test rejects the consistency of the random effects model. The fixed effects
model provides estimates of about 0.03 for both lectures and classes, but the
coefficients are no longer statistically significant. This result indicates that,
once we control for omitted variable bias, as we should, it is not possible to
identify separately the effects of lecture and class attendance.
6 Discussion and conclusions
The results of the empirical analysis reported in this paper suggest two main
conclusions. First, neither proxy variables nor instrumental variables pro-
vide a viable solution to the omitted variable bias in estimating the effect of
attendance on academic performance. The alternative solution proposed in
this paper is to exploit the panel structure of the data set to explicitly take
into account the effect of unobservable factors correlated with attendance,
such as student ability, effort and motivation. Second, after controlling for
unobservable factors, attendance of either lectures or classes is found to have
a smaller but significant impact on test scores in an Introductory Microeco-
nomics course. On the basis of this evidence, can we conclude that teaching
has a positive impact on student learning?
One possible objection could be that test scores are not a good measure of
learning: attendance could affect exam performance because students learn
how to do well in the exam, without any actual effect on the quality of learn-
ing (see e.g. Deere, 1994). This can be true if, for example, lectures only
improve exam-taking skills, or provide information on the topics and type of
questions that will be in the exam or, more generally, lectures present exam-
inable material that is not covered in the textbook.26 This kind of critique,
however, does not apply to the data set investigated in this paper: all stu-
dents had access to detailed lecture notes and past exam papers on the course
web site, so that attendance did not reveal any private information. In addi-
tion, lectures and classes followed very closely the textbooks, so that all exam
questions could be answered correctly by students not attending lectures or
classes, who had relied exclusively on the texts to prepare for the exams. It
should also be observed that the marking scheme was fully objective, so that
test scores could not be used to reward students for attendance.
and classes in the RE model is not statistically significant.
26It is also possible that grades are used, either explicitly or implicitly, to reward for
attendance.
16