Pre-registration
Botswana's divergence from Sub-Saharan African averages post-1966 is attributable primarily to retained pre-colonial Tswana chieftaincy institutions plus post-independence resource-rent management, rather than resource endowment alone.
Falsification criterion — what would disprove this
This hypothesis is considered falsified if:
PRIMARY (dispositive): SUPPORTED if (a) Botswana's synthetic-control mean post-1976 log-gap on real GDP per capita is ≥ +0.30 log-points (~+35% level outperformance) AND (b) the annual log-growth advantage over the donor-pool mean over 1966-2023 is ≥ +2.0 pp/yr. REFUTED if the post log-gap is ≤ +0.10 OR the growth advantage is ≤ +0.5 pp/yr. INFORMATIVE (not gating): placebo permutation p-value, V-Dem polyarchy gap, WGI government-effectiveness gap. Reported in diagnostics. METHOD_VALID: Maddison gdppc available for BWA + ≥4 SSA donors over the full 1966-2023 window; pre-fit RMSE on log-GDP not pathological (<1.0).
formal test & threshold
test: synthetic_control_bwa_vs_ssa_resource_exporters_1966_2023 threshold: PRIMARY: SC_post_log_gap >= 0.30 AND growth_gap_pp_per_yr >= 2.0 INFORMATIVE: placebo p, V-Dem polyarchy gap, WGI gov-effectiveness gap
Method
- Template
synthetic_control- Fixed effects
country, year- Clustering
country- Sample
- 11 countries · 1966 – 2023
- Evidence type
- associational
Synthetic-control with BWA as treated unit and SSA resource-exporter donor pool. Pre-treatment fit window 1966-1975 (early independence pre-diamond-boom); post-treatment 1976-2023. Primary outcome = log real GDP per capita (Maddison). Placebo permutation across donor pool. Secondary spec interacts treatment with institutional-quality (V-Dem) and resource-rent (mineral rents % GDP) covariates to decompose divergence into institutional vs resource channels. Caveat: BWA is sole treated unit; small-population and commodity-price-luck channels are flagged caveats not formally identified.
Data
| Variable | Source | Transform |
|---|---|---|
log_real_gdp_per_capita outcome | maddison:rgdpnapctier 3 | log |
institutional_quality_index outcome | vdem:v2x_polyarchytier 4 | level |
governance_indicator_composite outcome | wgi:GovernmentEffectivenesstier 4 | level |
botswana_independence_indicator treatment | constructed:indicator = 1 for BWA from 1966 (independence); 0 for SSA donor pool.tier 5 | indicator |
tswana_chieftaincy_institutions treatment | polity5:durabletier 4 | level |
log_population control | world_bank_wdi:SP.POP.TOTLtier 2 | log |
diamond_export_share_gdp control | world_bank_wdi:NY.GDP.MINR.RT.ZStier 2 | level |
● ready · ● pending · ● reconstruct-needed
Detailed result card
Result card — Botswana institutional exceptionalism, 1966-2023
Verdict: SUPPORTED — Botswana's synthetic-control mean post-1976 log-gap = +2.031 (~+662% level), clearing the +0.30 log threshold. Annual log-growth advantage +3.99pp/yr (≥ +2pp threshold). Pre-fit RMSE (log) = 0.355. V-Dem polyarchy gap (post-1976) = 0.3157645833333333; WGI gov-effectiveness gap (1996+) = 1.1414447947999988.
Pre-registration
- Claim: Botswana's divergence from SSA averages post-1966 is attributable primarily to retained pre-colonial Tswana institutions plus post-independence resource-rent management.
- Falsification (PRIMARY 1): SC mean post-1976 log-gap ≥ +0.30 (~+35% level).
- Falsification (PRIMARY 2): BWA − donor mean annual log-growth ≥ +2.0pp/yr.
- Informative (METHOD_VALID-style): V-Dem polyarchy and WGI government effectiveness gaps; pre-fit RMSE; placebo permutation rank.
Synthetic control
- Donor pool (post-fit-coverage filter): ['ZMB', 'ZWE', 'NGA', 'AGO', 'COD', 'GAB', 'GHA', 'CIV', 'NAM', 'ZAF']
- Pre-fit window: 1966-1975; treatment year: 1976
- Pre-fit RMSE (log): 0.35478974303094146
- Donor weights (>0.5%): COD=0.85, NGA=0.15
- Post-treatment mean log-gap: 2.031377907067469
- Post-treatment mean level gap: 662.4585093665457%
- 2023 terminal log-gap: None
Growth gap
- BWA mean annual log-growth 1977-2023: 0.05176978064670174
- Donor pool mean annual log-growth 1977-2023: 0.011881411061973221
- Gap (BWA − donor mean): 3.9888369584728522 pp/yr (threshold ≥ +2.0)
Placebo permutation
- BWA rank among donor pool (largest positive post-treat gap): (0, 11)
- One-sided permutation p-value: 0.09090909090909091
Informative covariate gaps (post-treatment)
- V-Dem polyarchy: BWA=0.6725, donor mean=0.3567354166666667, gap=0.3157645833333333
- WGI gov-effectiveness (1996+): BWA=0.46241573599999997, donor mean=-0.6790290587999989, gap=1.1414447947999988
Method note (downgrade from spec)
Spec called for synthetic_control with diamond-rent and V-Dem covariates
as decomposition channels. Implemented classic Abadie-Diamond-Hainmueller
synthetic control on log GDP-pc level matched in the pre-1976 window.
Diamond-rent channel (WDI mineral rents, NY.GDP.MINR.RT.ZS) is NOT on disk;
reported as a data gap rather than estimated. Channel decomposition is
left as a v2 question — for now V-Dem and WGI gaps are reported descriptively
to colour but not gate the verdict.
Data
- maddison:mpd2020 (primary outcome, log real GDP per capita)
- vdem:vdem_cy_full (v2x_polyarchy — informative)
- wgi:GOV_WGI_GE.EST (informative, 1996+)
Reproduces from vintages in manifest.yaml. See replication.py.
Notes
Stub seeded from an institutionalist school prediction about Botswana's post-1966 divergence being driven by retained Tswana chieftaincy institutions plus resource-rent management. Decomposing institutions vs resource-endowment effects is delicate; needs human review of identification.