Replicating Acemoglu, Johnson, and Robinson (2001): “The Colonial Origins of Comparative Development”
1 Introduction: The Primacy of Institutions in Economic Development
A central question in the literature on economic growth is the origin of the vast differences in per capita income across countries. While theories emphasizing geography, culture, and integration have been proposed, Acemoglu, Johnson, and Robinson (2001) [hereafter AJR] advance the institutions hypothesis: that differences in economic performance are primarily determined by differences in economic institutions that shape the incentives for investment and innovation. Specifically, institutions that secure property rights and constrain the power of elites are argued to be conducive to long-run prosperity.
A fundamental challenge in empirically testing this hypothesis is endogeneity. A simple OLS regression of income on a measure of institutional quality is likely to yield biased estimates due to omitted variables and, most critically, reverse causality—wealthier nations may be able to afford or may prefer to establish higher-quality institutions. To overcome this, AJR propose a robust identification strategy using an instrumental variable.
2 The Instrumental Variable Strategy
To establish a causal link, a valid instrument (Z) must satisfy two core conditions: relevance (\(Cov(Z, X) \neq 0\)) and the exclusion restriction (\(Cov(Z, \epsilon) = 0\)). AJR propose potential mortality rates of European settlers during the colonial period as an instrument for current institutional quality.
Their identification strategy rests on a causal chain that can be represented by a set of structural equations:
- The main outcome equation, where economic performance is a function of current institutions:
\[ \log y_i=\mu+\alpha R_i+\mathbf{X}_{\mathbf{i}}^{\prime} \gamma+\epsilon_i \] Here, \(y_i\) is income per capita, \(R_i\) s the measure of current institutions (protection against expropriation risk), and \(\mathbf{X}_{\mathbf{i}}^{\prime}\) is a vector of covariates.
The assumption of institutional persistence, where current institutions (\(R_i\)) are a function of early colonial institutions (\(C_i\)): \[ R_i=\lambda_R+\beta_R C_i+\mathbf{X}_{\mathbf{i}}^{\prime} \gamma_R+\nu_{R i} \]
The link between settlements and the formation of early institutions, where early institutions (\(C_i\)) depend on the extent of European settlements (\(S_i\)):
\[ C_i = \lambda_c + \beta_C S_i + \mathbf{X}_{\mathbf{i}}^{\prime} \gamma_C + \nu_{C i}\] The first link in the causal chain, where the nature of settlements (\(S_i\)) was determined by the mortality risk faced by settlers (\(log M_i\)):
\[ S_i = \lambda_S + \beta_S log M_i + \mathbf{X}_{\mathbf{i}}^{\prime} \gamma_S + \nu_{S i}\] The full causal path is thus:
\[\begin{align*} \text{Settler Mortality} \rightarrow \text{Settlements} & \rightarrow \text{Early Institutions} \\ & \rightarrow \text{Current Institutions} \rightarrow \text{Current Economic Performance} \end{align*}\]
The logic is that where the disease environment was favorable to European settlement (low mortality), colonists established “neo-Europes” with inclusive institutions focused on protecting property rights. Conversely, where mortality rates were high, they established “extractive states” with limited constraints on power, designed to plunder resources. AJR posit that these institutional frameworks persisted long after independence. The historical disease environment is argued to have no direct effect on modern income, satisfying the exclusion restriction.
The datasets used in this replication can be found on Daron Acemoglu’s MIT faculty page. The paper can be accessed from here.
3 Descriptive Statistics
The following code presents summary statistics for the key variables, offering prima facie evidence for the authors’ hypothesis by stratifying the base sample of 64 former colonies by quartiles of settler mortality.
# --- Load Data ---
ajr_data <- read_dta("D:/replication-ajr/maketable1.dta")
# --- Prepare Data Samples ---
base_sample <- ajr_data %>% filter(baseco == 1)
quartile_sample <- base_sample %>%
filter(!is.na(extmort4)) %>%
mutate(quartile = ntile(extmort4, 4))
# --- Define Helper Functions to Calculate Stats ---
get_means <- function(df, vars) {
df %>% summarise(across(all_of(vars), ~mean(.x, na.rm = TRUE)))
}
get_sds <- function(df, vars) {
df %>% summarise(across(all_of(vars), ~sd(.x, na.rm = TRUE)))
}
# --- Calculate All Statistics ---
all_vars <- c("logpgp95", "loghjypl", "avexpr", "cons00a",
"cons1", "democ00a", "euro1900", "logem4")
# Calculate means and sds for all groups
means_w <- get_means(ajr_data, all_vars)
sds_w <- get_sds(ajr_data, all_vars)
means_b <- get_means(base_sample, all_vars)
sds_b <- get_sds(base_sample, all_vars)
means_q <- quartile_sample %>% group_by(quartile) %>% get_means(all_vars)
sds_q <- quartile_sample %>% group_by(quartile) %>% get_sds(all_vars)
# Get observation counts
obs_w <- nrow(ajr_data)
obs_b <- nrow(base_sample)
obs_q <- quartile_sample %>% group_by(quartile) %>% count()
# --- Assemble the Final Table ---
final_table <- tibble(
` ` = c("Log GDP per capita (PPP) in 1995", "(Std. Dev.)",
"Log output per worker in 1988", "(Std. Dev.)",
"Average protection against expropriation risk, 1985-1995", "(Std. Dev.)",
"Constraint on executive in 1900", "(Std. Dev.)",
"Constraint on executive in first year of independence", "(Std. Dev.)",
"Democracy in 1900", "(Std. Dev.)",
"European settlements in 1900", "(Std. Dev.)",
"Log European settler mortality", "(Std. Dev.)",
"Number of observations", ""),
`Whole world` = c(means_w$logpgp95, sds_w$logpgp95, means_w$loghjypl, sds_w$loghjypl, means_w$avexpr, sds_w$avexpr,
means_w$cons00a, sds_w$cons00a, means_w$cons1, sds_w$cons1, means_w$democ00a, sds_w$democ00a,
means_w$euro1900, sds_w$euro1900, means_w$logem4, sds_w$logem4, obs_w, NA),
`Base sample` = c(means_b$logpgp95, sds_b$logpgp95, means_b$loghjypl, sds_b$loghjypl, means_b$avexpr, sds_b$avexpr,
means_b$cons00a, sds_b$cons00a, means_b$cons1, sds_b$cons1, means_b$democ00a, sds_b$democ00a,
means_b$euro1900, sds_b$euro1900, means_b$logem4, sds_b$logem4, obs_b, NA),
`(1)` = c(means_q$logpgp95[1], sds_q$logpgp95[1], means_q$loghjypl[1], sds_q$loghjypl[1], means_q$avexpr[1], sds_q$avexpr[1],
means_q$cons00a[1], sds_q$cons00a[1], means_q$cons1[1], sds_q$cons1[1], means_q$democ00a[1], sds_q$democ00a[1],
means_q$euro1900[1], sds_q$euro1900[1], means_q$logem4[1], sds_q$logem4[1], obs_q$n[1], NA),
`(2)` = c(means_q$logpgp95[2], sds_q$logpgp95[2], means_q$loghjypl[2], sds_q$loghjypl[2], means_q$avexpr[2], sds_q$avexpr[2],
means_q$cons00a[2], sds_q$cons00a[2], means_q$cons1[2], sds_q$cons1[2], means_q$democ00a[2], sds_q$democ00a[2],
means_q$euro1900[2], sds_q$euro1900[2], means_q$logem4[2], sds_q$logem4[2], obs_q$n[2], NA),
`(3)` = c(means_q$logpgp95[3], sds_q$logpgp95[3], means_q$loghjypl[3], sds_q$loghjypl[3], means_q$avexpr[3], sds_q$avexpr[3],
means_q$cons00a[3], sds_q$cons00a[3], means_q$cons1[3], sds_q$cons1[3], means_q$democ00a[3], sds_q$democ00a[3],
means_q$euro1900[3], sds_q$euro1900[3], means_q$logem4[3], sds_q$logem4[3], obs_q$n[3], NA),
`(4)` = c(means_q$logpgp95[4], sds_q$logpgp95[4], means_q$loghjypl[4], sds_q$loghjypl[4], means_q$avexpr[4], sds_q$avexpr[4],
means_q$cons00a[4], sds_q$cons00a[4], means_q$cons1[4], sds_q$cons1[4], means_q$democ00a[4], sds_q$democ00a[4],
means_q$euro1900[4], sds_q$euro1900[4], means_q$logem4[4], sds_q$logem4[4], obs_q$n[4], NA)
)
# --- Generate and Style the HTML Table ---
final_table %>%
# Use a single, robust mutate with case_when to format all numbers
mutate(across(where(is.numeric), ~ case_when(
# Condition 0: If the value is NA, make it a blank string first.
is.na(.) ~ "",
# Condition 1: Handle the "Number of observations" row
` `[row_number()] == "Number of observations" ~ sprintf("%.0f", .),
# Condition 2: Handle the standard deviation rows
grepl("Std. Dev.", ` `[row_number()]) ~ sprintf("(%.2f)", .),
# Condition 3 (Default): Handle all other numeric rows (the means)
TRUE ~ sprintf("%.2f", .)
))) %>%
# Replace any other NA values with a blank string (mostly for safety)
mutate(across(everything(), ~ replace_na(., ""))) %>%
# Generate the kable table
kable("html", caption = "<b>TABLE 1—DESCRIPTIVE STATISTICS</b>", align = "r") %>%
kable_styling(bootstrap_options = c("striped", "condensed"), full_width = FALSE) %>%
add_header_above(c(" " = 3, "By quartiles of mortality" = 4)) %>%
column_spec(1, extra_css = "padding-left: 2em;") %>%
row_spec(seq(1, 16, 2), bold = TRUE)| Whole world | Base sample |
|
|
|
|
|
|---|---|---|---|---|---|---|
| Log GDP per capita (PPP) in 1995 | 8.30 | 8.06 | 8.84 | 8.39 | 7.83 | 7.18 |
| (Std. Dev.) | (1.07) | (1.04) | (1.23) | (0.64) | (0.79) | (0.61) |
| Log output per worker in 1988 | -1.71 | -1.93 | -1.06 | -1.49 | -2.11 | -3.05 |
| (Std. Dev.) | (1.08) | (0.98) | (0.82) | (0.42) | (0.75) | (0.47) |
| Average protection against expropriation risk, 1985-1995 | 6.99 | 6.52 | 7.74 | 6.45 | 6.13 | 5.74 |
| (Std. Dev.) | (1.83) | (1.47) | (1.51) | (1.01) | (1.19) | (1.39) |
| Constraint on executive in 1900 | 1.85 | 2.25 | 3.86 | 3.06 | 1.13 | 1.00 |
| (Std. Dev.) | (1.79) | (2.11) | (2.91) | (2.02) | (0.52) | (0.00) |
| Constraint on executive in first year of independence | 3.63 | 3.40 | 4.57 | 2.62 | 3.27 | 3.27 |
| (Std. Dev.) | (2.39) | (2.39) | (2.95) | (1.96) | (2.28) | (2.15) |
| Democracy in 1900 | 1.12 | 1.64 | 4.07 | 2.47 | 0.20 | 0.00 |
| (Std. Dev.) | (2.54) | (3.00) | (4.30) | (2.88) | (0.41) | (0.00) |
| European settlements in 1900 | 30.10 | 16.18 | 31.50 | 25.00 | 8.18 | 1.00 |
| (Std. Dev.) | (41.86) | (25.53) | (42.22) | (15.60) | (12.19) | (2.73) |
| Log European settler mortality | 4.61 | 4.66 | 3.16 | 4.29 | 4.88 | 6.29 |
| (Std. Dev.) | (1.30) | (1.26) | (0.69) | (0.05) | (0.39) | (0.76) |
| Number of observations | 376 | 64 | 16 | 16 | 16 | 16 |
The descriptive statistics in Table 1 reveal a strong, negative raw correlation between the instrument (settler mortality) and both the outcome (log GDP per capita) and the endogenous regressor (institutional quality). As we move from the lowest mortality quartile to the highest, average income and the average quality of institutions systematically decline. This provides initial, non-causal support for the relationships underlying the IV strategy.
4 OLS Regressions
The following code establishes a baseline correlation using OLS, regressing log GDP per capita on a measure of institutional quality (average protection against expropriation risk). These results serve as a benchmark against which the 2SLS estimates can be compared.
\[ log(GDO_i)= \beta_0 + \beta_1Risk_i + \gamma^{\prime}X_i + \epsilon_i\]
# --- Load Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable2.dta")
# --- Create Data Subsets ---
base_sample <- ajr_dta %>% filter(baseco == 1)
# --- Run All Regressions ---
# Note on results: The public data has minor differences from the paper's,
# so coefficients may not match perfectly.
model_list <- list(
"(1)" = feols(logpgp95 ~ avexpr, data = ajr_dta, se = "hetero"),
"(2)" = feols(logpgp95 ~ avexpr, data = base_sample, se = "hetero"),
"(3)" = feols(logpgp95 ~ avexpr + lat_abst, data = ajr_dta, se = "hetero"),
"(4)" = feols(logpgp95 ~ avexpr + lat_abst + africa + asia + other, data = ajr_dta, se = "hetero"),
"(5)" = feols(logpgp95 ~ avexpr + lat_abst, data = base_sample, se = "hetero"),
"(6)" = feols(logpgp95 ~ avexpr + lat_abst + africa + asia + other, data = base_sample, se = "hetero"),
"(7)" = feols(loghjypl ~ avexpr, data = ajr_dta, se = "hetero"),
"(8)" = feols(loghjypl ~ avexpr, data = base_sample, se = "hetero")
)## NOTE: 52 observations removed because of NA values (LHS: 15, RHS: 42).## NOTE: 52 observations removed because of NA values (LHS: 15, RHS: 43).
## NOTE: 52 observations removed because of NA values (LHS: 15, RHS: 43).## NOTE: 55 observations removed because of NA values (LHS: 40, RHS: 42).## NOTE: 3 observations removed because of NA values (LHS: 3).# --- Define Table Components ---
gof_map <- list(
list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)
)
coef_map <- c("avexpr" = "Average Expropriation Risk",
"lat_abst" = "Distance from Equator",
"africa" = "Africa",
"asia" = "Asia",
"other" = "Other continents")
modelsummary(
model_list,
output = "gt",
title = "Table 2: OLS Regressions",
coef_map = coef_map,
gof_map = gof_map,
stars = c('*' = .1, '**' = .05, '***' = .01),
add_rows = tribble(
~term, ~"(1)", ~"(2)", ~"(3)", ~"(4)", ~"(5)", ~"(6)", ~"(7)", ~"(8)",
"Base Sample", "No", "Yes", "No", "No", "Yes", "Yes", "No", "Yes",
"Continent Dummies", "No", "No", "No", "Yes", "No", "Yes", "No", "No"
),
notes = "Notes: Robust standard errors are in parentheses."
) %>%
# Use gt's tab_spanner to create the correct headers
tab_spanner(
label = "Dependent variable: Log GDP per capita, 1995",
columns = 2:7 # Selects columns for models (1) through (6)
) %>%
tab_spanner(
label = "Dependent variable: Log output per worker, 1988",
columns = 8:9 # Selects columns for models (7) and (8)
)|
Dependent variable: Log GDP per capita, 1995
|
Dependent variable: Log output per worker, 1988
|
|||||||
|---|---|---|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
| Average Expropriation Risk | 0.532*** | 0.522*** | 0.463*** | 0.390*** | 0.468*** | 0.401*** | 0.446*** | 0.457*** |
| (0.029) | (0.050) | (0.052) | (0.051) | (0.063) | (0.064) | (0.029) | (0.050) | |
| Distance from Equator | 0.872* | 0.333 | 1.577** | 0.875 | ||||
| (0.499) | (0.442) | (0.651) | (0.614) | |||||
| Africa | -0.916*** | -0.881*** | ||||||
| (0.154) | (0.156) | |||||||
| Asia | -0.153 | -0.577* | ||||||
| (0.181) | (0.299) | |||||||
| Other continents | 0.304* | 0.107 | ||||||
| (0.174) | (0.223) | |||||||
| Num. Obs. | 111 | 64 | 111 | 111 | 64 | 64 | 108 | 61 |
| R-squared | 0.611 | 0.540 | 0.623 | 0.715 | 0.574 | 0.714 | 0.554 | 0.486 |
| Base Sample | No | Yes | No | No | Yes | Yes | No | Yes |
| Continent Dummies | No | No | No | Yes | No | Yes | No | No |
| * p < 0.1, ** p < 0.05, *** p < 0.01 | ||||||||
| Notes: Robust standard errors are in parentheses. | ||||||||
The OLS estimates consistently show a positive and statistically significant coefficient on institutional quality. However, as noted, these results are likely biased. The direction of the bias is ambiguous: omitted variables could bias the coefficient upwards or downwards, while reverse causality would likely lead to an upward bias. The subsequent IV estimates will be crucial for identifying the direction and magnitude of this net bias.
5 Determinants of Institutions
The following code provides the empirical evidence for the structural model outlined in Equations (2), (3), and (4),thereby establishing instrument relevance.It estimates OLS models such as the following, which directly tests Equation (2):
\[ Risk_i = \delta_0 +\delta_1\text{Early Institutions}_i + \gamma^{\prime}X_i + \nu_i\]
# --- Load and Prepare Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable3.dta")
# Create the main sample used in most regressions for Table 3
main_sample <- ajr_dta %>%
filter(excolony == 1, !is.na(extmort4)) %>%
mutate(euro1900 = euro1900 / 100)
# Create the smaller subsample for regressions that also require log GDP data
lpgp_sample <- main_sample %>%
filter(!is.na(logpgp95))
# --- Run Regressions for Panel A ---
panel_A_models <- list(
"(1)" = feols(avexpr ~ cons00a, data = main_sample),
"(2)" = feols(avexpr ~ cons00a + lat_abst, data = main_sample),
"(3)" = feols(avexpr ~ democ00a, data = main_sample),
"(4)" = feols(avexpr ~ democ00a + lat_abst, data = main_sample),
"(5)" = feols(avexpr ~ indtime + cons1, data = main_sample),
"(6)" = feols(avexpr ~ indtime + cons1 + lat_abst, data = main_sample),
"(7)" = feols(avexpr ~ euro1900, data = main_sample),
"(8)" = feols(avexpr ~ euro1900 + lat_abst, data = main_sample),
"(9)" = feols(avexpr ~ logem4, data = lpgp_sample), # Uses smaller sample
"(10)" = feols(avexpr ~ logem4 + lat_abst, data = lpgp_sample) # Uses smaller sample
)## NOTE: 22 observations removed because of NA values (LHS: 18, RHS: 10).
## NOTE: 22 observations removed because of NA values (LHS: 18, RHS: 10).## NOTE: 23 observations removed because of NA values (LHS: 18, RHS: 13).
## NOTE: 23 observations removed because of NA values (LHS: 18, RHS: 13).## NOTE: 22 observations removed because of NA values (LHS: 18, RHS: 12).
## NOTE: 22 observations removed because of NA values (LHS: 18, RHS: 12).## NOTE: 19 observations removed because of NA values (LHS: 18, RHS: 2).
## NOTE: 19 observations removed because of NA values (LHS: 18, RHS: 2).## NOTE: 16 observations removed because of NA values (LHS: 16, RHS: 5).
## NOTE: 16 observations removed because of NA values (LHS: 16, RHS: 5).# --- Generate Table for Panel A ---
modelsummary(
panel_A_models,
output = "gt",
title = "Table 3, Panel A: Determinants of Institutions",
coef_map = c("cons00a" = "Constraint on Executive in 1900", "democ00a" = "Democracy in 1900",
"cons1" = "Constraint on Executive at Independence", "indtime" = "Date of Independence",
"euro1900" = "European Settlements in 1900", "logem4" = "Log Settler Mortality",
"lat_abst" = "Distance from Equator"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE,
notes = "Notes: Dependent Variable is Average Expropriation Risk, 1985-95. Standard errors in parentheses."
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Constraint on Executive in 1900 | 0.321*** | 0.262** | ||||||||
| (0.081) | (0.089) | |||||||||
| Democracy in 1900 | 0.244*** | 0.209** | ||||||||
| (0.057) | (0.066) | |||||||||
| Constraint on Executive at Independence | 0.247** | 0.220** | ||||||||
| (0.076) | (0.075) | |||||||||
| Date of Independence | 0.009** | 0.006+ | ||||||||
| (0.003) | (0.003) | |||||||||
| European Settlements in 1900 | 3.177*** | 2.996*** | ||||||||
| (0.612) | (0.776) | |||||||||
| Log Settler Mortality | -0.607*** | -0.510*** | ||||||||
| (0.127) | (0.141) | |||||||||
| Distance from Equator | 2.179 | 1.552 | 2.718+ | 0.580 | 2.002 | |||||
| (1.416) | (1.471) | (1.429) | (1.514) | (1.337) | ||||||
| Num. Obs. | 63 | 63 | 62 | 62 | 63 | 63 | 66 | 66 | 64 | 64 |
| R-squared | 0.203 | 0.233 | 0.236 | 0.251 | 0.194 | 0.241 | 0.296 | 0.298 | 0.270 | 0.296 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
| Notes: Dependent Variable is Average Expropriation Risk, 1985-95. Standard errors in parentheses. | ||||||||||
# --- Run Regressions for Panel B ---
panel_B_models <- list(
# Dep Var: Constraint on Executive in 1900
"(1)" = feols(cons00a ~ euro1900, data = lpgp_sample),
"(2)" = feols(cons00a ~ euro1900 + lat_abst, data = lpgp_sample),
"(3)" = feols(cons00a ~ logem4, data = main_sample),
"(4)" = feols(cons00a ~ logem4 + lat_abst, data = main_sample),
# Dep Var: Democracy in 1900
"(5)" = feols(democ00a ~ euro1900, data = lpgp_sample),
"(6)" = feols(democ00a ~ euro1900 + lat_abst, data = lpgp_sample),
"(7)" = feols(democ00a ~ logem4, data = lpgp_sample),
"(8)" = feols(democ00a ~ logem4 + lat_abst, data = lpgp_sample),
# Dep Var: European Settlements in 1900
"(9)" = feols(euro1900 ~ logem4, data = lpgp_sample),
"(10)" = feols(euro1900 ~ logem4 + lat_abst, data = lpgp_sample)
)## NOTE: 10 observations removed because of NA values (LHS: 9, RHS: 2).## NOTE: 10 observations removed because of NA values (LHS: 9, RHS: 2).## NOTE: 10 observations removed because of NA values (LHS: 10, RHS: 5).
## NOTE: 10 observations removed because of NA values (LHS: 10, RHS: 5).## NOTE: 13 observations removed because of NA values (LHS: 12, RHS: 2).
## NOTE: 13 observations removed because of NA values (LHS: 12, RHS: 2).## NOTE: 12 observations removed because of NA values (LHS: 12, RHS: 5).
## NOTE: 12 observations removed because of NA values (LHS: 12, RHS: 5).## NOTE: 7 observations removed because of NA values (LHS: 2, RHS: 5).
## NOTE: 7 observations removed because of NA values (LHS: 2, RHS: 5).# --- Generate Table for Panel B ---
modelsummary(
panel_B_models,
output = "gt",
title = "Table 3, Panel B: Determinants of Early Institutions",
coef_map = c("euro1900" = "European Settlements in 1900", "logem4" = "Log Settler Mortality",
"lat_abst" = "Distance from Equator"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE,
notes = "Notes: Standard errors in parentheses."
) %>%
# Add spanners to clarify the dependent variable for each set of columns
tab_spanner(label = "Dep. Var: Constraint on Executive in 1900", columns = 2:5) %>%
tab_spanner(label = "Dep. Var: Democracy in 1900", columns = 6:9) %>%
tab_spanner(label = "Dep. Var: European Settlements in 1900", columns = 10:11)|
Dep. Var: Constraint on Executive in 1900
|
Dep. Var: Democracy in 1900
|
Dep. Var: European Settlements in 1900
|
||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
| European Settlements in 1900 | 5.491*** | 5.385*** | 8.566*** | 8.056*** | ||||||
| (0.727) | (0.929) | (0.932) | (1.190) | |||||||
| Log Settler Mortality | -0.821*** | -0.654*** | -1.221*** | -0.877*** | -0.112*** | -0.071*** | ||||
| (0.167) | (0.181) | (0.242) | (0.253) | (0.020) | (0.020) | |||||
| Distance from Equator | 0.333 | 3.634* | 1.605 | 7.569** | 0.871*** | |||||
| (1.809) | (1.716) | (2.315) | (2.425) | (0.189) | ||||||
| Num. Obs. | 70 | 70 | 75 | 75 | 67 | 67 | 68 | 68 | 73 | 73 |
| R-squared | 0.456 | 0.456 | 0.248 | 0.292 | 0.565 | 0.568 | 0.278 | 0.372 | 0.305 | 0.467 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
| Notes: Standard errors in parentheses. | ||||||||||
This table provides strong evidence for the first stage of the IV strategy. Panel A shows that settler mortality is a strong, negative, and significant predictor of current institutional quality. Panel B demonstrates the mechanism: settler mortality is a significant predictor of early institutional choices and the extent of European settlement. These results confirm the relevance of the instrument.
6 2SLS Estimates of Instituitons on Performance
The following code presents the central findings of the paper, reporting the two-stage least squares (2SLS) estimates of the causal effect of institutions on economic performance.
First Stage: The first stage regresses the endogenous variable (institutions) on the instrument (settler mortality) and any exogenous controls.
\[ Risk_i= \pi_0 + \pi_1log(Mortality_i) + \pi_2^{\prime}X_i + \nu_i\]
Second Stage: The second stage regresses the outcome (log GDP) on the predicted values of institutions from the first stage, controlling for other exogenous factors.
\[ log(GDP_i)= \beta_0 + \beta_1\widehat{Risk_i} + \beta_2^{\prime}X_i + \epsilon_i\]
# --- Load and Prepare Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable4.dta")
# Create the base sample and the `other_cont` dummy variable
base_sample <- ajr_dta %>%
filter(baseco == 1) %>%
mutate(other_cont = if_else(shortnam %in% c("AUS", "MLT", "NZL"), 1, 0))
# Create the required subsamples
no_neo_europes_sample <- base_sample %>% filter(rich4 != 1)
no_africa_sample <- base_sample %>% filter(africa != 1)
# --- Run All IV Regressions ---
iv_models <- list(
"(1)" = feols(logpgp95 ~ 1 | avexpr ~ logem4, data = base_sample),
"(2)" = feols(logpgp95 ~ lat_abst | avexpr ~ logem4, data = base_sample),
"(3)" = feols(logpgp95 ~ 1 | avexpr ~ logem4, data = no_neo_europes_sample),
"(4)" = feols(logpgp95 ~ lat_abst | avexpr ~ logem4, data = no_neo_europes_sample),
"(5)" = feols(logpgp95 ~ 1 | avexpr ~ logem4, data = no_africa_sample),
"(6)" = feols(logpgp95 ~ lat_abst | avexpr ~ logem4, data = no_africa_sample),
"(7)" = feols(logpgp95 ~ africa + asia + other_cont | avexpr ~ logem4, data = base_sample),
"(8)" = feols(logpgp95 ~ lat_abst + africa + asia + other_cont | avexpr ~ logem4, data = base_sample),
"(9)" = feols(loghjypl ~ 1 | avexpr ~ logem4, data = base_sample)
)## NOTE: 3 observations removed because of NA values (LHS: 3).# --- Generate Table for Panel A (2SLS Results) ---
modelsummary(
iv_models,
output = "gt",
title = "Table 4, Panel A: Two-Stage Least Squares Estimates",
coef_map = c("fit_avexpr" = "Average protection against expropriation risk 1985-1995",
"lat_abst" = "Latitude",
"africa" = "Africa dummy",
"asia" = "Asia dummy",
"other_cont" = "'Other' continent dummy"),
gof_map = "nobs",
stars = TRUE,
notes = "Notes: 2SLS estimates with standard errors in parentheses."
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Average protection against expropriation risk 1985-1995 | 0.944*** | 0.996*** | 1.281*** | 1.212** | 0.578*** | 0.576*** | 0.982** | 1.107* | 0.981*** |
| (0.157) | (0.222) | (0.358) | (0.354) | (0.098) | (0.117) | (0.299) | (0.464) | (0.171) | |
| Latitude | -0.647 | 0.939 | 0.038 | -1.178 | |||||
| (1.335) | (1.463) | (0.835) | (1.755) | ||||||
| Africa dummy | -0.464 | -0.437 | |||||||
| (0.358) | (0.424) | ||||||||
| Asia dummy | -0.924* | -1.047+ | |||||||
| (0.400) | (0.525) | ||||||||
| 'Other' continent dummy | -0.941 | -0.990 | |||||||
| (0.848) | (0.998) | ||||||||
| Num.Obs. | 64 | 64 | 60 | 60 | 37 | 37 | 64 | 64 | 61 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||
| Notes: 2SLS estimates with standard errors in parentheses. | |||||||||
# --- Generate Table for Panel B (First Stage Results) ---
first_stage_models <- purrr::map(iv_models, ~.$iv_first_stage$avexpr)
# --------------------------------
modelsummary(
first_stage_models,
output = "gt",
title = "Table 4, Panel B: First Stage for Average Protection Against Expropriation Risk",
coef_map = c("logem4" = "Log European settler mortality",
"lat_abst" = "Latitude",
"africa" = "Africa dummy",
"asia" = "Asia dummy",
"other_cont" = "'Other' continent dummy",
"(Intercept)" = "Constant"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE,
notes = "Notes: First-stage OLS regressions. Dependent variable is Expropriation Risk. Standard errors in parentheses."
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Log European settler mortality | -0.607*** | -0.510*** | -0.391** | -0.394** | -1.210*** | -1.137*** | -0.432* | -0.340+ | -0.629*** |
| (0.127) | (0.141) | (0.133) | (0.141) | (0.219) | (0.245) | (0.173) | (0.183) | (0.129) | |
| Latitude | 2.002 | -0.106 | 0.990 | 2.009 | |||||
| (1.337) | (1.487) | (1.428) | (1.391) | ||||||
| Africa dummy | -0.269 | -0.258 | |||||||
| (0.413) | (0.410) | ||||||||
| Asia dummy | 0.333 | 0.472 | |||||||
| (0.498) | (0.503) | ||||||||
| 'Other' continent dummy | 1.241 | 1.062 | |||||||
| (0.842) | (0.844) | ||||||||
| Constant | 9.341*** | 8.529*** | 8.184*** | 8.216*** | 11.844*** | 11.344*** | 8.538*** | 7.729*** | 9.456*** |
| (0.611) | (0.812) | (0.657) | (0.800) | (0.898) | (1.157) | (0.783) | (0.957) | (0.625) | |
| Num. Obs. | 64 | 64 | 60 | 60 | 37 | 37 | 64 | 64 | 61 |
| R-squared | 0.270 | 0.296 | 0.130 | 0.130 | 0.466 | 0.473 | 0.304 | 0.328 | 0.287 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||
| Notes: First-stage OLS regressions. Dependent variable is Expropriation Risk. Standard errors in parentheses. | |||||||||
# --- Run Regressions for Panel C ---
panel_C_models <- list(
"(1)" = feols(logpgp95 ~ avexpr, data = base_sample),
"(2)" = feols(logpgp95 ~ avexpr + lat_abst, data = base_sample),
"(3)" = feols(logpgp95 ~ avexpr, data = no_neo_europes_sample),
"(4)" = feols(logpgp95 ~ avexpr + lat_abst, data = no_neo_europes_sample),
"(5)" = feols(logpgp95 ~ avexpr, data = no_africa_sample),
"(6)" = feols(logpgp95 ~ avexpr + lat_abst, data = no_africa_sample),
"(7)" = feols(logpgp95 ~ avexpr + africa + asia + other_cont, data = base_sample),
"(8)" = feols(logpgp95 ~ avexpr + lat_abst + africa + asia + other_cont, data = base_sample),
"(9)" = feols(loghjypl ~ avexpr, data = base_sample)
)## NOTE: 3 observations removed because of NA values (LHS: 3).# --- Generate Table for Panel C ---
modelsummary(
panel_C_models,
output = "gt",
title = "Table 4, Panel C: OLS Regressions",
coef_map = c("avexpr" = "Protection Against Expropriation Risk", "lat_abst" = "Distance from Equator",
"africa" = "Africa", "asia" = "Asia", "other_cont" = "Other Continents"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE,
notes = "Notes: OLS regressions with standard errors in parentheses."
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Protection Against Expropriation Risk | 0.522*** | 0.468*** | 0.487*** | 0.471*** | 0.482*** | 0.466*** | 0.424*** | 0.401*** | 0.457*** |
| (0.061) | (0.064) | (0.076) | (0.074) | (0.065) | (0.071) | (0.057) | (0.059) | (0.061) | |
| Distance from Equator | 1.577* | 1.802* | 0.462 | 0.875 | |||||
| (0.710) | (0.852) | (0.731) | (0.628) | ||||||
| Africa | -0.918*** | -0.881*** | |||||||
| (0.169) | (0.170) | ||||||||
| Asia | -0.631** | -0.577* | |||||||
| (0.230) | (0.231) | ||||||||
| Other Continents | 0.216 | 0.107 | |||||||
| (0.377) | (0.382) | ||||||||
| Num. Obs. | 64 | 64 | 60 | 60 | 37 | 37 | 64 | 64 | 61 |
| R-squared | 0.540 | 0.574 | 0.414 | 0.456 | 0.611 | 0.616 | 0.704 | 0.714 | 0.486 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||
| Notes: OLS regressions with standard errors in parentheses. | |||||||||
Panel A shows a large, positive, and statistically significant causal effect of institutional quality on log GDP per capita. Notably, the 2SLS coefficient on institutions is substantially larger than the OLS coefficient from Table 2. This suggests that the OLS estimates were biased downwards, potentially due to measurement error in the institutions variable, which can cause attenuation bias. The results are robust across various subsamples. Panel B confirms the first-stage relationship is strong and significant in all specifications.
7 IV Regressions with Additional Controls
The following code begins to test the exclusion restriction by subjecting the baseline IV estimate to the inclusion of controls for other historical factors that could be correlated with colonial experience, such as colonial origin, legal origin, and religion.
\[ log(GDP_i)= \beta_0 + \beta_1\widehat{Risk_i}+ \gamma_1Latitude_i + \gamma_2\text{Colony Dummies}_i + \gamma_3Religion_i +...\]
# --- Load and Prepare Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable5.dta")
# Create the base sample
base_sample <- ajr_dta %>% filter(baseco == 1)
# Create the British colonies subsample
brit_colonies_sample <- base_sample %>% filter(f_brit == 1)
# --- Run IV and OLS Regressions ---
# Run IV Models (for Panels A and B)
iv_models <- list(
"(1)" = feols(logpgp95 ~ f_brit + f_french | avexpr ~ logem4, data = base_sample),
"(2)" = feols(logpgp95 ~ lat_abst + f_brit + f_french | avexpr ~ logem4, data = base_sample),
"(3)" = feols(logpgp95 ~ 1 | avexpr ~ logem4, data = brit_colonies_sample),
"(4)" = feols(logpgp95 ~ lat_abst | avexpr ~ logem4, data = brit_colonies_sample),
"(5)" = feols(logpgp95 ~ sjlofr | avexpr ~ logem4, data = base_sample),
"(6)" = feols(logpgp95 ~ lat_abst + sjlofr | avexpr ~ logem4, data = base_sample),
"(7)" = feols(logpgp95 ~ catho80 + muslim80 + no_cpm80 | avexpr ~ logem4, data = base_sample),
"(8)" = feols(logpgp95 ~ lat_abst + catho80 + muslim80 + no_cpm80 | avexpr ~ logem4, data = base_sample),
"(9)" = feols(logpgp95 ~ f_french + sjlofr + catho80 + muslim80 + no_cpm80 | avexpr ~ logem4, data = base_sample),
"(10)" = feols(logpgp95 ~ lat_abst + f_french + sjlofr + catho80 + muslim80 + no_cpm80 | avexpr ~ logem4, data = base_sample)
)
# Run OLS Models (for Panel C)
ols_models <- list(
"(1)" = feols(logpgp95 ~ avexpr + f_brit + f_french, data = base_sample),
"(2)" = feols(logpgp95 ~ avexpr + lat_abst + f_brit + f_french, data = base_sample),
"(3)" = feols(logpgp95 ~ avexpr, data = brit_colonies_sample),
"(4)" = feols(logpgp95 ~ avexpr + lat_abst, data = brit_colonies_sample),
"(5)" = feols(logpgp95 ~ avexpr + sjlofr, data = base_sample),
"(6)" = feols(logpgp95 ~ avexpr + lat_abst + sjlofr, data = base_sample),
"(7)" = feols(logpgp95 ~ avexpr + catho80 + muslim80 + no_cpm80, data = base_sample),
"(8)" = feols(logpgp95 ~ avexpr + lat_abst + catho80 + muslim80 + no_cpm80, data = base_sample),
"(9)" = feols(logpgp95 ~ avexpr + lat_abst + f_french + sjlofr + catho80 + muslim80 + no_cpm80, data = base_sample)
)
# --- Generate Tables for Each Panel ---
# Panel A: 2SLS with Additional Controls
modelsummary(
iv_models,
output = "gt",
title = "Table 5, Panel A: IV Regressions of Log GDP Per Capita with Additional Controls",
coef_map = c("fit_avexpr" = "Average Expropriation Risk", "lat_abst" = "Latitude",
"f_brit" = "British Colony Dummy", "f_french" = "French Colony Dummy",
"sjlofr" = "French Legal Origin", "catho80" = "Catholic Religion Dummy",
"muslim80" = "Muslim Religion Dummy", "no_cpm80" = "Other Religion Dummy"),
gof_map = "nobs",
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 1.078*** | 1.155** | 1.066*** | 1.339* | 1.080*** | 1.181*** | 0.917*** | 1.006*** | 1.012*** | 1.212** |
| (0.218) | (0.337) | (0.244) | (0.516) | (0.191) | (0.291) | (0.147) | (0.252) | (0.187) | (0.395) | |
| Latitude | -0.751 | -2.986 | -1.126 | -0.938 | -1.794 | |||||
| (1.675) | (3.214) | (1.560) | (1.503) | (2.133) | ||||||
| British Colony Dummy | -0.778* | -0.795* | ||||||||
| (0.354) | (0.393) | |||||||||
| French Colony Dummy | -0.117 | -0.058 | 0.290 | 0.396 | ||||||
| (0.355) | (0.419) | (0.378) | (0.495) | |||||||
| French Legal Origin | 0.887** | 0.962* | 0.234 | 0.294 | ||||||
| (0.324) | (0.394) | (0.411) | (0.519) | |||||||
| Catholic Religion Dummy | 0.009 | 0.008 | 0.007 | 0.005 | ||||||
| (0.009) | (0.010) | (0.011) | (0.014) | |||||||
| Muslim Religion Dummy | 0.000 | 0.000 | -0.002 | -0.003 | ||||||
| (0.009) | (0.010) | (0.010) | (0.012) | |||||||
| Other Religion Dummy | -0.007 | -0.010 | -0.008 | -0.014 | ||||||
| (0.010) | (0.012) | (0.011) | (0.015) | |||||||
| Num.Obs. | 64 | 64 | 25 | 25 | 64 | 64 | 64 | 64 | 64 | 64 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
# Panel B: First Stage
first_stage_models <- purrr::map(iv_models, ~.$iv_first_stage$avexpr)
# ---------------------------------
modelsummary(
first_stage_models,
output = "gt",
title = "Table 5, Panel B: First-Stage Regressions",
coef_map = c("logem4" = "Log Settler Mortality", "lat_abst" = "Latitude",
"f_brit" = "British Colony Dummy", "f_french" = "French Colony Dummy",
"sjlofr" = "French Legal Origin", "catho80" = "Catholic Religion Dummy",
"muslim80" = "Muslim Religion Dummy", "no_cpm80" = "Other Religion Dummy",
"(Intercept)" = "Constant"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Log Settler Mortality | -0.534*** | -0.426** | -0.593** | -0.419+ | -0.544*** | -0.443** | -0.579*** | -0.442** | -0.544*** | -0.379* |
| (0.140) | (0.158) | (0.189) | (0.213) | (0.127) | (0.141) | (0.130) | (0.151) | (0.141) | (0.165) | |
| Latitude | 1.970 | 3.202 | 2.079 | 2.496+ | 2.708+ | |||||
| (1.375) | (2.002) | (1.301) | (1.453) | (1.488) | ||||||
| British Colony Dummy | 0.629+ | 0.548 | ||||||||
| (0.366) | (0.368) | |||||||||
| French Colony Dummy | 0.047 | -0.117 | -0.004 | -0.163 | ||||||
| (0.430) | (0.441) | (0.524) | (0.521) | |||||||
| French Legal Origin | -0.674* | -0.688* | -0.408 | -0.375 | ||||||
| (0.328) | (0.324) | (0.552) | (0.542) | |||||||
| Catholic Religion Dummy | -0.018 | -0.012 | -0.012 | -0.005 | ||||||
| (0.013) | (0.013) | (0.016) | (0.016) | |||||||
| Muslim Religion Dummy | -0.016 | -0.012 | -0.011 | -0.007 | ||||||
| (0.012) | (0.012) | (0.013) | (0.013) | |||||||
| Other Religion Dummy | -0.009 | 0.000 | -0.006 | 0.004 | ||||||
| (0.015) | (0.016) | (0.016) | (0.016) | |||||||
| Constant | 8.747*** | 7.958*** | 9.617*** | 8.216*** | 9.471*** | 8.631*** | 10.557*** | 8.877*** | 10.196*** | 8.327*** |
| (0.690) | (0.878) | (0.835) | (1.192) | (0.599) | (0.791) | (1.292) | (1.604) | (1.363) | (1.685) | |
| Num. Obs. | 64 | 64 | 25 | 25 | 64 | 64 | 64 | 64 | 64 | 64 |
| R-squared | 0.308 | 0.331 | 0.300 | 0.373 | 0.317 | 0.345 | 0.321 | 0.354 | 0.331 | 0.369 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
# Panel C: OLS Regressions
modelsummary(
ols_models,
output = "gt",
title = "Table 5, Panel C: OLS Regressions with Additional Controls",
coef_map = c("avexpr" = "Average Expropriation Risk", "lat_abst" = "Latitude",
"f_brit" = "British Colony Dummy", "f_french" = "French Colony Dummy",
"sjlofr" = "French Legal Origin", "catho80" = "Catholic Religion Dummy",
"muslim80" = "Muslim Religion Dummy", "no_cpm80" = "Other Religion Dummy"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.528*** | 0.465*** | 0.606*** | 0.549*** | 0.563*** | 0.508*** | 0.533*** | 0.471*** | 0.471*** |
| (0.065) | (0.067) | (0.093) | (0.108) | (0.064) | (0.067) | (0.054) | (0.057) | (0.060) | |
| Latitude | 1.837* | 1.133 | 1.514* | 1.603* | 1.601* | ||||
| (0.700) | (1.103) | (0.697) | (0.636) | (0.648) | |||||
| British Colony Dummy | -0.307 | -0.372+ | |||||||
| (0.211) | (0.203) | ||||||||
| French Colony Dummy | -0.377 | -0.462* | 0.016 | ||||||
| (0.232) | (0.223) | (0.237) | |||||||
| French Legal Origin | 0.364+ | 0.345+ | -0.020 | ||||||
| (0.191) | (0.186) | (0.256) | |||||||
| Catholic Religion Dummy | 0.001 | 0.004 | 0.005 | ||||||
| (0.006) | (0.006) | (0.007) | |||||||
| Muslim Religion Dummy | -0.009 | -0.007 | -0.007 | ||||||
| (0.006) | (0.006) | (0.006) | |||||||
| Other Religion Dummy | -0.011 | -0.005 | -0.005 | ||||||
| (0.007) | (0.007) | (0.008) | |||||||
| Num. Obs. | 64 | 64 | 25 | 25 | 64 | 64 | 64 | 64 | 64 |
| R-squared | 0.565 | 0.610 | 0.648 | 0.664 | 0.566 | 0.597 | 0.691 | 0.722 | 0.722 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||
The estimated causal effect of institutions on income remains large, significant, and remarkably stable after including these additional controls. This provides evidence against the hypothesis that the instrument is working through these alternative channels. The stability of the coefficient strengthens the argument that institutions are the primary causal mechanism.
8 Robustness Checks for IV Regressions
The following code further scrutinizes the exclusion restriction by testing the baseline result against a battery of geographic and resource-based variables. This directly confronts the competing “geography hypothesis” which posits that climate, disease, and resource endowments are the fundamental causes of prosperity.
\[ log(GDP_i)= \beta_0 + \beta_1\widehat{Risk_i}+ \gamma_1Latitude_i + \gamma_2Temperature_i + \gamma_3\text{Soil Quality}_i +...\]
# --- Load and Prepare Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable6.dta")
base_sample <- ajr_dta %>% filter(baseco == 1)
# --- Define Control Variable Sets ---
temp_humid_controls <- "temp1 + temp2 + humid1 + humid2"
resource_controls <- "steplow + deslow + stepmid + desmid + drystep + drywint + goldm + iron + silv + zinc + oilres + landlock"
all_controls <- paste("lat_abst", temp_humid_controls, "edes1975", "avelf", resource_controls, sep = " + ")
# --- Run All Regression Models ---
# Run IV Models (for Panels A and B)
iv_models <- list(
"(1)" = feols(logpgp95 ~ 1 | avexpr ~ logem4, data = base_sample),
"(2)" = feols(as.formula(paste("logpgp95 ~ lat_abst +", temp_humid_controls, "| avexpr ~ logem4")), data = base_sample),
"(3)" = feols(logpgp95 ~ edes1975 | avexpr ~ logem4, data = base_sample),
"(4)" = feols(logpgp95 ~ lat_abst + edes1975 | avexpr ~ logem4, data = base_sample),
"(5)" = feols(as.formula(paste("logpgp95 ~", resource_controls, "| avexpr ~ logem4")), data = base_sample),
"(6)" = feols(as.formula(paste("logpgp95 ~ lat_abst +", resource_controls, "| avexpr ~ logem4")), data = base_sample),
"(7)" = feols(logpgp95 ~ avelf | avexpr ~ logem4, data = base_sample),
"(8)" = feols(logpgp95 ~ lat_abst + avelf | avexpr ~ logem4, data = base_sample),
"(9)" = feols(as.formula(paste("logpgp95 ~", all_controls, "| avexpr ~ logem4")), data = base_sample)
)
# Run OLS Models (for Panel C)
ols_models <- list(
"(1)" = feols(as.formula(paste("logpgp95 ~ avexpr +", temp_humid_controls)), data = base_sample),
"(2)" = feols(as.formula(paste("logpgp95 ~ avexpr + lat_abst +", temp_humid_controls)), data = base_sample),
"(3)" = feols(logpgp95 ~ avexpr + edes1975, data = base_sample),
"(4)" = feols(logpgp95 ~ avexpr + lat_abst + edes1975, data = base_sample),
"(5)" = feols(as.formula(paste("logpgp95 ~ avexpr +", resource_controls)), data = base_sample),
"(6)" = feols(as.formula(paste("logpgp95 ~ avexpr + lat_abst +", resource_controls)), data = base_sample),
"(7)" = feols(logpgp95 ~ avexpr + avelf, data = base_sample),
"(8)" = feols(logpgp95 ~ avexpr + lat_abst + avelf, data = base_sample),
"(9)" = feols(as.formula(paste("logpgp95 ~ avexpr +", all_controls)), data = base_sample)
)
# --- Generate Table for Panel A ---
# This section manually builds the complex layout for Panel A
# Extract coefficients and p-values
tidy_results <- map_dfr(iv_models, tidy, .id = "model") %>%
mutate(estimate_str = sprintf("%.2f", estimate), se_str = sprintf("(%.2f)", std.error))
get_p_value <- function(model, vars) { sprintf("[%.2f]", wald(model, vars)[[4]]) }
temp_vars <- c("temp1", "temp2"); humid_vars <- c("humid1", "humid2")
soil_vars <- c("steplow", "deslow", "stepmid", "desmid", "drystep", "drywint")
resource_vars <- c("goldm", "iron", "silv", "zinc", "oilres")
p_values <- list(
temp2 = get_p_value(iv_models[['(2)']], temp_vars), humid2 = get_p_value(iv_models[['(2)']], humid_vars),
soil5 = get_p_value(iv_models[['(5)']], soil_vars), resource5 = get_p_value(iv_models[['(5)']], resource_vars),
soil6 = get_p_value(iv_models[['(6)']], soil_vars), resource6 = get_p_value(iv_models[['(6)']], resource_vars),
temp9 = get_p_value(iv_models[['(9)']], temp_vars), humid9 = get_p_value(iv_models[['(9)']], humid_vars),
soil9 = get_p_value(iv_models[['(9)']], soil_vars), resource9 = get_p_value(iv_models[['(9)']], resource_vars)
)## Wald test, H0: joint nullity of temp1 and temp2
## stat = 0.161882, p-value = 0.850931, on 2 and 57 DoF, VCOV: IID.Wald test, H0: joint nullity of humid1 and humid2
## stat = 0.025921, p-value = 0.974423, on 2 and 57 DoF, VCOV: IID.Wald test, H0: joint nullity of steplow, deslow, stepmid, desmid, drystep and drywint
## stat = 0.490958, p-value = 0.812014, on 6 and 50 DoF, VCOV: IID.Wald test, H0: joint nullity of goldm, iron, silv, zinc and oilres
## stat = 0.472412, p-value = 0.79502, on 5 and 50 DoF, VCOV: IID.Wald test, H0: joint nullity of steplow, deslow, stepmid, desmid, drystep and drywint
## stat = 0.398411, p-value = 0.876478, on 6 and 49 DoF, VCOV: IID.Wald test, H0: joint nullity of goldm, iron, silv, zinc and oilres
## stat = 0.395728, p-value = 0.84939, on 5 and 49 DoF, VCOV: IID.Wald test, H0: joint nullity of temp1 and temp2
## stat = 0.699118, p-value = 0.502585, on 2 and 43 DoF, VCOV: IID.Wald test, H0: joint nullity of humid1 and humid2
## stat = 0.424181, p-value = 0.657014, on 2 and 43 DoF, VCOV: IID.Wald test, H0: joint nullity of steplow, deslow, stepmid, desmid, drystep and drywint
## stat = 1.2792, p-value = 0.28702, on 6 and 43 DoF, VCOV: IID.Wald test, H0: joint nullity of goldm, iron, silv, zinc and oilres
## stat = 0.388541, p-value = 0.85393, on 5 and 43 DoF, VCOV: IID.nobs <- map_chr(iv_models, ~as.character(nobs(.)))
find_val <- function(m, t, type) { tidy_results %>% filter(model == m, term == t) %>% pull({{type}}) %>% first() }
# Manually construct the table data frame
final_table_A <- tribble(
~term, ~`(1)`, ~`(2)`, ~`(3)`, ~`(4)`, ~`(5)`, ~`(6)`, ~`(7)`, ~`(8)`, ~`(9)`,
"Average protection against expropriation risk, 1985-1995", find_val("(1)", "fit_avexpr", "estimate_str"), find_val("(2)", "fit_avexpr", "estimate_str"), find_val("(3)", "fit_avexpr", "estimate_str"), find_val("(4)", "fit_avexpr", "estimate_str"), find_val("(5)", "fit_avexpr", "estimate_str"), find_val("(6)", "fit_avexpr", "estimate_str"), find_val("(7)", "fit_avexpr", "estimate_str"), find_val("(8)", "fit_avexpr", "estimate_str"), find_val("(9)", "fit_avexpr", "estimate_str"),
"", find_val("(1)", "fit_avexpr", "se_str"), find_val("(2)", "fit_avexpr", "se_str"), find_val("(3)", "fit_avexpr", "se_str"), find_val("(4)", "fit_avexpr", "se_str"), find_val("(5)", "fit_avexpr", "se_str"), find_val("(6)", "fit_avexpr", "se_str"), find_val("(7)", "fit_avexpr", "se_str"), find_val("(8)", "fit_avexpr", "se_str"), find_val("(9)", "fit_avexpr", "se_str"),
"Latitude", NA, find_val("(2)", "lat_abst", "estimate_str"), NA, find_val("(4)", "lat_abst", "estimate_str"), NA, find_val("(6)", "lat_abst", "estimate_str"), NA, find_val("(8)", "lat_abst", "estimate_str"), find_val("(9)", "lat_abst", "estimate_str"),
"", NA, find_val("(2)", "lat_abst", "se_str"), NA, find_val("(4)", "lat_abst", "se_str"), NA, find_val("(6)", "lat_abst", "se_str"), NA, find_val("(8)", "lat_abst", "se_str"), find_val("(9)", "lat_abst", "se_str"),
"p-value for temperature variables", NA, p_values$temp2, NA, NA, NA, NA, NA, NA, p_values$temp9,
"p-value for humidity variables", NA, p_values$humid2, NA, NA, NA, NA, NA, NA, p_values$humid9,
"Percent of European descent in 1975", NA, NA, find_val("(3)", "edes1975", "estimate_str"), find_val("(4)", "edes1975", "estimate_str"), NA, NA, NA, NA, find_val("(9)", "edes1975", "estimate_str"),
"", NA, NA, find_val("(3)", "edes1975", "se_str"), find_val("(4)", "edes1975", "se_str"), NA, NA, NA, NA, find_val("(9)", "edes1975", "se_str"),
"p-value for soil quality", NA, NA, NA, NA, p_values$soil5, p_values$soil6, NA, NA, p_values$soil9,
"p-value for natural resources", NA, NA, NA, NA, p_values$resource5, p_values$resource6, NA, NA, p_values$resource9,
"Dummy for being landlocked", NA, NA, NA, NA, find_val("(5)", "landlock", "estimate_str"), find_val("(6)", "landlock", "estimate_str"), NA, NA, find_val("(9)", "landlock", "estimate_str"),
"", NA, NA, NA, NA, find_val("(5)", "landlock", "se_str"), find_val("(6)", "landlock", "se_str"), NA, NA, find_val("(9)", "landlock", "se_str"),
"Ethnolinguistic fragmentation", NA, NA, NA, NA, NA, NA, find_val("(7)", "avelf", "estimate_str"), find_val("(8)", "avelf", "estimate_str"), find_val("(9)", "avelf", "estimate_str"),
"", NA, NA, NA, NA, NA, NA, find_val("(7)", "avelf", "se_str"), find_val("(8)", "avelf", "se_str"), find_val("(9)", "avelf", "se_str"),
"Number of observations", nobs[1], nobs[2], nobs[3], nobs[4], nobs[5], nobs[6], nobs[7], nobs[8], nobs[9]
)
# Render Panel A with gt
gt(final_table_A, rowname_col = "term") %>%
tab_header(title = "Table 6, Panel A: Two-Stage Least Squares") %>%
sub_missing(missing_text = "") %>%
cols_align(align = "center", columns = -term) %>%
tab_options(table.border.top.style = "none",
column_labels.border.bottom.style = "solid",
column_labels.border.bottom.width = px(2),
table_body.border.bottom.style = "none")| Table 6, Panel A: Two-Stage Least Squares | |||||||||
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Average protection against expropriation risk, 1985-1995 | 0.94 | 1.00 | 0.96 | 0.99 | 1.26 | 1.36 | 0.74 | 0.79 | 0.78 |
| (0.16) | (0.25) | (0.28) | (0.30) | (0.44) | (0.62) | (0.13) | (0.17) | (0.22) | |
| Latitude | -0.63 | -0.67 | -0.94 | -0.89 | -2.05 | ||||
| (1.63) | (1.27) | (2.25) | (1.02) | (1.44) | |||||
| p-value for temperature variables | [57.00] | [43.00] | |||||||
| p-value for humidity variables | [57.00] | [43.00] | |||||||
| Percent of European descent in 1975 | -0.00 | 0.00 | 0.00 | ||||||
| (0.01) | (0.01) | (0.01) | |||||||
| p-value for soil quality | [50.00] | [49.00] | [43.00] | ||||||
| p-value for natural resources | [50.00] | [49.00] | [43.00] | ||||||
| Dummy for being landlocked | 0.75 | 0.87 | 0.93 | ||||||
| (0.73) | (0.92) | (0.48) | |||||||
| Ethnolinguistic fragmentation | -1.02 | -1.13 | -1.70 | ||||||
| (0.32) | (0.34) | (0.47) | |||||||
| Number of observations | 64 | 64 | 64 | 64 | 64 | 64 | 64 | 64 | 64 |
# --- Generate Table for Panel B ---
first_stage_models <- purrr::map(iv_models, ~.$iv_first_stage$avexpr)
modelsummary(
first_stage_models,
output = "gt",
title = "Table 6, Panel B: First-Stage Regressions",
coef_map = c("logem4" = "Log Settler Mortality"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE,
notes = "Notes: Only the coefficient on Log Settler Mortality is shown. All models include the full set of controls."
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Log Settler Mortality | -0.607*** | -0.546** | -0.413** | -0.404** | -0.368* | -0.304+ | -0.636*** | -0.556*** | -0.541** |
| (0.127) | (0.158) | (0.143) | (0.147) | (0.157) | (0.167) | (0.146) | (0.152) | (0.198) | |
| Num. Obs. | 64 | 64 | 64 | 64 | 64 | 64 | 64 | 64 | 64 |
| R-squared | 0.270 | 0.381 | 0.342 | 0.343 | 0.455 | 0.468 | 0.272 | 0.304 | 0.566 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||
| Notes: Only the coefficient on Log Settler Mortality is shown. All models include the full set of controls. | |||||||||
# --- Generate Table for Panel C ---
modelsummary(
ols_models,
output = "gt",
title = "Table 6, Panel C: OLS Robustness Checks",
coef_map = c("avexpr" = "Average Expropriation Risk"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE,
notes = "Notes: Only the coefficient on Average Expropriation Risk is shown. All models include the full set of controls."
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.454*** | 0.406*** | 0.386*** | 0.384*** | 0.446*** | 0.410*** | 0.462*** | 0.450*** | 0.398*** |
| (0.065) | (0.067) | (0.062) | (0.063) | (0.077) | (0.077) | (0.052) | (0.055) | (0.064) | |
| Num. Obs. | 64 | 64 | 64 | 64 | 64 | 64 | 64 | 64 | 64 |
| R-squared | 0.593 | 0.625 | 0.650 | 0.650 | 0.645 | 0.669 | 0.687 | 0.689 | 0.823 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||
| Notes: Only the coefficient on Average Expropriation Risk is shown. All models include the full set of controls. | |||||||||
The coefficient on institutions remains robust and significant even after controlling for temperature, humidity, soil quality, natural resources, and ethnolinguistic fragmentation. The table reports p-values for the joint significance of these groups of controls, which are generally insignificant. This result is a powerful rebuttal to the simple geography hypothesis, suggesting that geography’s primary influence on modern income was through its effect on historical institutional development.
8.1 Geography and Health
The following code takes a look at the most plausible threat to the exclusion restriction: that settler mortality is not just a historical instrument but a proxy for a persistent disease environment that directly impacts health and productivity today. The analysis controls for modern health indicators like malaria prevalence and life expectancy.
Second Stage: \[ log(GDP_i)= \beta_0 + \beta_1\widehat{Risk_i}+ \beta_1\widehat{Health_i} + \epsilon_i\]
First Stage:
\[ Risk_i= \pi_0 + \pi_1Z_1i + \pi_2Z_2i + \nu_i\] \[ Health_i= \delta_0 + \delta_1Z_{1i} + \delta_2Z_{2i} + \eta_i\] where \(Health_i\) is a modern health outcome (e.g., malaria) and \(Z_1\) and \(Z_2\) are the instruments.
# --- Load and Prepare Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable7.dta")
# Create the base sample and the `other_cont` dummy variable
base_sample <- ajr_dta %>%
filter(baseco == 1) %>%
mutate(other_cont = if_else(shortnam %in% c("AUS", "MLT", "NZL"), 1, 0))
# Create subsamples for specific regressions
ols_sample_789 <- base_sample %>%
filter(!is.na(logem4), !is.na(latabs), !is.na(lt100km), !is.na(meantemp))
ols_sample_1011 <- base_sample %>%
filter(!is.na(yellow))
# --- Run All Regression Models ---
# Run IV Models (for Panels A and B)
iv_models <- list(
"(1)" = feols(logpgp95 ~ malfal94 | avexpr ~ logem4, data = base_sample),
"(2)" = feols(logpgp95 ~ lat_abst + malfal94 | avexpr ~ logem4, data = base_sample),
"(3)" = feols(logpgp95 ~ leb95 | avexpr ~ logem4, data = base_sample),
"(4)" = feols(logpgp95 ~ lat_abst + leb95 | avexpr ~ logem4, data = base_sample),
"(5)" = feols(logpgp95 ~ imr95 | avexpr ~ logem4, data = base_sample),
"(6)" = feols(logpgp95 ~ lat_abst + imr95 | avexpr ~ logem4, data = base_sample),
"(7)" = feols(logpgp95 ~ 1 | avexpr + malfal94 ~ logem4 + latabs + lt100km + meantemp, data = base_sample),
"(8)" = feols(logpgp95 ~ 1 | avexpr + leb95 ~ logem4 + latabs + lt100km + meantemp, data = base_sample),
"(9)" = feols(logpgp95 ~ 1 | avexpr + imr95 ~ logem4 + latabs + lt100km + meantemp, data = base_sample),
"(10)" = feols(logpgp95 ~ 1 | avexpr ~ yellow, data = base_sample),
"(11)" = feols(logpgp95 ~ africa + asia + other_cont | avexpr ~ yellow, data = base_sample)
)## NOTE: 2 observations removed because of NA values (RHS: 2).
## NOTE: 2 observations removed because of NA values (RHS: 2).## NOTE: 4 observations removed because of NA values (RHS: 4).
## NOTE: 4 observations removed because of NA values (RHS: 4).
## NOTE: 4 observations removed because of NA values (RHS: 4).
## NOTE: 4 observations removed because of NA values (RHS: 4).## NOTE: 4 observations removed because of NA values (IV: 2/4).## NOTE: 5 observations removed because of NA values (IV: 4/4).
## NOTE: 5 observations removed because of NA values (IV: 4/4).# Run OLS Models (for Panel C)
ols_models <- list(
"(1)" = feols(logpgp95 ~ avexpr + malfal94, data = base_sample),
"(2)" = feols(logpgp95 ~ avexpr + lat_abst + malfal94, data = base_sample),
"(3)" = feols(logpgp95 ~ avexpr + leb95, data = base_sample),
"(4)" = feols(logpgp95 ~ avexpr + lat_abst + leb95, data = base_sample),
"(5)" = feols(logpgp95 ~ avexpr + imr95, data = base_sample),
"(6)" = feols(logpgp95 ~ avexpr + lat_abst + imr95, data = base_sample),
"(7)" = feols(logpgp95 ~ avexpr + malfal94, data = ols_sample_789),
"(8)" = feols(logpgp95 ~ avexpr + leb95, data = ols_sample_789),
"(9)" = feols(logpgp95 ~ avexpr + imr95, data = ols_sample_789),
"(10)" = feols(logpgp95 ~ avexpr, data = ols_sample_1011),
"(11)" = feols(logpgp95 ~ avexpr + africa + asia + other_cont, data = ols_sample_1011)
)## NOTE: 2 observations removed because of NA values (RHS: 2).## NOTE: 2 observations removed because of NA values (RHS: 2).## NOTE: 4 observations removed because of NA values (RHS: 4).
## NOTE: 4 observations removed because of NA values (RHS: 4).
## NOTE: 4 observations removed because of NA values (RHS: 4).
## NOTE: 4 observations removed because of NA values (RHS: 4).## NOTE: 1 observation removed because of NA values (RHS: 1).
## NOTE: 1 observation removed because of NA values (RHS: 1).# --- Generate Tables for Each Panel ---
# Panel A: 2SLS with Health Variables
modelsummary(
iv_models,
output = "gt",
title = "Table 7, Panel A: 2SLS Regressions with Geography and Health Variables",
coef_map = c("fit_avexpr" = "Average Expropriation Risk", "fit_malfal94" = "Malaria Index",
"fit_leb95" = "Life Expectancy", "fit_imr95" = "Infant Mortality",
"lat_abst" = "Latitude", "malfal94" = "Malaria Index",
"leb95" = "Life Expectancy", "imr95" = "Infant Mortality",
"africa" = "Africa Dummy", "asia" = "Asia Dummy", "other_cont" = "Other Continent Dummy"),
gof_map = "nobs",
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.687** | 0.721* | 0.629* | 0.677+ | 0.551* | 0.562+ | 0.689* | 0.737** | 0.675** | 0.914*** | 0.899** |
| (0.251) | (0.297) | (0.276) | (0.338) | (0.239) | (0.314) | (0.258) | (0.236) | (0.227) | (0.245) | (0.318) | |
| Malaria Index | -0.578 | -0.599 | -0.623 | ||||||||
| (0.467) | (0.473) | (0.685) | |||||||||
| Life Expectancy | 0.027 | 0.026 | 0.017 | ||||||||
| (0.021) | (0.023) | (0.023) | |||||||||
| Infant Mortality | -0.010+ | -0.010 | -0.007 | ||||||||
| (0.005) | (0.006) | (0.007) | |||||||||
| Latitude | -0.565 | -0.529 | -0.102 | ||||||||
| (1.043) | (0.968) | (0.949) | |||||||||
| Africa Dummy | -0.532 | ||||||||||
| (0.352) | |||||||||||
| Asia Dummy | -0.880* | ||||||||||
| (0.375) | |||||||||||
| Other Continent Dummy | -0.768 | ||||||||||
| (0.844) | |||||||||||
| Num.Obs. | 62 | 62 | 60 | 60 | 60 | 60 | 60 | 59 | 59 | 64 | 64 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||||
# Panel B: First Stage
first_stage_models_avexpr <- purrr::map(iv_models, ~.$iv_first_stage$avexpr)
# ---------------------------------
modelsummary(
first_stage_models_avexpr,
output = "gt",
title = "Table 7, Panel B: First-Stage Regressions for Expropriation Risk",
coef_map = c("logem4" = "Log Settler Mortality", "yellow" = "Yellow Fever Dummy",
"latabs" = "Latitude (abs)", "lt100km" = "Coastal Dummy",
"meantemp" = "Mean Temperature", "lat_abst" = "Latitude",
"malfal94" = "Malaria Index", "leb95" = "Life Expectancy", "imr95" = "Infant Mortality",
"africa" = "Africa Dummy", "asia" = "Asia Dummy", "other_cont" = "Other Continent Dummy",
"(Intercept)" = "Constant"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Log Settler Mortality | -0.417* | -0.376+ | -0.336+ | -0.302+ | -0.356+ | -0.291 | -0.409* | -0.399* | -0.399* | ||
| (0.186) | (0.188) | (0.172) | (0.178) | (0.179) | (0.187) | (0.169) | (0.172) | (0.172) | |||
| Yellow Fever Dummy | -1.083** | -0.805* | |||||||||
| (0.405) | (0.381) | ||||||||||
| Latitude (abs) | -0.811 | -0.836 | -0.836 | ||||||||
| (1.796) | (1.812) | (1.812) | |||||||||
| Coastal Dummy | 0.574 | 0.546 | 0.546 | ||||||||
| (0.507) | (0.516) | (0.516) | |||||||||
| Mean Temperature | -0.116* | -0.118* | -0.118* | ||||||||
| (0.052) | (0.053) | (0.053) | |||||||||
| Latitude | 1.684 | 1.134 | 1.587 | ||||||||
| (1.398) | (1.397) | (1.379) | |||||||||
| Malaria Index | -0.790 | -0.649 | |||||||||
| (0.538) | (0.548) | ||||||||||
| Life Expectancy | 0.047* | 0.044* | |||||||||
| (0.019) | (0.019) | ||||||||||
| Infant Mortality | -0.013* | -0.012* | |||||||||
| (0.006) | (0.006) | ||||||||||
| Africa Dummy | -0.736* | ||||||||||
| (0.358) | |||||||||||
| Asia Dummy | 0.571 | ||||||||||
| (0.500) | |||||||||||
| Other Continent Dummy | 1.759* | ||||||||||
| (0.798) | |||||||||||
| Constant | 8.757*** | 8.209*** | 5.139** | 4.987** | 8.892*** | 8.270*** | 11.063*** | 11.054*** | 11.054*** | 7.328*** | 7.267*** |
| (0.751) | (0.876) | (1.808) | (1.823) | (0.688) | (0.873) | (1.421) | (1.433) | (1.433) | (0.350) | (0.376) | |
| Num. Obs. | 62 | 62 | 60 | 60 | 60 | 60 | 60 | 59 | 59 | 64 | 64 |
| R-squared | 0.295 | 0.312 | 0.341 | 0.349 | 0.323 | 0.339 | 0.372 | 0.360 | 0.360 | 0.104 | 0.284 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||||
# Panel C: OLS Regressions
modelsummary(
ols_models,
output = "gt",
title = "Table 7, Panel C: OLS Regressions with Geography and Health Variables",
coef_map = c("avexpr" = "Average Expropriation Risk", "lat_abst" = "Latitude",
"malfal94" = "Malaria Index", "leb95" = "Life Expectancy",
"imr95" = "Infant Mortality", "africa" = "Africa Dummy",
"asia" = "Asia Dummy", "other_cont" = "Other Continent Dummy"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.351*** | 0.346*** | 0.281*** | 0.279*** | 0.289*** | 0.277*** | 0.344*** | 0.286*** | 0.293*** | 0.522*** | 0.424*** |
| (0.055) | (0.057) | (0.052) | (0.053) | (0.050) | (0.051) | (0.053) | (0.051) | (0.049) | (0.061) | (0.057) | |
| Latitude | 0.258 | 0.147 | 0.534 | ||||||||
| (0.629) | (0.556) | (0.527) | |||||||||
| Malaria Index | -1.135*** | -1.108*** | -1.127*** | ||||||||
| (0.190) | (0.203) | (0.181) | |||||||||
| Life Expectancy | 0.052*** | 0.051*** | 0.050*** | ||||||||
| (0.007) | (0.007) | (0.007) | |||||||||
| Infant Mortality | -0.016*** | -0.015*** | -0.015*** | ||||||||
| (0.002) | (0.002) | (0.002) | |||||||||
| Africa Dummy | -0.918*** | ||||||||||
| (0.169) | |||||||||||
| Asia Dummy | -0.631** | ||||||||||
| (0.230) | |||||||||||
| Other Continent Dummy | 0.216 | ||||||||||
| (0.377) | |||||||||||
| Num. Obs. | 62 | 62 | 60 | 60 | 60 | 60 | 60 | 59 | 59 | 64 | 64 |
| R-squared | 0.713 | 0.714 | 0.770 | 0.770 | 0.778 | 0.782 | 0.732 | 0.765 | 0.774 | 0.540 | 0.704 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | |||||||||||
The causal effect of institutions remains large and significant when controlling for modern health variables. Furthermore, when health variables themselves are instrumented for (columns 7-9), they are not significant, while institutions remain so. This suggests that the historical disease environment’s effect on income is not primarily through a persistent local disease environment, but rather through its impact on the development of institutions.
9 Overidentification Tests
The final code block provides further tests of the IV strategy’s validity. It employs alternative historical variables as potential instruments and, by using multiple instruments simultaneously, performs tests of overidentifying restrictions (Sargan tests).
First Stage: \[Risk_i= \pi_0 + \pi_1Z_{1i}+\pi_2Z_{2i}+\pi_3^{\prime}X_i + \nu_i\] \[ log(GDP_i)=\beta_0+\beta_1\widehat{Risk}_i+\beta_2^{\prime}X_i + \epsilon_i\] Here, \(Z_{1i}\) is the main instrument (settler mortality) and \(Z_{2i}\) s an additional instrument (e.g., early institutions). The Sargan test is then used to assess whether both instruments are uncorrelated with the error term, \(\epsilon_i\).
# --- Load and Prepare Data ---
ajr_dta <- read_dta("D:/replication-ajr/maketable8.dta")
# Create the base sample for the analysis
base_sample <- ajr_dta %>% filter(baseco == 1)
# --- Run All Regression Models ---
# Panel A & B Models: IV with alternative instruments
panel_AB_models <- list(
"(1)" = feols(logpgp95 ~ 1 | avexpr ~ euro1900, data = base_sample),
"(2)" = feols(logpgp95 ~ lat_abst | avexpr ~ euro1900, data = base_sample),
"(3)" = feols(logpgp95 ~ 1 | avexpr ~ cons00a, data = base_sample),
"(4)" = feols(logpgp95 ~ lat_abst | avexpr ~ cons00a, data = base_sample),
"(5)" = feols(logpgp95 ~ 1 | avexpr ~ democ00a, data = base_sample),
"(6)" = feols(logpgp95 ~ lat_abst | avexpr ~ democ00a, data = base_sample),
"(7)" = feols(logpgp95 ~ indtime | avexpr ~ cons1, data = base_sample),
"(8)" = feols(logpgp95 ~ lat_abst + indtime | avexpr ~ cons1, data = base_sample),
"(9)" = feols(logpgp95 ~ indtime | avexpr ~ democ1, data = base_sample),
"(10)" = feols(logpgp95 ~ lat_abst + indtime | avexpr ~ democ1, data = base_sample)
)## NOTE: 1 observation removed because of NA values (IV: 0/1).
## NOTE: 1 observation removed because of NA values (IV: 0/1).## NOTE: 4 observations removed because of NA values (IV: 0/4).
## NOTE: 4 observations removed because of NA values (IV: 0/4).## NOTE: 5 observations removed because of NA values (IV: 0/5).
## NOTE: 5 observations removed because of NA values (IV: 0/5).## NOTE: 4 observations removed because of NA values (RHS: 4, IV: 0/4).
## NOTE: 4 observations removed because of NA values (RHS: 4, IV: 0/4).## NOTE: 5 observations removed because of NA values (RHS: 4, IV: 0/5).
## NOTE: 5 observations removed because of NA values (RHS: 4, IV: 0/5).# Panel C Models: Overidentification tests (multiple instruments)
panel_C_models <- list(
"(1)" = feols(logpgp95 ~ 1 | avexpr ~ euro1900 + logem4, data = base_sample),
"(2)" = feols(logpgp95 ~ lat_abst | avexpr ~ euro1900 + logem4, data = base_sample),
"(3)" = feols(logpgp95 ~ 1 | avexpr ~ cons00a + logem4, data = base_sample),
"(4)" = feols(logpgp95 ~ lat_abst | avexpr ~ cons00a + logem4, data = base_sample),
"(5)" = feols(logpgp95 ~ 1 | avexpr ~ democ00a + logem4, data = base_sample),
"(6)" = feols(logpgp95 ~ lat_abst | avexpr ~ democ00a + logem4, data = base_sample),
"(7)" = feols(logpgp95 ~ indtime | avexpr ~ cons1 + logem4, data = base_sample),
"(8)" = feols(logpgp95 ~ lat_abst + indtime | avexpr ~ cons1 + logem4, data = base_sample),
"(9)" = feols(logpgp95 ~ indtime | avexpr ~ democ1 + logem4, data = base_sample),
"(10)" = feols(logpgp95 ~ lat_abst + indtime | avexpr ~ democ1 + logem4, data = base_sample)
)## NOTE: 1 observation removed because of NA values (IV: 0/1).## NOTE: 1 observation removed because of NA values (IV: 0/1).## NOTE: 4 observations removed because of NA values (IV: 0/4).
## NOTE: 4 observations removed because of NA values (IV: 0/4).## NOTE: 5 observations removed because of NA values (IV: 0/5).
## NOTE: 5 observations removed because of NA values (IV: 0/5).## NOTE: 4 observations removed because of NA values (RHS: 4, IV: 0/4).
## NOTE: 4 observations removed because of NA values (RHS: 4, IV: 0/4).## NOTE: 5 observations removed because of NA values (RHS: 4, IV: 0/5).
## NOTE: 5 observations removed because of NA values (RHS: 4, IV: 0/5).# Panel D Models: IV with mortality as an exogenous control
panel_D_models <- list(
"(1)" = feols(logpgp95 ~ logem4 | avexpr ~ euro1900, data = base_sample),
"(2)" = feols(logpgp95 ~ logem4 + lat_abst | avexpr ~ euro1900, data = base_sample),
"(3)" = feols(logpgp95 ~ logem4 | avexpr ~ cons00a, data = base_sample),
"(4)" = feols(logpgp95 ~ logem4 + lat_abst | avexpr ~ cons00a, data = base_sample),
"(5)" = feols(logpgp95 ~ logem4 | avexpr ~ democ00a, data = base_sample),
"(6)" = feols(logpgp95 ~ logem4 + lat_abst | avexpr ~ democ00a, data = base_sample),
"(7)" = feols(logpgp95 ~ logem4 + indtime | avexpr ~ cons1, data = base_sample),
"(8)" = feols(logpgp95 ~ logem4 + lat_abst + indtime | avexpr ~ cons1, data = base_sample),
"(9)" = feols(logpgp95 ~ logem4 + indtime | avexpr ~ democ1, data = base_sample),
"(10)" = feols(logpgp95 ~ logem4 + lat_abst + indtime | avexpr ~ democ1, data = base_sample)
)## NOTE: 1 observation removed because of NA values (IV: 0/1).## NOTE: 1 observation removed because of NA values (IV: 0/1).## NOTE: 4 observations removed because of NA values (IV: 0/4).
## NOTE: 4 observations removed because of NA values (IV: 0/4).## NOTE: 5 observations removed because of NA values (IV: 0/5).
## NOTE: 5 observations removed because of NA values (IV: 0/5).## NOTE: 4 observations removed because of NA values (RHS: 4, IV: 0/4).
## NOTE: 4 observations removed because of NA values (RHS: 4, IV: 0/4).## NOTE: 5 observations removed because of NA values (RHS: 4, IV: 0/5).
## NOTE: 5 observations removed because of NA values (RHS: 4, IV: 0/5).# --- Generate Tables for Each Panel ---
# Panel A: 2SLS with Alternative Instruments
modelsummary(
panel_AB_models, output = "gt", title = "Table 8, Panel A: 2SLS with Alternative Instruments",
coef_map = c("fit_avexpr" = "Average Expropriation Risk", "lat_abst" = "Latitude", "indtime" = "Date of Independence"),
gof_map = "nobs", stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.870*** | 0.917*** | 0.706*** | 0.677** | 0.719*** | 0.690*** | 0.595*** | 0.611*** | 0.549*** | 0.555*** |
| (0.139) | (0.200) | (0.145) | (0.197) | (0.136) | (0.186) | (0.144) | (0.169) | (0.119) | (0.136) | |
| Latitude | -0.474 | 0.339 | 0.312 | -0.405 | -0.161 | |||||
| (1.241) | (1.082) | (1.046) | (0.922) | (0.813) | ||||||
| Date of Independence | 0.005** | 0.005*** | 0.005*** | 0.005*** | ||||||
| (0.001) | (0.001) | (0.001) | (0.001) | |||||||
| Num.Obs. | 63 | 63 | 60 | 60 | 59 | 59 | 60 | 60 | 59 | 59 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
# Panel B: First Stage for Alternative Instruments
first_stage_AB <- purrr::map(panel_AB_models, ~.$iv_first_stage$avexpr)
# ---------------------------------
modelsummary(
first_stage_AB, output = "gt", title = "Table 8, Panel B: First Stage for Alternative Instruments",
coef_map = c("euro1900" = "European Settlements in 1900", "cons00a" = "Constraint on Executive in 1900",
"democ00a" = "Democracy in 1900", "cons1" = "Constraint on Executive at Independence",
"democ1" = "Democracy at Independence", "lat_abst" = "Latitude", "indtime" = "Date of Independence",
"(Intercept)" = "Constant"),
gof_map = list(list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "r.squared", "clean" = "R-squared", "fmt" = 3)),
stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| European Settlements in 1900 | 0.032*** | 0.029*** | ||||||||
| (0.006) | (0.008) | |||||||||
| Constraint on Executive in 1900 | 0.318*** | 0.255** | ||||||||
| (0.083) | (0.090) | |||||||||
| Democracy in 1900 | 0.242*** | 0.203** | ||||||||
| (0.057) | (0.067) | |||||||||
| Constraint on Executive at Independence | 0.249** | 0.220** | ||||||||
| (0.079) | (0.078) | |||||||||
| Democracy at Independence | 0.191*** | 0.172** | ||||||||
| (0.049) | (0.050) | |||||||||
| Latitude | 0.764 | 2.346 | 1.727 | 2.949* | 2.486+ | |||||
| (1.540) | (1.435) | (1.492) | (1.446) | (1.413) | ||||||
| Date of Independence | 0.008** | 0.005+ | 0.008** | 0.006+ | ||||||
| (0.003) | (0.003) | (0.003) | (0.003) | |||||||
| Constant | 5.993*** | 5.896*** | 5.756*** | 5.484*** | 6.099*** | 5.858*** | 4.890*** | 4.722*** | 5.241*** | 5.065*** |
| (0.187) | (0.271) | (0.254) | (0.301) | (0.195) | (0.285) | (0.469) | (0.464) | (0.352) | (0.360) | |
| Num. Obs. | 63 | 63 | 60 | 60 | 59 | 59 | 60 | 60 | 59 | 59 |
| R-squared | 0.299 | 0.301 | 0.203 | 0.239 | 0.238 | 0.255 | 0.190 | 0.246 | 0.257 | 0.296 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
# Panel C: Overidentification Tests
gof_map_sargan <- list(
list("raw" = "nobs", "clean" = "Num. Obs.", "fmt" = 0),
list("raw" = "iv_sargan_p", "clean" = "Sargan Test p-value", "fmt" = 3)
)
modelsummary(
panel_C_models, output = "gt", title = "Table 8, Panel C: Overidentification Tests",
coef_map = c("fit_avexpr" = "Average Expropriation Risk", "lat_abst" = "Latitude", "indtime" = "Date of Independence"),
gof_map = gof_map_sargan, stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.893*** | 0.946*** | 0.808*** | 0.833*** | 0.799*** | 0.820*** | 0.670*** | 0.705*** | 0.633*** | 0.654*** |
| (0.128) | (0.173) | (0.132) | (0.177) | (0.127) | (0.167) | (0.117) | (0.142) | (0.108) | (0.127) | |
| Latitude | -0.597 | -0.299 | -0.214 | -0.752 | -0.513 | |||||
| (1.186) | (1.109) | (1.065) | (0.914) | (0.843) | ||||||
| Date of Independence | 0.005** | 0.005** | 0.005** | 0.005** | ||||||
| (0.001) | (0.002) | (0.001) | (0.002) | |||||||
| Num. Obs. | 63 | 63 | 60 | 60 | 59 | 59 | 60 | 60 | 59 | 59 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
# Panel D: 2SLS with Mortality as Exogenous Control
modelsummary(
panel_D_models, output = "gt", title = "Table 8, Panel D: 2SLS with Log Mortality as Exogenous Variable",
coef_map = c("fit_avexpr" = "Average Expropriation Risk", "logem4" = "Log Settler Mortality",
"lat_abst" = "Latitude", "indtime" = "Date of Independence"),
gof_map = "nobs", stars = TRUE
)| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Average Expropriation Risk | 0.814*** | 0.879** | 0.454+ | 0.416 | 0.515* | 0.481+ | 0.485* | 0.494+ | 0.402* | 0.407* |
| (0.229) | (0.292) | (0.249) | (0.299) | (0.227) | (0.275) | (0.232) | (0.248) | (0.180) | (0.191) | |
| Log Settler Mortality | -0.067 | -0.050 | -0.253 | -0.259 | -0.214 | -0.222 | -0.136 | -0.143 | -0.187 | -0.190 |
| (0.165) | (0.183) | (0.164) | (0.169) | (0.154) | (0.161) | (0.155) | (0.149) | (0.125) | (0.122) | |
| Latitude | -0.522 | 0.383 | 0.278 | -0.375 | -0.167 | |||||
| (1.150) | (0.895) | (0.863) | (0.842) | (0.733) | ||||||
| Date of Independence | 0.004*** | 0.005** | 0.004*** | 0.004** | ||||||
| (0.001) | (0.001) | (0.001) | (0.001) | |||||||
| Num.Obs. | 63 | 63 | 60 | 60 | 59 | 59 | 60 | 60 | 59 | 59 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 | ||||||||||
Panel C is the most critical. When settler mortality is used as an instrument alongside other plausible instruments (like early political institutions), the Sargan test fails to reject the null hypothesis that the instruments are valid (i.e., uncorrelated with the error term). This provides statistical evidence supporting the exclusion restriction. Panel D further shows that the main result is not driven by an outlier effect of settler mortality itself.
10 Conclusion
Through a robust and comprehensively tested instrumental variable strategy, this replication confirms the central finding of Acemoglu, Johnson, and Robinson (2001). The evidence strongly supports the hypothesis that institutions that protect property rights and constrain elite power have a large, positive, and causal effect on long-run economic development. The paper effectively refutes purely geographic or cultural explanations, demonstrating instead that these factors likely had their greatest impact through their influence on the historical development of institutions. This work has fundamentally reshaped the field of development economics, placing institutional quality at the forefront of the discipline.
11 References
Acemoglu, Daron, Simon Johnson, and James A. Robinson. “The Colonial Origins of Comparative Development: An Empirical Investigation: Reply.” The American Economic Review 102, no. 6 (2012): 3077–3110. http://www.jstor.org/stable/41724682.