an ideology on the right. It is thin, and it attaches to hosts on both sides. Treating it as a synonym for the radical right makes left-wing populist movements invisible and the concept useless comparatively.
• Freezing meant nothing changed. It meant the cleavage structure persisted while parties, leaders and policies changed constantly. The claim was always about alignments, not about stability of government.
• A party system with many parties is fragmented. Count effectively rather than nominally, and then ask Sartori's question about distance. Five parties clustered near the centre and five spread across an ideological gulf are different systems with the same number.

What to carry forward

Party systems were built on four cleavages laid down by state building and industrialisation, and they froze because organisations encapsulated voters, not because opinions were fixed. Those organisations have hollowed out: membership below a twentieth of the electorate, identification falling, volatility at postwar records, and parties that once took forty percent reduced to single figures within a decade. Populism, defined precisely, is a thin ideology about the people against the elite that attaches to hosts on either flank, and whether the new divide amounts to a cleavage depends on whether anything durable is being built on it.

The core of it: a cleavage needs a division, an identity and an organisation. Two of the three are widely present today. The third, which is the one that made party systems last, is the one that has gone.

Sources

1. Lipset, S. M., & Rokkan, S. (1967). Cleavage structures, party systems, and voter alignments: An introduction. In Party Systems and Voter Alignments: Cross-National Perspectives. Free Press.
2. Mudde, C. (2004). The populist zeitgeist. Government and Opposition, 39(4), 541-563.
3. van Biezen, I., Mair, P., & Poguntke, T. (2012). Going, going, ... gone? The decline of party membership in contemporary Europe. European Journal of Political Research, 51(1), 24-56.
4. Sartori, G. (1976). Parties and Party Systems: A Framework for Analysis. Cambridge University Press.
5. Wikipedia contributors. (n.d.). PASOK. Wikipedia
6. Wikipedia contributors. (n.d.). Effective number of parties. Wikipedia

Eighty-one votes to forty-seven

On 22 August 1973 the Chilean Chamber of Deputies voted, 81 to 47, for a resolution declaring that the government of Salvador Allende had broken the constitution and calling on the armed forces to restore constitutional order. Allende replied that the Chamber was inviting a coup and that his mandate came from the electorate. Both were making a defensible constitutional argument. Neither had any procedure available to settle it. Twenty days later the military took the presidential palace and Allende was dead.

Chile in 1973 is the case around which the most consequential institutional argument in comparative politics was built. The question is simple to state and hard to settle: does electing a president separately from a legislature make democracy more likely to fail?

Linz's four perils

Juan Linz argued in 1990 that presidentialism carries four structural hazards that parliamentary systems avoid.

Dual democratic legitimacy. A president elected nationally and an assembly elected in districts both hold a mandate from the same voters. When they disagree fundamentally, no democratic principle determines which should prevail, and the only institution positioned to arbitrate is the one with the guns. That is exactly the Chilean deadlock: the Chamber's resolution and Allende's reply are both correct, given their premises.

Temporal rigidity. A president holds office for a fixed term. A prime minister who loses the confidence of a majority is replaced next week, without any regime crisis at all, and a popular one can call an early election. A president who has lost all authority stays; a president governing well must stop. Impeachment is not a substitute, because it is designed as a criminal remedy rather than a political one.

Winner-take-all. The presidency cannot be shared. A party that wins 51 percent takes the entire executive, and one that wins 49 takes none of it. Allende took office with 36.6 percent of the vote in a three-way contest and full executive power, which made the stakes of the contest enormous and the losers permanent.

The outsider and the style. Presidential systems reward candidates with no legislative career and no obligation to a party, and encourage a plebiscitary self-understanding in which the president speaks for the nation while the assembly represents mere interests. Someone who came to office over the parties has neither the votes nor the habits to bargain with them.

Why this matters: each peril is a claim about incentives created by a rule, and each is separable. A system could have fixed terms without a winner-take-all executive, or dual mandates without outsider candidates, and knowing which peril is operating in a given crisis is more useful than the label.

The numbers that made the case

The empirical support looked overwhelming. In the regime data covering 1950 to 1990, the expected life of a presidential democracy was on the order of twenty years, while the expected life of a parliamentary democracy was several times that. Latin America, overwhelmingly presidential, had produced repeated breakdowns. Western Europe, overwhelmingly parliamentary, had not. The advice that followed was concrete, and constitutional advisers gave it for two decades: if you are designing a new democracy, do not build a strong separately elected presidency.

Cheibub's reply: what kind of country adopts a presidency?

Jose Antonio Cheibub accepted the correlation and attacked the inference, in the exact form Lesson 1 warned about. The question is not whether presidential democracies die more often. It is why a country has a presidential constitution in the first place.

His answer: countries that emerge from military dictatorship overwhelmingly adopt presidential constitutions, and countries that emerge from civilian authoritarian rule or from colonial parliamentary institutions overwhelmingly do not. A military legacy leaves behind an officer corps accustomed to intervening, a politicised command structure and a set of unresolved questions about civilian control. Those are powerful predictors of democratic breakdown on their own.

So the presidential democracies of the late twentieth century were disproportionately drawn from the population of countries most likely to suffer a coup regardless of their constitution. Condition on the prior regime type and, in Cheibub's analysis, presidentialism has no independent effect on survival. The constitution was a marker of the country's history, not a cause of its fate.

He adds a second finding that undercuts the deadlock story directly. Minority presidents are supposed to produce paralysis, yet they form coalitions at rates comparable to parliamentary minority governments, and legislative gridlock is no more common under presidentialism once you measure it rather than assume it.

How Brazil governs

Brazil is the case that makes Cheibub's point concrete. Its president faces a legislature in which the largest party has often held well under a fifth of the seats and a dozen parties hold the rest. On the Linz account this should be permanent deadlock. Instead Brazilian presidents build multi-party cabinets, distribute ministries and budget amendments among coalition partners, and pass legislation, a practice known as coalitional presidentialism.

The qualification matters as much as the finding. The currency of those coalitions is patronage, and the Brazilian corruption prosecutions of the last two decades are the bill. Coalitional presidentialism solves the deadlock problem and creates a different one, which is a fair summary of most institutional fixes anywhere.

The third form, and why the subtype matters

Most of the world's new democracies since 1990 chose neither pure type. A semi-presidential system has a popularly elected president with a fixed term and also a prime minister and cabinet who depend on the confidence of the legislature. Matthew Shugart and John Carey showed that the label covers two very different machines.

France has lived through the premier-presidential version three times: cohabitation from 1986 to 1988, 1993 to 1995, and 1997 to 2002. It was uncomfortable and it was survivable, because the rule was unambiguous: a prime minister with an assembly majority governs domestically, whatever the president thinks. In 2000 France cut the presidential term from seven years to five and aligned the electoral calendar so that presidential and legislative elections fall close together, largely to make cohabitation unlikely. The president-parliamentary variant has no equivalent tie-breaker, and its record is markedly worse.

Bottom line: the useful unit of comparison is not the three-way label but the specific allocation of powers: who appoints the cabinet, who can dismiss it, who controls the legislative agenda, and whether the president can dissolve the assembly.

What would settle the dispute

Cheibub's argument is a selection argument, so the way to test it is to find variation in constitutional form that is not produced by regime history. Three routes exist. Countries that changed form without changing their authoritarian past, such as France in 1958 or Moldova moving to and from parliamentary election of the president, provide within-country variation. Colonial inheritance provides an instrument of sorts, since British and French administrative legacies pushed similar countries toward different constitutions for reasons unconnected to their later politics. And the subtype distinction is itself a test: if the president-parliamentary variant fails more often than the premier-presidential one among countries with similar histories, then constitutional detail is doing causal work after all, and the pure selection story is incomplete.

Common misconceptions

• Presidential systems produce gridlock and parliamentary systems do not. Measured rather than assumed, legislative productivity under minority presidents is comparable to that under parliamentary minority governments. Deadlock is a possibility in both, and coalition-building is the normal response in both.
• Impeachment is the parliamentary confidence vote of a presidential system. It is built as a remedy for wrongdoing, requires supermajorities, and is slow and destabilising precisely because it is not a routine political instrument.
• Cheibub showed presidentialism is harmless. He showed that its association with breakdown is confounded by regime history. That is a finding about the evidence for the claim rather than a demonstration that constitutional design does not matter, and the subtype results suggest it does.

Pulling it together

Linz identified four hazards built into separately electing an executive: two mandates with no arbitrator, fixed terms that cannot bend, an indivisible prize, and outsiders with no reason to bargain. Chile in 1973 exhibits all four. The survival statistics appeared to confirm the diagnosis until Cheibub asked how countries came to have presidential constitutions and found the answer was largely their military past, which predicts breakdown on its own. Semi-presidential systems split the difference, and the split matters: a prime minister answerable to one master is a workable arrangement, and one answerable to two is the machine that broke Weimar.

Remember: the Chilean Chamber and the Chilean president were both right. When a constitution can produce two correct and incompatible answers, the deciding vote goes to whoever is outside the constitution.

Sources

1. Linz, J. J. (1990). The perils of presidentialism. Journal of Democracy, 1(1), 51-69.
2. Cheibub, J. A. (2007). Presidentialism, Parliamentarism, and Democracy. Cambridge University Press.
3. Shugart, M. S., & Carey, J. M. (1992). Presidents and Assemblies: Constitutional Design and Electoral Dynamics. Cambridge University Press.
4. Wikipedia contributors. (n.d.). 1973 Chilean coup d'etat. Wikipedia
5. Wikipedia contributors. (n.d.). Semi-presidential system. Wikipedia

Three regions, then thirty-six states

Nigeria became independent in 1960 with three regions. Each was dominated by one of the country's largest ethnic groups, and each was large enough that its leaders could contemplate leaving. By 1967 there were twelve states, by 1976 nineteen, by 1991 thirty, and since 1996 there have been thirty-six states plus a federal capital territory. The multiplication was not administrative tidying. It was designed to ensure that no single unit could again mobilise a region against the centre, a lesson learned in the civil war of 1967 to 1970.

Federalism is usually taught as a way of dividing power between levels. It is more useful to treat it as a technology that can be engineered for particular purposes, and this lesson compares two federations built for opposite reasons.

Riker's bargain, and his later doubt

William Riker argued in 1964 that federations are not born of a preference for decentralised government. They are born of a bargain, and the bargain has two sides. Politicians at the centre want to expand territory or aggregate military strength, usually against an external threat, but cannot conquer the units outright. Politicians in the units want protection or market access, and will trade some autonomy for it. Federalism is the price both sides accept.

Riker's later and more provocative claim concerns what the constitution actually does. He argued that the degree of centralisation in a federation is determined not by the text but by the party system. Where national parties control nominations, careers and money in the constituent units, the federation centralises whatever the constitution says. Where regional parties control their own nominations, it decentralises. On his account American federalism in the century after Reconstruction functioned mainly to protect racial hierarchy in the South from national majorities, and he drew the conclusion bluntly: someone who dislikes what a federal veto was used to protect should be sceptical of the institution that supplied the veto.

Key idea: a federal constitution is a set of vetoes, and vetoes are morally empty. They protect whatever a subnational majority wants protected, which has included minority language rights and has also included segregation.

Coming together and holding together

Alfred Stepan added the distinction that organises modern work on the subject. Coming-together federations are formed by previously independent units pooling sovereignty: the United States in 1787, Switzerland, Australia. Their constitutions typically overrepresent small units in an upper chamber, because that was the price of joining.

Holding-together federations are formed when a unitary state devolves power to prevent a breakup: India, Belgium, Spain. They are asymmetric more often, granting different powers to different regions according to what each was demanding, and their upper chambers are usually weaker, because no unit had a pre-existing sovereignty to sell.

The distinction predicts behaviour. Coming-together federations tend to give small units disproportionate national veto power, which is why the Australian and American upper houses look the way they do. Holding-together federations tend to concentrate national policy in the centre while allowing regions to differ on language, education and culture, which is exactly what a state trying not to fragment needs.

India: language first, then the emergency clause

India's federation was designed for holding together, and its central engineering decision came in 1956. After a campaign that included a fatal hunger strike, the States Reorganisation Act redrew internal boundaries so that state lines followed major language communities. The reasoning was that a linguistic group with its own state, its own government and its own official language has far less reason to want a country of its own. Seventy years later, with 28 states and no significant secession outside the north-east and Kashmir, the design has an unusually strong claim to have worked.

It came with a centralising instrument. Article 356 permits the union government to dismiss a state government and impose President's rule. It has been used well over a hundred times, frequently against state governments run by opposition parties, until the Supreme Court's 1994 judgment in the Bommai case made its use reviewable and required that a state government's loss of majority be tested on the floor of the assembly rather than asserted by a governor.

Note what that history illustrates about Riker's party argument. During the decades when one party dominated at the centre and in most states, India's federalism operated as a highly centralised system despite a federal text. As regional parties grew strong enough to matter in national coalition-building, the same text produced far more genuine bargaining.

Nigeria: multiplying units to dilute a majority

Nigeria's engineering ran in the opposite direction. Three regions meant three ethnic power blocs. When the Eastern Region attempted secession as Biafra in 1967, the federal government's response included creating twelve states, several of which split the seceding region's own minorities away from its dominant group. Every subsequent round of state creation followed the same logic: smaller units, more of them, none capable of standing alone.

Two further devices reinforce it. The federal character principle requires that appointments to federal bodies reflect the country's geographic spread rather than merit alone or the preferences of a single group. And presidential election rules require the winner to obtain not only a plurality nationally but at least a quarter of the votes in two-thirds of the states, which makes a candidate with overwhelming support in one region and none elsewhere unelectable.

Whether it has worked is genuinely contested. Nigeria has not fragmented since 1970, which is the outcome the design targeted. It has also produced a politics in which the main business of many states is capturing a share of centrally distributed revenue, which brings us to the money.

The money: who taxes, who spends, who is rescued

The literature on fiscal federalism begins with Wallace Oates and a simple rule: provide a public good at the lowest level of government that internalises its benefits and costs. Local refuse collection is local; national defence is not; the arguments happen in the middle, over schools, roads and health.

The rule assumes something that often fails: that the level which spends the money also raises it. Where a subnational government spends money it did not raise, two things follow. Voters cannot hold it accountable for the tax burden, because it did not set one. And the government faces a soft budget constraint, since overspending can be met by lobbying for a larger transfer rather than by cutting or taxing.

The Nigerian column is what a rentier federation looks like. A derivation provision returns thirteen percent of oil revenue to the states where the oil is produced, and the rest flows into a pool divided among all thirty-six states and the local governments. The resulting politics is about the formula, not about taxation, which is Lesson 3's argument about state capacity arriving through a different door.

What matters here: to judge whether a country is really federal, do not read the constitution. Find out what share of subnational spending is financed by subnational taxes, and what happens when a state cannot pay its wage bill.

What would overturn the case for federalism in divided societies

The claim defended here is that territorial autonomy for concentrated minorities reduces the demand for secession, with India's linguistic states as the leading evidence. The strongest counter-argument holds that autonomy does the opposite: it gives a regional elite a government, a budget, a broadcaster and an education system with which to build precisely the national consciousness that secession requires. Quebec, Catalonia and Scotland are the cases usually offered.

What would settle it is a systematic study of every case of territorial autonomy granted since 1945, coding whether secessionist mobilisation rose or fell afterwards, and separating cases where autonomy was granted early from those where it followed violence. The existing findings point in both directions, with the balance depending on whether the autonomous unit contains its own minority, which is the condition Nigeria's state-splitting was designed to create and Spain's autonomy statutes were not.

Common misconceptions

• Federal countries are more decentralised than unitary ones. Not reliably. Some unitary states devolve more real fiscal power to local governments than some federations do to their constituent units, and the party system predicts centralisation better than the constitutional form.
• Federalism protects minorities. It protects territorially concentrated groups, which is not the same population. A dispersed minority gains nothing from a federal veto, and may lose, since the veto can be used by a regional majority against the minority inside it.
• More states means more local democracy. Nigeria's multiplication of states was a security measure, and where new units depend entirely on central transfers it multiplies claimants on a common pool rather than creating self-governing communities.

Recap

Federations begin as bargains, and what they do afterwards depends less on their texts than on whether parties are national or regional. Coming-together federations pay small units in veto power; holding-together federations buy unity with asymmetry and cultural autonomy. India drew its state lines along languages in 1956 and kept an emergency clause it used more than a hundred times. Nigeria went from three regions to thirty-six states precisely so that no unit could threaten the whole, and financed them from a common oil pool. The fiscal test cuts through all of it: a government that spends money it did not raise is administering somebody else's budget, whatever the constitution calls it.

The upshot: ask of any federation who taxes, who spends, and who pays when a unit runs out of money. Those three answers tell you more than the whole division-of-powers chapter.

Sources

1. Riker, W. H. (1964). Federalism: Origin, Operation, Significance. Little, Brown.
2. Stepan, A. (1999). Federalism and democracy: Beyond the U.S. model. Journal of Democracy, 10(4), 19-34.
3. Oates, W. E. (1972). Fiscal Federalism. Harcourt Brace Jovanovich.
4. Wikipedia contributors. (n.d.). States of Nigeria. Wikipedia
5. Wikipedia contributors. (n.d.). Fiscal federalism. Wikipedia

The country that cannot count itself

Lebanon has not conducted a census since 1932. Its political system distributes offices by religious community: the president is a Maronite Christian, the prime minister a Sunni Muslim, the speaker of parliament a Shia Muslim, and parliamentary seats are allocated by confession. The shares were fixed using the 1932 count, which gave Christians a six to five advantage over Muslims. Everyone involved understands that the demographic reality changed decades ago. Nobody counts, because counting would require reallocating everything at once.

Start with that, because it poses the problem this lesson works. If ethnic and religious difference causes conflict, a system that hardwires those categories into every public office should be a machine for producing it. Yet consociational designs of exactly this kind are the standard international prescription for divided societies. Both things cannot be straightforwardly true, and sorting out which part is wrong takes us through the best-evidenced argument in the field.

What an ethnic group is, and three answers

Primordialism treats ethnicity as ancient, inherited and deeply felt, a given rather than a choice. It has the advantage of matching how people describe their own identities and the disadvantage of being unable to explain change: it cannot say why Yugoslav identity was real for many people in 1985 and unavailable by 1993.

Constructivism treats ethnic and national categories as historically produced. Benedict Anderson gave the argument its most durable form in Imagined Communities in 1983: a nation is an imagined political community, imagined because its members will never meet most of the others yet hold an image of their communion, limited because it has boundaries beyond which other nations lie, and sovereign because it emerged as dynastic and religious orders lost their grip. His mechanism is print-capitalism. Publishers needed markets larger than a dialect and smaller than Latin, so they standardised vernaculars, and readers of the same standardised vernacular could imagine themselves as one public. Anderson adds three colonial instruments that hardened categories: the census, which forces every person into exactly one box; the map, which turns territory into a logo; and the museum, which supplies an ancestry.

Instrumentalism asks who benefits. Ethnic identity is available to be mobilised, and politicians mobilise it when the payoff is higher than the alternatives, typically when resources are distributed by group or when an electoral rule rewards a bloc.

The point: constructed does not mean fake. Anderson's argument is that all communities above the village are imagined, and that the interesting question is the style in which they are imagined, not whether they are real enough to die for. People plainly do.

Horowitz: comparison, worth and the census as a contest

Donald Horowitz published Ethnic Groups in Conflict in 1985 and made an argument that is often reduced to a slogan and should not be. Ethnic conflict, he argued, is driven less by material competition than by comparison: groups evaluate their worth against a specific rival, and the anxieties that follow are about relative position and legitimacy rather than about goods.

Two consequences matter for institutional design. First, in an ethnically divided society with ordinary electoral rules, parties tend to form along ethnic lines, and once they do, competition within each group pushes leaders toward the extremes, since a moderate can always be outbid by someone promising more. Second, anything that measures group size becomes a contest, because size determines entitlement. That is why Lebanon does not count, and why Nigeria's censuses have been fought over so bitterly.

The grievance finding that vanished

Through the 1990s the standard expectation was that ethnically diverse countries would be more prone to civil war. James Fearon and David Laitin tested it in 2003 across 161 countries and 127 civil wars between 1945 and 1999, and found that measures of ethnic and religious fractionalisation had essentially no relationship with the onset of civil war once income was controlled.

What did predict onset was a different list: low per capita income, large population, mountainous terrain, and recent state creation or political instability. Their interpretation is about opportunity rather than grievance. Insurgency is a technology, and what makes it feasible is a state too weak to police its own periphery, terrain that shelters small armed bands, and a population large enough to recruit from. Income in their model is not a measure of poverty as a motive; it is a proxy for the reach of the state, which is Lesson 3's argument arriving in a new place.

This was a genuinely uncomfortable finding, because grievance is what participants themselves say the fighting is about.

And came back at the right level of analysis

The resolution came from asking what a country-level fractionalisation index actually measures. It measures the probability that two randomly chosen citizens belong to different groups. That is a fact about a country and not about any group's position within it, and it is identical for a country where power is shared among all groups and one where a single group monopolises the state.

Lars-Erik Cederman, Andreas Wimmer and Brian Min rebuilt the data at group level in 2010, coding each politically relevant group's access to executive power. Their finding: groups excluded from state power are substantially more likely to rebel, the risk rises with the size of the excluded group, and it rises further where a group has lost power it previously held. Fractionalisation still predicts nothing, and exclusion predicts a great deal.

Note that this is not a contradiction but a lesson in units of analysis. Fearon and Laitin were right about the variables available to them. The grievance mechanism was invisible because it was averaged away at the country level.

Two designs for a divided society

Lebanon, worked

Return to the opening case with the framework in hand. Lebanon's National Pact of 1943 was an unwritten agreement distributing the top offices by confession and fixing the parliamentary ratio at six Christians to five Muslims. It held for three decades, then collapsed into a civil war from 1975 to 1990 in which the demographic mismatch was one thread among several, alongside the Palestinian armed presence, Israeli and Syrian intervention, and intra-communal rivalry.

The Taif Agreement of 1989 ended the war by adjusting the same machinery rather than replacing it: parliamentary seats moved to parity between Christians and Muslims, and executive power shifted from the Maronite presidency to a cabinet in which portfolios are shared. The system was preserved and rebalanced. What was not solved is that entitlement still depends on group size, so the census remains impossible, and any demographic change now has nowhere to go except into a renegotiation of the whole settlement.

That is the consociational trade in its clearest form. The design bought an end to a fifteen-year war, which is not a small thing. It bought it by making the categories permanent.

What would overturn this account

Two claims here are vulnerable. The first is that exclusion from power drives rebellion, which rests on coding decisions about which groups are politically relevant and who counts as included; if an independent team recoded those judgements from primary sources and the relationship weakened substantially, the resolution offered above would fail and the opportunity account would stand alone. The second is that consociational designs freeze identities. Northern Ireland is the test case, since its 1998 settlement requires members of the assembly to designate as unionist, nationalist or other, and the share designating as other has been growing. If that share keeps growing and the mandatory designations lose their force, the freezing claim is weaker than its critics assert.

Common misconceptions

• Ancient hatreds explain ethnic wars. The phrase does no explanatory work, because the hatreds are available for centuries and the wars are not. What varies is opportunity and political exclusion, and those are what the evidence tracks.
• Diverse countries are more war-prone. Country-level diversity indices do not predict civil war onset. What predicts it is a group being shut out of power, which diversity measures cannot see.
• Constructed identities can be talked away. Anderson's point is that all large communities are imagined and that this is compatible with their being worth dying for. Knowing an identity has a history tells you nothing about how easily it can be dissolved.

What to remember

Ethnic categories are historically produced, hardened by censuses, maps and standardised print, and mobilised when politics rewards mobilising them. Country-level diversity predicts nothing about civil war; income, terrain, population and state weakness predict a great deal, and so does the exclusion of a specific group from executive power once you measure at the level of the group. Institutional responses divide between including every group by right and giving politicians reasons to campaign across groups, and each buys its peace with a different currency.

In short: Lebanon's missing census is not an administrative failure. It is what a political settlement looks like when entitlement is tied to the size of a group, and it is the clearest available illustration of what consociational design costs.

Sources

1. Anderson, B. (1991). Imagined Communities: Reflections on the Origin and Spread of Nationalism (rev. ed.). Verso.
2. Horowitz, D. L. (1985). Ethnic Groups in Conflict. University of California Press.
3. Fearon, J. D., & Laitin, D. D. (2003). Ethnicity, insurgency, and civil war. American Political Science Review, 97(1), 75-90.
4. Cederman, L.-E., Wimmer, A., & Min, B. (2010). Why do ethnic groups rebel? New data and analysis. World Politics, 62(1), 87-119.
5. Wikipedia contributors. (n.d.). Consociationalism. Wikipedia
6. Wikipedia contributors. (n.d.). Taif Agreement. Wikipedia

Seventy thousand on the ring road

On Monday 9 October 1989 roughly seventy thousand people walked around the inner ring road in Leipzig. Hospitals had been told to prepare for casualties, extra blood supplies had been ordered, and armed units were in position. The demonstrators carried candles, which occupied both hands and made it visibly impossible to throw anything. Nobody fired.

Two weeks later the Monday demonstration in Leipzig drew about three hundred thousand. On 4 November perhaps half a million assembled in East Berlin. On 9 November the border was opened, and within a year the German Democratic Republic had ceased to exist. A state with one of the most thorough surveillance apparatuses ever built fell in five weeks to unarmed crowds.

This lesson runs from the structural theories that explain why revolutionary situations arise to the evidence on which methods of resistance succeed, and it uses 1989 as the thread because it is a hard case for every theory in the field.

Skocpol: revolutions are not made, they come

Theda Skocpol's structural account, which Lesson 1 used as a methodological example, is worth stating on its own terms. A social revolution, meaning a rapid transformation of a society's state and class structures carried through by revolt from below, requires two conditions to hold simultaneously.

The first is state breakdown. The old regime's administrative and coercive apparatus must lose its capacity to function, and Skocpol's cases show this happening through international pressure: a state competing militarily with more developed rivals, unable to extract enough from its landed classes to keep up, is paralysed by the attempt. The second is a peasantry with autonomous local organisation, capable of acting without leadership from outside.

Her sharpest claim is negative. Revolutionary ideology and the intentions of revolutionary vanguards do not explain outcomes, because no group in her cases achieved what it set out to achieve. Revolutions are not made, she wrote; they come. And the state that emerges is typically more centralised and more capable of extraction than the one destroyed, which is the opposite of what most participants wanted.

Why this matters: a structural account tells you where revolutionary situations become possible. It does not tell you when one will occur, and it explicitly denies that anyone's plan determines what follows.

Why nobody saw 1989 coming

Almost no specialist predicted the collapse of communism in eastern Europe, and Timur Kuran's 1991 explanation is the best available. In a repressive regime, people misrepresent their preferences in public. Each person holds a private threshold: the number of others they would need to see protesting before they would join. Someone with a threshold of ten thousand is invisible until ten thousand appear.

Because the thresholds are hidden, the distribution of them is unknown to everybody, including the regime, the opposition and the participants themselves. A society can look completely stable while containing a chain in which a small increase in visible protest triggers a much larger one. When the chain fires, the result is a cascade, and it feels sudden to everyone precisely because the underlying discontent was never observable.

This is why Leipzig on 9 October matters more than its size suggests. The regime did not shoot. That single observation revised everyone's estimate of the cost of participating, and the count went from seventy thousand to three hundred thousand in fourteen days.

Tilly's repertoires: what a protest is made of

Charles Tilly spent much of his career on a question that sounds trivial and is not: why do people protest in the particular forms they do? He studied 8,088 contentious gatherings in Britain between 1758 and 1834 and documented a transformation in what he called the repertoire of contention.

The older repertoire was parochial and particular: attacks on a specific miller's stores, the seizure and forced sale of grain at a customary price, the ritual humiliation of a named offender. Actions were local, aimed at immediate targets, and often used the forms of official ceremony. The newer repertoire, in place by the 1830s, was national and modular: the public meeting, the petition, the demonstration, the strike. Modular means transferable, since the same form can be used by any group, about any grievance, against any target.

The change tracks the rise of the national state and of parliament. When power concentrates nationally, protest addresses itself nationally, and it adopts forms legible to a national audience. That is why the candlelit march around a ring road is recognisable across the world, and why a grain seizure would not be.

The numbers on what works

Erica Chenoweth and Maria Stephan set out to test a claim they expected to refute: that nonviolent resistance is more effective than armed struggle. They assembled data on 323 major campaigns aimed at regime change, expulsion of an occupier or secession, from 1900 to 2006.

Their proposed mechanism is participation. Armed struggle recruits from a narrow pool: young, physically capable, willing to kill and to die. Nonviolent campaigns can absorb the elderly, children, the cautious, professionals and civil servants, and they can be joined for an afternoon rather than for years. Larger and broader participation matters because it raises the probability that members of the security forces have relatives in the crowd, which is the condition under which orders stop being carried out.

The most quoted finding is the threshold: in their data, no campaign that achieved the active and sustained participation of at least 3.5 percent of the population failed. Apply it to the German Democratic Republic, with about 16.4 million people. Three and a half percent is roughly 570,000. The demonstrations of early November were of that order.

What would overturn this, and partly has

The 3.5 percent figure is descriptive, drawn from a set of campaigns that mostly succeeded for other reasons too, and it should not be treated as a law. More seriously, Chenoweth's own later work reports that the success rate of nonviolent campaigns has fallen sharply since around 2010, to well below the historical average, even as such campaigns have become more numerous. Proposed explanations include better authoritarian counter-strategies, digital organisation that mobilises crowds quickly without building the durable structures that sustain a campaign, and increased external support for incumbents.

Take that seriously as a methodological point rather than a footnote. A finding drawn from 1900 to 2006 was used to make predictions about the 2010s and the predictions did worse than the historical base rate. If regimes learn, then every empirical regularity in this field has a shelf life, and the evidence that would overturn a claim may simply be the next decade.

Common misconceptions

• Revolutions happen when conditions become unbearable. Conditions are unbearable in many places for long periods without revolution. What varies is state capacity, elite cohesion and the visible cost of acting, which is why the timing surprises even the participants.
• Nonviolent resistance works only against liberal opponents. The Chenoweth and Stephan sample includes campaigns against colonial powers, military juntas and communist party-states. The pattern holds well beyond the cases where the regime was already restrained, though repression does reduce the odds.
• Revolutions replace an oppressive state with a weaker one. Skocpol's central finding is the reverse: the post-revolutionary states in France, Russia and China were more centralised, more extractive and more capable of coercion than what they replaced.

The takeaway

Skocpol explains where revolutionary situations become possible: a state broken by military competition it cannot finance, and a rural population able to organise itself. Kuran explains why the moment is unpredictable, since preferences are hidden until a cascade reveals them. Tilly explains why protest takes the forms it does, and why those forms became national and transferable as states did. Chenoweth and Stephan supply the comparative evidence: across 323 campaigns, unarmed ones succeeded roughly twice as often as armed ones, through participation broad enough to split the security forces, and the advantage has narrowed since 2010.

Bottom line: the decisive event in Leipzig was not the size of the crowd but the fact that nobody fired into it. Everything downstream, including the size of the next crowd, followed from what that revealed about the regime's willingness to pay the price of staying.

Sources

1. Skocpol, T. (1979). States and Social Revolutions: A Comparative Analysis of France, Russia and China. Cambridge University Press.
2. Kuran, T. (1991). Now out of never: The element of surprise in the East European revolution of 1989. World Politics, 44(1), 7-48.
3. Tilly, C., & Tarrow, S. (2007). Contentious Politics. Paradigm Publishers.
4. Chenoweth, E., & Stephan, M. J. (2011). Why Civil Resistance Works: The Strategic Logic of Nonviolent Conflict. Columbia University Press.
5. Wikipedia contributors. (n.d.). Monday demonstrations in East Germany. Wikipedia
6. Wikipedia contributors. (n.d.). Preference falsification. Wikipedia

One hundred and fifty-eight dollars

In 1960 the World Bank recorded South Korea's income per head at about 158 dollars and Ghana's at about 183. Korea was poorer. It had been partitioned, then wrecked by a war that ended seven years earlier, and it had almost no natural resources. Ghana had gold, cocoa, an educated administrative class and a leader admired across the continent.

Sixty years later Korean income per head is above thirty thousand dollars and Ghanaian income is a small fraction of that. Comparativists have argued about that divergence ever since, and the argument has produced the two most important bodies of work in the political economy of development: one about what the East Asian states did, and one about what resource wealth does to states that have it.

What a developmental state actually did

Chalmers Johnson's MITI and the Japanese Miracle, published in 1982, introduced the term. Johnson distinguished a market-rational state, which sets the rules of economic competition and does not care what gets produced, from a plan-rational state, which has substantive goals about which industries the country should have and organises itself to achieve them. A developmental state in his sense has four features: a small elite economic bureaucracy recruited by rigorous examination; a pilot agency with authority over industrial policy; a political leadership that gives the bureaucracy room to operate; and institutions for continuous consultation with private firms.

South Korea after Park Chung-hee's coup of 16 May 1961 built exactly that. The Economic Planning Board, created in 1961, controlled the budget, foreign borrowing and investment approvals in one place. The state nationalised the banks and thereby controlled who received credit and on what terms. The chaebol conglomerates grew inside that arrangement, not despite it.

Here is the feature that most accounts of industrial policy miss, and it is the decisive one. Alice Amsden's study of Korean industrialisation emphasises that subsidised credit was conditional on measurable export performance. Firms that met export targets got more; firms that did not lost access. Exports were the metric because they cannot be faked: a product that sells in a foreign market at a world price has met a standard the domestic government did not set. Industrial policy without such a mechanism is a transfer to whoever is best connected. With it, the state has a hard test it did not have to invent.

Key idea: the distinguishing feature of the successful cases is not that the state picked winners. It is that the state could stop paying losers, and had a criterion that made losing visible.

Embedded autonomy: the knife edge

Peter Evans compressed the institutional requirement into a phrase in 1995. An effective developmental state needs autonomy, meaning a bureaucracy insulated enough from short-term political and business pressure to withdraw support from a failing firm, and simultaneously embeddedness, meaning dense concrete ties to the firms it is trying to develop, without which it cannot know what is actually happening in an industry.

Failure of either half has a signature. Autonomy without embeddedness produces a competent bureaucracy issuing irrelevant plans, the classic failure of central planning. Embeddedness without autonomy produces capture, in which the state cannot stop funding a firm because the firm's owners are its patrons, and it is much the more common failure. Evans's argument is that the combination is rare, unstable and historically contingent, which is why the developmental state has proved so hard to copy.

The curse, stated precisely

Now the other half. The resource curse has an economic version, in which resource booms raise the exchange rate and crowd out manufacturing, and a political version, which is our concern. Michael Ross set out three mechanisms in 2001.

1. The rentier effect. A government funded by oil does not need to tax its citizens. With no tax bargain, the demand for accountability that Lesson 3 traced through English fiscal history never arises. Low taxes and high spending buy quiescence.
2. The repression effect. Resource revenue funds security services on a scale a poor economy could not otherwise support, and gives the regime something worth defending.
3. The modernisation effect. An enclave export sector generates revenue without the urbanisation, occupational specialisation and education that ordinarily accompany growth, so the social changes associated with democratisation do not occur.

The first mechanism is the strongest and the most testable, and it connects directly to the state capacity measures from Module 2. A state that collects a tenth of its revenue from its population and nine tenths from a terminal has different incentives from one that must persuade taxpayers every year.

Four countries, one table

Read the second row before the others. Botswana and Norway found their resources while their political arrangements were already settled and reasonably accountable; Nigeria found oil during a period of military rule and regional conflict. The resource did not create the institutions. It arrived into whatever institutions were there.

The reappraisal

Stephen Haber and Victor Menaldo pressed exactly that point in 2011, and their method matters as much as their conclusion. Most resource-curse studies compare countries: oil states are less democratic than non-oil states. That is a cross-national correlation, with all the problems Lesson 2 catalogued, and in particular it cannot rule out that the countries which happened to have oil were already less democratic before anyone drilled.

Haber and Menaldo built long time series reaching back into the nineteenth century and asked whether countries became less democratic after resource wealth appeared, using each country as its own comparison. They found no evidence of a negative effect, and in some specifications a positive one.

The dispute is not settled. Critics argue that country fixed effects absorb most of the variation of interest, since resource dependence changes slowly, and that the historical coding is uncertain. But the burden has shifted. The defensible current position is conditional: resource wealth is dangerous for states with weak institutions at the moment the revenue arrives, and manageable for states with strong ones, which is a claim about interaction rather than about oil.

What would overturn this

Two tests would move the argument. First, price shocks provide within-country variation that is not chosen by the country: if democracy scores and state capacity measures move systematically with world oil prices, the rentier mechanism is doing real work; if they do not, the correlation is about which countries have oil. Second, the developmental state claim predicts that industrial policy with a hard external performance criterion outperforms industrial policy without one. That is testable across the many countries that have tried subsidised credit, and if programmes with export conditionality do no better than those without, the Amsden mechanism is not what distinguished Korea.

Common misconceptions

• The East Asian states succeeded by leaving markets alone. They intervened massively: allocating credit, setting export targets, restricting imports and licensing entry. What they also did was withdraw support from failures, which is the part rarely copied.
• Authoritarian government explains East Asian growth. The region contains a great many authoritarian governments that presided over stagnation, and Lesson 1's discussion of selection on the outcome is exactly about this argument.
• Resource wealth is a curse. Norway and Botswana both had large resource sectors and did well. The conditional version, that resources are dangerous where institutions are weak when the money arrives, fits the evidence better than a blanket claim.

Where this leaves us

Korea started poorer than Ghana and built a state that allocated credit against a criterion firms could not fake. Johnson named the type, Amsden identified the discipline that made it work, and Evans specified the knife edge between insulation and capture that so few states have managed to stand on. Resource wealth acts through three mechanisms, of which the rentier effect is the best supported, and the strongest challenge to the whole literature comes from asking a within-country question instead of a cross-country one.

The core of it: nothing in this lesson is about oil or about Confucian culture. It is about whether a state has a test it can apply to the money it hands out, and whether it can bear the political cost of applying it.

Sources

1. Johnson, C. (1982). MITI and the Japanese Miracle: The Growth of Industrial Policy, 1925-1975. Stanford University Press.
2. Evans, P. B. (1995). Embedded Autonomy: States and Industrial Transformation. Princeton University Press.
3. Ross, M. L. (2001). Does oil hinder democracy? World Politics, 53(3), 325-361.
4. Haber, S., & Menaldo, V. (2011). Do natural resources fuel authoritarianism? A reappraisal of the resource curse. American Political Science Review, 105(1), 1-26.
5. World Bank. (n.d.). GDP per capita (current US dollars). data.worldbank.org
6. Wikipedia contributors. (n.d.). Resource curse. Wikipedia

The claim that looks obviously right

Countries with secure property rights, courts that enforce contracts and constrained executives are rich. Countries without them are poor. The correlation is enormous and nobody disputes it. So institutions cause growth, and the policy conclusion writes itself.

This lesson takes that reasoning apart. Not to conclude that institutions do not matter, which is almost certainly false, but because the most famous attempt to prove it rigorously is also the best available worked example of how an identification strategy can be admired for a decade and then found to be resting on data that were partly invented. Following the failure carefully will teach you more about causal inference than following a success.

Why the naive regression cannot work

Regress income per head on a measure of property rights protection across countries and you get a large, precisely estimated coefficient. It is worthless on its own, for three reasons.

Reverse causation. Rich countries can afford courts, land registries, professional police and tax administrations. The arrow may run entirely from wealth to institutions.

Omitted variables. Anything that produces both, such as geography, disease environment, culture or human capital, will show up in the coefficient.

Measurement. The institutional quality indices in wide use, including expropriation-risk ratings produced for investors, are partly assessments made by analysts who know how rich the country is. A measure contaminated by the outcome cannot identify an effect on it.

The instrument, and why it was admired

Daron Acemoglu, Simon Johnson and James Robinson proposed in 2001 a way around all three. An instrumental variable is something that shifts the suspect cause without affecting the outcome by any other route. Their instrument was the mortality rate faced by Europeans in the places they colonised.

The numbers are stark. In the figures they assembled, annualised European mortality was about 483 per thousand in Sierra Leone and about 8.55 per thousand in New Zealand. The argument runs in four steps.

1. Europeans died at very different rates in different colonies, for reasons of disease environment that had nothing to do with the local political economy.
2. Where they could survive, they settled in numbers and built institutions protecting property and constraining rulers, because they intended to live under them. Where they died, they built extractive institutions designed to move resources out.
3. Those early institutions persisted through independence to the present.
4. Therefore settler mortality shifts current institutions, and it can affect current income only through them.

Their two-stage estimates implied that institutional differences account for a very large share of the income gap among former colonies, with an effect several times larger than the ordinary regression suggested. The paper became one of the most cited in economics, and its logic spread through comparative politics.

The point: step four is the exclusion restriction, and it cannot be tested. No statistic tells you whether an instrument affects the outcome by another route. It is a historical claim, and it has to be defended with historical evidence.

Step one fails: where the numbers came from

David Albouy went back to the mortality sources in a comment published in 2012. Three findings.

First, mortality data did not exist for most of the countries in the sample. Of the 64 countries used, rates for more than half were not observed in that country at all; they were assigned from another country, sometimes a distant one, on the grounds of regional similarity. The instrument's variation is therefore substantially constructed rather than measured.

Second, the observed rates mix incomparable populations. Some figures come from soldiers on campaign, living in the field during a war; others from soldiers in peacetime barracks; others from bishops or civilian settlers. Campaign mortality can be several times barracks mortality in the same place. A dataset that mixes them is measuring military circumstance as much as disease environment.

Third, and decisively for the statistics, when the conjectured rates are dropped or campaign conditions are controlled for, the first-stage relationship between mortality and institutions weakens sharply and the second-stage estimates lose significance. The authors replied vigorously, defending their sources and coding, and the exchange is worth reading in full, because the disagreement is about historical sources rather than about econometrics.

Step four fails, or at least is not free

Now the untestable step. Settler mortality is a measure of the historical disease environment. Does the disease environment affect income today by any route other than institutions?

It plainly might. Malaria still suppresses productivity, school attendance and investment in the places where it is endemic, and those places are largely the places where Europeans died. Jeffrey Sachs and others argue the direct channel is substantial. The authors respond that in high-transmission areas malaria mostly kills young children and that surviving adults acquire partial immunity, so the effect on adult labour productivity is smaller than it appears. That is a real argument, and it is a debate about tropical epidemiology, not about statistics.

A second route runs through people. Edward Glaeser and colleagues argued in 2004 that Europeans who settled brought human capital as well as institutions, and that schooling predicts subsequent growth better than institutional measures do. On that reading, mortality shifted the number of educated settlers, and education rather than constitutional design is the channel. The instrument is fine and the interpretation of the second stage is wrong.

What a weak instrument does to you

The technical part is short and worth knowing, because it recurs everywhere.

Two-stage least squares uses only the variation in the suspect cause that the instrument explains. If the instrument explains very little, that usable variation is small, and two things go wrong at once. The estimate becomes biased in the direction of the ordinary regression, which is precisely the bias the instrument was supposed to remove. And the conventional standard errors become unreliable, so confidence intervals are too narrow and the result looks more precise than it is.

The standard diagnostic is the first-stage F statistic on the excluded instrument, with a rule of thumb that values below about 10 should worry you. The point of Albouy's third finding is not that the estimate changes sign. It is that once the questionable observations are removed, the first stage is no longer strong enough for the estimate to mean what it claimed.

What survives

Quite a lot, and it is important not to overcorrect.

The historical argument stands mostly on its own. That colonial administrations built different institutions in places they settled and places they extracted from is documented in administrative records, not inferred from a regression, and the contrast between settler colonies and extractive ones is visible in land law, taxation and the franchise.

Better designs exist. Comparisons within a single colonial empire hold the coloniser constant. Sharp boundary studies compare adjacent districts assigned to different institutional regimes for administrative reasons, which is a far more credible natural experiment than a cross-country instrument. And the division of Korea in 1945 is the closest thing the field has to a controlled trial: one people, one language, one economic starting point, two institutional regimes, and a divergence in income of more than an order of magnitude.

A checklist for any instrument

• Where did the instrument's values come from? Not the citation, the underlying observations. How many are measured and how many imputed?
• Is the first stage strong? Ask for the F statistic, not the significance of the second stage.
• What is the alternative route? Write down the most plausible way the instrument could affect the outcome directly, and see whether the authors address it with evidence rather than assertion.
• Whose effect is being estimated? The estimate reflects the units whose behaviour the instrument actually moves, which may be a narrow and unusual subset.
• Does the historical story survive without the statistics? If so, say so, and let the regression be corroboration rather than proof.

Common misconceptions

• A significant instrumental variables result establishes causation. It establishes causation conditional on an untestable assumption. The assumption is the argument, and the statistics cannot check it.
• If the critique is right, institutions do not matter. The critique undermines one estimate of how much they matter. The historical and boundary evidence is largely untouched by it.
• A weak instrument just makes results noisier. It also biases them toward the ordinary regression, which means a weak instrument can reproduce exactly the bias it was introduced to remove.

What to remember

The chain was clean: mortality shaped settlement, settlement shaped institutions, institutions persisted, income followed. Each link is a separate claim and each can fail separately. The first link depends on mortality figures that were substantially assigned rather than observed and that mix campaign with barracks conditions. The last link depends on the assumption that the historical disease environment does nothing to income today except through institutions, which is a claim about malaria and about human capital rather than about econometrics.

Remember: when you meet an instrument, do not ask whether it is significant. Ask where its numbers came from and what else it could be doing. Those two questions would have raised every objection in this lesson, and neither one requires any statistics at all.

Sources

1. Acemoglu, D., Johnson, S., & Robinson, J. A. (2001). The colonial origins of comparative development: An empirical investigation. American Economic Review, 91(5), 1369-1401.
2. Albouy, D. Y. (2012). The colonial origins of comparative development: An empirical investigation: Comment. American Economic Review, 102(6), 3059-3076.
3. Glaeser, E. L., La Porta, R., Lopez-de-Silanes, F., & Shleifer, A. (2004). Do institutions cause growth? Journal of Economic Growth, 9(3), 271-303.
4. Wikipedia contributors. (n.d.). Instrumental variables estimation. Wikipedia
5. World Bank. (n.d.). GDP per capita (current US dollars). data.worldbank.org

A skill you cannot take anywhere else

A trainee toolmaker in a German engineering firm spends about three and a half years in a dual apprenticeship, alternating between the workshop and a state vocational school, working to a curriculum negotiated between employers' associations, unions and the government. At the end she holds a certificate recognised by every firm in her industry and close to worthless outside it.

That is an odd investment. She has spent three years acquiring something that ties her to one narrow sector, in an economy where her firm could fail. Why would anyone do it, and why do so many Germans while so few Americans do? The answer connects a country's welfare state to its production system, and it is the best entry point into two literatures that are usually taught separately.

Esping-Andersen's two dimensions

Gosta Esping-Andersen argued in 1990 that comparing welfare states by how much they spend is close to useless, because identical spending can serve opposite purposes. He proposed two dimensions instead.

Decommodification is the degree to which a person can maintain a decent livelihood without depending on selling their labour. A benefit that replaces most of your wage, for a long time, without a means test and as a matter of right is highly decommodifying. A short, means-tested, conditional payment is barely decommodifying at all, since you must return to the labour market on almost any terms.

Stratification is what the system does to social divisions. Some welfare states deliberately preserve status differences by tying benefits to occupational category. Some divide the population into a majority buying private provision and a residual minority on public assistance. Some treat everyone identically and thereby build a shared constituency for the system.

Why this matters: two countries can spend the same share of GDP and produce completely different societies, because what matters is who is entitled, on what terms, and whether the entitlement releases them from the labour market or pushes them back into it.

Three worlds, compared

Read the last row against the third. Conservative regimes score moderately on decommodification and historically had low female employment, because the benefits presumed a male breadwinner and the services that would allow a second earner, above all childcare, were not provided. Social democratic regimes made the services the point.

What the three worlds miss

The typology's most important critique came from feminist scholarship in the early 1990s. Decommodification asks whether a person can survive without selling their labour, which presumes they are selling it. For a woman doing unpaid care work, the relevant question is nearly the opposite: whether she can enter the labour market at all, and whether she can form and maintain a household independently. Ann Orloff and others proposed adding dimensions to capture access to paid work and the capacity to sustain an autonomous household, and the addition changes the rankings, because conservative regimes score far worse on the new dimension than on the old.

Two more objections concern coverage. Southern European systems combine generous pensions for insiders with thin coverage for everyone else and heavy reliance on family, which is arguably a fourth type. And the East Asian systems subordinate social policy to economic growth, with low transfers, high private savings and heavy investment in education, which some call productivist welfare capitalism. A three-category scheme built from eighteen rich democracies was never going to travel everywhere unchanged.

Varieties of capitalism: the firm's problem

Peter Hall and David Soskice reframed the comparison in 2001 by starting from the firm rather than the state. A firm must solve coordination problems in five areas: industrial relations, vocational training, corporate governance, relations with other firms, and relations with its own employees. National institutions differ in how firms solve them, and the solutions come in two coherent bundles.

In a liberal market economy, firms coordinate mainly through competitive markets and internal hierarchy. Finance comes from equity markets that read published quarterly results. Labour markets are fluid. Skills are general and portable, acquired through formal education. Wages are set firm by firm.

In a coordinated market economy, firms coordinate through non-market relationships: employers' associations, industry-wide bargaining, cross-shareholding, long-term bank relationships and works councils. Finance is patient, supplied by banks with access to private information about the firm. Employment is long-term. Skills are industry-specific and certified collectively.

Complementarities, and the answer to the opening question

The framework's central concept is institutional complementarity: the return to one institution rises when another is present, so the bundles hang together rather than being a menu.

That answers the toolmaker. She acquires a non-portable skill because employment protection, industry-wide wage floors and replacement income during unemployment make the downside bearable. Remove any one of those and the rational choice is a general degree instead. Estevez-Abe, Iversen and Soskice made this argument explicitly: social protection is not merely a cost imposed on the economy, it is part of the machinery that produces a particular kind of skilled workforce.

From this Hall and Soskice derive comparative institutional advantage. Coordinated economies should be better at incremental innovation, improving complex products through the accumulated knowledge of long-tenured workers, and the classic examples are machine tools, specialty chemicals and precision engineering. Liberal economies should be better at radical innovation, which needs fast reallocation of capital and people, and the examples are software, biotechnology and finance. The prediction is testable against patenting patterns by technology class, and it broadly holds.

What would overturn this

Three lines of attack are live. First, convergence: if globalisation and financial liberalisation are dissolving coordinated institutions, the categories describe a past. German banks have retreated from long-term shareholding, and if patient capital disappears the complementarity argument predicts the skill system should follow. Watch apprenticeship enrolment. Second, the framework has trouble with the countries in between, particularly France and Italy, where the state coordinates rather than either markets or associations. Third, it is better at explaining stability than change, since complementarity implies bundles resist reform, yet the German labour market reforms of 2003 to 2005 changed a great deal without the system collapsing into a liberal one. A framework that cannot say which changes are survivable is incomplete.

Common misconceptions

• Bigger welfare states are more decommodifying. Spending is a poor proxy. A large system of contributory pensions preserving occupational status can decommodify less than a smaller universal one, which is exactly why Esping-Andersen introduced the dimension.
• Coordinated market economies are simply more regulated. The distinction is about how coordination happens, not how much. Much coordination in Germany occurs through employers' associations and works councils rather than through statute.
• Employment protection is a straightforward cost to competitiveness. In the complementarity account it is a precondition for the specific-skill investments that give those economies their export advantage in engineering.

Summing up

Compare welfare states by what they do rather than what they cost: how far they release people from dependence on the labour market, and what divisions they create or erase. That yields three types, plus a serious feminist objection that the whole framework asks the wrong question for anyone doing unpaid care, plus southern European and East Asian cases that fit awkwardly. Varieties of capitalism approaches the same countries from the firm's side and finds two coherent bundles held together by complementarities, which is why social protection and export specialisation turn out to be parts of one system.

In short: the German toolmaker's certificate is worthless outside her industry, and she acquires it anyway because a set of other institutions makes that risk bearable. Change one of them and the training system unravels, which is the strongest claim this framework makes and the one most worth watching.

Sources

1. Esping-Andersen, G. (1990). The Three Worlds of Welfare Capitalism. Polity Press.
2. Hall, P. A., & Soskice, D. (Eds.). (2001). Varieties of Capitalism: The Institutional Foundations of Comparative Advantage. Oxford University Press.
3. Orloff, A. S. (1993). Gender and the social rights of citizenship: The comparative analysis of gender relations and welfare states. American Sociological Review, 58(3), 303-328.
4. Wikipedia contributors. (n.d.). Varieties of Capitalism. Wikipedia
5. Ortiz-Ospina, E., & Roser, M. (n.d.). Government Spending. Our World in Data. ourworldindata.org

From a topic to a question

Here is a topic: democratic backsliding in central Europe. It is not yet a research question, because it names a subject rather than something that varies. Turn it into one by finding difference.

Poland and Hungary both had governments that spent years reshaping courts, media regulators and electoral rules. In October 2023, on a turnout of 74.4 percent against 61.7 percent four years earlier, the Polish governing party lost its majority and a new government took office. In April 2022, facing an opposition that had united behind a single list, the Hungarian governing alliance took 54.1 percent of the party-list vote and 135 of 199 seats. Two countries doing the same thing, and one of them stopped.

That is a question: why was backsliding reversed at the ballot box in Poland and not in Hungary? This lesson builds the whole comparison around it, step by step, and then breaks it deliberately to show how fragile such a design is to a single decision.

Step one: define the outcome so that it varies

Write down what you are explaining and how you would know it when you see it. The outcome here is electoral reversal of an incumbent-led erosion, which has three parts: the incumbent had measurably eroded democratic institutions, the incumbent lost office through an election, and a successor government took power. Poland scores yes on all three. Hungary scores yes on the first and no on the other two.

Now choose an indicator. V-Dem's electoral democracy index gives a continuous score, and its components let you say what fell: clean elections, freedom of expression, freedom of association. Freedom House gives a categorical status. Use both, and report the disagreement, because Lesson 4's point applies here: the instruments diverge exactly on regimes like these.

Key idea: if you cannot say what value your outcome takes in each case, and what would count as an intermediate value, you do not have a dependent variable. You have a topic with a verb attached.

Step two: choose the design and say what it buys

This is a most-similar systems design. Poland and Hungary share a great deal: post-communist transitions negotiated in 1989, accession to the European Union in the same enlargement of 2004, membership of NATO, comparable size and income levels, parliamentary systems, and governing parties that used similar methods against similar targets.

The design therefore eliminates a specific list. Post-communist legacy cannot explain the difference, because both have it. European Union membership cannot explain it. Nor can the general strategy of executive aggrandisement, since both governments pursued it. What survives is whatever the two do not share, and the job of the rest of the design is to enumerate that honestly.

Step three: select cases, and admit what they cannot settle

State the selection rule in one sentence: two European Union member states that experienced sustained incumbent-led democratic erosion after 2010, chosen because they differ on whether that erosion was reversed electorally.

Now the admission. This is a comparison of two cases with several surviving differences, which is exactly Lijphart's problem from Lesson 1. It cannot establish that any one difference caused the divergence. What it can do is eliminate the shared conditions, generate a ranked list of candidates, and specify what further evidence would discriminate among them. Anyone who tells you a two-case comparison did more than that is overselling.

Step four: derive implications that discriminate

Four candidate explanations survive. The point of this step is to write down, for each, something we should observe if it is right and should not observe if it is wrong.

Notice what the third column does. It converts a plausible story into a claim about evidence somebody could gather. The fourth row is especially useful, because the Hungarian opposition did unite and did lose, which is prima facie evidence against coordination as the decisive factor and pushes it down the ranking. That is a two-case comparison doing real work: not proving anything, but reordering the candidates.

Step five: name what would overturn it

Suppose you rank the constitutional threshold first, on the reasoning that statutes can be repealed by the majority that wins the next election while a rewritten constitution cannot. Three findings would damage that.

• If Poland's reversal stalls broadly rather than specifically, the mechanism is wrong: the argument predicts obstruction concentrated at entrenched appointments with fixed terms, not everywhere at once.
• If a Hungarian opposition ever obtains a simple majority and finds the constitutional order easy to unwind through ordinary legislation and litigation, entrenchment was doing less work than claimed.
• If a third case appears with a constitutional supermajority and a successful electoral reversal, or with only statutory change and no reversal at all, the pattern was coincidence in a sample of two.

Write these down before you gather the evidence. A hypothesis with no specified failure condition cannot be tested, only illustrated.

The second run: change one input and watch it invert

Now break the design deliberately, in the way Lesson 2 showed Przeworski breaking the income and democracy correlation.

Keep both countries. Change the outcome from reversal of erosion to level of democracy across 2015 to 2023. On that specification, Poland and Hungary look similar: both declined substantially on the V-Dem index, both were the subject of European Union rule-of-law procedures, both retained competitive elections throughout. The outcome no longer varies, which means the design is no longer a most-similar comparison at all. It has become a most-different attempt with a constant outcome, and the conclusion it supports is entirely different: that European Union membership is compatible with sustained democratic erosion, since two members sustained it.

Both conclusions are defensible. They come from the same two countries and the same data, and the only thing that changed was the definition of the dependent variable. That is the single most important lesson in this course, arriving for the third time: specify the outcome first, because it determines what your cases can possibly tell you.

How far it travels

The last step is to say what population your finding claims to be about. A ranked explanation drawn from Poland and Hungary applies, at most, to parliamentary democracies with proportional or mixed electoral systems, entrenched constitutions requiring supermajorities, and external membership constraints. It says nothing directly about presidential systems, about backsliding under first-past-the-post, or about cases with no external anchor.

And test the boundary honestly. Slovakia, which shares a great deal with both countries, elected in 2023 a government whose programme was closer to the Hungarian pattern. A framework built on a pair should be tried on the neighbours immediately, because that is where it will fail if it is going to.

Common misconceptions

• A good comparison ends in a cause. A good small-N comparison ends in a shorter list of candidates and a specified test. Treating a two-case design as if it had identified a cause is the commonest overreach in the field.
• You should pick cases you know well. Knowledge helps, and familiarity is also where selection on the outcome creeps in, because the cases we know are usually the dramatic ones. State the selection rule before you name the countries.
• Naming disconfirming evidence weakens the argument. It is the only thing that makes it an argument. Every lesson in this course ended by doing it, and if you take one habit away, take that one.

What to carry forward

Start from variation, not from a subject. Define the outcome so precisely that you could code an unfamiliar case, and check it against at least two instruments. Choose a design and state what it eliminates. Write the selection rule before the case names. List every surviving explanation and give each one an observable implication that separates it from the others. Name the evidence that would overturn your ranking. Then say what population the finding is about, and try it immediately on the nearest case you did not use.

Worth holding on to: the same two countries supported opposite conclusions in this lesson, and nothing changed except the definition of what was being explained. Every method in this course exists to protect you from that, and none of them works if the first step is skipped.

Sources

1. King, G., Keohane, R. O., & Verba, S. (1994). Designing Social Inquiry: Scientific Inference in Qualitative Research. Princeton University Press.
2. Gerring, J. (2007). Case Study Research: Principles and Practices. Cambridge University Press.
3. Levitsky, S., & Ziblatt, D. (2018). How Democracies Die. Crown.
4. V-Dem Institute. (n.d.). The V-Dem Dataset. University of Gothenburg. v-dem.net
5. Herre, B. (n.d.). Electoral Democracy Index. Our World in Data. ourworldindata.org