All tables and numbers in this report are generated from the saved analysis files (data/analysis/ and outputs/), so they update when scripts/build_leader_spell_panel.R, scripts/spell_regressions.R and scripts/jones_olken_events.R are rerun. The written interpretation reflects the runs of 11 November 2026.

1. Core question

Does dynastic national leadership change a country’s economic and political trajectory? We study the consequences of dynastic rule, not its causes. A leader is dynastic when a family member previously held political office. The outcomes are:

economic growth (GDP per capita growth),
income inequality (disposable-income Gini),
human welfare (infant mortality), and
democracy (electoral democracy and executive corruption).

Why this is a gap. The closest study, Besley & Reynal-Querol (2017), covers hereditary leaders and growth only, with a narrow definition (a relative was previously head of state). Sub-national work (George & Ponattu 2018 for India; Mendoza et al. 2013 for the Philippines) finds that dynastic descendants worsen local living standards and public goods. No study estimates the effect of dynastic national leadership on inequality, infant mortality or democracy across countries. The GDD lets us do this for 169 countries, 1946–2020, with a broader definition of dynasty.

2. Data

2.1 Treatment: dynastic leadership

Dynasty = GDD pred_bin == 1: a predecessor in the leader’s family held political office at the national, state or local level. In the corrected GDD this is identical to pred_num ≥ 1.
Validation. In country-years covered by both datasets, GDD agrees with Paths to Power’s leader_politicalfamily in 88.8% of 6,520 cases. Most disagreements are cases where PtP counts family members who entered politics after the leader, or non-political family ties.
Ruling monarchs. Spells in GDD’s “Royal Dictatorship” category (56 spells, 53 of them dynastic) are flagged and excluded from the main sample. There, dynastic status is set by the succession rule, and the dynastic-to-non-dynastic “switches” are mostly revolutions (Iraq 1958, Libya 1969, Ethiopia 1974, Iran 1979). PtP’s leader_royal is not used for this flag because it records royal or noble background (e.g. Mandela, V.P. Singh), not reigning monarchs.

2.2 Outcomes

tbl(data.frame(
  Outcome = c("GDP per capita growth", "Income inequality", "Infant mortality",
              "Electoral democracy", "Executive corruption"),
  Source = c("Maddison (via QoG)", "SWIID 9.92 (Solt 2020)", "WHO / WDI (via QoG)",
             "V-Dem `v2x_polyarchy`", "V-Dem `v2x_execorr`"),
  `Spell-level measure` = c("Mean annual growth during the spell (%)",
                            "Change in Gini per year of the spell (points)",
                            "Change in log infant mortality per year (%)",
                            "Change per year from the first to the last spell year (×100)",
                            "Change per year from the first to the last spell year (×100; higher = more corrupt)"),
  `Event-study measure` = c("Annual growth", "Gini level", "Annual % change",
                            "Level ×100", "Level ×100"),
  check.names = FALSE), caption = "Outcomes")

Outcomes
Outcome	Source	Spell-level measure	Event-study measure
GDP per capita growth	Maddison (via QoG)	Mean annual growth during the spell (%)	Annual growth
Income inequality	SWIID 9.92 (Solt 2020)	Change in Gini per year of the spell (points)	Gini level
Infant mortality	WHO / WDI (via QoG)	Change in log infant mortality per year (%)	Annual % change
Electoral democracy	V-Dem `v2x_polyarchy`	Change per year from the first to the last spell year (×100)	Level ×100
Executive corruption	V-Dem `v2x_execorr`	Change per year from the first to the last spell year (×100; higher = more corrupt)	Level ×100

V-Dem’s liberal, egalitarian, participatory and deliberative democracy indices are all built on polyarchy and correlate above 0.9 with it, so they add little and would multiply the number of tests. Polyarchy is the democracy outcome. Executive corruption covers a separate dimension that links to the rent channel in Folke, Persson & Rickne (2017). Vote buying and clean elections are components of polyarchy and are kept for mechanism checks only.

2.3 Moderators and controls

Executive constraints at entry. Polity5 xconst. “Strong” means xconst = 7, following Besley & Reynal-Querol (2017). V-Dem legislative constraints (v2xlg_legcon) serve as a continuous alternative. Constraints are used only as moderators, never as outcomes.
Baseline level of each outcome when the spell begins, plus log GDP per capita in the year before entry.
Leader characteristics: age at entry and gender.
Log population at entry: UN WPP from 1950, extended to 1946–49 with Gleditsch population growth rates.

2.4 Building the leader-spell panel

Spell. A run of consecutive GDD years in one country with the same leader ID (code_7digit). A leader who returns later starts a new spell.
Outcome windows. Growth is the mean over the spell’s years. Gini and infant mortality are measured from the year before entry to the last spell year. Polyarchy and executive corruption use a post-entry window (first to last spell year), so the jump in the year a leader takes office is not credited to the leader (Section 4.3).
Leader exits. Spells are matched to Archigos 4.1 by country, overlapping years and name, using both GDD’s nominal and effective leader. An Archigos exit is credited to a GDD spell only if it falls in the spell’s last year or the next, and only once per Archigos spell.
Birth years missing in GDD are filled from Archigos (28 spells), only where Archigos matched the nominal leader by name.
Data issues found (handled in the analysis copy only; the GDD file is unchanged):
- some Japan rows carry the country code "\nJPN" (a stray line break);
- wid_data.rds mixes sub-national WID areas under the same country-year, so the national top-10% share is taken from QoG instead.

2.5 Coverage and sample sizes

cov_row <- function(v, lab) {
  ok <- !is.na(spn[[v]])
  data.frame(Outcome = lab, `Spells (all)` = sum(ok), `Dynastic spells` = sum(ok & spn$dynasty == 1),
             Countries = n_distinct(spn$country_isocode[ok]),
             `First entry year` = min(spn$first_year[ok]), check.names = FALSE)
}
tbl(bind_rows(
  cov_row("y_growth", "GDP pc growth"), cov_row("y_gini", "Gini"),
  cov_row("y_infmort", "Infant mortality"), cov_row("y_poly", "Polyarchy (post-entry)"),
  cov_row("y_execorr", "Executive corruption (post-entry)"),
  cov_row("xconst_pre", "Polity xconst at entry")),
  caption = sprintf("Outcome coverage, leader spells excluding ruling monarchs (%d spells in total; %d dynastic)",
                    nrow(spn), sum(spn$dynasty == 1)))

Outcome coverage, leader spells excluding ruling monarchs (1906 spells in total; 355 dynastic)
Outcome	Spells (all)	Dynastic spells	Countries	First entry year
GDP pc growth	1751	344	146	1946
Gini	1067	206	143	1961
Infant mortality	1641	308	157	1946
Polyarchy (post-entry)	1375	288	156	1946
Executive corruption (post-entry)	1375	288	156	1946
Polity xconst at entry	1694	314	152	1946

tbl(sp %>% count(transition, name = "Spells") %>% arrange(desc(Spells)),
    caption = "Transitions between consecutive spells (D = dynastic, ND = non-dynastic)")

Transitions between consecutive spells (D = dynastic, ND = non-dynastic)
transition	Spells
ND -> ND	1210
ND -> D	224
D -> ND	215
first spell in data	169
D -> D	144

3. Empirical strategy

3.1 Leader-spell regressions

Following Besley & Reynal-Querol (2017), each observation is a leader spell \(s\) in country \(c\) beginning in year \(t_0\):

\[ \Delta y_{s} = \alpha_c + \tau_{t_0} + \beta_1 D_s + \beta_2 S_s + \beta_3 (D_s \times S_s) + \lambda\, y_{s,0} + \gamma' X_s + \varepsilon_s \]

where \(D_s\) is dynastic status, \(S_s\) is strong executive constraints at entry, \(y_{s,0}\) is the baseline outcome, and \(X_s\) contains log GDP per capita before entry, leader age, gender and log population. Country fixed effects \(\alpha_c\) and entry-year fixed effects \(\tau_{t_0}\) are included, with standard errors clustered by country.

Model	Specification
M1	Dynasty + baseline outcome + log GDP pc
M2	M1 + age, gender, log population
M3	M2 + Dynasty × Strong constraints (`xconst = 7`); reports the effect under weak (\(\beta_1\)) and strong (\(\beta_1+\beta_3\)) constraints
M4	M2 + Dynasty × V-Dem legislative constraints (continuous, centred)

With country fixed effects, the estimate is identified from the countries that have both dynastic and non-dynastic spells.

3.2 Jones-Olken event study

Following Jones & Olken (2005), we use leader exits whose timing is plausibly random: natural death or retirement due to ill health (Archigos). For each event \(z\) in country \(c\) at year \(T\):

\[ \Delta_z = \frac{1}{5}\sum_{k=1}^{5} \tilde y_{c,T+k} - \frac{1}{5}\sum_{k=1}^{5} \tilde y_{c,T-k} \]

Here \(\tilde y\) is the outcome net of country and year fixed effects, the event year is excluded, and each window needs at least three years of data. We regress \(\Delta_z\) on the type of transition (non-dynast to dynast, dynast to non-dynast, dynast to dynast; non-dynast to non-dynast is the reference) and, symmetrically, on the change in dynastic status (−1, 0, +1). Events with another random exit in the same country within five years are dropped, as are events involving ruling monarchs (main sample). Placebo: the same calculation around fake event dates six years earlier.

3.3 Why these are conditional correlations, not causal estimates

A causal reading needs dynastic status to be as good as random once country, period and controls are held fixed. That is unlikely:

Dynasts take office in particular moments. They often arrive after crises, assassinations or democratic openings (Rajiv Gandhi, Corazon Aquino, Benazir Bhutto). Those circumstances also drive what follows.
Spell length depends on performance. Hereditary rule is more likely to end after poor growth (Besley & Reynal-Querol 2017), and outcomes are averaged over a spell whose end is itself endogenous.
Within-country changes over time. Weakening parties, clientelism and shifts in state capacity can raise both the chance of a dynast and the outcomes. Country fixed effects remove only permanent differences.
Bundled treatment. “Dynast” comes with wealth, networks, education and party machinery, so the estimate is not the effect of the family tie alone.
Identification from switchers. Only countries with both kinds of spells contribute, and they may not be representative.

The event study removes selection on when transitions happen, but not on who succeeds.

4. Results

4.1 Spell regressions: main sample (ruling monarchs excluded)

show_cols <- c("outcome", "M1 dynasty", "M2 dynasty", "M3 dynasty (weak xconst)",
               "M3 dynasty (strong xconst)", "M3 dynasty x strong", "M4 dynasty x legcon",
               "N_M2", "countries", "switching_countries", "dynastic_spells")
tbl(res %>% filter(sample == "no_monarchs") %>% select(all_of(show_cols)) %>%
      rename(Outcome = outcome, `M3 weak` = `M3 dynasty (weak xconst)`,
             `M3 strong (β1+β3)` = `M3 dynasty (strong xconst)`, `M3 interaction` = `M3 dynasty x strong`,
             `M4 × legcon` = `M4 dynasty x legcon`, N = N_M2, Countries = countries,
             `Switching countries` = switching_countries, `Dynastic spells` = dynastic_spells),
    caption = "Dynasty coefficients, leader-spell regressions. Country and entry-year FE; SEs clustered by country. * p<0.1, ** p<0.05, *** p<0.01")

Dynasty coefficients, leader-spell regressions. Country and entry-year FE; SEs clustered by country. * p<0.1, ** p<0.05, *** p<0.01
Outcome	M1 dynasty	M2 dynasty	M3 weak	M3 strong (β1+β3)	M3 interaction	M4 × legcon	N	Countries	Switching countries	Dynastic spells
GDP pc growth (%/yr)	0.426 (0.491)	0.251 (0.484)	0.585 (0.797)	-0.836*** (0.317)	-1.421* (0.813)	-1.970* (1.046)	1490	143	90	300
Gini, disposable (pts/yr)	0.005 (0.030)	-0.017 (0.031)	0.002 (0.039)	-0.034 (0.043)	-0.036 (0.053)	-0.207** (0.089)	884	128	70	184
Infant mortality (%/yr)	0.065 (0.206)	0.067 (0.193)	0.009 (0.236)	0.106 (0.202)	0.097 (0.298)	0.412 (0.660)	1405	141	88	276
Polyarchy, post-entry (x100/yr)	-0.251 (0.244)	-0.097 (0.256)	0.242 (0.395)	-0.599** (0.295)	-0.841* (0.482)	-0.274 (0.674)	1091	142	85	247
Exec. corruption, post-entry (x100/yr)	0.027 (0.214)	0.038 (0.256)	0.257 (0.533)	-0.268 (0.173)	-0.525 (0.578)	-0.072 (0.992)	1091	142	85	247

cp <- res %>% filter(sample == "no_monarchs") %>%
  transmute(outcome, Weak = `M3 dynasty (weak xconst)`, Strong = `M3 dynasty (strong xconst)`, Average = `M2 dynasty`) %>%
  pivot_longer(-outcome, names_to = "Constraints", values_to = "s") %>%
  mutate(e = est(s), se = se_(s), Constraints = factor(Constraints, c("Average", "Weak", "Strong")),
         outcome = factor(outcome, unique(res$outcome)))
ggplot(cp, aes(x = Constraints, y = e)) +
  geom_hline(yintercept = 0, colour = "grey60") +
  geom_pointrange(aes(ymin = e - 1.96 * se, ymax = e + 1.96 * se), colour = "#2c3e50", linewidth = 0.6) +
  facet_wrap(~outcome, scales = "free_y", nrow = 1, labeller = label_wrap_gen(18)) +
  labs(x = NULL, y = "Dynasty coefficient (95% CI)",
       title = "Dynastic leadership and outcomes, by executive constraints at entry") +
  theme_minimal(base_size = 12) + theme(panel.grid.minor = element_blank(), strip.text = element_text(face = "bold"))

Reading the results.

Growth. There is no average difference (M2: 0.251 (0.484)). Under strong executive constraints, dynastic spells grow 0.84 percentage points a year slower (-0.836*** (0.317)). Under weak constraints the estimate is positive but imprecise (0.585 (0.797)). This differs from Besley & Reynal-Querol, who find hereditary leaders do better under weak constraints and no different under strong ones.
Polyarchy (post-entry). No average effect. Under strong constraints, polyarchy declines during dynastic spells (-0.599** (0.295) points ×100 per year).
Inequality, infant mortality, executive corruption. Small, insignificant estimates in every specification. The only exception is the continuous legislative-constraints interaction for the Gini (-0.207** (0.089)), which is not mirrored in the xconst specification.

4.2 Robustness: including ruling monarchs

cmp <- res %>% select(sample, outcome, `M2 dynasty`, `M3 dynasty (weak xconst)`, `M3 dynasty (strong xconst)`, N_M2) %>%
  mutate(sample = ifelse(sample == "no_monarchs", "Excl. monarchs", "Incl. monarchs")) %>%
  pivot_wider(names_from = sample, values_from = c(`M2 dynasty`, `M3 dynasty (weak xconst)`, `M3 dynasty (strong xconst)`, N_M2),
              names_glue = "{.value} | {sample}")
tbl(cmp %>% select(outcome, starts_with("M2"), starts_with("M3 dynasty (strong"), starts_with("N_M2")) %>% rename(Outcome = outcome),
    caption = "Average (M2) and strong-constraints (M3) dynasty effects, with and without ruling monarchs")

Average (M2) and strong-constraints (M3) dynasty effects, with and without ruling monarchs
Outcome	M2 dynasty \| Excl. monarchs	M2 dynasty \| Incl. monarchs	M3 dynasty (strong xconst) \| Excl. monarchs	M3 dynasty (strong xconst) \| Incl. monarchs	N_M2 \| Excl. monarchs	N_M2 \| Incl. monarchs
GDP pc growth (%/yr)	0.251 (0.484)	0.216 (0.476)	-0.836*** (0.317)	-0.828*** (0.318)	1490	1522
Gini, disposable (pts/yr)	-0.017 (0.031)	-0.016 (0.031)	-0.034 (0.043)	-0.033 (0.043)	884	889
Infant mortality (%/yr)	0.067 (0.193)	0.084 (0.190)	0.106 (0.202)	0.114 (0.202)	1405	1434
Polyarchy, post-entry (x100/yr)	-0.097 (0.256)	-0.093 (0.250)	-0.599** (0.295)	-0.597** (0.294)	1091	1120
Exec. corruption, post-entry (x100/yr)	0.038 (0.256)	0.043 (0.253)	-0.268 (0.173)	-0.274 (0.171)	1091	1120

Including the 56 ruling-monarch spells changes coefficients only in the second or third decimal place. The exclusion is a conceptual choice and does not drive any result.

4.3 The entry-year artifact in polyarchy

Measured from the year before entry, dynastic spells appeared to bring democratic improvement. That window includes the year the leader takes office, and polyarchy rises more in the entry year of dynastic spells than of non-dynastic ones:

ej <- spn %>% mutate(jump = 100 * (v2x_polyarchy_first - v2x_polyarchy_pre)) %>%
  group_by(Leader = ifelse(dynasty == 1, "Dynastic", "Non-dynastic")) %>%
  summarise(`Mean polyarchy jump in entry year (×100)` = round(mean(jump, na.rm = TRUE), 2),
            `Share of spells with jump > 10` = sprintf("%.1f%%", 100 * mean(jump > 10, na.rm = TRUE)),
            Spells = sum(!is.na(jump)))
tbl(ej, caption = "Change in polyarchy between the year before entry and the first spell year")

Change in polyarchy between the year before entry and the first spell year
Leader	Mean polyarchy jump in entry year (×100)	Share of spells with jump > 10	Spells
Dynastic	1.48	7.6%	354
Non-dynastic	0.90	6.2%	1534

ctrl <- "+ age_entry + female + log_pop_entry | country_isocode + first_year"
win <- function(y, base) {
  m2 <- fe_m(as.formula(paste(y, "~ dynasty +", base, "+ log_gdppc_pre", ctrl)), spn)
  m3 <- fe_m(as.formula(paste(y, "~ dynasty * strong +", base, "+ log_gdppc_pre", ctrl)), spn)
  st <- lincom(m3, "dynasty", "dynasty:strong")
  c(Average = fmt(coef(m2)["dynasty"], se(m2)["dynasty"], pvalue(m2)["dynasty"]),
    Weak = fmt(coef(m3)["dynasty"], se(m3)["dynasty"], pvalue(m3)["dynasty"]),
    Strong = fmt(st["est"], st["se"], st["p"]), N = nobs(m2))
}
tbl(rbind(
  `Polyarchy: year before entry → end` = win("y_poly_pre", "v2x_polyarchy_pre"),
  `Polyarchy: first year → end (used)` = win("y_poly", "v2x_polyarchy_first"),
  `Exec. corruption: year before entry → end` = win("y_execorr_pre", "v2x_execorr_pre"),
  `Exec. corruption: first year → end (used)` = win("y_execorr", "v2x_execorr_first")),
  caption = "Dynasty effect on democracy outcomes under the two windows (×100 per year)")

Dynasty effect on democracy outcomes under the two windows (×100 per year)
	Average	Weak	Strong	N
Polyarchy: year before entry → end	0.869*** (0.313)	1.911*** (0.471)	0.014 (0.341)	1496
Polyarchy: first year → end (used)	-0.097 (0.256)	0.242 (0.395)	-0.599** (0.295)	1091
Exec. corruption: year before entry → end	0.022 (0.279)	0.419 (0.396)	-0.384 (0.360)	1496
Exec. corruption: first year → end (used)	0.038 (0.256)	0.257 (0.533)	-0.268 (0.173)	1091

The positive “effect” on polyarchy disappears once the entry year is excluded. It reflected the change around the moment dynasts came to power, not what they did in office. The post-entry window is therefore the main specification for both democracy outcomes.

4.4 How the weak-constraints growth result changed with the larger sample

The first run used WDI population (from 1960) and GDD birth years only, which dropped about 400 spells. In that sample, dynasts grew faster under weak constraints (significant at the 10% level). With the spliced population series and Archigos birth years, that result disappears, while the strong-constraints penalty holds:

g_m3 <- function(d) {
  m <- fe_m(as.formula(paste("y_growth ~ dynasty * strong + log_gdppc_pre", ctrl)), d)
  st <- lincom(m, "dynasty", "dynasty:strong")
  c(Weak = fmt(coef(m)["dynasty"], se(m)["dynasty"], pvalue(m)["dynasty"]),
    Strong = fmt(st["est"], st["se"], st["p"]), N = nobs(m))
}
tbl(rbind(
  `Spells from 1960 with GDD birth year (approximates the first run)` =
    g_m3(filter(spn, first_year >= 1960, birth_year_source == "GDD")),
  `Full sample (current)` = g_m3(spn)),
  caption = "Growth: dynasty effect under weak and strong constraints (M3), by sample")

Growth: dynasty effect under weak and strong constraints (M3), by sample
	Weak	Strong	N
Spells from 1960 with GDD birth year (approximates the first run)	1.383* (0.776)	-0.879** (0.345)	1175
Full sample (current)	0.585 (0.797)	-0.836*** (0.317)	1378

The weak-constraints estimate depends on which spells are included, so it should not be reported as a finding. The strong-constraints estimate is stable across both samples.

4.5 Jones-Olken event study

The main sample has 85 random exits. Of these, 22 are non-dynast to dynast, 14 dynast to non-dynast, 5 dynast to dynast, and 44 non-dynast to non-dynast.

tbl(jo %>% filter(sample == "main") %>% select(-sample) %>%
      rename(Outcome = outcome, Window = window, `Into dynasty (Δ status)` = `into dynasty (d_dyn)`,
             `SD of change` = sd_change, `Switch events` = n_switch),
    caption = "Change in outcome (5 years after minus 5 years before), relative to ND → ND transitions. SEs clustered by country. * p<0.1, ** p<0.05, *** p<0.01")

Change in outcome (5 years after minus 5 years before), relative to ND → ND transitions. SEs clustered by country. * p<0.1, ** p<0.05, *** p<0.01
Outcome	Window	Into dynasty (Δ status)	ND -> D	D -> ND	D -> D	N	Switch events	SD of change
GDP pc growth (%)	actual	1.08 (0.75)	-1.40 (1.02)	-4.13*** (1.32)	-1.93 (1.25)	74	32	3.92
GDP pc growth (%)	placebo (t-6)	0.43 (1.00)	1.45 (1.86)	0.74 (1.23)	-1.36 (1.33)	65	27	4.68
Gini, disposable	actual	-0.15 (0.55)	-0.67 (0.99)	-0.41 (1.17)	0.22 (0.76)	33	15	2.41
Gini, disposable	placebo (t-6)	0.39 (0.84)	0.40 (1.28)	-0.38 (1.49)	-0.11 (0.73)	29	12	2.51
Infant mortality, annual % change	actual	0.25 (0.50)	1.16 (0.79)	0.94 (0.67)	-0.92 (1.27)	73	30	2.26
Infant mortality, annual % change	placebo (t-6)	-0.36 (0.35)	-0.22 (0.59)	0.45 (0.57)	4.13 (4.00)	62	25	2.47
Polyarchy (x100)	actual	-1.61 (1.43)	-2.82 (1.94)	0.04 (3.38)	-2.10 (1.95)	83	35	9.18
Polyarchy (x100)	placebo (t-6)	0.18 (1.48)	0.84 (2.69)	0.75 (1.96)	-4.62 (3.12)	73	32	8.14
Exec. corruption (x100)	actual	0.64 (0.87)	-1.24 (2.26)	-3.25 (2.66)	1.18 (3.97)	83	35	10.38
Exec. corruption (x100)	placebo (t-6)	0.99 (0.74)	2.41** (0.96)	0.54 (1.69)	10.07 (8.64)	73	32	7.75

ev_plot <- ev_main %>% filter(!is.na(D_growth_mad)) %>%
  mutate(switch = factor(switch, c("ND -> ND", "ND -> D", "D -> ND", "D -> D")))
ggplot(ev_plot, aes(x = switch, y = D_growth_mad)) +
  geom_hline(yintercept = 0, colour = "grey60") +
  geom_jitter(width = 0.12, height = 0, colour = "#7f8c8d", alpha = 0.7, size = 2) +
  stat_summary(fun.data = mean_se, fun.args = list(mult = 1.96), colour = "#c0392b", linewidth = 0.8, size = 0.6) +
  labs(x = "Transition after the random exit", y = "Growth: 5 years after − 5 years before (pp)",
       title = "Growth around random leader exits, by change in dynastic status",
       subtitle = "Each dot is one event; red = mean with 95% CI. Net of country and year effects.") +
  theme_minimal(base_size = 12) + theme(panel.grid.minor = element_blank())