A second thing to note is that in the H(t) function, the
time in school (t – t
0
) is just elapsed time, conditional
on the choice made regarding the time to first enroll
(t
0
).
In Ghana this is pretty reasonable because grade
repetition is not common (2-3% per year).
3

Lastly, it is inevitable that some of the components of
these vectors, such as child motivation to learn and
parental tastes for schooling, are not observed.
In the second model discussed in the last lecture, if
credit markets are perfect and schooling is valued
only for the higher income it can bring in later years
(and not as a consumption good in its own right),
then household income/wealth would have no effect
in any of these three equations.
If an effect is
detected (after fixing all possible estimation
problems), than one or both of these assumptions is
false.
Under the first assumption it is also the case
that there is no reason for children to delay their
schooling, and schooling would also be “full-time”
until the day comes to quit school.
The empirical model estimated for the determinants
of learning of child i in school j is:
H
ij
=
β
0
+
β
1
S
i
+
X
Hi
′
β
X
+
Z
Hij
′
β
Z
+
C
Hi
′
β
C
+ ε
Hij
where S
i
is years of schooling.
As always, estimates of this equation may be biased
if ε
Hij
is correlated with the observed variables.
First,
such correlation could come about due to a potential
selection bias problem: not all kids are enrolled in
4