This report examines whether changes in Nigeria’s economy (GDP per capita, inflation, unemployment) are associated with primary school enrolment between 1990 and 2023. [Add 2 sentences on why this matters, for example for education policy or planning.]
ng <- read_csv("data/nigeria_wdi_raw.csv")
glimpse(ng)
## Rows: 34
## Columns: 8
## $ country <chr> "Nigeria", "Nigeria", "Nigeria", "Nigeria", "Nigeria", "N…
## $ iso2c <chr> "NG", "NG", "NG", "NG", "NG", "NG", "NG", "NG", "NG", "NG…
## $ iso3c <chr> "NGA", "NGA", "NGA", "NGA", "NGA", "NGA", "NGA", "NGA", "…
## $ year <dbl> 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 199…
## $ gdp_pc <dbl> 1577.196, 1541.591, 1571.190, 1498.959, 1433.389, 1395.07…
## $ inflation <dbl> 7.364400, 13.006973, 44.588843, 57.165253, 57.031709, 72.…
## $ unemployment <dbl> NA, 3.831, 3.790, 4.023, 4.064, 4.152, 4.043, 3.964, 3.92…
## $ enrolment <dbl> 83.10286, 82.32250, 86.74438, 91.29586, 91.53790, 87.4475…
ng_clean <- ng %>%
select(year, gdp_pc, inflation, unemployment, enrolment) %>%
arrange(year)
tibble(variable = names(ng_clean),
missing_values = colSums(is.na(ng_clean))) %>%
knitr::kable()
| variable | missing_values |
|---|---|
| year | 0 |
| gdp_pc | 0 |
| inflation | 0 |
| unemployment | 1 |
| enrolment | 5 |
rows with missing values were dropped automatically in the regression.
ng_clean %>%
pivot_longer(-year, names_to = "indicator", values_to = "value") %>%
ggplot(aes(year, value)) +
geom_line() +
facet_wrap(~indicator, scales = "free_y") +
labs(title = "Nigeria: key indicators, 1990-2023",
x = NULL, y = NULL, caption = "Source: World Bank WDI")
Inflation spiked sharply in the mid-1990s, peaking above 70%, then stayed mostly below 20% before rising to about 24% by 2023. GDP per capita fell slightly in the 1990s, rose steadily from about 2000 to a peak around 2014, then declined. Enrolment stayed mostly between 80% and 95%, but shows an unusually sharp drop around 2015 and several missing years, which is examined in section 6.1.
model <- lm(enrolment ~ gdp_pc + inflation + unemployment, data = ng_clean)
summary(model)
##
## Call:
## lm(formula = enrolment ~ gdp_pc + inflation + unemployment, data = ng_clean)
##
## Residuals:
## Min 1Q Median 3Q Max
## -50.903 -1.668 2.329 6.260 9.663
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 126.080846 20.571122 6.129 2.48e-06 ***
## gdp_pc -0.014946 0.006601 -2.264 0.0329 *
## inflation -0.084178 0.160344 -0.525 0.6044
## unemployment -2.469296 4.367566 -0.565 0.5771
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 12.26 on 24 degrees of freedom
## (6 observations deleted due to missingness)
## Multiple R-squared: 0.2207, Adjusted R-squared: 0.1232
## F-statistic: 2.265 on 3 and 24 DF, p-value: 0.1067
vif(model)
## gdp_pc inflation unemployment
## 1.403639 1.361158 1.045978
dwtest(model)
##
## Durbin-Watson test
##
## data: model
## DW = 1.8987, p-value = 0.2239
## alternative hypothesis: true autocorrelation is greater than 0
par(mfrow = c(2, 2)); plot(model); par(mfrow = c(1, 1))
In the full model, GDP per capita was the only statistically significant variable (p = 0.033). Its coefficient of -0.015 means each additional US$100 of GDP per capita was associated with a 1.5 percentage-point lower enrolment rate, holding other variables constant. Inflation (p = 0.604) and unemployment (p = 0.577) were not significant. The model as a whole was not significant (F-test p = 0.107) and explained about 22% of the variation in enrolment (adjusted R-squared = 12%). VIF values below 1.5 show no multicollinearity problem, and the Durbin-Watson test (p = 0.224) shows no significant autocorrelation. However, the residual plots show one extreme observation (2015), where enrolment falls far below all other years and which may be driving the GDP result.
infl_ts <- ts(ng_clean$inflation, start = 1990, frequency = 1)
fit <- auto.arima(infl_ts)
summary(fit)
## Series: infl_ts
## ARIMA(2,0,0) with non-zero mean
##
## Coefficients:
## ar1 ar2 mean
## 0.9693 -0.3153 18.1591
## s.e. 0.1602 0.1586 4.7757
##
## sigma^2 = 107.7: log likelihood = -126.72
## AIC=261.43 AICc=262.81 BIC=267.54
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 0.1956489 9.908401 6.24735 -17.07859 37.20106 0.9621552
## ACF1
## Training set -0.03063995
checkresiduals(fit)
##
## Ljung-Box test
##
## data: Residuals from ARIMA(2,0,0) with non-zero mean
## Q* = 8.7662, df = 5, p-value = 0.1188
##
## Model df: 2. Total lags used: 7
autoplot(forecast(fit, h = 5))
The automatic model selection chose an ARIMA(2,0,0) model for inflation. The forecast suggests inflation of about 25% in 2024, moving to about 20% by 2028. However, the uncertainty range is wide: by 2028 the 95% interval runs from 0% to 50%. This means the forecast should be read as a rough indication of direction, not a precise prediction, especially because inflation in Nigeria has been affected by sudden policy and exchange-rate changes that a simple model cannot anticipate.
In this analysis, GDP per capita was weakly and negatively associated with primary school enrolment in Nigeria between 1990 and 2023, while inflation and unemployment showed no significant link. The overall model explained little of the variation in enrolment, and the GDP result depends on one extreme observation in 2015. This suggests national economic indicators alone do not explain enrolment well. Inflation is forecast to ease to about 20% by 2028, but with a very wide uncertainty range.
Recommendation: Check the unusual 2015 enrolment value against UNESCO Institute for Statistics or National Bureau of Statistics data before relying on it. Future work should use state-level or household-level data, such as the Nigeria Demographic and Health Survey, to identify what drives enrolment, since national averages hide large regional differences.
Data: World Bank WDI. Analysis in R.