Regression diagnostics of goodness-of-fit, a residual normality test (based on a
small sample correction to the Jarque-Bera test; see Hendry and Doornik [21]), and
relatively less collinear local parameter estimates (ρ(βzi)) give preference to models
H6 and S7 among alternative GWR specifications and different initial reference periods,
for healthcare and primary schools respectively (see Table 3). Based on econometric
results of these models, improvements in public healthcare access have been
relatively less responsive on average to initial coverage rates, but more consistent
in terms of numbers effect and less spatially unstable, compared with the primary
education sector. To better understand GWR results, nearest-neighbour grids of
local parameters and related t-statistics for the numbers effect can be visualised in
Figures 3 and 4, for H6 and S7 respectively. The maps highlight heterogeneous local
impacts on public service access in terms of numbers of uncompensated losers: ceteris
paribus, while for healthcare some central-western districts (Illela and Tahoua,
among a few others) turn out to have ’benefited’ most, for primary education the
numbers effect appears to have been relatively better targeted in the south-east (in
districts such as Diffa and Nguigmi). 10 Relative to the education sector, local population
size turns out to be especially relevant as a factor associated with improved access coverage for districts in the south-central regions of Maradi and Zinder, such as Tessaoua and Tanout (Fig. 5).