Of course, religious affiliation may itself be related to educational
achievement and to economic growth. Consequently this IV specification
conditions on Catholic shares in the modern population. Thus the identifying assumption of this IV model is that, holding constant the effects
of modern religious affiliation, historical religious affiliation is not otherwise related to modern growth, apart from the indirect effect through
its bearing on competition and thus productivity in the modern school
system.
The IV estimation reported in column 4 oftable 4.1 provides additional support for our underlying model of cognitive skills and growth.
The share of Catholic population in 1900 is indeed positively related to
our measure of cognitive skills in the first stage, while in 1970 it is negatively related, which is in line with the previous literature.16 At the same
time, the 1970 Catholic share does not enter the second-stage growth
model significantly. But importantly, the variation in cognitive skills that
is related to historical Catholic shares has a significant positive effect
on economic growth that is close to the OLS estimate. TheF-statistic of the instruments in the first stage is just above 10, and LIML estimates
and Moreira bands confirm the result is not driven by weak-instrument
problems.