Center for Global Development

January 15, 2008

Draft 1.0

Measuring progress with tests of learning:

Pros and Cons for “Cash on Delivery Aid” in Education1

Marlaine E. Lockheed
Center for Global Development

This paper reviews, in non-technical terms, the case for and against using tests of learning
for measuring annual educational progress within programs of “Cash on Delivery aid.” It
examines the evidence supporting the first of two main assumptions behind Cash on
Delivery aid in education -- that progress in learning can be measured validly and reliably
–by examining three questions: whether valid and reliable measures of student learning
are currently available in developing countries, whether existing tests are capable of
registering the changes in educational results called for under “Cash on Delivery aid,”
and whether developing countries have the administrative capacity to undertake annual
assessments of learning.

The paper does not address a second key assumption, that developing countries have the
technical capacity to effect improvement in their education systems, if progress
incentives are in place. The evidence regarding the impact of progress incentives, such as
are found in the No Child Left Behind (NCLB) Act of 2002, is mixed. Earlier studies
tended to find no impact of performance incentives on increased student learning (e.g.
Amrein and Berliner 2002) while more recent analyses find a positive effect (Braun 2004,
Carnoy and Loeb 2003, Hanushek and Raymond 2005, Rosenshine 2003). But several
analysts have questioned whether student learning outcomes can improve within a reform
framework that lacks the material and pedagogical support required by the new
curriculum (Goertz and Duffey 2003; Kelley, Odden, Milanowski and Heneman 2000).
And one study found accountability less cost-effective than other approaches for boosting
achievement (Yeh 2007).

The paper is organized as follows: the next section provides a brief description of existing
testing activities in developing and transition countries. Section 2 outlines the essential
requirements for a measure of learning “progress” at the national level and compares
these requirements with what is known about the measures typically provided by

1
Thanks to Carol Owen (Educational Testing Service) and Abigail Harris (Fordham University) for their
constructive comments on an early version of this paper. Support for this paper was provided by the Center
for Global Development. Any errors of fact or interpretation are solely those of the author. Comments are
welcome.
 Measuring Progress with Tests of Learning

national, regional and international learning assessments, as carried out in developing and
transition countries. Section 3 discusses some technical topics in greater detail. The final
section discusses testing costs and options for the use of learning assessments in the
context of “Cash on Delivery aid.”

1. Building capacity for measuring learning outcomes

Nearly two decades ago, the first recommendation of the World Bank’s 1990 Primary
Education Policy Paper called for education systems to:

“Emphasize learning. Developing countries need to increase the

number of children who acquire the skills specified in their nation’s
curriculum and who successfully complete the primary cycle. To
this end, countries must emphasize students’ learning as the key
policy objective.” (World Bank 1990: 54).

Regrettably, back in 1990, few developing countries had the capacity to measure student
learning, so the donor community undertook to build that capacity through loans and
grants. At the World Bank, for example, the number of education projects providing
support for national learning assessments increased from no projects funded before 1988
to 27 percent of projects funded in 1991, 70 percent funded during 1990-1994 and around
60 percent of projects funded since 1995 (Larach and Lockheed 1992; Nielsen 2006).
Other multilateral donors, including UNESCO and IDB, and bilateral donors, including
USAID and CONFEMEN, have supported capacity building for student assessments.
These efforts have met with some success, and the context for a heightened interest in
“results based aid” where learning is an explicit indicator of results has greatly changed.

This change is clearest at the country level. Whereas in 1990, only a handful of
developing countries regularly carried out national learning assessments at the primary
level 2 , none had participated in regional learning assessments at the primary level, and
fewer than a half dozen had participated in any international learning assessment, in 2008
a variety of assessment systems flourish in developing countries. Increasing shares of
developing and transition countries 3 are implementing national learning assessments. The
percentages of developing and transition countries carrying out at least one national
learning assessment have risen dramatically: from 28 percent of developing and 0 percent
of transition countries in 1995-1999, to 51 percent of developing and 17 percent of
transition countries in 2000-2006 (Benevot and Tanner 2007).

2
In 1988, Chile established SIMCE, which assessed math, Spanish and writing at Grade 4; in 1990,
Colombia established a sample-based assessment for Grades 3 and 5 in math and Spanish; and in 1985,
Thailand established a Grade 6 assessment in math, Thai language and science (Murphy and others 1996).
Many countries, often former British colonies, had well-established systems of selection and certification
examinations, but these were “high stakes” tests designed as gatekeepers for further education, rather
assessments of learning. For example, in 1950, Jamaica instituted an 11+ (Common Entrance Examination)
for selection into lower secondary schools.
3
Developing and transition countries include those countries designated as low and middle-income
countries by the World Bank.

2
Lockheed

In addition, many developing and transition countries are participating in international

learning assessments; 17 developing countries participated in the 2006 Progress in
Reading Literacy Study (PIRLS 2006), 26 in the Programme for International Student
Assessment (PISA 2006) and 37 in the 2007 Trends in Mathematics and Science Study
(TIMSS 2007). Furthermore, regional learning assessments are ongoing in several
regions: 15 countries in Southern Africa participate in the Southern African Consortium
for Monitoring Education Quality (SACMEQ III), 22 countries in Francophone Africa
participate in the Programme d’Analyse des Systemes Educatifs des Pays de la
CONFEMEN (PASEC) and 16 countries in Latin America participate in the Laboratoria
Latinoamericano de Evaluacion de la Calida de la Education (LLECE). Can such
national, regional or international learning assessments provide suitable indicators of
progress, for a program of “Cash on Delivery aid” at the primary level? Or, as some
argue, should the international community prepare measurement instruments for this
purpose (Annual Status of Education Report 2006, RTI International 2007). This paper
will address this question.

2. Measuring learning progress

“If you want to measure change, don’t change the measure”. This observation, attributed
to Otis Dudley Duncan in 1969 and applied to national assessments of learning by Albert
Beaton in 1988, is central to any discussion of measuring progress. Measuring learning
progress (that is, positive change) requires measurement instruments that are stable over
time, in at least six ways:
• Testing the same cohorts (e.g. age cohort) for T1 and T2
• Measuring the same academic content or competencies at T1 and T2
• If sampling is used, using the same sampling procedures for T1 and T2
• Using measurement instruments having the same levels of difficulty at T1 and T2
• Using measurement instruments having the same reliability at T1 and T2
• Confirming the equivalence of the tests through empirical equating (Linn 2005)
In addition, to measure progress annually, a country must have the capacity to construct
psychometrically valid and empirically equated tests, and to administer, score and report
test results on an annual basis.

Achieving stability in learning assessments is generally carried out through the process of
standardization. As Braun and Kanjii (2006: 317) note: “Standardization is a prerequisite
for fairness when scores must be comparable. It demands, at a minimum, that the tests be
administered under uniform conditions and graded according to a fixed set of rules or
rubrics.” Few developing countries can ensure that this is the case, and lack of
experience leads to underestimation of the complexity of the processes for designing,
developing, administering, scoring, analyzing and reporting the results of standardized
tests (Braun and Kanjii 2006).

Much has been written about the difficulty of measuring change. Ragosa (1995)
identified nine widely-held myths about measuring change, and concluded that such

3
 Measuring Progress with Tests of Learning

measurement is possible, if individual growth curves are taken as the starting point.
However, Braun and Bridgeman are skeptical, observing that models of individual
growth

“assume that the psychometric meaning of test scores stays stable

over time, which is a dubious proposition at best. Physical growth
may be easily measured because exactly the same ruler can be used
to measure the height of a person at 2 years old and then at 25 years
old, but the same test could not be used to measure the reasoning or
reading skills of the child and the adult.” (Braun and Bridgeman
2005: 4).

Rock, in discussing “the uncritical use of gain scores and their interpretation” and arguing
for the use of adaptive testing, observes that:
“Gain scores that are not based on an adaptive testing approach are
likely to give erroneous results because of floor and ceiling effects.
Tests that are too hard (floor effects) or too easy (ceiling effects)
will be unreliable in both tails of the score distribution and are likely
to underestimate the amount of gains for those children in both tails”
(Rock 2006:1).

In the United States, state-based accountability systems, in particular, have been

questioned for their capacity to evaluate annual growth in student performance
(Martineau 2006, Way 2006). One paper notes that volatility of test score measures “can
wreak havoc in school accountability systems.” (Kane and Staiger 2002: 236 as quoted in
Way 2006), although others debate this conclusion (Ragosa 2002).

Alternatives to individual growth monitoring include:

• status models (that describe the status of a school or educational system and
report average scores or the share of students that have reached a particular
performance standard rather than reporting change),
• successive groups models (that compare the performance of the same grade of
students across consecutive years, e.g. grade 4 students in 2007 with the
performance of grade 4 students in 2006; differences in cohort composition need
to be taken into account), and
• longitudinal models (that follow a particular cohort as they age or progress
through the grades).

Longitudinal models place a huge burden on education systems, which are plagued by
three main technical limitations (Way 2006): (a) vertical scaling of tests 4 , (b) that all
children be followed, and (c) an infrastructure and data capable of tracking students
longitudinally. Student dropout and mobility and the lack of a suitable infrastructure lead
to serious missing data problems. Because of difficulties in implementing longitudinal
assessments, in the U.S., implementation of state-based accountability systems has
4
Vertical scaling simply means that a common scale exists against which student performance at
increasingly older ages or higher grades can be measured.

4
Lockheed

generally relied on the “successive groups” approach for measuring year to year change,
despite the technical limitations associated with measuring different cohorts of students
(Marion et al 2002 as cited in Way 2006). Successive group models are also used for
most national, regional and international learning assessments, and can provide
information regarding differential learning outcomes among subgroups of students (such
as gender, ethnicity, socio-economic status or geographical location). Successive group
models require, at a minimum, that the tests be horizontally equated from one year to the
next. 5

What do we know about the tests used in learning assessments in developing countries?
Are they aligned with respect to the six dimensions of stability noted above? How do the
tests used in national, regional and international assessments differ? Regrettably, little
public information is available about the technical characteristics of national learning
assessments. Some information is provided in the Statistical Annex of the 2008
UNESCO Global Monitoring Report, the single most comprehensive listing of countries
with national learning assessments (UNESCO 2007). It presents information regarding
117 countries’ national learning assessments, including the target populations, the
academic content/ competencies assessed and the regularity of the assessments. Detailed
information about the tests used and the degree to which they are equated from
administration to administration, however, is not provided in this report. Another recent
report on testing in Latin America, specifically, is also relatively silent about the
technical dimensions of the tests and notes that “Technical validation of test items or
questions is a critical element in developing assessment instruments; unfortunately,
national technical reporting on this subject is not particularly detailed” (Ferrer 2006: 28).

Table 1. Major regional and international learning assessments

Countries
Assessment (most Target Years
(Sponsor) recent) population Content tested Frequency implemented
TIMSS 37 Grade 4, 8 Math, Science 4 year cycle 1995, 1999,
(IEA) 2003, 2007
PIRLS 17 Grade 4 Reading 5 year cycle 2001, 2006
(IEA)
PISA 26 15-year olds Math, Science, 3 year cycle 2000, 2003,
(OECD) Reading 2006
SACMEQ 15 Grade 6 Math, Language variable 1995-97,
(IIEP and African 2000-2002,
ministers of education 2007
LLECE 16 Grade 3,6 Math, Reading, 10 year cycle 1997, 2007
(OREALC/UNESCO) Writing, Science*
PASEC 22** Grades 2,5 Math, French, variable 1993-95,
(CONFEMEN) National Language 1997-2001,
2003-06

5
That is, although the test takers from one year to the next are the same age or grade, they are actually
different individuals who take different versions of the test; these versions need to be empirically equated
for the scores to have the same meaning from one year to the next.

5
 Measuring Progress with Tests of Learning

*optional, **member countries; assessments in 1-4 countries per year; 9 countries have participated in one
assessment, 6 countries have participated in 2 assessments and 2 countries have participated in 3
assessments; in all 17 of the 22 member countries have participated in an assessment.

The major regional assessments are Laboratoria Latinoamericano de Evaluacion de la

Calidad de la Education (LLECE) in Latin America, the Southern African Consortium for
the Measurement of Education Quality (SACMEQ) in Southern Africa, and Programme
d’Analyse des Systemes Educatifs des Pays de la CONFEMEN (PASEC) in Francophone
Africa. The major international assessments are the Trends in International
Mathematics and Science Study (TIMSS), Progress in Reading Literacy Study (PIRLS)
and Programme for International Student Assessment (PISA). Websites of regional and
international learning assessments provide the most up-to-date information about these
exercises (table 1). Such regional and international assessments utilize standardized tests
with high degrees of reliability achieved through modern psychometric methods.

Criteria that National Learning Assessments often meet

Three of the criteria mentioned above appear to be met by national learning assessments:
stability of target populations, stability of content, and administrative capacity. All are
met by regional and international learning assessments. One other criterion, stability in
sampling methods, is rarely applicable to national learning assessments, as most test
entire grade cohorts.

Stability of target populations. Most countries target the same age or grade for testing,
from one year to the next, within a narrow range of primary grades. 6 All countries in
Sub-Saharan Africa, East Asia and the Pacific, Latin America and the Caribbean and
Central Asia that carried out any national learning assessment, 2000-2006, tested children
in enrolled in at least one of grades 4-6 and many countries also tested children in grades
8 or 9 (Benevot 2007). Many countries also conduct annual assessments, but alternate
among grades to be tested. For example, Chile tested 4th grade students in 2002, 2nd grade
students in 2003 and 8th grade students in 2004 (Ferrer 2006). By comparison, regional
and international assessments typically test students in the same grades or age cohorts
consistently from one assessment to the next.

Content stability. Most countries with national learning assessments report stability in
terms of test content. This means that the tests cover the same general curricular content
areas, typically literacy and numeracy, from one assessment to the next. Over 90 percent
of countries with national learning assessments test mathematics, reading and writing,
and about 50 percent of countries test science, with some regional variations (Benavot
and Tanner 2007). However, there is no guarantee that the tests cover exactly the same
content from one year to the next, that test questions covering the same content are
comparable with respect to difficulty from one year to the next, or that the curriculum
includes “clear (or even operational) definitions of what students are expected to be able

6
Age cohorts are more comparable over time than are grade cohorts, since variations in grade cohorts can
result from demographic changes in grade enrollments, and differences in repetition and dropout rates from
one year to the next.

6
Lockheed

to do with the conceptual knowledge contained in the curricula” (Ferrer 2006:20). Thus,
although the content may appear to be stable, it may also change dramatically. Change
can come from the process of annual test development, the absence of curricula content
and performance standards to guide test development, and curriculum reforms. Ferrer
(2006) notes that Colombia, Ecuador and Uruguay have made efforts to specify curricular
standards and link them with assessments, but these are exceptions in Latin America. 7

Regional and international assessments also measure student learning in reading, writing
mathematics and science, but the tests are typically constructed (and empirically equated)
so as to be stable over time. TIMSS and PIRLS also are constructed to reflect the
academic curricula of the participating countries. International reports from TIMSS,
PIRLS and PISA already include sections that document change in achievement over
time for participating countries.

Administrative capacity. Few countries have the administrative capacity to undertake

annual assessments of students at the primary level. UNESCO’s 2008 EFA Global
Monitoring Report identifies 20 developing or transition countries that report carrying out
annual national learning assessments (table 2), and 10 others that report having
implemented national learning assessments for two consecutive years over the period
2005-2007. 8 These numbers may be overstated, however. In reporting out the countries
in Western Europe and North America that have annual national learning assessments,
UNESCO classifies United States’ National Assessment of Education Progress (NAEP)
as an annual assessment, although it actually takes place biannually (NAEP website).
Regional and international learning assessments typically take place at intervals of 3 or
more years, reflecting the greater technical demands of such studies.

Table 2. Low and middle-income countries reporting yearly national learning

assessments for a minimum of three consecutive years since 2000

Region Name of Countries

Sub-Saharan Africa Gambia, Seychelles
Arab states Jordan
East Asia and the Pacific Indonesia, Malaysia, Thailand
Latin America and the Caribbean Anguilla, Bahamas, Belize, Brazil, Chile, Guyana,
Jamaica, Mexico
Central and Eastern Europe and Central Asia Estonia, Hungary, Mongolia, Poland, Turkey
Source: UNESCO 2007

Sampling. Most national learning assessments do not utilize scientific sampling methods
for selecting schools or children for assessment. Rather, entire cohorts or populations of
students are assessed, a practice that reduces technical complexity regarding sampling
and weighting of results while increasing administrative complexity and cost. Among
countries in Latin America, those that utilized sample-based assessments in the 1990s
have generally shifted to the use of national censuses in the 2000s. Regional and

7
Jamaica, not reviewed by Ferrer, also has a long-standing effort in this regard.
8
Malawi, Madagascar, Uganda, Egypt, Mauritania, Myanmar, Pakistan, Philippines, El Salvador, Albania

7
 Measuring Progress with Tests of Learning

international assessments, by comparison, utilize scientific samples, although the actual

sampling is done by international experts rather than by experts within the country.

Criteria that National Learning Assessments rarely meet

Criteria that ensure stability of measurement instruments over time are rarely met in
national learning assessments. Constructing instruments capable of measuring change
over time is technically complex 9 . National learning assessments in developing or
transition countries rarely employ such complex measurement instruments because such
countries rarely have the requisite domestic capacity or can afford to purchase expertise
from abroad. 10 Even in countries where the capacity exists, the results of the assessment
may not be able to be used effectively. For example, with respect to the National System
of Basic Education Evaluation (SAEB) in Brazil, Wilcox and Ryder (2002:217, citing
Crespo and others 2000) observe that “the SAEB, though a world-class performance
measurement, has failed to fulfill its potential,” largely due to the absence of technically
qualified personnel to interpret results, and little capacity to communicate the results to
policy-makers and the general public. By comparison, recent regional and international
learning assessments, which typically employ technical expertise from OECD countries,
use measurement instruments that often are explicitly designed to measure change and
provide interpretive materials aimed at the policy makers.

Three areas where national learning assessments are likely to fall short are in terms test
stability in difficulty, reliability and comparability over time.

Stability of difficulty. Available studies of national learning assessments do not discuss

technical issues of test construction, including how levels of difficulty are held constant
year to year. However, some evidence from developing countries suggests that even
relatively sophisticated national assessment units construct tests with “volatile” levels of
difficulty. For example, average mathematics and language scores of grade 6 students in
Jamaica showed remarkable variation, 1999-2004. 11 The main explanation for this
variation was that the test was not stable with respect to difficulty, and the test developers
had not designed the test in such a way that differences in difficulty could be empirically
adjusted post hoc (Lockheed and others 2005). By comparison, all three major
international assessments develop tests that are designed to use modern test theory and
psychometric methods to ensure stability with respect to difficulty.

Reliability. National learning assessments rarely report information about the reliability
of the tests, whereas regional and international assessments often include such
information in technical manuals accompanying the assessment. Classical methods of test
construction typically report internal consistency reliability (Cronbach’s alpha) or test-
9
Educational Measurement, 4th Edition, (Brennan 2006) discusses classical and modern techniques for
establishing test stability over time. Topics include validity, reliability, item response theory (IRT), test
bias, scaling, norming and equating.
10
There are exceptions; Qatar, for example, employs Educational Testing Service to conduct its national
assessments of learning
11
For a test of written communication graded on a scale of 1-6, average scores were significantly different
across the four years: 5.0 in 2001, 3.7 in 2002, 4.5 in 2003 and 3.0 in 2004.

8
Lockheed

retest reliability, while modern approaches emphasize Item Response Theory (IRT) for
establishing tests that measure the same constructs reliably. For example, PIRLS 2001
reports the classical Cronbach’s alpha reliability scores for each country in the
assessment, with a high median overall reliability of .88 (Mullis, Martin, Gonzalez and
Kennedy 2003). Reliability is not reported for tests equated through IRT, which provides
other, more sophisticated indicators of reliability.

Empirical equating. In order for tests to have the same meaning from one administration
to the next, they must be equated (see Braun and Holland 1982 for a discussion of test
equating) and tests must be designed in advance for this purpose, using calibrated items.
Applying the same exact test from one time to the next does not guarantee their
equivalence, as individual questions may change in their degree of difficulty from one
year to the next. 12 Psychometric professional consensus is that equating can occur only
when tests measure the same constructs and have the same reliability and when the
equating process is symmetrical, equitable and population invariant (Linn 2005). Because
equating requires the application of complex psychometric and statistical techniques, this
is the area in which most national learning assessments show greatest weaknesses, and
where IRT is generally applied in regional and international assessments, such as
SACMEQ, PIRLS, PISA and TIMSS. Regional and international learning assessments
expose participants to many of the technical issues for ensuring the stability of
measurement instruments over time. In addition, specific programs for building national
assessment capacity have been established by donors, such as the World Bank, and
international testing bodies, such as Educational Testing Service and the International
Association for the Evaluation of Educational Achievement (IEA).

But building the capacity to develop valid and reliable measures of student learning
achievement faces significant obstacles in developing countries. Among these are: (a) the
technology and psychometrics of test development are evolving continuously, requiring
continuous professional development for test developers, (b) test development software,
including item banking software, is not supported after newer versions are available, (c)
the lack of ongoing doctoral programs in psychometrics in developing countries means
that specialists are often sent overseas for training, but receive little support when they
return home, and (d) the recurrent costs for training and upgrading test development staff
are often unsustainable. Moreover, as the complexity of testing and assessment has
grown – with greater numbers of domains, higher levels of performance, and more
variety in performance measures -- all the processes required to construct, score and
equate tests become more difficult and more expensive, and less within the capacity of
developing countries to achieve.

To summarize, the minimum requirements for monitoring change over time are
rarely satisfied by existing national learning assessments in developing countries. A
few middle-income countries have the technical and administrative capacity to measure
learning progress on an annual basis. Specifically, 11 countries, indicated in bold in Table
3, have been exposed to the technical skills needed to build stable tests, through

12
For example, a question regarding the distance between the earth and the moon was difficult before any
astronauts had been to the moon, and easy immediately after the first moon landing.

9
 Measuring Progress with Tests of Learning

participation in a recent regional or international assessment study, and also have

demonstrated their administrative capacity to administer, score and report the results of
tests on an annual basis, through having done so for a minimum of three consecutive
years. These countries are: Brazil, Chile, Estonia, Hungary, Indonesia, Jordan, Malaysia,
Mexico, Poland, Thailand and Turkey. Egypt and El Salvador could also be included, as
they have administered national learning assessments for two consecutive years during
2005-2007 and have participated in a recent international assessment. Two countries,
Brazil and Chile, are also recognized for the high quality of their national learning
assessment.

Table 3. Low and middle-income countries participating in recent international

learning assessments

Region PIRLS 2006 PISA 2006 TIMSS 2007

Sub-Saharan Africa South Africa Botswana, Ghana, South
Africa
Arab states Morocco Jordan, Tunisia Algeria, Egypt, Jordan,
Morocco, Oman, Syria,
Tunisia, Yemen
East Asia and the Pacific Indonesia Indonesia, Thailand Indonesia, Malaysia,
Thailand
Latin America and the Trinidad and Tobago Argentina, Brazil, Colombia, El Salvador,
Caribbean Chile, Colombia, Trinidad and Tobago
Mexico, Uruguay
Central and Eastern Bulgaria, Georgia, Azerbaijan, Bulgaria, Armenia, Bosnia, Bulgaria,
Europe and Central Asia Hungary, Iran, Croatia, Czech R., Czech Rep, Georgia,
Latvia, Lithuania, R. Estonia, Hungary, Hungary, Iran, Kazakhstan,
of Macedonia, Kyrgyz R., Latvia, Latvia, Lithuania, R. of
Moldova, Poland, Lithuania, Montenegro, Macedonia, Moldova,
Romania, Russian F. Poland, Serbia, Slovak Poland, Romania, Russian
Slovak R. Slovenia R., Romania, Russia F. F., Serbia, Slovak R.
Slovenia, Turkey Slovenia, Turkey, Ukraine
Source: IEA 2007, OECD 2007

3. A discussion of some terminology

National learning assessments typically utilize tests that purport to measure what has
been learned. I now digress with a brief discussion of terminology related to national (or
regional or international) learning assessment, beginning with the term “test.” Any
discussion of “tests” must take into account that the term covers a wide range of
assessment instruments, used for a variety of purposes. Webster’s defines a test as “any
series of questions or exercises or other means of measuring the skill, knowledge,
intelligence, capacities, or aptitudes of an individual or group.” Anderson and others
(1986: 425) observe that “common elements seem to be (a) an experience that is
reproducible across two or more people or groups and (b) some means of characterizing
individuals or groups in comparable terms on the basis of that experience.”

These simple definitions hide the complexity of tests, which can be described in a great
variety of ways, among them the 16 dimensions outlined in table 4. Other analysts

10
Lockheed

provide additional dimensions (see Educational Measurement, 4th Edition, Brennan

2006). National, regional or international assessments typically use a common
constellation of test characteristics (table 4, column 3).

Table 4. Dimensions of tests as applied to national, regional or international

assessments

Application to
Dimension Example assessments
Proposed use Individual diagnosis, selection, Evaluation
certification, program evaluation
About whom test-based decisions are Individuals or groups None (group
to be made implications)
What construct is to be measured Personality, aptitude, mental abilities, Mental abilities and
interests, skills skills
What subject matter or content is to Mathematics, reading, art Mathematics, reading
be measured
Whether the focus is on maximal or Maximal
typical performance
How heterogeneous are the test Batteries of tests, single subject tests Batteries of tests
questions or exercises in terms of
constructs or subjects
How the score or performance is to Subjective standards, criterion Criterion referenced,
be interpreted referenced, norms norm referenced
Type of response the student is to Performance (an essay, a drawing, a Performance and
provide recital), recognition (multiple choice, recognition
true-false, item matching)
How the student’s response is scored Objective vs. subjective, “machine” vs. Objective and
hand, quantitative vs. qualitative subjective
Whether there are standards for the Correct or “best” answers, scales or Correct, best answers
acceptability of the response agreement
Whether the student’s and the tester’s Usually for clinical purposes only Congruent
perceptions of the tests are congruent
When the test is administered Annually, periodically, before or after Annually or
an instructional program periodically
Emphasis on the speed of response Speeded with shorter time limits vs. Power
power with longer time limits
To whom the test is administered Individuals or groups Groups
Who constructs the test Teacher made vs. professional Professional
Source: adapted from Anderson and others 1986

Standardized tests are tests that are “administered under uniform conditions and graded
according to a fixed set of rules or rubrics”; they are not simply tests that use multiple-
choice test items (or questions). They are tests whose results are not contingent on the
time or location of the test or on the scoring of the results. Standardized tests are often
administered with multiple choice (recognition) formats, to ensure consistent scoring as
well as lower costs. Detailed scoring guides, or rubrics, combined with improvements in
scanning technology has meant that consistent scoring can be achieved for constructed
response items (i.e. open-ended questions), allowing standardized tests to include a
higher share of such items, although the cost of professional scoring of constructed
response items is many time greater than that for scoring multiple choice items. Major

11
 Measuring Progress with Tests of Learning

international assessments use both types of items in their standardized tests. For example,
about one-third of TIMSS 2003 items and two-thirds of PISA 2003 items were
constructed-response items (Hutchison and Schagen 2007).

Standardized tests can be referenced to norms, to criteria or to both. Norm referencing,

in its original sense, refers to a process whereby a random sample of individuals is drawn
from the reference population and these individuals are tested. If the test is well
constructed, scores will be distributed according to a normal curve. Results from
subsequent administrations of the test to different individuals or groups will be compared
with the “norm reference group.” For example, infant growth curves are norm-referenced
with respect to the height and weight progress of a norm-reference sample of infants.
More recently, with the advent of large-scale testing, the term norm-referencing is used
when an individual score is compared with the distribution of all scores (such scores are
often referred to as percentile scores). Norm-referenced tests are constructed to spread
questions across a broad range of difficulty. Criterion referencing is completely
different, as it refers to categories of performance that are ordered, such as hurdles of
different heights; individuals either meet or do not meet the criterion at different levels.
Much debate can surround decisions regarding the cut-off points for various criteria, and
criteria-referenced tests are constructed to have greater discrimination around cut-off
points. A single test can provide both “norm referenced” and “criterion-referenced”
information, although there is debate about this practice. 13

Standards based assessment is related to criterion referencing, and pertains to a test that
seeks to assess performance relative to a set of pre-established performance standards or
criteria (Tognolini and Stanley 2007). Performance standards are intrinsic to criterion
referenced tests (Berk 1986, Hambleton and Plake 1995). Standards based assessments
(or tests) comprise items that sample from a broad range of performance, but even
meaningful standards-based assessments “cannot represent the depth and breadth of skills
reflected in the standards documents for any one domain or grade level” (Rupp and
Lesaux 2006: 316-7). Establishing standards can require decades of deliberation, and
building standards-based assessments requires considerable technical skill on the part of
test developers:

“The challenge for test developers is that standards-based

assessment must be broad enough to address in sufficient detail the
complex aspects of proficiency that are laid out in the standards –
those that require complex reasoning and problem-solving skills—
while still addressing basic knowledge and skills” (Rupp and Lesaux
2006:317).

The US experience with standards-based assessment shows that different jurisdictions

can have very different definitions of standards, with “high standards” in one jurisdiction
barely meeting “average standards” in another (Fuller and Wright 2007). A New York
Times editorial observes that “many states have gamed the system – and misled voters –

13
Consider household income. A household can be described as being in the top quintile of the distribution
(norm referencing) or as below the poverty line (criterion referencing).

12
Lockheed

devising weak tests, setting low passing scores or changing tests from year to year to
prevent accurate comparisons over time” (New York Times 2007).

Some discussions of assessment distinguish between “competency” assessments and

“curriculum” assessments. These distinctions, however, suggest greater differences
than have been empirically verified. A well-designed school learning program will
address both curriculum content and cognitive demands, and good tests will include
questions that assess both.

For example, consider the content of two international mathematics tests (in TIMSS 2003
and PISA 2003) that are widely believed to be very different with respect to what they
measure: TIMSS2003 measures “curriculum” and PISA 2003 measures ”competency.”
Comparing the actual content of the test questions on the two tests, however, Hutchinson
and Schagen (2007) find that both tests include questions that measure competencies in
mathematics (using concepts, reasoning) and both tests include questions that cover the
curriculum (knowing facts and solving routine problems). About a quarter of questions
on both tests measure reasoning competency, while TIMSS 2003 has a higher share of
questions that measure knowing facts, and PISA 2003 has a higher share of questions that
measure using concepts. Moreover, scores on TIMSS 2003 can be summarized into levels
of student performance ranging from advanced (students can organize information, make
generalizations, solve non-routine problems, and draw and justify conclusions from data)
to low (students have some basic mathematical knowledge), with such benchmarks
indicating levels of competencies (Mullis and Martin 2007).

Another term is “high stakes” testing. An important feature of sample-based learning

assessments is that they are not used to make decisions about either individuals or groups
(although they do provide information that can inform decisions about groups). Because
students (and teachers and schools) have no direct incentives to perform well on these
tests, test performance is not influenced by potentially distorting incentives. Sample-
based learning assessments are typically considered “low stakes” tests, in sharp contrast
with “high stakes” tests that are used for selection or certification purposes, and – in the
United States--for accountability under the No Child Left Behind (NCLB) legislation. In
a seminal paper, “Will national tests improve student learning?” first presented at the
1991 Annual Meeting of the American Educational Research Association and later
published, Laurie Sheppard delineated six reasons why “high stakes” tests may fail to
reform education. She argued that high stakes tests: (a) become inflated without actual
improvement in learning, (b) narrow the curriculum, (c) misdirect instruction even for
basic skills, (d) deny students opportunities to develop thinking and problem-solving
skills, (e) result in hard-to-teach children being excluded from the system, and (f) reduce
professional knowledge and the status of teachers. In addition, high stakes tests are
subject to distortions due to cheating by students, teachers and schools. Testing agencies
in both developed and developing countries have gone to great lengths to prevent such
from occurring and to identify situations in which they have occurred.

Many developing countries have well-established institutions for constructing,

administering, scoring and reporting the results from “high stakes” selection and

13
 Measuring Progress with Tests of Learning

certification tests (often referred to as “examinations”), and many of the national

learning assessments reported by UNESCO are, in fact, examinations. Such tests are not
used for purposes of educational accountability, and are “high stakes” only for the
students taking the tests. Regional bodies, such as the Caribbean Examinations Council
and the West African Examinations Council, conduct or provide support for national
examinations. Selection examinations were once commonly administered at the end of
primary education (Primary School Leaving Examinations) to regulate the flow of
students into junior secondary education. Examinations at this level, however, have
typically disappeared since junior secondary education has been incorporated into EFA.

Selection examinations are widely administered at the end of secondary education to

regulate the flow into tertiary (higher) education, and some countries still have
examinations at the end of junior secondary education, which are used for certifying
completion of schooling at this level and for purposes of “guidance” – that is,
determining which course of study the student should pursue at the secondary level.
Selection examinations often serve dual purposes, both for screening and for certifying
successful completion of the level of schooling. In general, when two different tests are
used, selection examinations contain a higher share of difficult test questions than do
certification examinations, since the purpose of the selection examination is to limit
access.

4. Using testing for Cash on Delivery aid

The above discussion is intended to draw attention to the administrative and technical
complexities surrounding the application of tests intended to measure progress in
learning. We have not yet discussed the costs associated with testing, and will do so in
this section. What are some of the implications of this discussion for using test results for
Cash on Delivery aid, and what might some alternatives entail?

Costs of testing

Testing comprises a relatively small share of total education expenditures. For example,
Ilon and Harris (1992) used the “ingredients” method (Levin and McEwan 2004) to
estimate costs associated with test development, registration, production, administration,
scoring and reporting for a sample of 20,000 test takers in Jamaica in 1992. The total cost
amounted to US$196,250 ($9.80 per test taker) which was less than 7 percent of
Jamaica’s average annual expenditure on education, 1993-96. 14 More recent data on
costs suggest that participation in international or regional assessments, such as TIMSS
or PIRLS, is also relatively inexpensive. The annual fee for countries participating in the
international activities that support capacity development in assessment was $30,000 per
grade assessed, or $120,000 for the four-year testing cycle, about US$20-25 per test

14
Education expenditure as a share of GNP, 1993-1996 = 7.5%; GNP per capita in 1995 = $1,510;
population in 1995 = 2.5 million.

14
Lockheed

taker. 15 At the same time, there is ample evidence that testing units within governments
are only weakly funded and that only strong political incentives – such as, for example,
the incentive that EU or OECD membership provides governments for participating in
PISA – shake loose adequate funds. Underfunding is particularly widespread in low-
income countries, including those with long histories of participation in regional
examinations councils, such as the West African Examinations Council or the Caribbean
Examinations Council.

Implications

The major implications of the above are that (a) cost is not a determining impediment for
using testing in Cash on Delivery aid, but that (b) the technical and administrative
requirements for using a valid and reliable measure of learning progress pose significant
impediments for testing in developing countries. Political will is also a constraint, but
even with strong political will, capacity issues remain. While the capacity for carrying
out national learning assessments has improved over the past two decades in developing
countries, it remains fragile in most low-income countries. Even in the US, where the
NCLB has created a strong incentive for regular learning assessment, the assessment
capacity for meeting NCLB requirements has been severely stretched. In addition, three
other factors argue against measuring progress directly with tests: the volatility of change
scores, the risk of non-participation and “test pollution.”

Volatility in scores leads to incorrect rewards. National learning assessments utilize, for
the most part, measurement instruments that are poorly suited for registering change, and
are consequently volatile over time. Such volatility means that, on an annual basis,
countries (or schools) could be rewarded or penalized erroneously. While a few, middle
income, countries may have the technical capacity for developing, administering, scoring
and reporting the results from valid measures of student learning on an annual basis,
these countries are not the main focus of the donor community’s concern with respect to
education. Countries for whom Cash on Delivery aid is most salient lack the requisite
technical and scores will not reflect actual progress.

A risk of public embarrassment when scores are released can threaten participation.
Countries have been embarrassed by their performance on international learning
assessments, and this embarrassment has reduced participation and led to restrictions in
publication and dissemination of results (for example, Mexico declined to release its
scores on TIMSS 1999). The publicity that could surround “payments for progress”
could also be embarrassing for some education authorities. The challenge would be to
publicly disclose results without specifically identifying or penalizing schools, which
may not be realistic. Although schools often report the results of “high stakes” tests – e.g.
how many of their students passed an examination or qualified for a merit scholarship –
the “high stakes” are for the students rather than the schools. While a school might
perceive strong incentives in having many students perform well, resources flowing to
the school generally are not contingent on the students’ performance.

15
Based on World Bank Development Grant Facility support to the IEA for 17 developing countries
participating in TIMSS2007 and an average sample size of 5,000 students per country.

15
 Measuring Progress with Tests of Learning

Public disclosure of test scores does not necessitate embarrassment, however, and can
help target resources where they can be effective. In Chile, for example, a program
provided the lowest performing 10 percent of schools, as identified from test scores, with
a package of school inputs, including textbooks, in-service teacher training and tutoring
for low-achieving children; between 1997 and 2000 the test score gap between
indigenous and non-indigenous students had dropped nearly 30 percent due to these
school reforms (McEwan 2006).

“Test pollution” when low stakes tests become high stakes tests. Linking test
performance to budgetary support to education would, inevitably, make any test that is
used to monitor change into a “high stakes” test, at least for school administrators.
Regional and international learning assessments are, by their very nature, low stakes
tests. But high stakes tests are subject to numerous distortions, which Pearson and others
(2001) refer to as “test score pollution,” beginning with those associated with cheating,

Countries with weak testing capacity are simply not able to guarantee that the results of
high stakes tests will not be distorted through cheating. Cheating occurs at all levels, and
parents, students, teachers and schools all have been found to cheat on high stakes tests.
Some countries apply heroic measures to avoid cheating, with astronomical costs. For
example, to reduce cheating on a college admissions test, the National Assessment and
Evaluation Center in Georgia established secure testing facilities and purchased video-
cameras and monitors to observe both test takers and the environment surrounding the
test facility (Maia Miminoshvili personal communication May 18, 2005). In other
countries, cheating is routinely detected through statistical methods, and the cost is also
substantial.

Cheating can occur at all stages of testing. Tests or test questions can be stolen and
released to prospective test takers. Test takers can hire others to take their place for
testing or use cell phones to communicate information about the test questions. Teachers
can coach students or correct their answers after the test administration. Administrators
can invite “low performing” students to skip school on the testing day. The list is endless,
and cheating is caused by having so much ride on the results. As Pearson and others
(2001:177) remark: “Given what we know about test score pollution, we are forced to
believe that nothing is fair or objective and to trust no one.”

Alternatives to using national learning assessments in Cash on Delivery aid

Given the technical complexities involved in measuring change in learning, the lack of
the necessary measurement and implementation capacity in most developing countries,
and the inevitable “test pollution” arising from high-stakes testing, alternatives to using
the results of national learning assessments for monitoring progress could be considered.
The most promising approach would be for multiple indicators of progress to be adopted
into a “country report card” in which test results would be only one of several indicators
of education quality. Some alternatives are discussed in the following paragraphs.

16
Lockheed

Use results from assessments implemented by testing experts. One alternative would be
for countries participating in Cash on Delivery aid programs to contract international
measurement bodies to carry out their annual learning assessments, which could be based
on nationally developed standards. Two advantages of this option are that technical
capacity in measurement is no longer an issue and that the independence of the
assessment could be ensured. Three disadvantages are that this option might generate
little ownership of the assessment by the countries, would not necessarily help national
policy makers understand the test results, and would entail significant costs. But donors
could finance the costs, and there is precedent. The World Bank’s Development Grant
Facility and UNESCO have provided support to countries participating in IEA’s
international learning assessments (TIMSS and PIRLS), and USAID has supported
countries participating in RTI’s Early Grade Reading Assessment studies.

Use donor-developed tests of competency in early grades. A related alternative would be

for international donors to develop tests of reading and math competencies in the early
grades, which tests could be used to measure the results of Cash on Delivery aid. Any
such tests, of course, would need to meet the key criteria listed above. Such tests have
been developed, with reasonable evidence of internal consistency reliability. The earliest
examples are tests of English and mathematics developed by Educational Testing Service
and used in a study of educational achievement in Kenya and Tanzania in 1980 (Knight
and Sabot 1990). These tests were subsequently adapted and used in Ghana, along with a
test of local language developed by the University of Ghana in association with the
University College of Education at Winneba, to evaluate World Bank lending for
education in Ghana (Operations Evaluation Department 2004). 16 During the 1990s,
curriculum-based, tests were developed for use in a longitudinal evaluation of USAID
supported projects in Malawi and Ghana; these tests were roughly equated, and data on
the development of the tests are available (Dowd and Harris 2008, Calahan and Harris
2008). In India, an NGO called Pratham developed a reading test to assess literacy, but
their reports provide little information about the quality of the test, including such basics
as measures of dispersion or reliability (Annual Status of Education Report 2006).

Currently, a beginning reading test (Early Grade Reading Assessment, or EGRA) is under
development by RTI International, with funding from USAID, the World Bank and
various NGOs. EGRA has been applied in pilot versions for various languages in
Afghanistan, Bangladesh, the Gambia, Haiti, Kenya, Mali, Nicaragua, Niger, and
Senegal, with results reported for a few countries; estimates of internal consistency
reliability are available for Kenya 17 and Senegal (RTI International 2007, Springer-
Charolles 2007). The developers of the EGRA, however, are quite explicit that their test
is not suitable for accountability purposes, but is intended only for diagnostic use by and
for teachers in early grades (Grove 2007).

16
All tests had reasonable internal consistency reliabilities as measured by Cronbach’s alpha, although the
reliabilities of tests developed by ETS and adapted for use in Ghana (.72 - .79 for English and .75 - .82 for
mathematics) were slightly higher than those developed for local languages in Ghana (.64 and .70).
17
Cronbach’s alpha of .72 - .89 for English and .88 - .92 for English and Kiswahili combined.

17
 Measuring Progress with Tests of Learning

The advantages and disadvantages of such tests are similar to those associated with using
the results of test implemented by international measurement bodies.

Use participation in regional or international assessments. A third alternative would be

for these countries’ participation in ongoing regional or international assessments to
count as “progress.” Collaborating with international measurement experts would enable
“testing units” in developing countries to strengthen their capacity by forming ongoing
relationships with professional colleagues while providing “low stakes” measures capable
of tracking change over longer periods of time (3-5 years, for example). Performance
would be measured by progress in implementing the technical steps associated with the
regional or international assessment, supported by external technical assistance, rather
than by annual testing of students. This approach has been successful in bringing more
developing countries into such regional and international assessments, and could be
expanded to low-income countries, in particular.

Publish national assessment data disaggregated by key indicators. A fourth alternative

would focus on the gaps in performance among various population subgroups in the
country, rather than on the actual level of performance. While this approach also involves
technical complexities (related to weighting of results, for example) and tests would still
need to meet the technical criteria listed in section 2, it could generate national discussion
about the quality of education for “disadvantaged” groups. Care would need to be taken,
however, to avoid reinforcing gender, ethnic, socio-economic or regional stereotypes
regarding student abilities. Some countries might resist publication of disaggregated
student test scores, from such a concern. For example, Malaysia has not published the
results of national or international assessments, disaggregated by ethnicity (Chinese,
Malay, Tamil) for this reason, and few countries publish results disaggregated by and
indicator of “language spoken at home.”

Measure progress in terms of the development of concrete curriculum standards. Most

developing countries lack concrete curriculum standards, and many teachers in
developing countries do not have a clear vision of what performance is to be expected
from their students (Ferrer 2006). An important first step for improving the quality of
education in developing countries could be initiating a process for developing and putting
into place curriculum standards and their associated measures of performance. Progress
could be indicated by reaching certain specific objectives, such as: (a) defining specific
language and mathematics skills, (b) presenting them to the educational community and
general public for discussion, as Colombia did in 2003 (Ferrer 2006), (c) establishing
learning progress maps for students, as in Chile, (d) providing data on test specifications
related to the skill objectives for review by technical experts, and (e) carrying out an
assessment based on such learning standards.

Adopt non-testing alternative indicators of progress. Another alternative would be for

Cash on Delivery aid to select an indicator of progress that does not involve the technical
and administrative considerations surrounding testing. Such indicators include: (a)
community monitoring of teacher attendance, (b) random checks on average daily student
attendance, and (c) student placements at next higher level of education. As these

18
Lockheed

measures would also become “high stakes” care would need to be taken in collecting and
reporting the relevant data.

The advantages of community monitoring of teacher attendance are two-fold: it involves

the community in the school and is relatively simple to administer. This is also the case
for random checks on average daily student attendance. Monitoring student progress in
terms of placement at the next higher level of education has significant disadvantages
insofar as access to the next higher level of education is frequently determined by supply
factors that are not quickly amenable to change; in particular, since upper secondary
education typically requires teachers having higher levels of skills and qualifications, it is
difficult to expand secondary education without prior investment in tertiary education and
teacher training, with considerable lag times before the results of these investments are
visible.

Use multiple indicators of progress. Perhaps the most effective alternative would be a
mix of indicators that combine testing and assessment with other, observable, indicators
that are uncomplicated by the technical dimensions of testing. Tests are an important part
of this package of indicators, since they can help focus attention on what learning
objectives are important and what signals student accomplishment in attaining these
objectives. But relying on tests alone places too much importance on an indicator that is
easily “corruptible.”

In Short

Existing national learning assessments are poorly suited for measuring annual educational
progress in developing countries, for both technical and administrative reasons. These
tests could, however, be used in combination with other indicators of education quality to
measure educational status and to identify within-country variations in student learning.
“Test pollution” is an issue that would need to be addressed, as performance incentives
can change low stakes tests into high stakes tests. Currently available technical evidence
regarding tests of early reading is insufficient to judge whether such tests offer a high
quality alternative to national assessments. Regional and international assessments offer
high quality alternatives, but are poorly suited for monitoring annual improvements.
Countries could purchase technical expertise, but the costs may be too high to consider.

Alternative indicators of progress suffer from fewer technical difficulties, but are also
possibly subject to distortions caused by becoming “high stakes” indicators. Progress in
implementing regional or international assessments, including independent sampling,
data collection and analysis, could be an essential element of performance-based aid for
education. Progress in establishing clear and measurable performance standards could be
a measure of progress toward quality. Country “report cards” that include multiple
indicators of progress – including test scores – may be more appropriate than a single
indicator.

19
 Measuring Progress with Tests of Learning

References

Amrein, A. L and D. C. Berliner. 2002. “High-Stakes Testing, Uncertainty, and Student

Learning.” Education Policy Analysis Archives 10 (18). Retrieved December 2, 2007
from http://epaa.asu/epaa/v10no18.

Annual Status of Education Report. 2006. New Delhi. www.Indiatogether.org.

Berk, R. 1986. “A consumer’s guide to setting performance standards on criterion

referenced tests.” Review of Educational Research, 56, 137-172.

Braun, H. 2004. “Reconsidering the impact of high-stakes testing.” Educational Policy

Archives, 12 (1). Retrieved December 2, 2007 from http://epaa.asu.edu/epaa/v12n1/.

Braun, H. and Bridgeman, B. 2005. An introduction of the measurement of change

problem. RM-05-01. Princeton, NJ. Educational Testing Service

Braun, H. and A. Kanjii. 2006. “Using assessment to improve education in developing

nations.” In J. Cohen, D. Bloom and M. Malin (Eds.) Educating All Children: A Global
Agenda. Cambridge, MA: American Academy of Arts and Sciences.

Brennan, R. (Ed.) 2006. Educational Measurement (4th Edition). Westport, CT: Praeger.

Cahalan, C. and A. Harris. 2008. “Passage difficulty and reading performance in Ghana.”
Paper to be presented at the Annual Meeting of the Comparative and International
Education Society, New York, March 17-21.

Carnoy, M. and S. Loeb. 2002. “Does external accountability affect student outcomes? A
cross-state analysis.” Educational Evaluation and Policy Analysis 24 (4), 305-331

Dowd, A. and A. Harris. 2008. “Measuring reading proficiency in Malawi, 1999-2006”

Paper to be presented at the Annual Meeting of the Comparative and International
Education Society, New York, March 17-21.

Ferrer, G. 2006. Educational Assessment Systems in Latin America: Current Practice and
Future Challenges. Washington, DC: PREAL

Fuller, B. and J. Wright. 2007. “Diminishing returns? Gauging the achievement effects of
centralized school accountability.” Paper presented at the Annual Meeting of the
American Educational Research Association, Chicago, April 11.

Goertz, M. and M. Duffy. 2003. “Mapping the landscape of high-stakes testing and
accountability programs.” Theory into Practice 42 (1), 4-11.

20
Lockheed

Greene, J.P, M. A. Winters, and G. Forster. 2003. “Testing high stakes tests: Can we
believe the results of accountability tests?” Civic Report 33 (February). Retrieved
December 2, 2007 from http://www.manhattan-institute.org/html/cr_33.htm.

Grove. A. 2007. RTI website article on Early Grade Reading Assessment.

Hambleton, R. and Plake, B. 1995. “Using an extended Angoff procedure to set standards
on complex performance assessments.” Applied Measurement in Education, 8 (1), 41-55

Hutchison, D. and I. Schagen. 2007.”Comparisons between PISA and TIMSS—Are We

the Man with Two Watches?” In. T. Loveless (Ed.) Lessons Learned: What International
Assessments Tell Us about Math Achievement. Washington, DC: The Brookings
Institution.

Ilon, L. and A. Harris. 1992. “Economic analysis of testing system for Jamaica.” Paper
prepared for World Bank.

Ilon, L. 1992. “A framework for costing tests in Third world settings.” Population and
Human Resources Department Discussion Paper PHREE-92-65. Washington, DC: The
World Bank.

Kane, T. J. and D. O. Staiger. 2002. “Volatility in school test scores: Implications for
test-based accountability systems.” In D. Ravich (Ed.) Brookings papers on education
policy, Washington, DC: Brookings Institution. (pp 235-260).

Koski, W. S. and H. A. Weis. 2004. “What educational resources do students need to

meet California’s educational content standards? A textual analysis of California’s
educational content standards and their implications for basic educational conditions and
resources.” Teachers College Record 106 (10), 1907-1935

Lockheed, M., A. Harris, P. Gammill, K. Barrow, T. Jayasundera. 2006. “New Horizons

for Primary Schools in Jamaica: Inputs, Outcomes and Impact” Washington, DC:
Academy for Educational Development

Lockheed, M., A. Harris, P. Gammill, and K. Barrow. 2006. “Impact of New Horizons
for Primary Schools on Literacy and Numeracy in Jamaica: Inputs.” Washington, DC:
Academy for Educational Development

Martineau, J. A. 2006. “Distorting value-added: The use of longitudinal vertically scaled

student achievement data for growth-based, value-added accountability.” Journal of
Educational and Behavioral Statistics. 31 (1), 35-62

McEwan, P. 2006. “The fortuitous decline of ethnic inequality in Chilean schools.”

Wellesley College, Wellesley, MA.

21
 Measuring Progress with Tests of Learning

Mullis, I. and M. Martin. 2007. “TIMSS in perspective: Lessons learned from IEA’s four
decades of international mathematics assessments.” In. T. Loveless (Ed.) Lessons
Learned: What International Assessments Tell Us about Math Achievement. Washington,
DC: The Brookings Institution.

Murphy, P., V. Greaney, M. Lockheed and C. Rojas (Eds.). 1996. National Assessments:
Testing the System. Washington, DC: The World Bank Economic Development Institute.

New York Times. 2007 (November 26). “Test and Switch”. Editorial.

Operations Evaluation Department. 2004. Books, buildings and learning outcomes: An

impact evaluation of World Bank support to basic education in Ghana. Washington, DC:
The World Bank.

Pearson, P. D., S. Vyas, L. Sensale and Y. Kim. 2001. “Making our way through the
assessment and accountability maze: Where do we go now?” The Clearing House 74 (4),
175-182

Porter, A. C., M. Chester, and M. Schlesinger. 2004. “Framework for an effective

assessment and accountability program: The Philadelphia example.” Teachers College
Record 106 (6), 1358-1400

Ragosa, D. 1995. Myths and methods: “The myths about longitudinal research” plus
supplemental questions. In J. M. Gotman (Ed.) The analysis of change. Mahwah, NJ:
Erlbaum.

Ragosa, D. 2002, October. Irrelevance of reliability coefficients to accountability

systems: Statistical disconnect in Kane-Staiger “Volitility in School test scores.”

Resnick, L.B. 2006. “Making accountability really count.” Educational Measurement:

Issues and Practice. Spring, 33-37

Riffert, F. 2005. “The use and misuse of standardized testing: A Whiteheadian point of
view.” Interchange 36 (1-2), 231-252

Rock, D. 2006 “Some thoughts on gain scores and their interpretation in developmental
models designed to measure change in the early school years.” Princeton, NJ:
Educational Testing Service Center for Global Assessment. Draft August 14, 2007.

RTI International. 2007. “Early Grade Reading Kenya Baseline Assessment: analyses and
Implications for Teaching Interventions Design.” Draft for discussion, August 20, 2007.

Sprenger-Charolles, L. 2007. “Senegalese and Gambian Early Grade Reading

Assessments (EGRA): Analyses of the results and suggestions for future EGRA
applications.” Paris: CNRS and Descartes University.

22
Lockheed

Tognolini, J. and G. Stanley. 2007. “Standards-based assessment: a tool and means to the
development of human capital and capacity building in education.” Australian Journal of
Education, 51 (2), 129-145.

Way, W. 2006. Precision and volatility in school accountability systems. RR-06-26.

Princeton, NJ: Educational Testing Service.

Wilcox, K. C. and R. Ryder. 2002. “Standardized testing and improving educational

opportunity in Brazil.” The Educational Forum 66 (spring), 214-219

Yeh, S. 2007. “The cost-effectiveness of five policies for improving student

achievement” American Journal of Evaluation 28 (4), 416-436.

23