Modernization Theory, Lipset (1960)

Development causes democracy

The Threat of Revolution

Democracy Spurs Development

Dictatorship, Democracy, and Development (Olson, 1993)

Democratic Survival

Democracy and Development in Research

• Are the theoretical arguments for \(Democracy \rightarrow Development\) convincing?

• Are the theoretical arguments for \(Development \rightarrow Democracy\) convincing?

Democracy and Development in Research

• Paglayan, 2020

  • Reconsiders whether democratization increases the provision of education. Finds that democratization can increase education provision, but that education is also a powerful for autocrats

• Fujiwara, 2015

  • Electronic voting in Brazil reduced mistakes when voting from poorest, resulting in better accountability and more spending in health care

Democracy and Development in Research

• Ferraz and Finan, 2008

  • Random audits of Brazilian majors reduces the chances that corrupt politicians are re-elected

• Mori, et. al., 2024

  • Use gender quotas to examine if female politicians use local budget more or less effectively

• Pailler, 2018

  • Democratic competition increases deforestation!

That Correlation Again

From Acemoglu et. al. 2008

Acemoglu et. al. 2008

• The criticism: correlation of democracy and development does not mean causation

Which is the correct DAG?

The Problem

• If the confounded DAG is correct, we need to control for U

• What U do the authors propose?

Strategy 1: Fixed Effects

“We utilize two strategies to investigate the causal effect of income on democracy. Our first strategy is to control for country-specific factors affecting both income and democracy by including country fixed effects”

From Between to Within

• The standard regression compares rich countries to poor countries (between variation)

“The idea of fixed effects is to move beyond this comparison and investigate the”within-country variation,” t”

• When a country gets richer than its own average, does it become more democratic than its own average?

Why Does This Matter? Simpson’s Paradox

Simpson’s Paradox

A trend in pooled data can reverse when the data is split into groups. The overall trend is driven by differences between groups, not by the relationship within groups

FE = Unit-Level Intercepts

• FE give each unit its own intercept, just like splitting by group

• The pooled regression mixes between and within variation

• FE isolates only within-group variation, which can flip conclusions

Fixed Effects in Practice

Add a dummy variable for each country (one intercept per country, one shared slope)

Go from

To

Why Time-Invariant Only?

“If these omitted characteristics are, to a first approximation, time-invariant, the inclusion of fixed effects will remove them and this source of bias.”

• Colonial history and geography are in the past. They do not change. FE absorbs them

• But new things happen: wars, oil discoveries, coups. These change the within-country link between income and democracy in new, unexpected ways

Load and Prepare Data

pacman::p_load(readxl, tidyverse, here)

d <- read_xls(here("./slides/code/Income-and-Democracy-Data-AER-adjustment.xls"),
              sheet = 2) %>%
  filter(!is.na(lrgdpch), !is.na(polity4)) %>%
  arrange(country, year) %>%
  group_by(country) %>%
  mutate(lag_dem = lag(polity4),
         lag_income = lag(lrgdpch)) %>%
  ungroup()
# polity4: standardized democracy score (-10 to 10)
summary(d$polity4)

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  0.000   0.150   0.700   0.573   0.950   1.000 

# lrgdpch: log real GDP per capita
summary(d$lrgdpch)

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  6.094   7.297   8.149   8.167   8.960  10.413 

Cross-Sectional Correlation

This is the pattern that modernization theory predicts

ggplot(d, aes(y = polity4, x = lag_income)) +
  geom_point(alpha = 0.3) +
  geom_smooth(method = lm, se = FALSE) +
  labs(x = "Lagged Log GDP per capita", y = "Polity Score") +
  theme_bw()

OLS Regression (no FE)

mod_ols <- lm(polity4 ~ lag_dem + lag_income, data = d)
modelsummary::modelsummary(mod_ols, stars = TRUE,
  coef_omit = "Intercept",
  gof_omit = "AIC|BIC|Log.Lik|F")

Why Lag?

• Why use lagged GDP and not current GDP?

Adding Fixed Effects with lm

mod_fe <- lm(polity4 ~ lag_dem + lag_income +
               factor(country) + factor(year), data = d)
modelsummary::modelsummary(
  list("OLS" = mod_ols, "Country + Year FE" = mod_fe),
  stars = TRUE,
  coef_omit = "factor|Intercept",
  gof_omit = "AIC|BIC|Log.Lik|F|R2")

What Changed?

• In OLS, lag_income is positive and significant

• With FE, the income coefficient drops to almost zero, not distinguishable from zero

• This is Acemoglu et al.’s key finding

A Better Package: estimatr

pacman::p_load(estimatr)
# cluster-robust SEs + efficient FE
mod_robust <- lm_robust(polity4 ~ lag_dem + lag_income,
  fixed_effects = country + year,
  clusters = country, data = d)
modelsummary::modelsummary(mod_robust, stars = TRUE,
  coef_omit = "Intercept",
  gof_omit = "AIC|BIC|Log.Lik|F|R2")

Visualizing FE in Our Data

Why Are Rich Countries Democratic?

“These results naturally raise the following important question: why is there a cross-sectional correlation between income and democracy? In other words, why are rich countries democratic today?”

The Authors’ Answer

“At a statistical level, the answer is clear: even though there is no relationship between changes in income and democracy in the postwar era or over the past 100 years or so, there is a positive association over the past 500 years.”

Historical Roots of the Correlation

“In particular, for the whole world sample, the positive association is considerably weakened when we control for date of independence, early constraints on the executive, and religion.”

Do We Buy It?

• Acemoglu et al.’s claim: the income-democracy correlation reflects deep historical divergence, not a modern causal effect of income on democracy

• What do you think? Is this convincing?

• What would it take to change your mind?

Summing up

• Good theoretical reasons to think democracy is causally related to development

• Most extant work focuses on micro, sub-national dynamics

• Fixed effects control for time-invariant unobserved confounders

• If confounders change over time, FE cannot help

Strategy 2: Instrumental Variables

“Our second strategy is to use instrumental-variables (IV) regressions to estimate the impact of income on democracy”

• When FE is not enough (time-varying confounders remain), IV provides another approach

• Core idea: find a variable Z that shifts income but has no direct effect on democracy

IV Intuition

• We need a variable Z (the “instrument”) that:

  1. Is correlated with income (relevance)
  2. Affects democracy only through income (exclusion restriction)

• Z isolates the part of income variation that is “as-if random”

• Acemoglu et al. use trade-weighted world income as instrument: global economic shocks shift a country’s income but arguably do not directly cause democratization

The Exclusion Restriction

• The critical assumption: Z affects democracy ONLY through income

\[Z \rightarrow Income \rightarrow Democracy\]

• There must be no direct arrow from Z to Democracy

• This is untestable; it must be argued on substantive grounds

• If the exclusion restriction fails, IV estimates are biased

IV in Acemoglu et al.

• Instrument: trade-weighted world income

• Logic: when your trading partners get richer, your income goes up (relevance)

• Exclusion: your partners’ income growth does not directly change your political institutions

• Result: IV estimates also show no causal effect of income on democracy

Limitations of IV

• Exclusion restriction is never testable

• Results depend heavily on instrument choice

• IV estimates a local effect (for observations affected by the instrument)

• Combined with FE evidence, strengthens the case against modernization theory