The organisation of university education is increasingly in the
spotlight, both in academic and policy circles. Recent research has
stressed the importance of higher education in providing positive
externalities within firms (Moretti 2004), within local labour markets (Glaeser et al. 1992), and in fostering
economy wide growth (Aghion et al. 2007).
Concurrently, most OECD countries have adopted policies that have
led to dramatic increases in university enrolment during the last
decade. The average annual increase in university enrolment in OECD
countries during the period 1995-2005 was above 4%. In the UK, this
growth has occurred at both the undergraduate and graduate level
and across a wide range of universities. The UK is actually at the
low end of enrolment growth within the OECD. Between 1998 and 2005,
the US experienced a 30% increase in student enrolment, for
example.
Such breathtaking increases in enrolment inevitably lead to
university students facing larger class sizes. The effect of
increasing class size in tertiary education is not well
understood.
The established literature on class size effects in primary and
secondary schools provides useful guidance (Krueger 1999, Angrist and Lavy 1999, Hoxby 2000), in universities the range of class
sizes is typically larger than at other tiers of the education
system, and different mechanisms driving class size effects might
operate. Although tertiary education may involve more self-learning
than primary or secondary education, class size remains solidly at
the top of the policy agenda and concerns of both faculty and
students.1
Identifying class-size effects from within-student
variation
To address this policy question, we estimate the impact of
class size on the final exam marks of graduate students in a
leading UK university between 1999 and 2004 (Bandiera et
al. 2010). As we observe the same student being exposed to
very different class sizes, we estimate the effects of class size
on students' exam performance by comparing the same
student's performance to her own performance in courses with small
and large class sizes.2
It is important to stress that, on average, most of the variation
in marks is due to fixed students' characteristics and not
university inputs.3 On average the
performance of a given students only varies by around 7% of the
average mark across her courses. We shed light on how much of this
within-student difference is attributable to differing class sizes
the student faces.
The effect of class size on students' performance is - as expected
- negative; students do worse in big classes. Namely, a given
student receives lower marks in courses with
larger classes, everything else equal.
To get a sense of the magnitude of this effect, our estimates imply
that a one standard deviation increase in class size from the mean
(that is going from the average class of 56 to a class size of 89)
would decrease the mark by 9% of the observed variation in marks
within a given student. These estimates, however, mask two
important forms of heterogeneity: (i) the impact of class size
varies across the range of class sizes; (ii) the effect of class
size varies across students.
On the first form of heterogeneity, the negative effect of class
size on student exam performance is large and negative only in the
smallest and the largest classes. There is no class size effect
across a wide range of intermediate class sizes. The magnitudes
imply that moving the average student from a class of 10 to a class
of 25 leads to a drop in exam performance of around 12.5% of
within-student standard deviation. Increasing the class size from
25 to 45 determines a further 12.5% drop.
In contrast, there is no impact in a wide intermediate range, while
moving from 80 to 150 determines a further drop of 25% in the
within-student standard deviation. If moved from a very small class
(of size 10) to a very large class (of size 150), the average
student can be expected to suffer a loss corresponding to about 50%
of the overall variation in exam marks the average student
experiences across all of her courses.
The second form of heterogeneity concerns students' ability.
Students at the top of the mark distribution are those most
affected by class size. The effect is almost four times larger for
students in the top 10% of the distribution of exam marks than for
students at the bottom 10%, and about 50% larger than the average
student. This heterogeneity is most apparent in the largest classes
and virtually non-existent for a range of intermediate class sizes.
This implies the highest-ability students would benefit the most,
in terms of academic achievement, from any reduction in class
sizes, when class sizes are initially very large.
To shed light on the underlying mechanisms for the class-size
effect, our analysis uses information on teachers' assignments to
classes and on students' characteristics. We find no evidence that
departments purposefully assign faculty of differing quality to
different class sizes, and we find no evidence that faculty members
alter their behaviour when exposed to different class sizes. It
appears that the preparation and delivery of lectures is
independent of the number of students taught.
On student characteristics, the class-size effect does not vary
with proxies for students' wealth. Hence if larger classes resulted
in lower grades because students had more limited access to library
books or computer laboratories, the effect should have been smaller
for students who can purchase these inputs privately.
Moreover, the class-size effect does not vary with student's
familiarity with this particular university as an undergraduate or
with the UK system generally. This casts doubts on the relevance of
mechanisms that work through the information students have, such as
their awareness of other local resources (for example other
libraries in the area), or their knowledge of the characteristics
of faculty, courses, or departments.
Discussion and policy implications
Against a backdrop of rapidly increasing enrolment rates
in tertiary education, our analysis has important policy
implications. Class size matters for student performance and
particularly for the most able students.
However, reducing class size is not always an effective strategy
and is certainly not effective for all students in the same way.
Reducing the size of very large modules (above 100) could be a
cost-effective way to improve students' performance. For modules in
the range 30-100 reducing class size could be a rather ineffective
strategy, while for classes below 30 it could be a valid but not
necessarily cost-effective strategy.
Attention should be devoted to other inputs in such cases, and more
refined and cost-effective solutions than pure number counting
should be identified. To this end, it is important to have a better
understanding of the mechanisms that link class size and
performance.
Although student-to-staff ratio is a commonly used indicator of
quality both in national and international comparisons, this might
be a noisy measure of quality over this intermediate range of class
sizes.4 Given the mechanisms our data
rule outs, there appear to be at least two ways that larger classes
reduce students' performance. First, changes in student behaviour
such as their attentiveness or participation. Second, reduced
resource availability, such as library books or faculty time during
office hours.
As the best students are the most affected, that could imply that
large classes induce a reduction in tutoring activity rather than a
substantial deterioration in classroom conditions. It is reasonable
to expect that the best students are able to compensate classroom
deterioration at least as well as other students. However, the best
students are also those that benefit the most (in terms of both
learning and motivation) from contact with teachers. They,
therefore, suffer the most in terms of reduced performance when
such contacts or tailored feedback is less frequent.
Footnotes
1 This is particularly evident in the UK, where concerns
on the increasing student-to-staff ratios in higher education
institutions have recently been expressed in a report of the
Department for Innovation, Universities, and Skills and by the most
important unions of university teachers.
2 Our estimates are therefore purged from confounding effects that
arise because students choose which modules to take. For instance,
if more able students were to choose smaller classes, a
cross-student comparison would capture both the effect of class
size and the effect of student ability. The within-student
comparison only captures the former.
3 Characteristics like ability and motivation are certainly
affected by previous experiences and schooling, but they can be
taken as given for what concerns university policy.
4 This choice is somehow justified by the fact that, in order to
evaluate teaching quality, the student to staff ratio probably
remains the only globally available and comparable indicator.
Auteurs: Oriana Bandiera, Valentino Larcinese en Imran
Rasul