Do Smaller Classes Always Improve Students’ Long‐run Outcomes?

AuthorAstrid Marie Jorde Sandsør,Bjarne Strøm,Torberg Falch
DOIhttp://doi.org/10.1111/obes.12161
Published date01 October 2017
Date01 October 2017
654
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 79, 5 (2017) 0305–9049
doi: 10.1111/obes.12161
Do Smaller Classes Always Improve Students’
Long-run Outcomes?*
Torberg Falch†, Astrid Marie Jorde Sandsør‡ and Bjarne Strøm
Department of Economics, Norwegian University of Science and Technology Dragvoll,
N-7491 Trondheim, Norway (e-mails: torberg.falch@svt.ntnu.no; bjarne.strom@svt.ntnu.no)
Department of Economics, University of Oslo, Postboks 1095 Blindern, N-0317 Oslo,
Norway (e-mail: a.m.j.sandsor@econ.uio.no)
Abstract
We exploit the strict class size rule in Norway and matched individual and school register
information for 1982–2011 to estimate long-run causal effects on income and educational
attainment. Contrary to recent evidence from the US and Sweden, we do not find any
significant average effect on long-run outcomes of reduced class size. We further use the
large register data set and quasi-experimental strategy to estimate whether the class size
effect depends on external conditions facing students and schools, such as teacher quality,
extent of upper secondary school choice, school district size, local fiscal constraints and
labour market conditions. Overall, we find that the class size effect does not depend on
these factors measured at the school district level. The absence of class size effects on
long-run outcomes in Norway is consistent with earlier findings for short-run outcomes,
using comparable data and empirical strategies.
I. Introduction
The impact of school resources on student performance has been disputed since the pub-
lication of the Coleman et al. (1966) report. Although availability of data and empirical
strategies to uncover causal effects have increased substantially in recent years, the evi-
dence on the effect of resources on education outcomes is still inconclusive.1The literature
is not conclusive even for more narrow and popular policytools as class size. Although the
results from the large famous randomized experiment in Tennessee (STAR) suggest that
smaller classes are beneficial in terms of test scores,2studies, using quasi-experimental
JEL Classification numbers: I2, H7.
*We greatly acknowledge comments from two referees, Karl Ove Moene and participants at the Workshop on
Applied Economics of Education in Catanzaro, the CESifoArea Conference on Economics of Education, and seminars
at University of Lancaster and the Institute of Social Research in Oslo.
1Summaries of the literature on the relationship between school resources and student achievement include
Hanushek (1996, 2003, 2006), Krueger (2003) and Webbink(2005).
2See Krueger and Whitmore (2001) and Chetty et al. (2011) on evidence from the STARexperiment. In contrast
to the STARexperiment, field experiments on class size conducted before WW II provided little evidence in support
Smaller classes and long-run outcomes 655
approaches to identify causal effects differ substantially in their conclusions.3One inter-
pretation is that extra resources and reduced class size are effective tools in some contexts,
while ineffective in other contexts.
Testscores only measure cognitive skills, while class size may also affect non-cognitive
skills. In addition, evidence based on test scores may be biased in settings where teachers
systematically manipulate test scores as recentlydemonstrated in Angrist, Battistin and Vuri
(2015).4Both arguments suggest that analysesof long-r un outcomes in terms of educational
attainment and income in adulthood as used in our empirical study would provide the most
credible evidence of the effect of school resources. Such studies will embed all short-run
effects, including effects on non-cognitive skills that are difficult to measure directly.
Three recently published papers analyse long-run effects of class size. Chetty et al.
(2011) and Dynarski, Hyman and Schanzenbach (2013) study long-run outcomes for par-
ticipants in the STARexperiment, while Fredriksson, Oosterbeek and ¨
Ockert (2013) exploit
a class size rule in Sweden to estimate both short-run and long-run outcomes. These papers
all find positive long-run effects of smaller classes, which suggests that the mixed effects
in the literature on short run effects are related to imperfect measurement of student skills.5
However, the findings for the long run are also consistent with the findings in the short
run, using test scores within the same contexts.6Of particular interest is Fredriksson et al.
(2013) who find a positive short-run effect on non-cognitive ability, which is an outcome
rarely available for researchers.These results motivate studies on long-run outcomes from
contexts where the evidence indicates no class size effect on short-run outcomes.
In this paper, we estimate long-run effects of class size for Norway where previous
research has not been able to provide evidence of short-run gains from smaller classes
in terms of student achievement.7We investigate whether the class size effect in lower
secondary education depends on characteristics of the environment in which the schools
of the hypothesis that smaller classes increase student achievement, see Rockoff (2009) for an interesting review of
these early field experiments.
3The seminal paper by Angrist and Lavy (1999) initiated a literature exploiting class size rules in a regression
discontinuity framework, Hoxby (2000) uses idiosyncratic variationin cohor t size, andW ¨oßmann and West(2006)
employ a within-school across classes strategy. While Angrist and Lavy (1999) find the expected negative effect
of class size on student achievement for Israel, Hoxby (2000) and W¨oßmann and West (2006) find zero effects in
Connecticut and for most OECD countries respectively. In a recent paper, Denny and Oppedisano (2013) even find
positiveeffects for the US and the UK. They use the same empirical strategy as W¨oßmannand West(2006) in addition
to an approach based on restrictions on higher order moments.
4Angrist et al. (2015) exploit a class size rule in Italy and find a strong negative relationship between test scores
and class size in Southern Italy.This relationship is, however, entirely driven by manipulation of the test scores by
the teachers.
5Chetty et al. (2011) find positive effects of smaller classes on college attendance and earnings, but the latter is
imprecisely estimated and insignificant.
6See for example Krueger and Whitmore (2001) for the STAR experiment. In addition to Fredriksson et al.
(2013), several studies from Sweden find that increased school resources increase student performance in the short
run, including Bj¨orklund et al. (2004, ch. 4), Lindahl (2005) and Fredriksson and ¨
Ockert (2008). Browning and
Heinesen (2007) and Heinesen (2010) find that lower class size in Danish compulsory education increases student
performance in terms of both student test scores and educational attainment.
7The Norwegian studies exploiting the class size rule in short-run studies are Bonesrønning (2003), Woessmann
(2005), Leuven, Oosterbeek and Rønning (2008), and Iversen and Bonesrønning (2013). They find small or zero
average effects of class size. Hægeland, Raaum and Salvanes (2012) exploit variation in school resources across
school districts with different income from local taxes on hydropower plants in Norway. They find that higher
resources increase student achievement.
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
656 Bulletin
and students operate. Severalhypotheses of heterogeneous effects across school districts are
derived and tested. Leuven and Løkken (2015) explore similar Norwegian data, estimating
the effect of class size both in primary and lower secondary education. Their analysis
utilizes that some schools include grade 1–10 and estimates the effect of class size in
primary education in addition to the effect in lower secondary education, assuming that
the students stayed in the same school during all school years.We find qualitatively similar
average effects of class size as they do.
The findings for short-run outcomes differ substantially betweenthe Scandinavian coun-
tries Sweden, Denmark and Norwaywith apparently similar educational and labour market
institutions. All countries have small income differences, generous welfare state arrange-
ments, and comprehensive public school systems seeking to equalize opportunities across
families and students. Nevertheless, closer inspection reveals that important institutional
differences prevail with regard to, e.g. school district size and teacher shortages.8
We first exploit the strict class size rule in Norway and match individual and school
register information from 1982 through 2011 to estimate causal effects on educational
attainment and income. While experimental studies are often viewedas the ‘gold standard’
in empirical research, exploiting the class size rule in a quasi-experimental approach makes
it possible to circumvent the potential Hawthorne effect that might plague experimental
studies (Ehrenberg et al., 2001). In contrast to Fredriksson et al. (2013), we are able to use
register data for the whole population of schools for cohorts born between 1966 and 1984,
representing almost 1 million students and 1,150 schools with separate catchment areas.9
Secondly, information on the whole population of schools and students offers a unique
possibility to use the quasi-experimental strategy to study whether the class size effect
depends on characteristics of the environment in which the schools and students operate.
Wefocus on dimensions that mirror differences in external conditions indicated by previous
studies to be important for school efficiency and student performance, such as teacher
quality, extent of upper secondary school choice, school district size, local fiscal constraints,
and labour market conditions.
We find insignificant effects of class size in grade 8–10 on educational attainment
and income. While this is in contrast to the previous papers on long-run effects, it is in
accordance with the findings in the short run for Norway and the long-run effect in Leuven
and Løkken (2015). Moreover, we find no evidence that class size effects vary with school
district characteristics.
The paper is organized as follows. In section II, we present arguments as to why the
effect of resources may depend on characteristics of the external environment in which
schools and students operate. Section III describes the institutions and the data, while
8The institutional differences increased after the major reforms in Sweden in the mid-1990s. Our focus here
is on institutional differences that have prevailed for several decades since several of the Swedish studies, includ-
ing Fredriksson et al. (2013), use data on individuals graduating compulsory education before these reforms. See
Bj¨orklund et al. (2004, ch. 4) for a description of the Swedish reforms in the 1990s and Nusche et al. (2011) and
Bonesrønning (2013) for a description of recent Norwegian reforms.
9Fredriksson et al. (2013) use data for a roughly 10% sample of the cohorts born 1967, 1972 and 1982, and a
5% sample of the cohort born 1977. In addition, to ensure exogenous catchment areas for schools, they only include
school districts (‘rektorsomr˚ader’) with one school in their main analysis, implying that they are left with a sample of
about 6,000 students and 191 schools. Since we include all schools in our analysis, the statistical powerin robustness
and heterogeneity analyses should be higher.
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT