Findings

Six hypotheses organized the empirical program. Two were supported cleanly, one is the book’s central result, two came back more complicated than the theory predicted, and one did not survive its own test. What follows is the ledger.

The hypotheses

What separates individuals inside a country

Chapter 5 estimates ordinary least squares with country fixed effects, so every between-country difference is absorbed and the question becomes narrower and harder: among people living in the same country, what distinguishes those with higher and lower authoritarian governance preference? All continuous predictors are standardized, so the coefficients can be read against each other directly.

Education is the largest effect, then age, then emancipative values. All three survive each other’s inclusion, so none is simply standing in for the others, and none is an artifact of the well-known correlations among them.

The trust result is the one that repays a second look. Split the composite into its three items and the single positive coefficient turns into two opposite findings: institutional trust makes people less likely to reject democracy as a system, exactly as the theory says, and more likely to endorse a strong leader or army rule. The composite is dominated by the latter two.

Chapter 5 offers a legitimation reading — that people who trust the institutions they have may be comfortable endorsing executive authority or military discipline as tools, particularly where those institutions already concentrate power — and then says the cross-sectional design cannot adjudicate between that and the obvious alternative, that people who like strong authority also tend to trust powerful institutions. The reading is labeled speculative in the book, and it is left that way here.

Whether demand predicts regime change

Chapter 6 is the hinge. If mass preference is only atmosphere, it should carry no information about what institutions do next. A discrete-time hazard model tests that directly, taking country-mean AGP from one survey wave and asking whether it moves the probability that a V-Dem autocratization episode begins in the following wave period — so the measure always precedes the outcome it is predicting.

Chapter 9 then spends a chapter trying to break it. Thirty-one specification tests: alternative item combinations for the construct, alternative latent-variable estimators, alternative cross-wave adjustments, alternative lag structures, subsample restrictions, alternative outcome codings, placebo tests, and four families of rival explanation — economic grievance, civilizational and religious accounts, elite supply-side cues, and polarization with trust disaggregated. The coefficient never reverses sign, and outside the placebos it stays significant throughout.

When waves ebb

The word “ebb” gets used loosely, and it can mean at least three incompatible things: fewer new autocratization episodes, improving average democratic quality, or a political climate less hospitable to authoritarian appeals. Chapter 8 picks one and defines it so it can be tested.

A persistent, meaningful decline in the incidence of elevated authoritarian risk — the proportion of countries whose model-estimated probability of autocratization onset within a forward ten-year window exceeds a prespecified threshold — sustained across multiple consecutive time bins, and exceeding plausible sampling uncertainty as assessed by clustered bootstrap resampling.

Chapter 1, §1.5

The definition deliberately targets incidence rather than attitude levels. A decline in mean AGP concentrated in countries that were never at risk is detectable but not consequential; it is not an ebb in the problem the book is about.

Measured that way, ebb turns out to be common rather than exceptional. Across the wave-to-wave transitions in the study period, 71 countries recorded at least one decline of at least 0.20 standardized AGP units — the book’s ebb threshold — and declines outnumber rises. The book also tabulates the rates by wave pair; that table is being re-derived and is not reproduced here.

The sequencing matters more than the totals. Countries where emancipative values were already deep when economic and institutional recovery arrived show faster and more durable declines. Where value change is shallow, the same recovery produces a short-term dip that erodes at the next shock, because there is no buffer underneath it. The practical implication is uncomfortable: the time horizon for a genuine demand-side ebb is longer than the horizon of the programs meant to produce one.

One further pattern. The countries with the most durable declines are not those with the fastest growth or the most dramatic reforms. They are the ones where economic security improved broadly, across the income distribution, rather than at the top — because the mechanism runs through individual threat appraisal, and rising aggregate GDP that leaves the median household no more secure does not reduce the input.

What the book does not claim

The association between AGP and autocratization onset is predictive, not identified. The instruments available — insecurity and institutional trust — are themselves substantive causal mechanisms, so the exclusion restriction cannot be assumed to hold. The book says this in Chapter 9 and repeats it in Chapter 10, and this site repeats it here.

Nor does the analysis show that ebb is coming. It shows that ebb has happened, that its conditions are identifiable, and that it is neither automatic nor universal. A conclusion that ended in reassurance would misrepresent the evidence; so would one that ended in fatalism.

• The full set of limitations
• Reproduce any of these estimates