Photo credits
  • Moneta Sleet Jr. - public domain, Saint Louis Art Museum
  • Heinrich-Boell-Stiftung - CC BY-SA 2.0
  • Ukpai Jeffrey - CC BY-SA 4.0

Reference

Technical details

Every variable in the analytical base, every model the project fitted, and the equation each one estimates. The methodology page carries the argument; this page carries the material.

Variables

The twenty-five metrics the base carries: twelve catalogue outcomes (each with model cards) and thirteen covariates and context series. The unit, direction and coverage status are the catalogue's own fields, never retyped here.

MetricMeaningUnitDirectionSourceCoverage
u5mrUnder-5 mortality rate outcomeper 1,000 live birthslower is betterNDHS / UN IGMEpartial
mmrMaternal mortality ratio outcomeper 100,000 birthslower is betterWorld Bank (UN MMEIG estimates)ok
immun_penta3Penta3 immunisation coverage outcome% surviving infantshigher is betterDHIS2 / WUENICpartial
skilled_birthSkilled birth attendance outcome% of birthshigher is betterNDHS / MICSpartial
literacy_youthYouth literacy (15-24) outcome%higher is betterNBS / UNESCO UISdeclared_gap
primary_completionPrimary completion rate outcome%higher is betterNEMIS / World Bankdeclared_gap
ner_primaryNet enrolment rate, primary outcome%higher is betterNEMIS / World Bankdeclared_gap
fatalitiesReported security fatalities outcomecountlower is betterUCDP GEDok
incidentsReported security incidents outcomecountlower is betterUCDP GEDok
gdp_growthReal GDP growth (y/y) outcome%higher is betterNBS / World Bankok
cpi_yoyHeadline inflation (y/y) outcome%lower is betterNBS / World Bankok
unempUnemployment rate outcome%lower is betterNBS / World Bank (ILO modelled)partial
neonatal_mrNeonatal mortality rateper 1,000 live birthslower is betterWorld Bank
measles_immMeasles immunisation coverage% of children 12-23mhigher is betterWorld Bank
health_spend_gdpCurrent health expenditure% of GDPhigher is betterWorld Bank
life_expectancyLife expectancy at birthyearshigher is betterWorld Bank
adult_literacyAdult literacy (15+)%higher is betterWorld Bank
educ_spend_gdpGovernment expenditure on education% of GDPhigher is betterWorld Bank
homicidesIntentional homicidesper 100,000 peoplelower is betterWorld Bank
gdp_per_capitaGDP per capita (constant)constant US$higher is betterWorld Bank
populationPopulation, totalpersonslower is betterWorld Bank
fxOfficial exchange rateNGN per US$lower is betterWorld Bank (CBN for monthly)
donor_odaNet official development assistance receivedcurrent US$lower is betterWorld Bank (OECD DAC)
oil_rentsOil rents% of GDPlower is betterWorld Bank
oilBrent crude spot priceUSD per barrellower is betterFRED (Brent crude)

Treatments, outcomes and design variables

The columns a design uses, beyond the catalogue metrics.

ColumnRoleMeaning
faac_gross_total_nairaTreatment (A)The state's gross FAAC allocation for the month, naira. The dose Design A estimates against.
log_doseTreatment transform (A)log of the allocation. In a specification with state fixed effects the elasticity is invariant to a state-constant denominator, which is why the per-capita axis is not needed for the slope.
d1_faac_per_capitaDeclared dose axis (A)The per-capita axis the estimand is written on. The column is typed and empty: the state-population denominator is an unmade decision (X-09), so every specification in the grid is nominal.
d2_faac_share_of_state_revenueDeclared instrument (A)FAAC divided by FAAC plus internally generated revenue. The IGR half has no landed series, so the non-oil instrument does not exist yet.
fatalities, incidents, fatalities_civiliansOutcomes (A)UCDP GED counts by state and month. An event count and a death toll are different objects, which is why all three are published.
oilTreatment (B)Brent crude spot price, monthly. Design B uses its first difference in logs: the shock, not the level.
gross_total_nairaOutcome (B)The national monthly FAAC allocation, the same dose aggregated across states.
u5mrOutcome (C)Under-5 mortality per 1,000 live births, annual native observations. The treated unit of the synthetic control.
w_jEstimand (C)The simplex weight on donor country j. Weights are negative-free and sum to one; the post-treatment gap is what the design reports.

Models

One row per design. The estimator, treatment and outcome strings are read from the committed specs; where each design's numbers are stored is stated once, on the methodology page.

DesignEstimatorOutcomeTreatmentControlsInference
APPML (Poisson pseudo-maximum-likelihood)fatalities, incidents, fatalities_civilianslog(faac_gross_total_naira)state_id, year_month (fixed effects)cluster by state
Blocal projection (Jorda), OLS per horizon, statsmodels-equivalent HACnational gross_total_naira (log cumulative change); national cpi_index (log cumulative change)Δ log Brentmonth_of_yearNewey–West, 12 lags
Csimplex-constrained least squares on the pre-treatment outcome path (quadprog), fixed objectiveu5mr (annual native observations)transition window (1 July year rule)simplex donor weights; screening predictors as balance onlyin-space and in-time placebos
GridPPML (A) and local projection (B)three A outcomes, national FAAClog FAAC at lag 0–2; Δ log Brent3 adjustment sets (A); 4 control sets (B)clustered by state; Newey–West

Mathematical specification

Notation is the specs' own: a state-month panel for A and the grid, a national month for B, an annual country panel for C.

A · State dose-response, PPML

μs,t = exp( β · log(FAACs,t) + αs + γt )
fatalitiess,t ~ Poisson( μs,t )
β
elasticity of conflict fatalities to the state's allocation
αs
state fixed effect: every time-invariant difference, including size and the level of the allocation
γt
month fixed effect: the country-wide month, including national shocks
s, t
state and month

B · National transmission, local projection

Δh log FAACt = βh · Δ log Brentt + δm(t) + εt,h
Δh log FAACt = log FAACt+h − log FAACt−1
βh
cumulative pass-through at horizon h: the fraction of a 1 % oil move that has reached the states by h
δm(t)
month-of-year dummy: the disbursement calendar
h
horizon in months, 0 to 12
εt,h
residual, autocorrelated by construction; Newey–West with 12 lags

C · Synthetic control on under-5 mortality

ŷNGA,t = Σj wj · yj,t
w = argminw ≥ 0, Σj wj = 1 ‖ yNGA,pre − Ypre w ‖²
gapt = yNGA,t − ŷNGA,t
wj
simplex weight on donor country j
pre
every year before the window's first post year
gapt
the difference between Nigeria and its synthetic counterfactual, in deaths per 1,000 live births
placebo
the same fit with each donor treated in turn, and with the treatment date moved back 3, 5 and 7 years

Grid · The specification curve

ys,t = βk · log(FAACs,t−k) + αs + γt + us,t
k
dose lag in months, 0 to 2
adjustment set
state only; state and month; state and month with a state trend
window
full; excluding COVID; post-2019
note
the grid varies Design A's axes on the state panel and Design B's on the national month; every fitted cell is published, and the preferred cell is the pre-registered one

Registry vintage 2026-11-01