Вы находитесь на странице: 1из 14

J Econ (2010) 101:199212

DOI 10.1007/s00712-010-0145-0
A signalling model of school grades under different
evaluation systems
Maria De Paola Vincenzo Scoppa
Received: 31 March 2009 / Accepted: 18 May 2010 / Published online: 5 June 2010
Springer-Verlag 2010
Abstract We examine the role of school grades as a signal of worker productivity
under different examination systems in relation to errors that may affect student per-
formance. Firms use school grades as a signal of workers effective skills, taking
into account that these evaluations are effected by stochastic shocks. We show that
more precise evaluation systems, being associated with a higher reactivity of wages
to school grades, induce students to provide more effort. Low ability students tend
to react less than high ability students. Moreover, individuals with low abilities may
prefer less accurate evaluation systems. Nevertheless, when productivity increases,
these systems become less convenient and the number of individuals preferring them
diminishes. Our analysis highlights an important trade-off between centralised and
decentralised evaluation systems. On the one hand, frequent evaluations, typical of
decentralised systems, weaken the impact on grades of those errors which inuence
student performance and, so, reduce signal noise, while, on the other hand, differ-
ent teachers generally adopt different performance assessment standards, leading to
noisier evaluations.
Keywords Education Signalling School grades Evaluation systems
JEL Classication D02 H42 I28
M. De Paola (B) V. Scoppa
University of Calabria, Cosenza, Italy
e-mail: m.depaola@unical.it
V. Scoppa
e-mail: v.scoppa@unical.it
123
200 M. De Paola, V. Scoppa
1 Introduction
External exit examinations and high standards are often advocated as the appropriate
method for evaluating students in order to improve the educational systemand enhance
student skills. Anumber of works emphasise the positive consequences that centralised
exams may produce on agents involved in the educational process. First of all, since
grades in centralised exams provide employers with a better signal of student skills
than locally-graded examinations do, wages tend to be higher where central exams are
in place and students have greater incentives to learn (see Wmann 2005). Secondly,
according to Bishop and Wmann (2004), centralised exams provide parents with
information about the performance of their children against an absolute standard and
relative to other students in the educational system. This information allows parents to
understand whether it is the whole class which is performing badly or if poor perfor-
mance is limited to their own child, and, in this way, it enables them to exert pressure
both on students and teachers (Bishop 1996, 1997, 2006). Thirdly, central exams also
affect teachers behaviour, since student performance in standardised tests can be used
to monitor teaching quality on a regular basis and to develop output based incentive
schemes (Lavy 2004; Glewwe and Kremer 2006; Jrges et al. 2005). Finally, with
centralised examinations, student achievement becomes crucial for school reputation
and for attracting good students.
In spite of the favour encountered by centralised exams among economists, they are
highly criticised by teachers and pedagogical specialists, who question their efcacy.
The opponents of central exit exams argue that pedagogical discretion is necessary to
deal with students with heterogeneous characteristics. According to this view, external
exams induce educators to teach to the test and to ignore important areas of knowl-
edge. Teachers are able toevaluate, throughdirect observations, aspects of the students
knowledge that are difcult to measure through standardised exams, which typically
consist of a number of very specic questions. Lazear (2006) examines this issue con-
cluding that standardised tests should be used when learning and monitoring are very
costly, but should be discarded for high ability students.
An important issue, which we analyse in this paper, is that of the effect produced
by different evaluation systems on measurement errors affecting the evaluation of
students in exams and the use of these evaluations as a signal of effective skills. We
consider two different types of errors, those that inuence student performance and
those deriving fromdifferent evaluation standards. According to the relevance of these
errors, grades obtained by individuals during their educational career are a more or less
reliable signal of their effective skills. In fact, we assume that rms, in the initial stage
of workers careers, are not able to observe individual abilities or the effort provided
in the educational process, but that they simply observe each students school leaving
grades and try to infer students effective skills from these through a signal extraction
process.
Starting from these assumptions, we show that more precise evaluation systems are
associated with a higher reactivity of wages to school grades and, therefore, induce
a higher level of student effort. However, when ability and effort are complements,
the effect is heterogeneous across students, since the incentive effect of more accurate
evaluation systems increases together with student ability.
123
A signalling model for school grades 201
From our analysis, it emerges that individuals with low abilitytypically charac-
terised by weak social and economic backgroundmay prefer less accurate evaluation
systems, since, due to the lack of effective signals of individual skills, rms tend to
pay more uniform wages which are based on average abilities and effort. When the
distribution of abilities in the population is not symmetrical, a majority made up of low
ability subjects may oppose the adoption of a precise evaluation system. Interestingly,
the number of individuals preferring less precise evaluation systems diminishes when
productivity increases. In fact, noisier evaluation systems produce two effects: on the
one hand, they lead to a more egalitarian pay structure, while, on the other hand, they
lead to a reduction in student effort and, as a consequence, lower the total output of
the economy. With a high level of productivity of skills, the wage reduction deriving
from lower effort tends to weigh against the positive effect that low ability agents
obtain from the redistribution of income. Therefore, the higher the productivity of
skills, the greater the support offered by society to evaluation systems which produce
more accurate signals for the labour market.
Using this framework, we compare advantages and disadvantages of centralised
and decentralised evaluation systems. Decentralised evaluation carried out by teach-
ers allows students to be evaluated more frequently, since teachers and students interact
on a daily basis, in comparison to centralised systems. Frequent evaluations reduce the
impact of errors which inuence student performance, such as being lucky enough to
have studied the material relevant for the examination or being in good health during
the examination. However, different teachers may adopt different criteria and evalua-
tion standards, introducing noise into student evaluation. On the other hand, centralised
exams are more complex to organise and their frequency is limited, implying that the
evaluations which they produce may be more affected by stochastic shocks inuencing
student performance. However, since centralised exams rely on a common standard,
they may lead to more reliable and comparable evaluations.
Depending on the importance that these two types of errors play in distorting stu-
dent performance at exams in comparison with their effective skills, it may be efcient
to adopt a centralised evaluation system rather than a decentralised one. We show that
grades in centralised exams only provide a better signal of student competence than
locally-graded exams do when the variance in errors affecting student performance is
lower than the variance of errors deriving from the adoption of different assessment
methods.
The paper is organised as follows. Section 2 briey discusses the related litera-
ture. Section 3 presents a model showing how the signalling value of school grades is
inuenced by the accuracy of the evaluation system and its effects on student effort
and welfare. In Sect. 4, we compare advantages and disadvantages of centralised and
decentralised evaluation systems. Section 5 offers some concluding remarks.
2 Related literature
Educational literature has recently pointed out that student effort represents a funda-
mental input into the educational production function. Bishop and Wmann (2004)
show that student effort is related to a number of institutional factors, such as the
type of examination, the degree of school autonomy, teacher quality, and so on (see
123
202 M. De Paola, V. Scoppa
also Bonesrnning 2004; De Paola and Scoppa 2007). Students take these aspects
into account when deciding their level of effort and react strategically. De Fraja and
Landeras (2006) show, for example, that resources devoted to the improvement of
teaching quality may not lead to better educational results if student effort and school
inputs are substitutes in the educational production function.
Evaluations are used in our model as a signal of effective student skills. The idea
that rms cannot observe worker skills but only their educational attainment is at the
heart of the models proposed by Costrell (1994) and Betts (1998) who analyse howthe
setting of educational standards inuences the effort provided by students. From their
analysis, it emerges that the definition of a high standard increases the educational
effort of students who attain a certain degree, but, on the other hand, discourages low
ability subjects from acquiring the credential.
1
Following the path taken by these works, we assume that student effort is deter-
mined by the perspective of obtaining higher wages on the labour market. The novel
aspects we introduce focus on the relationship between evaluation systems and student
performance, considering that rms form their expectations regarding an employees
productivity on the basis of the evaluation the latter obtained at school. The informa-
tive value of school grades depends on measurement errors which, in turn, are affected
by the type of evaluation adopted. We show that the effects of more precise evaluation
systems are heterogeneous according to individual ability: high ability individuals
obtain an advantage while low ability students may suffer a welfare reduction.
The effects of student effort and of a number of institutional features on academic
performance have also been investigated empirically. Fuchs and Wmann (2007)
show that student effort has a positive effect on student performance in international
tests. Stinebrickner and Stinebrickner (2004), Stinebrickner and Stinebrickner (2008),
using a sample of US college students, nd a positive and statistically significant rela-
tionship between study-time and rst-year college grades. Instrumenting study-time
with whether a students rst-year roommate brought a video game to school with
him, they are able to disentangle a causal effect of studying on grade performance.
A considerable amount of literature analyses how student effort reacts to different
types of educational inputs. Leuven et al. (2009), Kremer et al. (2009) and Angrist
et al. (2009) analyze, through experimental studies, the effects of nancial incentives
on student effort showing that students tend to provide more effort when they can
obtain a nancial reward. Figlio and Lucas (2003) showthat higher standards improve
performance and behavioural outcomes due both to greater student effort and greater
parent involvement. Bishop (1997, 2006) and Bishop and Mane (2001) point out that
centralised exit examinations significantly improve student performance. Similarly,
Wmann (2005) and Fuchs and Wmann (2007) using, respectively, the interna-
tional TIMSS and PISA micro data, estimate the effect of centralised exams, suggest-
ing that students in countries imposing a centralised examination at the end of upper
1
Fernandez (1998) and Gary-Bobo and Trannoy (2008) analyse why the allocation of students to schools
is not exclusively based on a price system, but also relies on the use of examinations. Fernandez (1998),
under the assumption of borrowing constraints, shows that exams dominate markets in terms of matching
efciency, while Gary-Bobo and Trannoy (2008) show that a mix of pricing and selection in examinations
is optimal when educational institutions observe a private signal of student ability.
123
A signalling model for school grades 203
secondary education outperformstudents in countries with no centralised examination.
Jrges et al. (2005) nd analogous results considering the German educational system.
However, inline withthe predictions of our theoretical model, some recent empirical
investigations showthat the effects of centralised exit examinations are heterogeneous
and vary according to student ability. Wmann (2005) nds that the positive effects
of centralised examinations on student performance are greater for high ability stu-
dents, while low ability students react less. Dee and Jacob (2006) nd that exit exams
increase student performance, but significantly reduce the probability of completing
high school, particularly for black students, and increase the dropout rate in urban and
high-poverty school districts. Similar results emerge from a study conducted by Dean
and De Cicca (2001).
3 School grades as a signalling device of student skills
In this section, we propose a theoretical model showing how rms use school grades
as a signal of worker skills and how students react to the market wage schedule in
terms of effort provided at school.
We assume that individuals are risk-neutral and live for two periods. They go to
school in the rst period, sustaining the cost of effort, and, in the second period, they
enter the labour market, obtaining a wage W. There is no discounting. Individuals are
identical in every respect except for their ability which is distributed according to a
probability density function with mean a and variance
2
a
.
Students attend school and attain an educational qualication. Schools also provide
an evaluation of student skills. This evaluation, denoted by v, is affected by student
effort e, ability a and by two different types of errors. One of these errors is related to
factors that may inuence student performance, indicated by , with mean zero and
variance
2

, and the other, indicated by , with mean zero and variance


2

, is related
to aspects concerning the evaluation systemadopted by teachers. The rst type of error
depends on whether students were lucky enough to have studied precisely the subject
relevant for the examination and on how they felt that day, while the second type of
error is related to elements which affect teacher evaluation, for example the time he
spends correcting the examination or whether he is more or less demanding. These
errors are not correlated. It is plausible to think that stochastic variables affecting stu-
dent behaviour are not related to those affecting teacher evaluations. Moreover, we
assume that errors are not correlated to an individuals ability or to his effort. Formally:
Cov (e, ) = Cov(e, ) = Cov(a, ) = Cov(a, ) = Cov(e, ) = Cov (, ) = 0.
For the sake of simplicity, we assume the following specic functional form for the
qualication obtained by individual i, v
i
:
v
i
= e
i
+ a
i
+
i
+
i
(1)
where effort and ability are perfectly substitutable.
The expected value of the qualication is equal to: E(v) = E(e

) + a, where
E(e

) represents the expected optimal level of effort in equilibrium (which, as we


showbelow, is uniformacross individuals) and its variance is Var(v) =
2
a
+
2

+
2

.
123
204 M. De Paola, V. Scoppa
The lifetime individual utility functionbased on the idea that individuals live
for two periods and that there is no discountingtakes the following simple form:
E(U
i
) = E(W
i
) c(e). In order to avoid cumbersome analytical expressions, we use
an explicit function given by: c(e) = e
2
_
2, to represent the cost of effort where
is a parameter measuring the disutility of effort.
3.1 Labour market
We assume that output y
i
produced by individual i in the labour market is related to his
skills, which in turn depend on his innate ability and on the effort provided at school.
In order to focus on the relationship between student effort and evaluation systems,
we neglect other important educational inputs, such as teacher effort and quality and
other educational resources.
Assuming substitutability between effort and ability, we suppose that skills are
equal to s
i
= e
i
+ a
i
. Output is then related to skills according to the following
production function:
y
i
= s
i
(2)
where is a productivity parameter.
In Sect. 3.4 we use an alternative skill function, in which effort and ability are
linked by a complementarity relationship, to show that our main results are robust to
the assumption of the functional form of skills.
With perfect information and perfectly competitive labour markets, wage W
obtained by individual i will be equal to his output: W
i
= (e
i
+ a
i
). However,
we assume that, in the initial stage of the workers career, rms are not able to observe
individual abilities or the effort they provided during the educational process, but only
the evaluation v
i
obtained by each student at school. In fact, it requires time for rms
to nd out about the workers effective abilities and only later on in the workers career
is it possible to pay him a wage based on his effective skills (Lange 2007; Lange and
Topel 2006). For the sake of simplicity, we do not explicitly model the stage at which
rms learn about the workers effective skills, but we focus on the initial stage and
assume that rms pay a wage based on the expected abilities of individuals who have
obtained a given educational qualication.
It is worthwhile noting that in our framework, differently from a pure signalling
model, student effort contributes both to improving the signal and to increasing indi-
vidual productivity.
Firms seek to understand workers effective skills of on the basis of the evaluation
v
i
. This is a typical signal extraction problem and rms form their expectations
regarding worker skills on the basis of school evaluations, i.e. E(s
i
|v
i
). An important
result in statistics is that when two variables are jointly normally distributed, E(s
i
|v
i
)
is a weighted sum of the unconditional mean E(s
i
) and the signal v
i
. Therefore,
E(s
i
|v
i
) =
0
+
1
v
i
, where the parameters
0
and
1
can be determined using the
standard OLS formulae. It follows that the two parameters
0
and
1
are given by the
following expressions:
123
A signalling model for school grades 205

1
=
Cov (s, v)
Var (v)
=
Cov (e + a, e + a + + )
Var (e + a + + )
(3)

0
= E (s)
1
E (v) (4)
Given our assumptions on variance and covariance of variables, it is possible to show
that Cov (s, v) = Var (a) =
2
a
. Therefore:

1
=

2
a

2
a
+
2

+
2

(5)

0
=
_
E
_
e

_
+ a
_
(1
1
) (6)
It follows that the wage paid by employers to each employee depends on the
expected skills according to the following function:
W
i
= E (e
i
+ a
i
|v
i
) = (
0
+
1
v
i
) (7)
This equation tells us how information on v
i
updates rms expectations of individual
skills and, therefore, how v
i
affects wages.
2
3.2 Student behaviour
We assume that students are perfectly informed about their own abilities and that they
get utility from the evaluation they obtain from school in terms of future gains on the
labour market.
The expected utility of a student i with ability a
i
, who provides a level of effort e
i
and obtains an evaluation v
i
is equal to:
E (U
i
|a
i
) = E (W
i
|a
i
) c (e
i
) = [
0
+
1
E (v
i
)]
e
2
i
2
= [
0
+
1
(a
i
+ e
i
)]
e
2
i
2
(8)
Students decide upon the level of effort which will maximise their utility function,
taking as given their own ability and how the market rewards skills. By maximis-
ing E(U
i
|a
i
) with respect to effort, we obtain the following rst order condition:

1
e
i
= 0, from which the optimal effort is:
e

1
(9)
The level of effort provided by each student depends on the parameter represent-
ing labour productivity, on the cost of effort and on the reactivity of wages to school
2
Empirical evidence of the signalling value of school grades and its inuence on wages paid by rms is
provided by Schwerdt and Wmann (2009).
123
206 M. De Paola, V. Scoppa
evaluation,
1
. Since these factors are the same for all individuals, it is clear that they
will provide an identical level of effort.
3
Substituting Eqs. (5) into (9), we obtain:
e

_

2
a

2
a
+
2

+
2

_
(10)
The optimal level of effort is higher when the variances of and are lower (the
evaluation system is less affected by stochastic variables) since employers are willing
to pay a higher wage premiumon grades obtained at school as a result of their receiving
a better signal of students abilities. When evaluation is not affected by errors, that is
(
2

+
2

) 0, then
1
1 and the optimal effort is equal to the rst best level,
e

.
On the other hand, if measurement errors increase to innity, that is,
2

or
2

tend
to , student effort decreases to zero, because the evaluation is less informative of
individual skills and rms pay a wage based on average abilities, W = a.
Moreover, if the variance in abilities is higher, then
1
increases, positively affect-
ing the effort provided by students. This is due to the fact that school evaluation is
more important when variability in abilities is higher. Intuitively, when the variance
of individual abilities tends to zero, the signalling role of schooling is less relevant
since the average ability is a good signal of students effective skills. On the other
hand, when the variance of individual abilities increases, the average ability is not
more a reliable signal of effective skills and there is room for the provision of useful
information by schools.
3.3 The effects of accuracy in the evaluation system on student welfare
In this section we evaluate whether student welfare improves when the evaluation
scheme adopted by the educational system becomes more precise.
Considering the optimal level of effort e


1
, student expected utility, given
by Eq. (8), can be written as:
E (U
i
|a
i
) =
_

0
+
1
_
a
i
+

1

__

(
1
)
2
2
(11)
Substituting
0
into (11), we obtain the following:
E (U
i
|a
i
) =
_
a +
1
_
a
i
a +

__

(
1
)
2
2
(12)
3
In Sect. 3.4, we show that when effort and ability are complements, the effort provided by individuals
depends on their ability. When the evaluation system becomes more precise, effort increases especially for
high ability subjects.
123
A signalling model for school grades 207
We are now able to analyse the effect produced by an increase in the variance of error
terms and , respectively
2

and
2

, on student utility. Let us denote the variance of


errors and with
2
=
2

+
2

. By differentiating the individual expected utility


with respect to
2
, we obtain:
E (U)

2
=
_

2
(a
i
a)
_
+
_

2
__
1
1
2

1
_
+
_

1
__

1
2

2
_
(13)
After some rearrangement, Eq. (13) becomes:
E (U)

2
=

1

2
_
(a
i
a) +
_

_
(1
1
)
_
(14)
It is easy to see that individual utility diminishes when
2
increases, when:
a
i
> a = a
_

_
(1
1
) (15)
since

1

2
< 0.
It follows that students whose ability is above the threshold level a are negatively
affected by less accurate evaluation systems, while students with abilities below this
threshold are positively affected. In fact, a wage system based on average skills, redis-
tributing resources from highly skilled individuals to low skilled ones, tends to favour
the latter category.
The threshold value a decreases when the disutility of effort becomes lower. More
interestingly, the threshold decreases when productivity increases, implying that, in
highly productive economic systems, the individuals with relatively low abilities also
prefer more accurate evaluation. In fact, less accurate evaluation systems have two
effects. On the one hand, they lead to a more egalitarian pay structure, while, on the
other hand, they reduce effort and, as a consequence, reduce the total output produced
in the economy. When takes a high value, the wage reduction deriving fromthe lower
level of effort tends to counterbalance the positive effect that low ability individuals
obtain from the income redistribution deriving from a low value of
1
.
One implication of our analysis is that if the accuracy of evaluations is decided upon
by aggregating individual preferences through a voting system, a political majority
opposing precise evaluations might emerge. This would happen when the ability dis-
tribution in the population is characterised by a median ability which is lower than the
threshold ability level a. This voting equilibrium is less likely to arise in economies
with high levels of productivity.
123
208 M. De Paola, V. Scoppa
3.4 A more general skill function assuming a complementarity relationship between
ability and effort
In this section, we provide a framework in which effort and abilities are complemen-
tary in the determination of skills, with the aim of showing that students of different
abilities react differently to an increase in the precision of the evaluation system.
More specifically, we assume that skills acquired by students at school are given by
s = ae. Similarly, school evaluations are: v = ae + + .
As shown above, rms set wages according to a level of worker skills inferred
from school evaluation. As a consequence, rms pay a wage W = E(e + a|v) =
(

0
+

1
v). Given the skill acquisition function,

1
and

0
are equal to:

1
=
Cov (s, v)
Var (v)
=
Cov (ea, ea + + )
Var (ea + + )

0
=
_
e

a
_
_
1

1
_
Taking into account the wage rms are willing to pay for the qualication v
i
, the
student expected utility is given by:
E (U
i
|a
i
) =
_

0
+

1
E (v
i
)
_

e
2
i
2
=
_
e a
_
1

1
_
+

1
(a
i
e
i
)
_

e
2
i
2
(16)
Maximising with respect to effort, we obtain the following rst order condition:
E (U
i
|a
i
)
e
i
=
_

1
a
i
_
e
i
= 0 (17)
Denote the marginal utility of effort as function F (e, ):
F =
E (U
i
|a
i
)
e
i
=

1
(e) a
i
e
i
= 0 (18)
We use the implicit-function theorem to analyse the effect on student effort produced
by an increase in the variance of a, and . We obtain:
e
i

j
=
F
_

j
F
_
e
i
=
a
i

1

j
F
_
e
i
where j = a, , (19)
Since the denominator is always positive when SOC is respected, it emerges from
(19) that student effort diminishes when

and

increase. This result is in line with


that discussed in Sect. 3.3. However, it is possible to see from (19) that more accurate
evaluation systems have heterogeneous effects on student performance. For lowability
123
A signalling model for school grades 209
students, the incentive effect of more precise evaluation systems is relatively small. In
fact, the effects of accurate evaluation systems increase with student ability.
In the next section, we will turn to the simpler skill function s = e +a to compare
centralised and decentralised evaluation systems.
4 A comparison between centralised and decentralised evaluation systems
In this section, we compare centralised and decentralised evaluation systems by refer-
ring to the advantages deriving fromrepeated observations and the costs due to hetero-
geneous assessment methods. While decentralised evaluations, carried out by teachers
during their activity, often represents a by-product of teaching, as it derives from the
continuous interaction with students, centralised exams are less frequent since they
require a more complex organisation.
During the educational process, teachers and students interact on a daily basis and
teachers have rst hand knowledge regarding students psychological and physical
conditions. Both these facts help in reducing the effect of stochastic shocks effecting
students on the evaluations they obtain when the examination systemis based on grades
awarded at local level. On the other hand, teachers may adopt different measurement
criteria and this introduces noise into student evaluation, making it less convenient
for employers to base their pay systems on observed educational performance. Con-
versely, central exams adopt a common standard and provide a better signal. However,
they are undertaken less frequently and shocks affecting student performance may play
a more relevant role in shaping examination results.
To analyse this kind of trade-off, we assume that, with a centralised evaluation sys-
tem, students are evaluated according to a common standard and, as a consequence,
the error term deriving from heterogeneous evaluation methods is not relevant, so
implying that
2

= 0. On the other hand, we assume that this type of examination is


only undertaken at the end of the educational process and shocks affecting students
may inuence their performance at exams. It follows that the variance in evaluations
awarded by the centralised system is equal to Var(v) =
2
a
+
2

.
4
When evaluation is at a decentralised level, delegated to teachers, it is possible to
evaluate student performance on a large number of occasions, which we denote with
n. Therefore, effects deriving from stochastic variables which inuence student per-
formance are reduced, and the variance of this type of error is equal to
2

_
n. On the
other hand, since different teachers adopt different evaluation methods, the variance
in evaluations is equal to: Var (v) =
2
a
+

2

n
+
2

.
It follows that the individual i

s expected utility under each examination system is


equal to:
4
Although typically central exit exams are used jointly with decentralised examinations, for the sake of
simplicity we assume that students just undertake a central exam at the end of schooling. Our assumption
does not lead to results that are substantially different from those that we would have obtained assuming
that student evaluation is a weighted average of the results obtained at the central exam and at decentralised
examinations.
123
210 M. De Paola, V. Scoppa
E
_
U
K
i
|a
i
_
=
_
a +
K
1
_
a
i
a +

__

_

K
1

_
2
2
(20)
where K = C under the centralised evaluation system and K = D under the decen-
tralised one.
The centralised systemproduces a higher utility compared to decentralisation when
E(U
C
i
|a
i
) > E(U
D
i
|a
i
), which corresponds to the following:

_
a +
C
1
_
a
i
a +

__

_

C
1

_
2
2
>
_
a +
D
1
_
a
i
a +

__

_

D
1

_
2
2
(21)
By simplifying some terms, we obtain the following condition:
_

C
1

D
1
_
__
a
i
a +

_


2
_

C
1
+
D
1
_
_
> 0 (22)
For individuals of average ability a
i
= a, the previous equation becomes:
_

C
1

D
1
_

_
1

C
1
+
D
1
2
_
> 0 (23)
Since
C
1
< 1 and
D
1
< 1, the termin square brackets is always positive. Therefore, a
centralised evaluation systemis preferable when
C
1
>
D
1
, that is when

2

n
+
2

>
2

(the variance of the decentralised system is greater than the variance of centralisation)
and gains obtained by reducing the variance of do not compensate for the losses
deriving from adding into the evaluation the variance of error term , typical of the
decentralised system (
2

>
2

n
).
It is interesting to note that the difference
2

n
increases when
2

increases,
implying that the advantage of decentralised systems is greater when shocks affecting
student performance have a higher variance.
For individuals whose abilities are above average ability, a
i
> a, the same result
holds. On the other hand, the term in square brackets in expression (23) may be neg-
ative for low ability individuals and, in this case, individuals may prefer evaluation
systems with a higher variance.
It is worthwhile noting that the number of evaluations undertaken under the decen-
tralised examination system may be inuenced by class size. When teachers face very
large classes, it may be difcult for them to judge students on the basis of daily inter-
action, participation in classwork, etc. As a consequence, when class size increases,
since the cost of evaluating students increases, teachers may decide to award grades
to students on the basis of a lower number of evaluations, rendering these grades
less precise. This interaction between class size and measurement errors, arising in
the decentralised, but not in the centralised, system, should also be considered when
comparing the two.
123
A signalling model for school grades 211
5 Concluding remarks
In this paper, we analyse a labour market with imperfect observability of worker skills
in which rms use grades obtained by students during their educational careers in order
to infer their productivity and, hence, determine their wages. Therefore, school grades
are used by rms as a signal of abilities. We assume that rms form expectations of
individual abilities by solving a signal extraction problem.
In this framework, we study the effects of different evaluation systemscharacter-
ised by different measurement errorson the reliability of this signal and, therefore,
on the relationship between wages and skills at an individual level. We, then, analyse
the effect that more precise evaluation systems have on student effort in their studying
activities and on their welfare. We show that systems with lower measurement errors
encourage students to provide a higher level of effort in studying activities. Employers,
in receiving a more precise signal of student skills, are willing to pay a higher wage
premium on the grade attained at school and, as a result, students are induced to make
more effort.
Stimulated by growing theoretical and empirical literature which considers central-
ised examinations as an instrument to improve students performance and educational
quality (Bishop and Wmann 2004; Wmann 2005), we use our framework to
compare costs and benets of centralised vs. decentralised evaluations. From the per-
spective adopted in this paper, the advantage of a decentralised system is that it allows
repeated evaluation of student performance and, hence, is able to reduce measurement
errors due to shocks effecting students. On the other hand, a centralised system, refer-
ring to a common standard, avoids errors deriving from different evaluation standards
that typically characterise locally graded exams. We showthat the advantage of decen-
tralised systems is greater when shocks affecting student performance have a higher
variance and teachers adopt similar grading standards.
An important result of our paper is the heterogeneous impact of the accuracy of
evaluation on individuals of different abilities. Whereas high ability individuals typi-
cally prefer more precise evaluation systems, lowability individuals may oppose them.
We also show, though, that individuals with relatively low abilities may prefer more
accurate evaluation systems when labour productivity increases. In fact, measurement
errors, on the one hand, lead to a more egalitarian pay structure, improving welfare
obtainedbyless able individuals, but, onthe other hand, theyreduce effort and, as a con-
sequence, reduce the total output produced in the economy. When labour productivity
is high, the wage reductionderivingfroma lower level of effort might outweighthe pos-
itive effect that low ability individuals obtain from a more egalitarian wage structure.
These results pose some important equity concerns. Policies aimed at the improv-
ing of the signalling value of school grades may increase wage inequality and reduce
the welfare of low ability students who come disproportionately from economically
disadvantaged groups. To reduce these negative effects, it might be useful to accom-
pany these policies with interventions supporting the accumulation of human capital
of weaker students.
Acknowledgments We would like to thank for useful comments three anonymous referees, Giorgio
Brunello, Maurizio Franzini, Anna Giunta, Giovanni Mastrobuoni, Eleonora Patacchini and the participants
123
212 M. De Paola, V. Scoppa
at the VI Workshop BRUCCHI LUCHINO (Salerno, December 2007) and at the Economics of Education
Conference, Universitt Zrich, June 2008.
References
Angrist J, Lang D, Oreopoulos P (2009) Incentives and services for college achievement: evidence from a
randomized trial. Am Econ J Appl Econ 1:11361163
Betts J (1998) The impact of educational standards on the level and distribution of earnings. Am Econ Rev
88:266275
Bishop J (1996) The impact of curriculum-based external examinations on school priorities and student
learning. Int J Educ Res 23(8):653752
Bishop J (1997) The effect of national standards and curriculum based exams on achievement. Am Econ
Rev 87(2):260264
Bishop J (2006) Drinking fromthe fountain of knowledge: student incentive to study and learn-externalities,
information problems and peer pressure. In: Handbook of the economics of education. North Holland,
New York
Bishop J, Mane F (2001) The impacts of minimum competency exam graduation requirements on high
school graduation, college attendance and early labor market success. Labour Econ 8:203222
Bishop J, Wmann L (2004) Institutional effects in a simple model of educational production. Educ Econ
12(1):1738
Bonesrnning H (2004) Do the teachers grading practices affect student achievement? Educ Econ
12(2):151167
Costrell R (1994) A simple model of educational standards. Am Econ Rev 84(4):956971
Dean L, De Cicca P (2001) Higher standards, more dropouts? across time. Econ Educ Rev 20:459473
Dee T, Jacob B (2006) Do high school exit exams inuence educational attainment or labor market perfor-
mance? NBER Working Paper, N. W12199
De Fraja G, Landeras P (2006) Could do better: the effectiveness of incentives and competition in schools.
J Public Econ 90:189213
De Paola M, Scoppa V (2007) Returns to skills, incentives to study and optimal educational standards.
J Econ 134
Fernandez R (1998) Education and borrowing constraints. NBER Working Paper, p 6588
FiglioD, Lucas M (2003) Dohighgradingstandards affect student performance? J Public Econ 88(9):1815
1834
Fuchs T, Wmann L (2007) What accounts for international differences in student performance?
A re-examination using PISA data. Empir Econ 32:433464
Gary-Bobo R, Trannoy A (2008) Efcient tuition fees and examinations. J Eur Econ Assoc 6(6):12111243
Glewwe P, Kremer M (2006) Schools, teachers, and educational outcomes in developing countries.
In: Hanushek EA, Welch F(eds) Handbook of the economics of education. North Holland, Amsterdam
pp 9431017
Jrges H, Bchel F, Schneider K (2005) The effect of central exit examinations on student achievement:
quasi-experimental evidence from TIMSS Germany. J Eur Econ Assoc 3(5):11341155
Kremer M, Miguel E, Thornton R (2009) Incentives to learn. Rev Econ Stat 91(3):437456
Lange F (2007) The speed of employer learning. J Labor Econ 25(1):135
Lange F, Topel R (2006) The social value of education and human capital. Handbook of the Economics of
Education
Lavy V(2004) Performance pay and teachers effort, productivity and grading ethics. NBERWorking Paper
No. 10622
Lazear E (2001) Educational production. Q J Econ 777808
Lazear E (2006) Speeding, terrorism, and teaching to the test. Q J Econ 121:10291061
Leuven E, Oosterbeek H, van der Klaauw B (2009) The effect of nancial rewards on students achieve-
ments: evidence from a randomized experiment. J Eur Econ Assoc (forthcoming)
Schwerdt G, Wmann L (2009) The signaling value of central school exams. Mimeo
Stinebrickner R, Stinebrickner T (2004) Time-use and college outcomes. J Econom 121(12):243269
Stinebrickner R, Stinebrickner T (2008) The causal effect of studying on academic performance. BE J Econ
Anal Policy 8(1) (Frontiers)
Wmann L (2005) The effect heterogeneity of central exams: evidence from TIMSS, TIMSS-repeat and
PISA. Educ Econ 13(2):143169
123

Вам также может понравиться