Vignoles et al. (2000) present an extensive discussion of methodological issues
and review the findings of the recent studies from the USA and the United
Kingdom in the context of the impact of various inputs on educational
outcomes. In the US, they find that the evidence of aggregate expenditure is
not reliable, so that one needs to consider different inputs separately. They say
higher quality studies do find a significant impact on class-size reduction on
student achievement, but this impact would be too small to justify the implied
increase in expenditure on cost-efficiency grounds. The evidence on teacher
characteristics is also rather ambiguous. They found teacher experience to
have a positive impact on achievement, but this is non-linear, as only the first
few years of experience are associated with significant, positive coefficients.
However, they say that, there is some robust evidence of a positive impact of
110
precise estimates are found when using detailed input measures (e.g.
Distinguishing between initial and further years of teaching experience, or
teacher qualifications by subject areas) rather than less informative summary
measures.
Table 4-5: Effect of Key Variables
Author Location Input effect of 1% inc. in
expenditure Angrist and Lavy (1999) Israel Class Size 3.6%, 1.7-1.9% Barro and Lee (1996) C. Country Class Size 2.70%
C. Country Teacher Salary Insignificant
Cooper and Cohn (1997) US Class Size 1-4%
US Teacher Experience Insignificant
US Teacher
Qualification 0.5-2.5%
Dewey et al.(2000) US Expenditure 0.15% - 0.12%
US Teacher Experience 0.008% - 0.04%
US Teacher
Qualification Insignificant
US Teacher Salary 0.04%
Figlio (1977) US Expenditure 2.5-6.4%
Gold Haber et al. (1999) US Class Size -6.6% to -7.2%
US Teacher
Qualification 2.30%
US Teacher Salary Insignificant
Gupta et al. (1999) C. Country Expenditure 3.30%
Hanushek et (1998) US Class Size 0.001- 0.05%
US Teacher Experience -7% to -15%
US Teacher
Qualification -4%
US Teacher Salary 0.76-1.2%
Hoxby (1999) US Class Size Insignificant
Karuger (1999) US Class Size 7-9%
US Teacher Experience 3%
US Teacher
Qualification Insignificant Kirjavainen and Loikkanen (1998) Finland Class Size 25% of inefficient
Kruger Whitmore(1999) US Class Size 20%
Marlow (2000) US Expenditure -0.01 to +0.002
Source : Compiled from Vignoles et al., pp. 23 to 29.
They find UK studies as typical of the literature pertaining to other developed
countries than the US, with relatively few methodologically strong studies. They
further contemplate that it is also patchy and lacks both depth and breadth of
coverage with respect to the different phases of education and data sets used.
111
[….]‘ (p.36).Their review of the selected studies related to three major inputs; expenditure per pupil; class size and teacher characteristics are given in table
4.5, which summarizes the results of 26 studies. Four studies examine the
expenditure as an independent variable. Dewey et al. (2000), found positive results between expenditure per pupil and SAT scores. Figlio‘s results indicate that expenditure reduction were harmful to student achievement in
mathematics, reading, science and social studies. Gupta et al. (1999), get a
positive response for countries that invest a greater proportion of national
income in education have higher enrolment rates.
Marlow (2000) focuses mainly on the effect of competition on school
performance, where competition is measured by an index based on the number
and concentration of different school districts within a particular country. His
premise is that a larger number of school districts, with more equal shares of
the market, will stimulate greater competition in that country and raise performance. They conclude, ―Overall, sufficiently significant concerns about that data and methodological tools employed in the four papers examined suggest that the ‗Hanushek view‘- that increased expenditure in itself does not raise student achievement – cannot be seriously challenged by the results of these studies‖ (p.22). Nine studies examine class size as an independent variable. The issue of class size remains contested one owing to conflicting
findings. Hanushek (1997) suggested that smaller class sizes do not
systematically lead to improve student achievement. Similarly, Hoxby (1998)
while using two quasi-experimental techniques in panel framework to examine
the influence of class size on test scores in Connecticut district school finds
112
National Educational Longitudinal Study (NLES) of 1998, Goldhaber and Brewer (1997) found that class size was significant but had the ‗wrong sign‘, i.e. suggesting that largest classes are associated with better student outcomes. In
a subsequent paper using the same data, Goldhaber, Brewer, and Anderson
(1999) found a similar result. On the other hand, Cooper and Cohn (1997),
using stochastic frontier estimation, found that smaller class sizes reduce
student test scores. However, three studies reflected in the above table do find
a positive link between smaller classes and student achievement; two are from
the US (Krueger, 1999; Hanushek, Kain, and Rivkin, 1998), and the third one
from Israel (Angrist and Lavy, 1999). Krueger found that the most substantial
gains occurred in the initial year of class size reduction, with the difference
between children in small and large classes increasing by a single percentage
point in subsequent years. A subsequent analysis by Krueger and Whitmore
(1999) suggests that the student achievement associated with the STAR experiment had permanent effects. Krueger‘s analysis suggests that a 1 percent decrease in class size would lead to a-level gain of 4 percent followed
by a 1 percent growth per annum, which appears to be an upper bound on the
potential return from reducing calls size (Vingoles et al., 2000, p.26).
Hanushek, Kain, and Rivkin (1998) provide alternative estimates of class size
effects using the Harvard / UTD Texas Schools Project database. Hanushek et
al. (1998), finds that class size effects are statistically significant for 4th and 5th
formers but not for 6th formers. However, although significant, the magnitude of the effects was considerably smaller than those obtained by Krueger (1999),
and account for less than 0.1 % of the total variation in student achievement.
113
classes, is an examination of class size reductions in Israel in the early 1990s
by Angrist and Lavy(1990).
For teacher characteristics, they examined three most popularly employed
measures of teacher quality viz; teacher experience; teacher education and
teacher salary in three studies. A summary of these studies is given in table 4-5
which will be seen in comparison to Hanushek (1997a) survey which gives the
percentage of studies finding significant results for the most commonly included teacher characteristics teacher‘s education (9 % positive and significant), experience (29 %), and salary (27 %). In the context of teachers experience
they consider three studies for analyzing the input - teacher experience and its
impact on student achievement. Hanushek et al. (1998), found a significant and
positive effect on achievement for teachers with up to 2 years experience, as
compared to teachers with no experience (with the exception of 4th and 5th form mathematics). However, no effects were found from greater levels of teacher
experience. As summarized in table 6-5, other studies surveyed in this paper
(other than Dewey et al.) all found that teacher experience lacked explanatory
power. However, these alternative studies do not directly contradict Hanushek
et al. In the context of teacher education, they refer to six studies done to
examine the effect of teacher qualification on student achievement. For teacher
education, only three studies find significant results. In particular Goldhaber
and Brewer (1997) detect robust results that being taught by a teacher with a degree in mathematics has a positive impact on pupils‘ mathematics scores. By contrast however, Hanushek et al. (1998) , found that 4th form students appear to suffer through having more highly educated teachers. Neither study found
114
education and outcomes-results that are consistent with Goldhaber et al’s recent re-examination of the NLES database. The only paper that appears to have had some success with an ‗aggregate‘ teacher education variable is Cooper and Cohn (1997), however, their results are not robust at all the
specifications. In the context of impact of teacher salary on student
achievement they present four studies. Hanushek et al. (1999), found that
when student fixed effects and teacher mobility are accounted for, a higher
salary exert a positive influence on student achievement. Dewey et al. (2000),
obtain positive and robust results that relative salary differentials matter in
determining student outcomes. Other analyses, summarized in table 4-5 find no significant relationship. They sum up, ―there is some robust evidence that teacher experience and teachers‘ salaries have significant effects but that teacher‘s education level (with the exception of teacher with a qualification in mathematics) do not‖ (Vingoles et al., 2000, p.32).