Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Forecasting Education
4
Patterns of Potential Human Progress Volume 2: Advancing Global Education 54
There are many ways to approach the
discussion of tools and models for forecasting
the expansion of global education. It is
possible to consider them in terms of their
coverage and aggregationwhether they are
country-specic or multicountry; whether
they focus on primary education or look
also at other levels of education; whether
they consider only enrollment levels or also
the underlying intake and survival patterns;
whether they attend to student ows and/or
to adult education attainment; or whether
they forecast for ten years, twenty-ve years,
or more.
It is also possible to talk about tools and
models in terms of their concern with, and
treatment of, related issue areaswhether
they consider demographics and economics
explicitly and dynamically in interaction with
education, whether they consider primarily the
impact of these other systems on education,
or whether they also look to the implications
of education for other aspects of human
development. And it is possible to talk about
tools and models in terms of their basic
methodological characteristicswhether they
are largely extrapolative of select variables or
more broadly structural in their representation
of multiple, interacting facets of educational
systems; whether they tend primarily to
be accounting systems with exogenously
(externally) provided assumptions about
change; or whether they more dynamically
represent households, governments, and other
potential agents in interaction.
As important as all such characteristics are,
perhaps the most fundamental distinguishing
characteristic is something elsenamely, the
purpose and desired outcome of the use of the
tool or model. Most broadly, forecasting tools
and models are organized around two purposes.
Exploratory tools seek simply to understand
the path of a system, whereas normative tools
identify a desired future and then assess the
likelihood of attaining that future and/or
identify means by which the path toward the
Forecasting
tools and
models can be
exploratory,
normative,
or both.
Forecasting Education 55
desired outcome might be accelerated, redirected
(if the current path is not congruent with the
goal), or otherwise enhanced.
1
This volume combines exploratory and
normative purposes. As Chapter 1 indicated,
the questions we seek to address are: (1) What
path does the formal global education system,
as a collection of countries, appear to be on as
we look forward fty years? (2) Is an aggressive
but still reasonable acceleration of that path
possible? (3) What might be the broader
consequences of such a normative but attainable
acceleration? Given those purposes and
questionsand understanding that all models
are simplications of reality and therefore fall
short of being ideal toolswhat are some of the
general characteristics of the ideal tool that
we might want for such investigation?
Characteristics of Ideal Education
Forecasting Models and Tools
There is a considerable distance between
the characteristics of the simplest possible
exploratory and normative education forecasting
tools and the characteristics that would be found
in an ideal model or tool for our purposes in this
volume. Here, we list some desirable elements
of a tool with a mid- to long-term temporal
reach, beginning with the characteristics that an
exploratory model would include:
n An accounting system that tracks student
ows by education level across all levels and
grades as well as education attainment in
the adult population, with as much detail as
possible regarding elements that vary from
one component of a population to another
(e.g., sex, age, rural-urban residence, income
status, and ethnicity).
n Representation of the dynamics that are
the immediate drivers of student ows (and
hence ultimately of attainment levels),
including separate representations of demand
and supply dynamics and constraints.
n With respect to demand, the ideal
exploratory system represents the
dynamics of enrollment patterns (intake
and survival) in the context of family
circumstances and demographic and
economic trends.
n With respect to supply, the ideal tool
has the capability to estimate the costs
and resource requirements associated
with various enrollment dynamics and
demographic patterns and to forecast the
likely need for, and availability of, public,
private, and international funds.
n Representation not only of demographic
and economic impacts and constraints on
education but also of educations impacts
on demography and on economic systems,
as well as bidirectional feedback loops
between education and other aspects of
human development systems, such as
poverty reduction and the characteristics of
sociopolitical systems.
2
n Transparency of structures, equations,
algorithms, and data; availability to others
for use and analysis; exibility and simplicity
of use.
These same elements would characterize an ideal
mid- to long-range normative forecasting model
or tool. However, the ideal normative tool would
also include the following:
n Specication of points of intervention and an
assessment of their reasonableness.
3
n Evaluation of the impacts of the interventions
not just on education participation and
attainment but also on broader systems
(demographic, economic, and sociopolitical).
n At least some elements of a cost-benet
analysis.
This chapter describes the IFs modeling system,
which can be used for both exploratory and
normative analyses, and considers its particular
strengths and limitations relative to the ideal.
First, however, we briey introduce signicant
global education modeling and forecasting
approaches that others have developed in
recent years.
Recent Education Modeling and
Forecasting Approaches
Over the last several years, a number of models
have been developed by others with similar
interests in understanding the education
transition and the transitions likely continued
unfolding, requirements, and/or consequences.
We identify the models or tools here that have
informed our efforts and comment very briey
on some of their features that had special
Certain
characteristics
are associated
with ideal models,
even though no
model can attain
them all.
relevance for our own work; more information
about these other models and approaches appears
in the Appendix to this chapter. Later, in Chapter
5, we will consider forecasts from some of these
tools in comparison with those from IFs.
McMahon (1999) focused heavily on
exploring the social benets of primary
and secondary education, thus connecting
educations expansion to economic,
demographic, and sociopolitical change. He
used a cross-sectional approach to drive much
of the dynamic analysis in his econometric
model and developed a base case as well as two
normative scenarios to forecast the impact of
specic education policy changes.
4
Delamonica, Mehrotra, and Vandemoortele
(2001), using an accounting-centric approach
and UN population projections, conducted a
normative analysis of the incremental costs
of moving to universal primary education
by 2015. Their estimates of costs included
measures intended to enhance education
quality as well as capital costs for needed
increases in capacity. Bruns, Mingat, and
Rakotomalala (2003) also explored the costs
of meeting universal primary education by
2015. They extended previous analyses by
developing best practice expenditure and
resource mobilization guidelines based on
the education policies and practices of low-
income countries making the best progress
toward universal primary education, and their
framework included the normative concept of
minimum adequate cost.
Clemens (2004) used historical data to
analyze transition paths in net primary
enrollment rates and found an S-shaped curve
that could be used to extrapolate the number
of years countries and regions might need
to reach 90 percent primary net enrollment.
He also compared the earlier experience of
currently high-income countries with the
recent experience of low- and middle-income
countries, and found that the speed of advance
in enrollment rates has accelerated considerably.
Wils, OConnor, and Somerville, as reported in
a paper authored by Wils, Carrol, and Barrow
(2005), also found S-shaped patterns in the
advance of primary education and used them
to project growth in primary entry rates and
completion rates separately (rather than
aggregate enrollment rates). They also used data
gathered in household surveys to provide more
extensive and longer estimates of historical
patterns of school participation.
Lutz, Goujon, and Wils (2005) built on the
multistate demographic methodology of the
International Institute for Applied Systems
Analysis (IIASA) to explore the future across
all levels of education, including tertiary.
Moreover, the principal focus of their work
was on adult attainment levels differentiated
by age and sex, so as to build the foundation
for looking at the relationship between those
levels and characteristics and other aspects of
global change. Similarly, Hilderink (2007) added
attention to adult attainment to his forecasting
of ows across all levels of formal education.
Also of interest to us, his formulations, like
ours, relate education demand and education
supply to GDP per capita.
An EPDC paper (2007b) reported the work of
Wils, Barrow, Oliver, Chaluda, Goodfriend, Kim,
and Sylla in the development and early use of
ProEnrol, a country-level, cohort-projection model
for use at the primary and secondary levels. The
paper described the model as the rst effort
to make cohort or grade-by-grade projections
in an international, global series, including
representations of promotion and repetition.
The IFs project, as will be seen in the
discussion that follows, has both independently
developed and also built upon many of the
features of these other models: the cross-
sectional analysis and attention to sociopolitical
impacts of education found in McMahon (1999);
the computation of costs required to meet goals
of Delamonica, Mehrotra, and Vandemoortele
(2001), as well as some of the attention to best
practice that Bruns, Mingat, and Rakotomalala
(2003) built into their analysis; the S-shaped
expectations for educations advance that
both Clemens (2004) and Wils, OConnor, and
Somerville cited in Wils, Carrol, and Barrow
(2005) found and used, as well as their explicit
recognition of country-specic circumstances;
the attention to adult attainment of education
by Lutz, Goujon, and Wils (2005); the use of GDP
per capita to drive the formulation of education
demand and supply made by Hilderink (2007);
and the country-level, cohort analysis of Wils
et al. (2007b). This is not to say, of course, that
one model can do everything as fully or as well
as more specialized studies and approaches, but
In recent years,
a number of
models have been
developed to help
in understanding,
forecasting,
and facilitating
the education
transition.
The IFs project
has built upon
features of other
models and has
also independently
developed
new features
and modeling
capabilities.
we do believe there is also value in our more
comprehensive approach. We will return to a
description of the IFs approach to modeling of
education after providing an introduction to the
larger IFs system.
The IFs Modeling System
As stated previously, the particular strengths of
IFs derive from the combination of its extended
time frame, its extensive geographic coverage
with capability to exibly group countries for
analysis and display, and its dynamic integration
of multiple human systems. In addition, its
global education model is the only one we know
of that represents all three levels of formal
education in grade-by-grade student ows or
cohorts, as well as the only one that represents
lower and upper secondary education separately.
IFs can be used both for exploratory analyses
of dynamic trends and patterns and for the
creation of normative scenarios and explorations
of their respective impacts.
In the sections that follow, we will rst
provide a brief overview of the broader IFs
forecasting system and then discuss the
IFs education model in more detail. In the
process of that discussion, we will attempt to
identify the strengths and weaknesses of IFs in
comparison with the characteristics of an ideal
global education forecasting tool.
General design considerations
International Futures is a large-scale, long-
term, integrated global modeling system. It
represents demographic, economic, energy,
agricultural, sociopolitical, and environmental
subsystems for 183 countries interacting in the
global system.
5
The central purpose of IFs is to
facilitate exploration of global futures through
alternative scenarios.
The issues of interest that motivated
the design of IFs fall generally into three
categories: human development, social
fairness and security, and environmental
sustainability (see Table 4.1). Across these
domains, the project especially looks to
Sen (1999) for his emphasis on freedom
and individual development, Rawls (1971)
for his emphasis on fairness within society,
and Brundtland (UN 1987) for her seminal
denition of sustainability. These emphases,
in combination, provide a philosophical
framework for the exploration of human
beings as individuals, of human beings
with each other, and of human beings with
the environment.
Human systems fundamentally involve
different types of agents (economists tend
to focus on households and rms; political
scientists add governments) interacting with
each other in various structures (economists
focus on markets; political scientists look to
action-reaction systems and international
regimes; sociologists add societies and
demographic structures; anthropologists focus
on cultures; and physical scientists extend the
reach to ecosystems). In general, scientists
seek to understand the complex cocreation and
evolution of agent behavior and the structural
characteristics of human and social systems.
IFs attempts to capture some of that
complexity and richness by being rooted
in the theory of various disciplines and
subspecializations. It is a structure-based, agent-
class-driven, dynamic modeling system. That is,
it tries to represent typical behavior patterns of
major agent classes (households, governments,
rms) interacting in a variety of global
structures (demographic, economic, social, and
environmental) with extensive representation
of underlying accounting systems.
6
IFs draws
upon standard approaches to modeling specic
issue areas whenever possible, and then, as
necessary, it extends and integrates these. For
instance, the IFs demographic model uses a
typical cohort-component representation,
tracking country-specic populations over
time by age and sex, further differentiated
in IFs by education. Within that structural or
accounting framework, the model represents the
fertility decisions of households (inuenced by
income and education) as well as mortality and
migration patterns.
The database underlying IFs (and integrated
with the system so it can be used by others)
includes a vast range of data for 183 countries,
represented over as many years since 1960
as possible on a country-by-country basis.
7

The desire
to explore
issues of human
development,
social fairness
and security, and
environmental
sustainability
motivated the
design of IFs.
Table 4.1 Human systems and issues of interest to the IFs project
Humans as individuals Personal development/freedom
Humans with each other Peace and security/social fairness
Humans with the environment Sustainable material well-being
IFs draws upon
standard modeling
approaches to
specic issue
areas whenever
possible and then,
as needed, extends
and integrates
them.
The model system itself runs in annual time
steps from its initial year (currently 2005),
8
and
the model interface facilitates user interventions
exibly across time, issue area, and geography.
The models of IFs
Figure 4.1 shows the major conceptual blocks.
Full issue-specic models represent most of
the blocks, including education. The elements
of the technology block are actually dispersed
throughout the system, and the named linkages
between blocks (and the identied linkages
themselves) are a small illustrative subset, by no
means an exhaustive listing.
The two models within the IFs system that
interact mostly closely with the education model
are the population and economic models. In
our representation of the human development
system, the sociopolitical model also interacts
quite closely with the education model (as well
as with the economic and demographic models).
We cannot here provide technical details of
these or other models that collectively make up
the IFs forecasting system that fully integrates
those models. We do, however, provide certain
summary information in the discussion ahead,
which will be too much technical detail for some
readers and far too little for others. Those who
want more information about the IFs system
(including the education model) will nd
extensive documentation at www.ifs.du.edu. In
particular, see Hughes (2004b, 2006; Hughes
et al. 2004) for a structural overview and for
discussions of scenario analysis and validation.
The model system is also freely available there in
both online and downloadable forms.
The demographic model has the standard
cohort-component structure that the UN and
other institutions use in population forecasting,
but it represents fertility and mortality as
functions of other variables in IFs. Some of the key
characteristics of the population model are that it
n Represents twenty-two age-sex categories to
age 100+ in the cohort-component structure
(but computationally spreads the ve-year
cohorts initially to one-year cohorts and
calculates change in one-year time steps)
n Calculates change in cohort-specic fertility
in response to income, income distribution,
education levels, and contraception use
n Calculates change in mortality rates in
response to income, income distribution,
education, and assumptions about
technological changes impact on mortality
n Separately represents the evolution of HIV
infection rates and deaths from AIDS
n Computes literacy rates, average life
expectancy at birth, and an overall measure
of human development (the Human
Development Index, or HDI)
n Represents migration and ties it to ows of
remittances.
The economic model has the multisector
equilibrium structure of models that most
forecasters of development processes use, but it
has extended representation of the production
side so as to facilitate long-term analysis and
to link productivity to other variables in IFs,
including education. Some of the most important
characteristics of the economic model are that it
Figure 4.1 The major models in the IFs modeling system and example
connections
Sociopolitical International political
Education Health
Population Economic
Agriculture Energy
Technology Environmental resources and quality
Government
expenditures
Conflict/cooperation
Stability/instability
Mortality
Income
Demand,
supply, prices,
investment
Fertility
Labor
Efficiencies
Land use
Water use
Resource use
Carbon production
Food
demand
Extensive
documentation
of the IFs system
of models is at
www.ifs.du.edu,
as is the model
itself in both online
and downloadable
forms.
n Represents the economy in six sectors:
agriculture, materials, energy, industry,
services, and information/communications
technology, or ICT; other sectors could also be
congured because the system uses raw data
from the Global Trade and Analysis Project
(GTAP)
n Computes and uses input-output matrices
that change dynamically with development
level
n Is a general equilibrium-seeking model that
does not assume exact equilibrium will exist
in any given year; rather, it uses inventories
as buffer stocks and to provide price signals so
that the model chases equilibrium over time
n Contains a Cobb-Douglas production function
that (following insights of Solow and Romer)
endogenously represents contributions to
growth in multifactor productivity from
human capital (education and health), social
capital and governance, physical and natural
capital (infrastructure and energy prices),
and knowledge development and diffusion
(research and development and economic
integration with the outside world)
n Uses a Linear Expenditure System to
represent changing consumption patterns
n Utilizes a pooled rather than bilateral trade
approach for international trade
n Is embedded in a social accounting
matrix (SAM) envelope that ties
economic production and consumption to
representation of intra-actor nancial ows
(it represents, however, only the skilled and
unskilled households from GTAP).
Few sociopolitical models exist except in the
form of highly specialized representations (such
as the forecasting of state failure). The model
in IFs has a relatively extensive treatment of
sociopolitical variables, including government
budgeting, which is important in representing
constraints upon expansion of education. Some
of the sociopolitical models relevant features are
that it
n Represents scal balances through taxing and
spending decisions
n Shows six categories of government spending:
military, health, education, research and
development, foreign aid, and a residual
category (as well as representing transfer
payments for pensions and social welfare)
n Represents changes in social conditions
of individuals (such as fertility rates,
literacy levels, or poverty), attitudes
of individuals (such as the level of
materialism/postmaterialism of a society
from the World Values Survey), and the
social organization of people (such as the
status of women)
n Represents the evolution of democracy
n Represents (in very basic fashion) the
prospects for state instability or failure.
The use of IFs
Although initially developed as an
educational tool, IFs is increasingly used in
research and policy analysis. For instance, it
was a core component of the TERRA project
sponsored by the European Commission to
explore the New Economy. More recently,
forecasts from IFs supported Project 2020 of
the National Intelligence Council (NIC) (USNIC
2004) as well as NICs subsequent study, Global
Trends 2025: A Transformed World (USNIC 2008).
IFs also provided driver forecasts and some
integrating analysis for Global Environment
Outlook4 of the United Nations Environment
Programme (2008).
The menu-drive interface of the
International Futures software system allows
display (in tables and standard graphical
formats) of historical data values since 1960,
in combination with forecasts from the base
case and from alternative scenarios over
time horizons from 2005 through 2100. It
includes a Geographic Information System
(GIS), or mapping capability, and also provides
specialized display formats, such as age/sex
and age/sex/education cohort structures and
social accounting matrices.
The system facilitates scenario development
and policy analysis via a scenario tree that
simplies changes in framing assumptions
and agent-class interventions. Users can save
scenarios for development and renement
over time, including the normative education
scenario developed and analyzed in Chapter 6
and Chapter 7. Standard framing scenarios, such
as those from the United Nations Environment
Programmes Global Environmental Outlook4, are
available with the model for users to explore and
potentially to develop further.
The IFs system
facilitates
explorations of
interventions and
scenarios.
The Education Model
The education model of IFs simulates patterns
of education participation and attainment in
183 countries over a long time horizon under
alternative assumptions about uncertainties
and interventions (Irfan 2008). Its purpose is to
serve as a generalized thinking and analysis tool
for educational futures within a broader human
development context.
In Figure 4.2, we display the major variables
and components that directly determine
education demand, supply, and ows in
the IFs system. We emphasize again the
interconnectedness of the components and their
relationship to the broader human development
system. For example, during each year of
simulation, the IFs cohort-specic demographic
model provides the school-age population to the
education model. In turn, the education model
feeds its calculations of education attainment
to the population models determination
of womens fertility. Similarly, the broader
economic and sociopolitical systems provide
funding for education, and levels of educational
attainment affect economic productivity and
growth and therefore also education spending.
Table 4.2 summarizes the most important
aspects of the accounting system, the dominant
relationships, and the key dynamics that our
education model represents. At the accounting
level, the major ows within the model
are student and budgetary ows, and the
major stock is that of gender-differentiated
educational attainment of the adult population.
The model structurally represents the formal
education system from the primary through
tertiary levels, and it further divides the
secondary level into lower secondary and upper
secondary levels and into general and vocational
categories within each of the secondary levels.
It tracks students by grade and by sex. Intake
(or transition to a higher educational level)
and persistence or survival rates are the two
variables that most immediately determine
the patterns of student participation and
progression through the grades.
The dominant relationships in the model are
those that determine the intake (or transition)
and survival rates and the costs of education per
student, all three of which are closely connected
to per capita income. The model also takes into
account the long-term nonincome drivers of
education in an aggregate fashion. As the model
simulates the gradual expansion of education,
the intake and survival rates saturate following
an S-shaped pattern.
9
With respect to key dynamics, the processes
of the demographic and economic models,
as sketched earlier, signicantly affect the
forecasting of education. Similarly, the dynamics
Figure 4.2 Direct drivers of education demand and supply in IFs
Demography
Sociopolitical
Economy
Government budget (sociopolitical)
Student flows and
budget balancing
Education
demand
Income
per capita
Human capital stock
(Nonincome) systemic shift
Education cost International transfers
Education
investment (supply)
Education
and budget
flow indicators
of the government budget process in the
sociopolitical model produce a key variable for
the education model, namely, the availability
of funding for education. Within the education
model itself, a central dynamic is nding the
balance between the demand for education and
its availability or supply and then adjusting
growth in intake (or transition) and survival
rates, as well as spending per student, to be
consistent with that balance.
Modeling of complex, integrated, dynamic
systems for long-term forecasting is seldom a
matter simply of specifying equations. It generally
requires development of algorithmic structures
(logical procedures for integrating calculations
and maintaining accounting systems), as well
as equations. In the IFs educational model, one
such algorithm manages student progression
through the grades. On the budgetary side,
another balances the forecasted funding demand
and funding availability in order to shape the
enrollment and spending levels. And still another
addresses the ows of graduates into and through
the adult population. Subsequent sections of this
explication of the education model provide basic
information on these processes.
10
Accounting system
As Chapter 2 outlined, a conceptual description
of student ows begins with entry in the rst
grade of primary school. At the end of each
year, students either progress to the next
grade, repeat the current grade, or drop out.
Eventually, some proportion of the entering
cohort reaches the beginning of the nal grade;
that proportion constitutes the survival rate.
Further, of those who persist to the beginning
of the last grade of primary school, most
subsequently graduate and become eligible
to continue to the lower secondary level. The
transition rate identies the portion of those
completing the primary level that actually
continues into general programs at the lower
secondary level,
11
following which a new
pattern of grade-level progression, repetition,
and dropout ensues. Conceptually, similar ows
(albeit at different rates) take place at the upper
secondary level and at the tertiary level.
IFs accounts for education participation
by simulating gender-specic grade-by-grade
student ows, using country-specic entry
ages and years of schooling at each level to
represent enrollments and to distinguish
gross and net ow indicators. We dynamically
forecast intake rates (or transition rates to
general programs at the lower and upper
secondary levels) and survival rates, and we
calculate enrollment rates as the combined
result of those ows, tracking students through
grades. Clearly, this approach provides more
useful information than a focus on enrollment
rates alone would, as the same enrollment
rate might result from different combinations
of intake and survival rates. It also provides
points for representing interventions that
shape the actual dynamics of enrollment.
Table 4.2 Foundational elements of the IFs education model
Education model aspect Key elements
Accounting system Flows of students into, through, and out of schools
Flows of public spending into education system
Stocks of adults with different levels of education attainment
Dominant relationships Intake demand is driven by household income and nonincome systemic factors
and follows an S-shaped pattern toward a saturation point
Survival rate is driven by the same factors as intake with income being the
most dominant
Education cost is driven by per capita income with a different cost function
at each level of education
Key dynamics Demographic change
Economic development
Public education spending constrained by revenue receipts, government
consumption, and demands from other public sectors
Equilibration between the demand and supply of education funds
Long-term
dynamic forecasting
often requires
development of
algorithms as well
as the specication
of accounting
structures and
equations.
Student ows
A truly full representation of student ows (see,
again, Figure 2.3) would represent movement
across grades over time with grade-by-grade and
country-by-country specication of repetition
rates, dropout rates, and rates of return and late
entry by overage students. It would culminate
with rates of completion and transition by
some to higher levels. In addition to being
very intensive with respect to initial data
and ongoing computations, there is a limited
basis for forecasting idiosyncratic patterns of
repetition, dropout, and reentry by grade.
Our grade-by-grade student ow model
therefore uses some simplifying assumptions in
its calculations and forecasts. We combine the
effects of grade-specic dropout, repetition,
and reentry into an average cohort-specic
grade-to-grade ow rate, calculated from the
survival rate for the cohort. Each year, the
number of new entrants is determined by the
forecasts of the intake rate and the entrance age
population. In successive years, these entrants
are moved to the next higher grades, one grade
each year, using the grade-to-grade ow rate.
The simulated gradewise enrollments are then
used to determine the total enrollment at the
particular level of education.
There are some obvious limitations to our
simplied approach. Although our model
effectively includes repeaters, we represent
them implicitly (by including them in our grade
progression) rather than representing them
explicitly as a separate category. Moreover,
by setting rst-grade enrollments to school
entrants, we exclude repeating students from
the rst-grade total. On the other hand, the
assumption of the same grade-to-grade ow
rate across all grades might somewhat overstate
rst-grade enrollment in a typical low-education
country, where rst-grade dropout rates are
typically higher than the dropout rates in
subsequent grades. Since our objective is to
forecast enrollment, attainment, and associated
costs by level rather than by grade, we do not
lose much information by accounting for the
approximate number of school places occupied
by the cohorts as they proceed and by focusing
on accurate representation of total enrollment.
Figure 4.3 juxtaposes the primary grade-
by-grade enrollment data in Bangladesh in
1988 as constructed from UIS-reported intake
and gradewise survival rates against the
grade-by-grade pattern that IFs simulates. It
illustrates that the net effect of our simplifying
assumptions generally produces reasonable
results with respect to overall enrollment
rates, headcounts, and hence also resource
requirements. The initialization of the model,
discussed later, further protects initial data on
enrollment. Even so, our approach results in an
(usually small but occasionally quite signicant)
initial discrepancy between reported and
calculated enrollment, as seen in Figure 4.3. IFs
computes that differential as an additive factor
so as to assure that our computation and the
data are consistent; the model carries forward
the additive factor but causes it to converge to
zero over time.
A separate algorithmic structure helps
represent gross enrollment patterns at the
primary level. Specically, the model tracks the
pool of potential students who are above the
entrance age (as a result of never enrolling or
of having dropped out), and it brings some back
as students (dependent on initial conditions
with respect to gross versus net intake) for the
dynamic calculation of total gross enrollments.
A generally similar grade-ow methodology
models student ows at the lower and upper
secondary levels, including country-specic
entrance ages and durations at each level. Two
adaptations were necessary. First, UIS provides
only gross enrollment data for lower and upper
IFs simulates
gender-specic
grade-by-grade
student ows,
using country-
specic entry
ages and years of
schooling.
Figure 4.3 Example comparing primary grade ow data and IFs simulation:
Bangladesh (1988)
100
40
60
4 5 1
Enrollment rate
for total primary:
Data
Enrollment rate
for total primary:
Ifs simulation Percent
Grades
Grade-by-grade
enrollment rates:
Data
90
70
80
2 3
Grade-by-grade
enrollment rates:
Ifs simulation
50
Source: UIS data and authors calculations.
secondary enrollment rates, so our core lower
and upper secondary forecasts are also gross
rates only (a relationship estimates total
secondary net enrollment from the gross values).
Second, although UIS provides transition
rate data from the primary to the secondary
levelwhich in effect is the transition rate into
lower secondaryit does not provide transition
rates from the lower secondary to the upper
secondary level. However, UIS does provide
grade-by-grade secondary headcount time-series,
from which the IFs model calculates historical
lower to upper secondary transition rates as the
starting point for forecasts of future rates.
In the ISCED taxonomy of educational
programs (see the Appendix of Chapter 2),
tertiary education displays the greatest
complexity. Not only are there two categories
(programs that lead to an advanced research
qualication and programs that do not),
there are also two subcategories within the
programs that do not lead to an advanced
research qualication. One subcategoryitself
quite broadencompasses theoretically based
programs and programs that prepare students
for practice in high-skill professions; the second
category includes programs that are practical,
technical, and occupationally specic.
To cut through some of this complexity,
UIS in some treatment of data and IFs in its
representation of student ows both make
simplifying assumptions at the total tertiary
level.
12
For example, rather than using
country-specic and tertiary category-specic
program durations to calculate ows, both
UIS and IFs base calculations of tertiary ow
rates on an assumed ve-year program period.
To initialize the model, we rst use the total
UIS headcount of graduates of all programs to
calculate an overall gross tertiary graduation
rate based on the assumed ve-year program
period. We then use our calculated overall
graduation rate with the total tertiary gross
enrollment rate from UIS to calculate an
overall tertiary gross intake rate.
13
Education attainment
The algorithm for the tracking of education
attainment is very straightforward. The model
maintains the structure of the population not
only by age and sex categories but also by years
and levels of completed education. In each year
of the models run, the youngest adults pick
up the appropriate total years of education
and specic levels of completed education. The
model advances each cohort in one-year time
steps after subtracting deaths. The primary
weakness of the approach, common to many
but not all other models, is that it does not
represent differential mortality rates associated
with different levels of education attainment
(generally lower for the more educated).
14
This
leads, other things being equal, to a modest
underestimate of adult education attainment,
growing with the length of the forecast horizon.
The method that IFs uses to advance adults
through the age/sex/education categories also
slightly misrepresents the level of education
attainment in each ve-year category.
15
Financial ows
In addition to student ows, and interacting
closely with them, we want to track nancial
ows. In IFs, we conceptualize those ows as
being the result of the interaction of demand-
and supply-side forces, a dynamic to which the
discussion will return. The accounting side is
relatively simple. Given forecasts of spending
per student by level of education and given
enrollments by level, an estimate of the total
demand for education funding is simply the
sum across education levels of the products of
spending per student and student numbers.
This so-called demand for educational funding
is, however, a crude conceptualization. The
ow structure of the model does not truly
represent a demand for education (see, again,
Figure 4.2) because initial conditions clearly
reect historical nancial constraints. As with
any other collective good, societies tend to
underprovide education relative to the point
at which expenditures would truly equal their
potential benets.
Hence, a more accurate conceptualization is
that IFs represents a demand-driven, supply-
constrained system. In the future, the extent of
supply constraint may wax or wane (sometimes
even providing largesse), and the differing
budget situations will affect both expenditures
per student and enrollment levels.
Turning to the budget, governments provide
most education funding. Public expenditures
on education as a portion of GDP vary greatly
across countries (see Figures 3.12 and 3.13),
IFs represents
population
structures by
age, sex, and
years and levels
of completed
education.
averaging around 5 percent in recent years and
ranging from under 2 percent to over 13 percent.
Although fewer than 25 percent of all countries
report data on private funding, it appears that
private funds account at most for about one-
fourth of all education expenditures and that
they are concentrated at the upper secondary
and tertiary levels (see Chapter 3 for further
discussion). Because of the scarcity of private
funding data, IFs specically represents public
funding only, and our formulations of public
funding implicitly assume that the public/
private funding mix will not change over time.
In reality, the picture is more complicated. At
the primary level, and perhaps also at the lower
secondary level, it is more likely that tuition
fees charged for public education will be phased
out over time. In fact, the targeted increases
in our formulations of per student costs in
those countries with low current per student
expenditures may be a proxy for a shift to public
support. However, at the upper secondary and
tertiary levels, private funding in the form of
tuition fees may increase in some countries in
order to expand capacity.
Dominant relationships
Before turning to the formulations of the model
for forecasting intake and survival, it is useful
to note that two alternative methodologies
frame effectively all long-term forecasting. The
rst is extrapolation, and the second is causal
analysis. Each has a variety of advantages and
disadvantages, and our earlier review showed
that both have been used to forecast the spread
of education.
A key advantage of extrapolation is the
relative simplicity involved in fundamentally
univariate analysis, relying only upon the
history of a variable in order to determine its
future. It should be noted, though, that the
frequent use of specialized formulations in
extrapolations (such as the S-curve) implicitly
builds in the effects of other variables, such
as the constraint of bringing difcult-to-
reach populations into school and the shift of
resources to higher levels of education, which
are both implicit in the slowing of growth as an
enrollment rate approaches 100 percent.
The fact that it is univariate is also an
important disadvantage of extrapolation. For
instance, when historical series are short and
especially when they are sparse, both of which
tend to be true for intake or transition and
survival rates, the basis for extrapolation is
liable to weaken (although turning to household
survey data can extend data series relative to
purely administrative data).
Chapter 2 already reviewed some of the
most signicant issues associated with causal
analysis, including problems in sorting out
the direction of causality in bivariate analysis,
the possibilities of spurious relationships and
complex interaction effects in multivariate
analysis, and complications introduced by long
lag effects.
However, among the important advantages
of a causal approach such as that in IFs is the
ability to play with the driving variables in a
causal analysis, allowing development of a range
of scenarios linked to important drivers, some of
which may in turn be linked to potential policy
levers. In long-term analyses, causal approaches
can sometimes more clearly represent the
structure of a system, incorporating interaction
effects and constraints such as that between the
supply of funds and the demand for funds in
the IFs education model. Education systems are,
in fact, subject to a variety of such interactions
and constraints. For instance, students cannot
enter higher levels of education unless they
complete lower ones. Moreover, there tend to
be patterns of relationships between intake and
survival rates, as well as between enrollment
rates at different levels of education, that purely
extrapolative formulations might, in long-term
forecasting, not reproduce.
Even much of traditional causal analysis,
if it were undertaken purely on the basis
of independent formulations for intake
and survival at different education levels,
would strain to maintain such relationship
patterns. Instead, causal analysis embedded
in algorithmic (rule-representing) logic and
attentive to the patterns of causal or dominant
relationships across levels of education can be
useful, and that is fundamentally the approach
of the IFs education model.
Intake and survival
As the discussion of student ow accounting
emphasized, the rates of intake of students into
primary education (or the rates of transition of
primary students to higher levels), the patterns
The government
budget submodel
of IFs determines
the amount of
funding available
for education.
A causal
approach, such as
that of IFs, allows
development
and analysis of a
range of scenarios
linked to drivers
of change.
of grade-by-grade progression, and the rates
of survival through the grade progression to
the nal grade (as well as rates of completion
of that nal grade) collectively determine
enrollment rates and numbers. The relationships
that underlie those forecasts of intake or
transition and of survival are especially
important or dominant ones.
The forecasting of adjusted primary net
intake rates begins to illustrate the IFs
causal approach (later in this chapter we
explain the concept and use of adjusted
primary net intake). It has several elements.
The rst is use of cross-sectional analysis
to specify the relationship between gross
domestic product per capita and adjusted
net intake at the primary level (see Figure
4.4).
16
Such cross-sectional representations,
looking at relationships between variables
across countries at a given time point, help
us understand something about the typical
long-term developmental patterns of countries
globally and thus give basic insight into likely
longitudinal dynamics.
17
There is a clear
tendency for primary intake rates to increase
with GDP per capita, particularly at lower
levels of GDP per capita (below about $5,000
at purchasing power parity). This relationship
reects, in part, changing economic structures
and changing demand for the skills acquired
through education, as well as the growing
ability of richer societies to provide education.
The specic basic function in Figure 4.4 is
ANIR
t
= 41.8 + 5.77 * ln GDPPCP
t
where
ANIR is adjusted primary net intake rate
GDPPCP is GDP per capita at purchasing
power parity
Although GDP per capita is a powerful driver
and/or correlate of a great many aspects of
social change (Hughes 2001), the relatively
low R-squared values in Figure 4.4 (and for
most such relationships between GDP per
capita and intake, transition, and survival
rates, the R-squared values fall in the range
of 0.150.35) suggest there is much room for
extended analysis of potential dynamics of
intake. We have explored the addition of other
factors, such as the education of women as
captured in the percentage of women fteen
and older who have completed secondary
education. On the whole, that factor tends
to be comparable in power to GDP per capita
(and is highly correlated with it), but such
factors tend not to add a great deal to the
multiple R-squared. Still, we know from many
empirical analyses that parents education is a
key determinant of intake rates (Clemens 2004:
4), and the omission of its explicit treatment
from the IFs formulation for intake is almost
certainly a weakness.
We estimated, and the IFs model uses, a
full set of gender-specic, cross-sectional
functions (see Figure 4.5) as the rst step in
forecasting the ow rates at different levels of
education. The functions for intake rates show
the expected progression with GDP per capita.
That is, at lower levels of income, countries
show higher typical rates of primary net intake
than they do rates of lower secondary gross
enrollment, which in turn exceeds rates of
upper secondary gross enrollment. The patterns
for survival rates are more complex, and it is,
of course, possible that countries have higher
survival rates at the secondary level than at
the primary level. The functions also show the
advantages that females tend to develop even
in middle-income countries at the tertiary
level, both in intake and survival rates.
Forecasting
rates of intake
or transition and
of survival is
foundational to
IFs forecasts of
enrollment.
Figure 4.4 Relationship of female adjusted primary net intake and
GDP per capita at PPP
100
20
30
40
50
4
0
,
0
0
0 0
Primary intake, net adjusted
GDP per capita at PPP (2000 $)
60
70
80
90
3
5
,
0
0
0
3
0
,
0
0
0
2
0
,
0
0
0
2
5
,
0
0
0
1
0
,
0
0
0
1
5
,
0
0
0
5
,
0
0
0
Guinea-Bissau
Comoros
Djibouti
Congo, Democratic Republic of
Uruguay
Oman
Equatorial Guinea
Eritrea
Kuwait
Hong Kong
Burkina Faso
Ethiopia
Chad
New Zealand
Korea, Republic of
Cuba
Note: Equation: y = 34.2401 + 6.5355 * log (x); R-squared = 0.19
Source: IFs Version 6.12 using UIS and WDI data (most recent by country).
This set of functions provides a fundamental
group of expectations for intake and survival
that provides an initial foundation for forecasts.
In addition, these functions help maintain
relational integrity of forecasts across education
levels. Again, however, the great variation of
countries around the functions makes clear the
need for additional steps.
Instead of adding additional variables to the
regression analysis, our formulation turns to
other factors and approaches. First, considerable
path dependency exists at the country level.
Returning to Figure 4.4, note that Cuba is
positioned well above the regression line, due
to socialist policies that support universal
education. Oman falls well below the line, as
other Middle East countries often do. Geographic
factors, ethnic and religious patterns, and
cultural traditions inuence intake rates,
helping to create such country-specic and
region-specic path dependencies. IFs partially
protects those patterns by computing additive
adjustment factors in the rst forecast year
that represent the position of empirical values
relative to the relationship. As GDP per capita
grows in forecasts, these adjustment factors
continue to maintain the position of countries
relative to the relationship.
At the same time, however, such differences
can be idiosyncratic and temporary, and
deviations from larger systemic patterns often
erode. Thus, the model uses a convergence
process to bring outliers gradually to the values
of the function. We have estimated the duration
of convergence periods subjectively based on
model behavior, and these periods vary across
levels of education. We generally anticipate
convergence to be faster at lower levels of
education, where emphasis is greater and
enrollment rates on average are higher, than at
higher levels.
18
We allow primary adjusted net
intake to converge only in an upward direction
Figure 4.5 Relationship of intake or transition and survival rates at different levels of
education with GDP per capita at PPP
GDP per capita at PPP (2000 $ in thousands)
1
0
2
0
3
0
4
0 0
5
0
120
0
20
40
60
Percent
80
100
Female survival
120
0
20
40
60
Percent
80
100
Male survival
120
0
20
40
60
Percent
80
100
Female intake or transition
120
0
20
40
60
80
100
Male intake or transition
Upper secondary Lower secondary Tertiary
1
0
2
0
3
0
4
0 0
5
0
Primary
Percent
Source: IFs Version 6.12 using UIS and WDI data (most recent by country).
on the assumption that demand for education
in countries will very seldom actually decline
(extreme budget pressures, as we shall see, can
sometimes force a decline).
19
A general upward bias in convergence
patterns is also consistent with another factor
built into the formulation for adjusted primary
net intake, namely, a systemic shift. Figure
4.6 shows the upward systemic shift of intake
rates for males in relation to GDP per capita
during the 1990s, reecting the additional
emphasis that individuals and governments have
placed on education in recent years regardless of
income levels. Some of this shift may be due to
a greater need for education in order to compete
for jobs in an increasingly knowledge-based
economy. Some, however, may also reect simple
competition for relative position by increasingly
well-educated individualsthe credentialism
or sheepskin effect (Hungerford and Solon
1987). It also seems reasonable to suspect that
the greater emphasis on education in recent
years has an ideational component, not solely
material ones. We compute an ongoing systemic
shift at the primary level based on the pattern
of recent years, subject, of course, to saturation
effects as levels move higher.
More generally, the education model
uses this constellation of elements (GDP per
capita, historic uniqueness of countries and
their movement toward convergence over an
extended time frame, and the representation
of systemically shifting patterns) in its basic
formulations for intake, transition, and survival
rates. Illustratively, that for adjusted primary
net intake is:
ANIR
t
= F (GDPPCP, ANIR
t=1
, Converge, SS)
where
ANIR is adjusted net intake rate
GDPPCP is GDP per capita at purchasing
power parity
Converge is a fractional movement toward the
estimated function
SS is systemic shift (upward) of the function
across time
On top of these formulations, used in exploratory
analysis, the model also makes it possible
for the user to target growth rates for more
normative analysisfor instance, by replacing
the function with specication of a 2 percentage
point annual increase of intake rates across a
forecast horizon (the 2 percentage points would
be effective at the midlevel in the range of
intake rates, tapering to zero as the intake rate
approaches 100 percent, because of an S-shaped
representation). Chapters 6, 7, and 8 reect this
type of normative use of the model.
Costs of education and public spending
The education of each student has a cost,
differing by level of education and generally
rising across levels. Countries vary greatly,
however, in what they spend per student at
each level, and patterns also can change rather
dramatically over time. Because spending per
student is a key variable, its determination in
forecasts is another relationship that greatly
inuences or dominates model behavior.
In the context of developing a normative
scenario for education futures, Chapter 6
provides an extended discussion of spending
on education, both per student and in the
aggregate. With respect to spending per student,
it attempts to tease reasonable target levels
out of existing analyses of good practice and
from cross-sectional patterns of spending by
educational level as a function of GDP per capita
(see Figures 6.56.7 for cross-sectional patterns).
Typically, and especially at the upper secondary
and tertiary levels, spending per student begins
at quite high proportions of GDP per capita when
income and enrollment levels are low, reecting
high cost structures. It falls as enrollment and
IFs reects global
patterns of change
through gradual
convergence of
country-specic
path dependencies
to typical paths
over long time
frames.
Figure 4.6 The changing relationship of adjusted primary net intake and
GDP per capita at PPP over recent years
120
70
80
90
100
110
2
0
4
0
3
0
5
0 0
Primary net intake rate (1992) Percent
Primary net intake rate (2000)
1
0
Source: IFs Version 6.12 using most recent UIS and World Bank data available in IFs.
income rise at higher levels of education, but it
can rise with income at the primary level.
For the purposes of both exploratory (where
we seem to be going) and normative (where we
might like to go) forecasting, the IFs project
draws upon these analyses of good practice and
cross-sectional patterns to anticipate future levels
of spending per student. For exploratory analysis,
we assume that levels of spending per student
will very gradually converge from empirical initial
conditions to target levels. For normative analysis,
we posit considerably more rapid convergence.
Key dynamics
Education and broader systems of human
development interact closely over time. For
that reason, the International Futures system
integrates the education model with the detailed
demographic, economic, and sociopolitical
models. The dynamics of those other models
or systems become, in essence, part of the
dynamics in the education model. Earlier
discussion sketched the structures and features
that determine the dynamics in those other
models. In this section, we return briey to note
some of the relationships across models.
With respect to important dynamics within
the education model itself, sound accounting
systems assure us that student ow patterns
are internally consistent and connected well
with tracking of adults education attainment.
They also help us to track government
revenues and expenditures and to identify how
much funding governments might direct to
education. The key specications of demand
for education at all levels and of potential
spending per student will, if resources are
available, dominate the forecasting patterns
and allow calculation of the funding required
if societies were to meet that demand. In an
integrated modeling system (and in the real
world of competing demands for resources),
however, initial calculations of supply and
demand seldom do balance. Instead, there will
be complex dynamics of interaction between
the two at any point in time and across time,
and this section sketches that process as well.
Linkages (backward and forward) of education
to other systems
The discussion of the education model to this
point has indicated the important linkages
from the economic model to education
demand and to spending per student, both
involving GDP per capita (backward linkages
from the perspective of the education model).
Enrollment rates translate, of course, into
student headcount only with the help of age-
sex structures from the demographic model.
This volume most often presents and discusses
forecasts of enrollment rates, but the tracking
in the IFs system of enrollment numbers is
essential to analysis of the ability of societies
to meet demand for education.
The government budget submodel of IFs
determines the amount of funding for education.
That submodel forecasts total government
revenues and expenditures, using a social
accounting matrix to embed them in the larger
system of domestic and international nancial
ows and to maintain accounting consistency.
Expenditures include both transfer payments
and direct spending on the military, health,
education, research and development, and
other programs. Spending on education tends
to increase as societies become wealthier, but
so does other spending, including that on
health. The government budget model balances
competing demands.
In addition to these backward linkages to
other systems, the IFs system represents a
number of forward linkages from education to
demographic variables (for example, fertility),
economic variables (for example, productivity),
and sociopolitical variables (for example,
democracy level). Chapter 8 will return to these
linkages. The existence of both backward linkages
from education to drivers in other systems and of
forward linkages from education to other systems
creates important feedback loops (see, again,
Figure 4.2), which Chapter 8 will also explore.
The reconciliation of budget demand
and supply
What if spending demand from the education
model and spending supply from the government
budget submodel do not match? The reality is,
of course, that they will not, which requires an
algorithmic process of reconciliation. Imbalances
between supply and demand for funding set up
a multistage problem of allocation in the model,
as in budgeting systems. Moreover, they set up
both the need for immediate balancing and the
need for incremental changes in longer-term
IFs uses
analyses of good
practice, cross-
sectional patterns,
and country-specic
starting points
to forecast per
student spending
at each education
level.
patterns. The theoretical framework for handling
budget reconciliation is fundamentally that of
incremental decisionmaking and budgeting.
20
The rst decision issue for immediate
resolution is determination of the total public
spending on education. Given the tendency
for governments to spend on average just
under 5 percent of GDP on education quite
independently of GDP per capita (except for the
poorest countries) and the persistence of that
average level across time, it is clear that there
is a substantial degree of top-down inuence in
determination of the ultimate budget. At the
same time, however, the need for governments
to be responsive to changing demographic
patterns and enrollment growth is obvious.
Thus, for exploratory analysis, we presume
that the balance of forces is predominantly
governmental but not exclusively so (with a
roughly sixty-forty weighting). One immediate
forecasting implication of this is that countries
such as the Democratic Republic of Congo, which
mobilize resources very poorly and direct very
little of them to education, cannot be expected
to have strong educational futures unless those
patterns change.
A second-stage decision issue for resolution
is the balance of spending across levels of
education. The model does that (whether the
budget is in decit or surplus) proportional to
the initial demand for funding (student numbers
times cost per student). Thus, the forecasting
of intake/transition and survival rates and the
student numbers to which they give rise, in
combination with the forecasting of costs per
student, determine this allocation.
A third-stage decision issue is allocation of
spending surpluses or decits at each level of
education between student numbers and costs
per student. The algorithmic structure at this
stage is somewhat complicated by the need
to contend with imbalances already existing
between enrollment and spending patterns and
targets for them; for instance, if enrollment
is already above the target and spending is
below, it would make no sense to adjust both
upward in case of a budget surplus. Some
preliminary adjustments incrementally correct
such imbalances, and then, on the whole,
proportionally comparable adjustments change
enrollment drivers (intake/transition and
survival) and spending per student.
21
Finally, in order to facilitate adjustments
in future years, we compute moving average
multipliers that carry forward the magnitude
of adjustment to each term over time. These
multipliers smooth the adjustment processes
across time, allowing the overall system to chase
equilibrium, even if it never completely nds it.
In normative analysis, the budgetary link
between demand and supply sides can be
turned off, and the demand can force the
spending on the supply side. Chapters 6
through 8 explore this use of the model. Even
in this situation, however, it is important that
the required spending on the demand side be
accounted for in the government budgeting
model, thereby reducing funds available for
expenditure in other areas and/or requiring
additional government revenues. The social
accounting matrix of the IFs economic and
sociopolitical models makes it possible to do
this and to trace the consequences (such as the
impact on health spending) of higher or lower
spending on education.
Initializing the model
Initializing forecasts when data are scarce is hard
work. But some algorithm ought to be able to do
it. The historical series that constitute the IFs
database begin with 1960, whereas the base year
of the IFs education simulations (and those of
the larger IFs system) is 2005. Data values from
2005 initialize the model for forecasting. The base
year values of student and education nancial
ows come from the UIS, base year demographic
data are primarily from the United Nations
Population Division, economic data come heavily
from the World Bank, and sociopolitical data
are from many disparate sources.
22
Before the
model can use these data, however, signicant
data extension, cleaning, and reconciliation are
necessary, including the estimation of base year
values when they are missing for a country.
Data extension
UIS provides student and nancial ow rate data
for many measures of education participation,
particularly at the primary level. In some
instances, we had to modify the data to make it
meaningful for a long-term model. For example,
our representation of net intake rates combines
the very strictly of-age rates that the UIS reports
with students one year over- and one year
Reconciling
demand for
education with
availability of
funding is a
multistage and
incremental
process.
When data are
missing or are
incongruent across
related variables,
algorithms help
with data extension,
estimation, and
reconciliation.
underage (also reported by the UIS) to create an
adjusted net intake rate. Box 4.1 provides more
detail on the concept and the rationale for our
use of the adjusted net intake rate.
23
In other instancesfor example, with
respect to the division between lower and upper
secondary levelsUIS provides raw data (e.g.,
headcounts) but fewer ow rates than at the
primary level. In the instances when UIS provides
ow rates, we import them directly into the IFs
historical series. When UIS provides only raw
data (e.g., grade-by-grade headcounts or total
expenditures and total number of students), we
calculate rates from these series ofine, using
spreadsheet applications or auxiliary programs.
We then enter the results into the IFs historical
series. Examples of data series handled this way
include the lower and upper secondary survival
rates, the transition rate from the lower to the
upper secondary level, the tertiary intake rate
and the overall tertiary graduation rate, and
differentiated per student costs at the lower
and upper secondary levels. We also use such
auxiliary processing to create the data series for
the adjusted primary net intake rate.
Data cleaning and reconciliation
Whenever possible, we use an important
automated subsystem of IFs that we call
the preprocessor to help prepare initial
conditions. The preprocessor uses algorithms
that simplify the preparation of initial
conditions from the raw data. Among other
benets, the preprocessor makes possible rapid
recomputation of initial conditions when a new
data update becomes available. The two major
functions of the preprocessor are (1) lling
missing base year values, and (2) reconciling
incongruent data or estimates.
Filling missing base year values
When 2005 data are missing for a country, the
IFs system estimates 2005 values rather than
excluding the country from forecasts. We apply
the following estimation techniques, normally
in the order listed: (1) using the most recent
data point for the country if it is temporally
proximate; (2) calculating an imputed data
point from a longitudinal temporal regression if
a recent data point is not available but a longer
historical series exists; and (3) estimating the
data value from a cross-sectional relationship
stored in the system, most often as a function of
GDP per capita at PPP.
There are some specialized algorithms
in the preprocessor or rst model year to
handle particular issues. For instance, one
Table 4.3 Primary intake rates by age categories (2005)
Of-age
Overage
1 year
Underage
1 year
Overage or
underage
2 or more
years
Arab States 62.9 9.4 17.7 6.6
Central and Eastern Europe 73.9 12.0 9.0 1.5
Central Asia 67.0 17.1 19.7 6.3
East Asia and the Pacic (Poorer) 59.8 18.8 14.9 12.8
Latin America and the Caribbean 69.6 16.7 14.4 7.6
South and West Asia 72.2 23.3 1.6 17.4
Sub-Saharan Africa 49.0 25.2 10.6 18.5
East Asia and the Pacic (Richer) 88.8 14.3 0.8 1.4
North America and Western Europe 78.9 16.4 1.8 2.2
World 64.3 17.8 12.4 9.9
Source: Compiled from UIS data (unweighted country averages).
In developing countries, the difference between gross and net intake
rates is often great, particularly in early stages of the transition to
broader education participation as overage students take advantage
of an increased emphasis on education and growing opportunities to
enroll. However, as is strikingly evident from Table 4.3, far more entering
children are just one year over or under system-dened entry ages rather
than two or more years older or younger. This entry pattern often persists
indenitely, even as the rates of children two or more years away from the
system-dened entry age decline over time.
Were our focus only on the entry of students of precisely ofcial entry
age, we would, in our view, discount the progress countries are making with
respect to the timely entry of appropriate-age students. Hence, in place
of the conventional net intake measure, we simulate an adjusted primary
net intake rate, which is the intake rate of children at the ofcial system-
dened entry age plus the children one year above or below that age. The
difference between the adjusted net intake rate and the gross intake rate
then becomes the indicator of divergence from age-appropriate universal
primary intake. In our model, simulated primary gross entry rates gradually
converge toward the adjusted primary net intake rate as more students
enter on time and the pool of potential late entrants diminishes.
Box 4.1 Adjusted primary net intake
segment of code computes the annual size of
an overage pool of out-of-school children
potentially available for primary intake. A
particularly important specialized process
takes the education attainment data (which
are not provided from original sources by age)
and spreads the attainment levels across age
categories in order to initialize the ongoing
calculation of attainment described earlier. This
spread process takes into account the percentage
of the adult population with a certain level of
education, the current completion rates at that
level of education, and the age structure of the
adult population. Knowing that completion rates
almost always exceed the average attainment
levels (that is, education participation is
increasing over time and therefore decreases
across progressively older cohorts), a factor for
age-related decline in attainment levels can be
computed in an iterative process.
24
Reconciling incongruent data or estimates
Incongruities among the base year primary ow
rates (intake, survival, and enrollment) can
arise either from reported data values that, in
combination, do not make sense or from the
use of stand-alone cross-sectional estimations
to ll holes. Such incongruities might arise
among ow rates within a single level of
education (e.g., primary intake, survival,
and enrollment rates that are incompatible)
or between ow rates across two levels of
education (e.g., primary completion rate and
lower secondary intake rate).
The IFs education model uses algorithms to
reconcile incongruent ow values. They work
by (1) analyzing incongruities, (2) applying
protocols that identify and retain the data or
estimations that are probably of higher quality,
and (3) substituting recomputed values for the
data or estimations that are probably of lesser
quality. For example, at the primary level, data
on enrollment rates are more extensive and more
straightforward than either intake or survival
data; in turn, intake rates have fewer missing
values and are arguably more reliable measures
than survival rates.
Conclusion
The purposes of our modeling are to enhance
understanding of where the global education
transition appears to be taking education
systems and to create and explore a normative
scenario that might accelerate that transition.
Past models and tools provide a strong basis for
insights into how to structure such a model,
and in building the education model in IFs,
we have drawn upon them. In the process, we
have created a model that is structurally based
and agent-class-driven, that represents formal
education at all levels, and that is integrated
quite tightly with models of demographic,
economic, and sociopolitical systems.
Although we believe our education model
to be a strong one, it is imperfect. Modeling
and forecasting systems simplify reality, in part
to allow us to better understand its dominant
structures and dynamics. In fact, this is the
fundamental reason we do modeling. Simplied
representations help us clarify and extend our
own mental models of the system of interest
to us. They also allow us to think about how
systems might be unfolding and therefore to
produce forecasts with and without modeled
interventions. However, we should never confuse
forecasting with prediction.
In this spirit, we will turn in succeeding
chapters to the use of the IFs modeling system
and to its forecasts of further advances in
global education, including some comparison
with forecasts produced by other tools and
approaches. Chapter 5 and subsequent chapters
will look explicitly to the future, presenting
both the base case of IFs and alternative
forecasts developed around it.
Appendix to Chapter 4: Education
Modeling and Forecasting Approaches
McMahon
Education and Development: Measuring the Social
Benets describes McMahons development and
implementation of an econometric interactive
macrodynamic model (McMahon 2002: ix), the
purpose of which was exploration of the social
benets of education (McMahon 2002: 179).
Analyses of the social benets of education
included both direct and indirect impacts of
primary and secondary education on economic
development; on population growth via health
and fertility; on democracy, human rights, and
political stability; on poverty and inequality; on
the environment; and on crime.
The book began with a base case or
exploratory analysis that assumed the
Our structurally
based, agent-
class-driven
model represents
formal education
at all levels and
is integrated with
demographic,
economic, and
sociopolitical
models.
continuation of past education policies and the
associated unfolding of education and its impacts.
Simulations of two specic normative education
policy changes and how they might enhance the
extension of education and its impacts moved
beyond the base case. The rst normative scenario
was built on a 2 percentage point increase in
public investment in education as a percentage
of gross national product (GNP); the second
normative scenario assumed a 20 percentage point
increase in male and female secondary education
enrollment rates (McMahon 2002: 185186).
The analysis used historical data for the period
from 1965 to 1995, and its forecasts extended to
2035; the countries included (78) were those for
which consistent data were available on all key
variables. The model incorporated empirically
tested varying time lags between changes in
primary and secondary education and changes
in other components of development; it included
bidirectional feedback loops; its mathematical
equations incorporated the concept of long-run
equilibrium relationships; and its parameters
were estimated from cross-sectional analyses
examining relationships between variables across
many countries at one point in time. The major
contributions of the model were (1) the inclusion
of both primary and secondary education, (2)
the placement of the broad direct and indirect
social impacts of education at the center of
analysis,
25
and (3) an effort to calculate a net
return associated with extending education
participation.
26
Delamonica, Mehrotra, and Vandemoortele
In a study published by the United Nations
International Childrens Emergency Fund
(UNICEF), Delamonica, Mehrotra, and
Vandemortele (2001) projected the incremental
costs associated with moving from the level
of each developing countrys net primary
enrollment rate in 2000 to universal primary
education for all developing countries (128) by
2015. The project was undertaken to update
global and regional cost estimates for the
period from 2000 to 2015, in distinction from
cost estimates based on enrollment patterns in
the early to mid-1990s (Delamonica, Mehrotra,
and Vandemortele 2001: 2). The purpose and
approach were normative.
Delamonica et al. used population projections
from the UN Population Division as a basis for
their cost projections. They then assumed the
increases needed to bring each countrys net
enrollment rate to 100 percent by 2015 would
occur in a linear fashion. They also made the
uniform assumption that all countries would
absorb any incremental costs arising from
population trends at constant enrollment rates
(i.e., that a country could nd the resources
to educate the then-current proportion of its
population of school-age children, no matter
what size that school-age population would be
over the 20002015 period). Their methodology
held GDP per capita constant for the period from
2000 to 2015.
Delamonica et al. estimated costs in four
discrete categories: (1) recurrent expenditures
related to net enrollment rate increases; (2)
quality improvements, as reected by an
adjustment in unit costs to allow 15 percent
of recurrent costs for nonwage items (e.g.,
instructional materials) without a reduction in
teacher salaries; (3) reducing pupil-to-teacher
ratios to an average of 40; and (4) capital
costs for those countries where the increase
in students from the expanded net enrollment
ratio would be greater than the decrease in the
school population from trends in the decline of
expected births. Items (1) and (4) were added
to the costs as increases in the net enrollment
ratio brought new students into the school
system, whereas items (2) and (3) were added
across the school population in the rst year
of estimated costs (Delamonica, Mehrotra, and
Vandemortele 2001: 1213). Although the only
dynamic element of the model was its use of
school-age population projections as the basis
for estimating costs of a linear increase to UPE,
the model provided a framework for considering
various cost components (including quality
improvements and capital outlays) and overall
resource requirements to be met domestically
(through economic growth or reallocation of
government funds) or by international donors.
Bruns, Mingat, and Rakotomalala
Bruns, Mingat, and Rakotomalala (2003)
authored a seminal normative study entitled
Achieving Universal Primary Education by 2015:
A Chance for Every Child. The study utilized
a simulation model developed by Ramahatra
Rakotomalala and subsequently adopted for use
by the countries selected to participate in the
Education for All Fast-Track Initiative sponsored
by the World Bank, UNESCO, UNICEF, and the
regional development banks.
The study provided a detailed analysis of
the forty-seven low-income countries that
were furthest in 2000 from the MDG goal of
universal primary education, with an estimate
added for Afghanistan (Bruns, Mingat, and
Rakotomalala 2003: 20). The study focused on
estimating, under certain normative targets
or benchmarks, the following: (1) what it
would cost to achieve the goal in terms of
incremental funding between 2000 and 2015,
27

(2) the portion of that funding that developing
countries could afford under the assumption of
a 5 percent economic growth rate applied across
all countries, and (3) where and how much
international assistance would be needed.
World Bank task teams collected enrollment
data for the then most recent year (usually
2000) directly from the education ministries of
the forty-seven low-income countries included
in the study. UNESCO-published data (usually
for 1997) were used when more recent data were
not available from the education ministries.
Population data were from the United Nations/
World Bank population database used by the
World Bank (Bruns, Mingat, and Rakotomalala
2003: 39, 41).
The study began with an exploratory
analysis of the characteristics of the low-
income countries that were making accelerated
progress toward UPE in 2000, compared to
countries that were not (2003: 8). From
this empirical analysis, a best practices or
normative framework was created to provide
guidelines for policy levers to achieve universal
primary completion at minimum adequate
cost (2003: 109). The framework included
benchmarks or targets for quality improvements,
for efciency improvements, and for domestic
resource mobilization (2003: 82). The nancing
benchmarks included a cap on the portion of
educational expenditures from government
revenues going to primary education in order
to avoid stripping resources from secondary
and tertiary education. The various benchmarks
were combined in different ways to produce
alternative scenarios, and incremental costs
(including estimates of gaps in domestic funding
capacity) were generated for each scenario.
The approach connected education to broader
systems via population projections, benchmarks
for funding and resource mobilization, and
the inclusion of an economic growth rate
assumption (albeit a single assumption for
all countries). The model also moved toward
inclusion of enrollment dynamics by measuring
costs associated with a targeted upper limit on
repetition rates and a targeted completion rate
rather than an overall enrollment rate alone.
Clemens
In 2004, Michael Clemens, through the Center
for Global Development, authored a provocative
background paper for the Millennium Project
Task Force on Education and Gender Equality.
His focus was on understanding if there is a
typical primary schooling transition pathway
in developing countries and also on the degree to
which the transition to mass primary education
can be accelerated by government policies.
28

Clemens raised these questions in the context
of exploring the feasibility of meeting the MDG
goal of universal primary education by 2015;
his approach was primarily exploratory, but it
included some elements of normative analysis.
To explore these questions, Clemens
developed an aggregated ow model that
focused on transition speeds as measured by
overall net primary enrollment rates, using
administrative data compiled by UNESCO eld
ofces from school registers for the years from
1960 to 2000 for over 100 developing countries.
Based on the typical developing country
experience between 1960 and 2000, he produced
S-curve extrapolations
29
of the number of years
individual countries and multicountry regions
might need in order to reach 90 percent net
primary enrollment (Clemens 2004: 42, 52). He
used the same UNESCO data to model a typical
gender transition speed in primary and
secondary enrollment.
Clemens explored three other dimensions in
his evaluation of the feasibility of achieving
UPE by 2015: (1) he compared the 19602000
transition rates in developing countries with
the rates of todays rich countries during
their earlier transitions to universal primary
education; (2) he estimated the necessary
transition speeds if todays developing countries
are to meet the 2015 goal of universal primary
education (Clemens 2004: 55); and (3) he
used cross-country data from 1980 to explore
relationships between education transition rates
and a number of social, economic, and education
policy variables at that point in time (Clemens
2004: 45). He also reviewed an extensive body of
literature on drivers of education participation,
pointing out in particular the importance of
the relationship between household income and
parental level of education in family schooling
decisions (that is, the role of demand factors).
Clemenss approach did not include costs
or the specics of enrollment dynamics (e.g.,
intake, repetition, or completion). Instead, it
focused needed attention on a number of other
critically important aspects of the education
transition: (1) the importance of policy
attention not just to education availability and
supply but also to circumstances that inuence
demand for education, (2) the importance of
placing the transition to mass primary education
in developing countries in a longer historical
context in order to set aggressive but realistic
goals with respect to time frames, and (3) the
need to take individual countries circumstances
into account in setting goals.
Wils, OConnor, and Somerville
A model developed by Wils, OConnor, and
Somerville, and published in a 2005 paper by
Wils, Carrol, and Barrow for the Education Policy
and Data Center,
30
focused on the concept of
growth paths toward universal primary education.
The context, as it was with Clemenss study, was
an exploration of the feasibility of meeting the
MDG goal of universal primary education by 2015
(Wils, Carrol, and Barrow 2005).
The model dealt exclusively with student
ows at the primary level, and it operated by
extrapolation of ow rates without regard to
population dynamics or resource requirements
and availability. However, it advanced the
conceptualization and implementation of ow
dynamics in modeling by using and comparing
two measures of primary education coverage
entry and completionrather than using a
single overall enrollment rate. Average entry and
completion rate patterns were estimated for each
of seventy low-income countries for the historical
period from 1950 to 2000, and then they were
projected forward in S-shaped extrapolations. The
descriptor used by Wils et al. for the trajectories
of the paths was the number of years it will take
each country to go from a primary completion
rate of 10 percent to a primary completion rate of
90 percent, represented as T10-90 (Wils, Carrol,
and Barrow 2005: 10).
A signicant difference between this
and earlier models was its use of household
surveys and population censuses (rather than
administrative data) as preferred data sources
(2005: 2).
31
Another difference was its use of a
backward-looking lens to establish historical
entry and completion rates. The authors
divided the population of 15- to 65-year-olds
participating in the household surveys between
1999 and 2001 into single-year age cohorts,
and they used the percentage of the cohort
that was 14 years old in each year from 1950
to 2000 and that reported having at least some
primary schooling in order to estimate primary
entry rates for each of those years. They applied
a similar methodology, using 19 as the age, to
estimate primary completion rates for the same
historical period. The use of household data also
allowed identication of out-of-school children
by various subcategories (e.g., in subpopulations
within the country) as well as analyses of
inequality across groupings, such as entry and
completion gaps between urban males and rural
females (2005: 39).
Lutz, Goujon, and Wils
A paper by Lutz, Goujon, and Wils, also published
by the Education Policy and Data Center in
2005, elaborated the application of what is
described as a multistate demographic method
to forecast the extent of education attainment
among adult populations.
32
The model used in
this paper was primarily exploratory but included
some normative aspects, and it differed from
the previously discussed models and tools in a
number of important ways. First, its focus was
not education system ows but rather the stock
of human capital as reected by the education
attainments of a population by age and sex across
four categories: no education, primary education,
secondary education, and tertiary education.
33

Second, other than purely extrapolative
enrollment trend projections produced by UNESCO
in the 1980s and 1990s,
34
it was, so far as we
know, the rst model or tool to look across all
levels of formal education. Third, by focusing on
education levels in the adult populationand,
further, on education levels by age and sexit
facilitated exploration of the relationships
between levels of education in a population and
other human development systems (e.g., fertility,
life expectancy, and economic growth).
The developers initialized the model with UN
population data by age and sex extending back
to 1937 and produced visual representations
of population characteristics in three pilot
countries (Guinea, Zambia, and Nicaragua) in
2000, in population pyramids by age and sex in
ve-year intervals. Education attainment levels
from USAID Demographic and Health Surveys
were superimposed upon the pyramids, as were
estimated fertility levels and infant and child
mortality levels by mothers education from
the Demographic and Health Surveys. Moving
forward, projections of the population pyramids
to 2030 reected expected impacts of changes
in education attainment, with both fertility and
child mortality decreasing as education increases.
At the time the paper was written, the model
used stylized rather than dynamically formulated
assumptions to advance the pyramids to 2030
with respect to initial entry rates and transition
rates between one level of education and the
next. The authors created three scenarios, using
differing stylized or normative assumptions
as follows: (1) constant (current) entry and
transition rates, (2) trend entry and transition
rates, and (3) MDG goal fullling entry rates
(with constant or trend transition rates).
The IIASA model and its pyramidal displays
clearly illustrated that education is a long-
term investment by showing the time lag
between increases in education attainment
among young members of a population
and increases in the overall structure and
pattern of human capital stock in the total
population. Further, the authors pointed out
that by using the distribution of educational
attainment as an indicator of human capital
(rather than a single population-wide measure
of average years of schooling), it was possible
to explore relationships between age, sex,
levels of education, and other variables (e.g.,
health, poverty, and economic growth). They
also pointed to the possibility of subnational
forecasts, as the methodology can be applied
to any population that is clearly dened and
for which there is the necessary information
by age, sex, and level of education (2005: 33).
Work continues at IIASA to extend the model
to a large number of countries and to further
exploration of relationships between age,
sex, levels of education, and other variables;
one example is a backward-reconstruction of
populations by age, sex, and level of educational
attainment for 120 countries for the period from
1970 to 2000 (Lutz et al. 2007).
Hilderink
A 2007 working paper authored by Hilderink
described an exploratory education module
being developed and embedded in the
established PHOENIX dynamic population and
health model at the Netherlands Ministry of
Health and Environment (Hilderink 2007). At
the time the paper was prepared, the PHOENIX
education module used enrollment rates as the
single measure of education ows; however, the
author stated the plan was to use intake and
drop-out rates in a subsequent phase. Education
attainment levels and literacy are other
components of the model. Geographic regions
are the unit of analysis, and the model extends
across primary, secondary, and tertiary levels.
The model was initialized with education
data from UIS and economic data from the
World Banks World Development Indicators.
Simulations for the period from 1950 to 2000
were being used to calibrate and validate the
model; the paper provided forecasts for the
period from 2000 to 2025, but it mentioned a
simulation period extending to 2050. The model
includes bidirectional connections between
education levels, mortality, and fertility. It also
introduces the concept of education demand and
education supply by dynamically connecting both
enrollment rates and education expenditures
to GDP per capita, with the assumption that
demand and supply are equal. Although still in a
developmental stage, the model is being designed
as a comprehensive tool for forecasting education
dynamics in conjunction with a number of
connections to broader systems.
Wils, Barrow, Oliver, Chaluda, Goodfriend,
Kim, and Sylla
An EPDC background paper prepared for the
2008 Education for All Global Monitoring Report
described and presented initial results from
ProEnrol, a country-level cohort-projection
model being developed by the Education Policy
and Data Center (EPDC 2007b). The measures
of student ows forecast in the paper were
primary and secondary enrollment rates, student
headcounts, and gender parity indices, under
the exploratory assumption of each countrys
continuation of its current education policies.
Cohort-projection models focus on the
grade-by-grade dynamics of student ows.
Countries often use them to meet education
system operational planning needs by
projecting numbers of students by grade
and level and the school resources therefore
needed.
35
Individual countries may also use
them to understand the dynamics of their
student ows and to assess education system
functioning with respect to these dynamics
(entry, promotion, repetition, dropout,
reentry, survival, and completion).
ProEnrol, however, is intended for a larger-
scale use. As the background paper noted: The
Cohort projection model developed by the EPDC
is the rst effort to make cohort projections
in an international, global series and is done
here on an experimental basis. The intention
of the GMR [Global Monitoring Report] at this
point is to test this method (2007b: 69).
The background paper included projections
of primary net enrollment rates and student
headcounts for 60 countries using ProEnrol,
projections of primary gross enrollment rates
and student headcounts for 129 countries using
ProEnrol, and projections of secondary net
enrollment rates and student headcounts for 82
countries using ProEnrol (2007b: 70). The EPDC
made projections for two points in time, 2015
and 2030, and calculated a gender parity index
for each projected series.
ProEnrol was initialized with historical
enrollment data (1999 forward) from UIS
on pupils by grade (females and both sexes
combined) at the primary and secondary levels.
It used United Nations medium population
projections for the period from 2000 to 2025
to calculate gross and net entering school
populations (headcounts) by multiplying the
projected population of school entryage
children by projected gross and net entry rates.
Extrapolations from past trends in intake rates
were used to project future intake rates, and
country-specic constant values (equal to the
most recent year data were available) were used
for promotion and repetition rates.
36
The models only linkage with systems
outside education at the time of the background
paper was its use of UN population projections
of school entryage children as a foundation
for projecting school enrollments. For example,
it did not calculate resource requirements or
compare potential enrollments with estimates
of resource availability. However, it would seem
those components might rather easily be added,
since the model calculates student headcounts,
and per student costs are widely available from
UIS. Perhaps more important at this stage, a
protocol needs to be developed for projecting
reasonable changes in promotion and repetition
rates over time, since they are ow components
subject to dynamic changes. However, the
existence of possible future improvements
should not detract from the contribution
ProEnrol has already made by developing and
now testing signicant aspects of a grade-by-
grade cohort projection methodology on a global
scale, including the rst specic representations
of promotion and repetition.
1 Either type of model might employ a simple or a
sophisticated methodology. At its simplest, an
exploratory tool might forecast future enrollment
trends by the extrapolation of recent patterns.
Likewise, at its simplest, a normative tool might
consist of a basic mathematical exercise, such as
the calculation of how much primary intake and
survival rates need to increase each year between
now and 2015 to meet the MDG goal of universal
primary education at that time.
2 The time lag between initial changes in intake
rates and the possibility of impacts from increased
education attainment on other aspects of human
development systems is the reason we stipulate
that a forecasting tool with a mid- to long-range
time frame is critical.
3 In an ideal system, the points of intervention are
sufciently actionable that policymakers can
readily discern implications for policy choices and
implementation strategies. However, even more
abstractly dened interventions (e.g., focusing
on increasing survival rates) can be helpful in
providing a course for improved outcomes.
4 McMahon (2009) extended his analysis to the
tertiary level in a book focused on the social and
private benets of higher education.
5 For an introduction to the character and use of the IFs
modeling system, see Hughes and Hillebrand (2006).
6 We emphasize that IFs is not an agent-based modeling
system because it focuses on the aggregated behavior
of agent classes, rather than on the behavior of
individual agents as agent-based models do.
7 The various member organizations of the United
Nations family are a primary data source, but
other sources, such as the World Banks World
Development Indicators, are also used extensively.
8 More technically, the model structure is recursive
(it computes equations sequentially in each time
step without simultaneous solution). It combines
features of systems dynamics (notably, the
accounting structures with careful attention to
both ows and stocks) and econometrics (using
estimated equations for the dynamic behavior of
the agent classes).
9 We have an explicit representation of the S-shaped
path in our normative scenario only. The more
implicit saturation behavior in our exploratory base
case results from the integration of various dynamic
drivers of education ows.
10 Again, the interested reader is referred to www.ifs.
du.edu for further documentation.
11 The rate is calculated as the proportion of those in
the last primary grade who enter general programs
at the lower secondary level the following year. At
the secondary level, UIS enrollment rates include
students in both general secondary programs and
vocational secondary programs. However, UIS
transition rates to lower secondary are for general
secondary programs only, and our model follows that
convention in our calculations and forecasts of both
lower and upper secondary transition and survival
rates. We maintain country-level gender-specic
vocational enrollment rates as a constant percentage
of ofcial lower and upper secondary school-age
populations, reect them in enrollment data, and use
them to bound transition and survival rates in lower
and upper secondary general programs.
12 A future volume in this series will focus on
infrastructure and will include more differentiated
analyses and forecasts of tertiary education,
making use of program-specic UIS tertiary data to
initialize the model.
13 UIS has a data series for tertiary entry rates.
However, we developed the procedure described
earlier because the UIS series has data for only
about 30 percent of all countries and for very few
developing countries.
14 The multistate demographic method developed and
utilized by IIASA does include education-specic
mortality rates.
15 The current IFs education model tracks adult
age-sex-education categories by ve-year intervals
rather than one-year intervals. In a model with
a one-year time step, as IFs is, this means that
one-fth of each cohort advances annually. In an
environment of increasing education participation
and attainment, the process creates some
degree of numerical diffusion as a portion of the
educational attainment assigned to the youngest
cohort advances too rapidly to the next cohort (a
process sometimes called numerical diffusion). This
means also that some of the stock of educational
attainment ages and dies too rapidly, slightly
exacerbating the underestimate.
16 IFs generates all such relationships for males and
females separately in order to capture sex-related
variations in education participation patterns vis--
vis GDP per capita.
17 See McMahon (1999: 1314) on the manner in
which cross-sectional analysis helps represent
patterns of long-term change.
18 Convergence periods in IFs range from 20 to 100
years; most are between 40 and 70 years.
19 We do not apply a similar constraint to primary
gross intake rates, as they in fact typically
overshoot 100 percent during a rapid education
transition and then either rapidly or slowly decrease
to just above or below 100 percent as adjusted net
intake rates approach 100 percent.
20 As early as 1940, V. O. Key drew attention to the
central question: On what basis shall it be decided
to allocate x dollars to activity A instead of activity
B? (1940: 1138). Later experts in the eld of
public nance (for example, Wildavsky [1988])
helped establish incrementalism as the dominant
paradigm to explain budgeting processes and
decisions. As Lindblom (1959: 81) put it, political
decisions are made more through successive
limited comparisons than through any square
one comparison among possible alternatives.
Allison (1971) contrasted three decisionmaking
models, and his models of organizational process
and bureaucratic politics are closer to reality for
social and budgetary policy than is the rational
actor model.
21 Even an algorithmic representation of an
incremental decisionmaking process requires
parameter specication. In contrast to the
statistical estimation procedures used for functions
such as those driving intake/transition and
survival, as well as those setting targets for per
student spending and total government spending
on education, that algorithmic parameterization is
done via analysis of the behavior of the model, a
process that modelers commonly call tuning.
22 The maintenance of the IFs database is an ongoing
process, and data from major sources are updated
at regular intervals. In the preparation of this
volume using IFs Version 6.12, our most recent
download of UNESCO data followed UISs September
2008 update. We also used the 2008 version of the
World Banks World Development Indicators and the
2006 Population Updates from the UN Population
Division. The model was initialized with 2005
values from those sources. Data from those sources
for more recent years, as available, were used in
cross-sectional analyses, where our convention is to
use the most recent years data available for each
individual country.
23 At one extreme, in Indonesia about 60 percent of
children one year below the ofcial age enter every
year, a higher entrance rate than for of-age children.
24 Weishuang Qu of the Millennium Institute provided
information on that approach, used also in the T-21
model.
25 We note that McMahon chose not to follow
economists more frequent convention of referring
to market and nonmarket returns. By instead
referring to social returns and including economic
development among them, McMahon applied
econometric analysis within a human development
framework.
26 The model used gross enrollment rates as
the single student ow measure in the 1999
publication. In a subsequent paper reporting the
use of the model to assess social outcomes of
education in Africa, primary completion rates were
added as a second student ow measure (Appiah
and McMahon 2002).
27 Like the study undertaken by Delamonica et al., this
study focused on costs associated with achieving
a normative target. Unlike the Delamonica study,
however, Bruns et al. focused on incremental costs
associated both with enrollment rate changes
and with population dynamics and then compared
these total incremental costs within a resource
framework that, although simple in its assumptions,
considered an impact from economic growth.
28 Transition in this context refers to the change
from low to high rates of participation in primary
education.
29 S-curves t broad-scale social changes because
change processes often start slowly, then build
rapidly in a middle range, and slow as they
approach a limit (such as 100 percent).
30 The Education Policy and Data Center was
established in 2004 to contribute to global
education policy and planning through data and
analysis, and it has rapidly become very important
in these roles. It is part of the Academy for
Educational Development (AED) and is funded
primarily by USAID and AED (Wils et al. 2007b: 6).
31 Most typically, these were the USAID-sponsored
Demographic and Health Surveys (DHS) and the
UNICEF-sponsored Multiple Indicator Cluster Survey
(MICS) (Wils, Carrol, and Barrow 2005: 2).
32 Demographic multistate projection models
reect and project the distribution of various
characteristics or states (such as levels of
educational attainment) across a population
(or subpopulations) segmented by age and by sex.
The paper stated that demographic multistate
projection methods were rst developed at the
IIASA in the 1970s (Lutz, Goujon, and Wils
2005: 9).
33 The denition of levels of education attainment
used by Lutz et al. (2005: 16) differs from the
denitions used by many other systems. Lutz et
al. dene no education as never having gone to
school or completing less than one year of primary
education. They place people in the category of
primary education if they complete at least one
year of primary school, in the category of secondary
education if they ever entered secondary school,
and in the category of higher education if they
ever entered tertiary education after completion
of secondary school. The use of these denitions
produces a higher prole of education attainment
than the use of completion measures would, and
it needs to be taken into account when comparing
their results with those of some other models and
analyses, including IFs.
34 Although purely extrapolative, these earlier UNESCO
projections were important and ambitious projects,
particularly because they included all levels of
formal education. One study released in 1989
provided trends and projections of enrollment by
level of education and by age for the period from
1960 to 2025 (UNESCO 1989), and another in 1993
provided updated trends and projections for the
same span of years (UNESCO 1993).
35 Porta and Wils (2007) described and compared
four such detailed education system planning
tools in a 2006 EPDC paper: (1) the World Bank
tool associated with the Bruns, Mingat, and
Rakotomalala project described earlier in this
chapter; (2) the UNESCO Education Policy and
Strategy Simulation Model (EPSSim); (3) the Modelo
de Necesidades de Financiamiento (MNF) model
used by Nicaragua and Guatemala; and (4) the EPDC
Demo Ed Model.
36 The report noted the model could also have used
trend values or user-set values for promotion and
repetition rates. However, the analysis did not use
trend rates because of the extreme projected values
they sometimes produced (EPDC 2007b: 77). The
analysis used grade-by-grade specic repetition
rates when they were available and otherwise
applied the average repetition rate to each grade
(EPDC 2007b: 77).

Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Uploaded by

Copyright:

Available Formats

Forecasting Education

You might also like

Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Uploaded by

Document Informationclick to expand document informationForecasting Education from Patterns of Potential Human Progress Volume 2: Advancing Global Education

Document Informationclick to expand document information

Copyright:

Available Formats

Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Forecasting Education: Forecasting Tools and Models Can Be Exploratory, Normative, or Both

Uploaded by

Copyright:

Available Formats

Forecasting Education

You might also like