W O R K I N G PA P E R

February 2017

Girls’ Schooling and Women’s

Literacy: Schooling Targets
Alone Won’t Reach Learning
Goals

Lant Pritchett and Justin Sandefur

RISE-WP-17/011
February 2017

Funded by:

The findings, interpretations, and conclusions expressed in RISE Working Papers are entirely those of the author(s). Copyright for RISE Working Papers remains with the author(s).
www.riseprogramme.org
 Girls’ Schooling and Women’s Literacy: Schooling Targets
Alone Won’t Reach Learning Goals
By: Lant Pritchett and Justin Sandefur

Abstract
Using the Demographic and Health Survey (DHS) data on the ability of women at various levels
of schooling attainment on the ability of women at various levels of schooling attainment to read
a simple sentence, we show that reaching universal completion of grade six among girls would
not bring the world anywhere close to the goal of universal female literacy. These calculations
are based on the empirical relationship between grades completed and ability to read, a
descriptive ‘learning profile.’ The large literature on schooling and life outcomes suggests simple
correlations are a reasonable guide to causal effects, and the typical concern is over-estimation
of the true return to schooling—implying our calculations using a descriptive and not causal
learning profile are a best-case scenario.

This best case is often not at all good: the learning profile is so weak in Nigeria that even if all
women had completed grade six, adult female illiteracy would only have fallen from 58 percent
to 53 percent. In contrast, children in many other countries do learn to read in much higher
numbers and enrolling out- of-school girls would dramatically reduce illiteracy. For instance, in
Ethiopia the same calculations yield a reduction in illiteracy from 82 to 25 percent. But across
nearly 50 developing countries with available data, our calculations suggest 40 percent of
women would be illiterate even if all women completed at least grade six.

Achieving new Sustainable Development Goal targets of universal literacy and numeracy will
require both achievement of universal schooling and dramatic improvements in the learning
profile in most developing countries.

1
Contents

Introduction ............................................................................................... 1

1. A learning profile from DHS data ......................................................... 5

1.1 Schooling and reading data from DHS ........................................... 5

1.2 A learning profile from DHS data .................................................... 7

2. How much would reaching UPC(6) help reach an SDG

of universal literacy ............................................................................ 10

2.1 Scenario of UPC(6) explained and illustrated............................... 10

2.2 Results for all countries ................................................................. 13

2.3 Aggregate calculations .................................................................. 18

Conclusion .............................................................................................. 20

References ............................................................................................. 21

2
Introduction
The Millennium Development Goals (MDGs) established goals in just eight domains. The target
was modest and only mentions schooling 1 (Pritchett and Kenny 2013):

Target 2.A: Ensure that, by 2015, children everywhere, boys and girls alike, will be able
to complete a full course of primary schooling. 2

The new Sustainable Development Goals (SDGs) shifted from low-bar goals to expressing
millennium development ideals, covering more domains (17 versus 8) with more, higher, and
broader goals in each domain. The education targets shifted from just schooling to schooling
with “effective learning outcomes” and emphasize not just completion of school but universal
literacy and numeracy being achieved for “all youth”:

Target 4.1: By 2030, ensure that all girls and boys complete free, equitable and quality
primary and secondary education leading to relevant and effective learning outcomes.

Target 4.6: By 2030, ensure that all youth and a substantial proportion of adults, both
men and women, achieve literacy and numeracy.

The implicit assumption in schooling targets was that there existed a tight link between “time
served” in schooling and education—the acquisition of learning, skills, competencies, abilities,
values, beliefs and dispositions and the general preparation for adult life that are the real
purpose of education. People treat “schooling” and “education” as interchangeable synonyms
on the implicit presumption that schooling nearly everywhere and for nearly everyone leads to
education.

Tragically, the accumulation of national and international data on learning assessments and the
direct estimates of learning profiles—the descriptive relationship between years of schooling
and a measure of skills or learning—shows the assumption that schooling and learning are
tightly linked is often false. In many countries today the largest fraction of the uneducated
youth—functionally illiterate and innumerate--are schooled.

Spaull and Taylor (2015) use schooling attainment profiles and language and mathematics
results from a cross-nationally comparable assessment in Southern Africa (SACMEQ) to show
that in South Africa 26 percent of 12-year-olds were both functionally illiterate and enrolled in
grade 6 while only 3 percent were illiterate but had not reached grade 6, so 90 percent of the

1 Modest especially by comparison to earlier goals, such as the Jomtien Education for All Declaration adopted in March 1990,

which is much more focused on learning. Article I reads: “Every person—child, youth and adult—shall be able to benefit from
educational opportunities designed to meet their basic learning needs. These needs comprise both essential learning tools (such as
literacy, oral expression, numeracy, and problem solving) and the basic learning content (such as knowledge, skills, values, and
attitudes) required by human beings to be able to survive, to develop their full capacities, to live and work in dignity, to participate
fully in development, to improve the quality of their lives, to make informed decisions, and to continue learning.”
2 http://www.un.org/millenniumgoals/education.shtml

3
cohort of 12-year-olds who were in school and illiterate had enrolled and persisted in school to
grade 6. Eliminating the “unschooled” would reduce illiteracy by just 10 percent. In Uganda 45
percent of 12-year-olds were both functionally illiterate but in grade 6 (Spaull and Taylor 2015).
The Education for All Global Monitoring Report (Rose 2014) reports that in Zambia more than
three-quarters of primary school age children stay in school beyond grade 4 but only 44 percent
of these children are able to read. Similarly, in Nigeria, approximately 80 percent of people aged
15 to 24 who left school after five to six years completed are unable to read a full sentence. 3
Asadullah and Chaudhury (2013) show that in rural Bangladesh knowing a child’s school
completion only very weakly predicts whether the child has even basic mastery of literacy and
numeracy. The household sample based ASER assessments in India (and similar assessments
in other countries like UWEZO) show similar results—that many children who have reached
primary school completion have not mastered basic literacy and numeracy (ASER 2014).

We add to this literature on learning by using DHS data to construct learning profiles which link
the ability to read a simple sentence and the year of schooling completed for women in 51
countries.4 The latest waves of the DHS ask women to read a simple sentence (in their chosen
language) in addition to recording women’s self-reported schooling experience. We use this
data to show that in many countries the learning profile is quite flat—literacy increases only
weakly with more years of schooling. The (unweighted) average across 51 countries was that
only about half of women who completed grade 6 (but no higher) could read a single sentence.

It is important to bear in mind that we estimate correlations between schooling and learning in
the DHS data, not causal parameters. However, the large literature on the Mincerian earnings
function—the empirical relationship between schooling and adult labor-market earnings—has
shown that correlations of this type are a reasonable guide to the true causal relationship
identified through natural experiments and instrument variables strategies (Card 2001). The
typical concern is upward bias in the simple correlations of schooling and outcomes, due to the
tendency for pupils of higher ability to get more schooling. If our descriptive learning profile is
biased upward because the more adept are more likely to persist to higher levels of schooling
this would imply the causal learning profile—the impact on a given individual of more schooling--
is less steep than our descriptive profiles and hence our estimates here are too optimistic and a
policy driven expansion of schooling would produce even less literacy than we suggest.5

Estimating the empirical relationship between schooling and learning allows us to address the
question: how much would the achievement of schooling goals like universal completion of
primary school for girls at the existing learning profiles contribute to a goal of universal female
literacy? In 22 of 51 countries achievement of universal grade 6 completion would still have left

3 http://unesdoc.unesco.org/images/0022/002256/225660e.pdf
4 The DHS data on functional literacy have been used by the UNESCO Dakar+7 report and by the Population Council in
assessing female education and these have generated evidence about literacy by level of schooling.
5 As a caveat, note that the opposite scenario is possible: as Card (2001) notes, out-of-school children may face a high marginal

cost of schooling and thus a higher potential return, though more recent work contends that upward ability bias remains a key
challenge, at least in the U.S. (Heckman et al. 2016).

4
over a third of adult women illiterate as the learning profiles were too flat. A population weighted
average suggests that even if universal grade 6 completion been achieved 40 percent of women
would have still remained illiterate. This is of course no argument against getting every girl in
school but efforts to “Let Every Girl Learn” have to focus as much on girls’ learning once in
school as just getting them to attend a school.

1. A learning profile from DHS data

Demographic and Health Surveys (DHS) have been conducted in over 90 countries, with more
than 300 surveys carried out since 1984. We made use of the Woman's Questionnaire, carried
out with a nationally representative sample of women in each country (covering both rural and
urban areas). We used DHS data from 51 countries, which includes 21 low-income, 24 lower-
middle-income, and 5 upper middle-income countries according to World Bank classification.
We use the latest available data for each country and 80 percent of the countries have data
from 2010 or later.

Because we are interested in women who have already completed their schooling but also want
to be a recent as possible and have an adequate sample size for each grade of school
completed we use data from just the cohort of women aged 25-34. This implies that although
the data are the most recent they reflect schooling conditions of some time ago. So, for
instance, the data from Nigeria are from 2014. Women who turned 30 (the middle of the 25-34
age range) in 2014 were born around 1984 and if they started school at age 6 they started
school in 1990. Some form of this lag is inevitable in using data about adults with complete
schooling and unfortunately there is very little evidence about the evolution of learning profiles
over time that would allow us to say how our estimates compared to current learning profiles. In
the case of ASER in India which has maintained a consistent set of data on learning profiles
over almost a decade the learning profiles appear to have gotten worse rather than better
(ASER 2014), so, while we don’t want to extrapolate from this evidence to other countries, we
also stress that, while enrollments have certainly expanded, there can be no general
presumption that learning profiles have improved over time.

1.1 Schooling and reading data from DHS

DHS recorded whether a woman attended school or not and if they attended, the highest level
(primary, secondary, tertiary). They also asked women what was the highest grade they
attended within that level. 6 Only women who reported their highest level of education was
primary or less were asked the reading question.
The number of years of schooling corresponding to complete primary varies across countries.
According to the ISCED 2011 (UNESCO 2012) the primary level “typically lasts six years,

6A typical questionnaire for DHS Phase 6 is at

http://dhsprogram.com/pubs/pdf/DHSQ6/DHS6_Questionnaires_5Nov2012_DHSQ6.pdf. These are questions 105 and 106 and

107 instructs the enumerator to ask the reading question only of women whose highest level of attendance (not completion) was
primary so that people who attended but did not complete secondary are not asked to read.

5
although its duration can range between four and seven years.” Since the actual years vary
from country to country our calculations are not strictly speaking about “universal primary
completion” as defined by each individual country but rather are about universal completion of
grade 6 (UPC(6)).

The DHS survey respondents who report not having attended secondary school are asked to
read a card with sentences such as:

Parents love their children.

Farming is hard work.
The child is reading a book.
Children work hard at school.

The surveyors had cards in different languages that were intended to span the array of possible
languages a woman in the sample area could read. Women can choose the language to read.
Women for whom there was no language appropriate card do not have a reading result and are
not in our results, but this was usually less than one percent of women. This is not a test of
reading English or even of reading in the dominant national language but of reading in the
language the respondents chose.

Surveyors indicated whether women (a) could not read at all, (b) were only able to read parts of
the sentence or (c) could read all of the test sentence. 7 We treat women as literate if they were
able to read the complete sentence or if they attended secondary school or more, even though
this has the obvious bias of assuming every woman with secondary school could read.

As with any binary categorization this line divides through a reality of a continuous variation
across individuals and reading and literacy proficiency. Ours is a very low bar for the definition
of functional literacy and does not measure any degree of comprehension or any of the other
complex skills that constitute literacy proficiency. For instance, the OECD’s Survey of Adult
Skills (PIAAC) “defines literacy as the ability to understand, evaluate, use and engage with
written texts in order to participate in society, achieve one’s goals, and develop one’s knowledge
and potential.” They define literacy as broader than “reading” as it “encompass(es) the range of
cognitive strategies (including decoding) that adults must bring into play to respond
appropriately to a variety of texts of different formats and types in the range of situations or
contexts in which they read.” By this standard the DHS measure of ability to read a single
simple sentence is a very rudimentary assessment of “reading” which is itself just one element
of literacy.
In the PIAAC data literacy is grouped into levels 1 to 5 and those “below level 1.” The
description of those “below level 1” is: “Individuals at this level can read brief texts on familiar
topics and locate a single piece of specific information identical in form to information in the
question or directive. They are not required to understand the structure of sentences or

7 The question is 108 and includes the instruction that if the woman cannot read the whole sentence to ask whether she can read

part of the sentence. So “part” might mean the woman could recognize as little as one word after prompting. We think this does not
meet a reasonable standard of “literate.” The UNESCO Institute of Statistics (2013) adopts this same definition for the primary
schooled when using the DHS data.

6
paragraphs and only basic vocabulary knowledge is required.” This is clearly much harder than
reading a single sentence as in the DHS assessment. Only 3.3 percent of adults in the OECD
scored below level 1.

One point of comparison between the DHS measure and the PIAAC data on literacy proficiency
is that PIAAC included just the city of Jakarta, which one assumes is likely to score substantially
higher than the Indonesian national average. In Jakarta’s PIAAC assessment 56.6 percent of
adults 25-65 with “less than upper secondary complete” were classified as “below level 1” in
literacy proficiency. In contrast in our estimates 89 percent of those with only Grade 8 complete
were classified as literate and hence only 11 percent illiterate. Hence many fewer people are
illiterate by our standard than even the lowest category (“below level 1”) of OECD literacy
proficiency, which again emphasizes that the measure we use is a very low bar for literacy.

A second point of comparison is the PISA assessment of reading, which defines reading as
having three elements: retrieving information, interpreting texts, reflection and evaluation. The
criteria for achieving even the lowest assessed level of interpreting texts is that a student
should: Recognize the main theme or author’s purpose in a text about a familiar topic, when the
required information in the text is prominent. In the 2000 PISA assessment of the reading ability
of 15-year-olds, who would have been 28 in the 2013 DHS, 80 percent were at level 1 or below
whereas in the DHS estimates 89 percent of those reporting completing grade 8 could read the
sentence. Again this confirms our definition of literacy with the DHS is a very low bar.

1.2 A learning profile from DHS data

A descriptive learning profile is the relationship between years (or levels) of schooling
completed and a measure of learning/skills/capabilities. Learning profiles are relatively scarce
and quite recent as most national or international assessments are school based samples which
assess only children in a given year of schooling (e.g. all students in grade 4) or of a certain age
(e.g. all children in grade 15). Recent survey instruments like ASER (and its offspring like
UWEZO) survey all school age children in a village and hence can produce learning profiles
showing reading proficiency at each grade of schooling completed (Oye, Sandefur and Pritchett
2016). Also surveys which ask respondents their highest grade completed and include some
measure of learning can construct a retrospective learning profile (Kaffenberger and Pritchett
2016). The MICS (Multiple Indicator Clusters Survey) and DHS have also been used for
learning profile analysis in Africa (UNESCO 2014).

We estimate the learning profile for the youngest cohort of women with their schooling
completed, those aged 25 to 34. In order to construct a learning profile we compute a local
polynomial regression of literacy on years of schooling for each of the 51 countries. A
regression was used to report the proportion at a given grade who can read. We use this
instead of using raw cell averages to smooth across grades as even though DHS samples are
large, the cell sizes can be quite small and hence estimates quite variable.

7
Figure 1a shows the DHS estimated learning profile of “ability to read a sentence” and grade
completed up to grade 6 for five countries and the average across all countries. This figure
illustrates two important facts.

First, the variation across countries in the percent of adult women who completed grade 6 can
read a sentence ranges almost from zero to 100 percent. While in Nigeria it is only 11 percent
(and there are five countries where learning is even worse than Nigeria) in Rwanda 97 percent
of women who completed grade 6 can read. There is also nearly everything in between. In
Bangladesh it is 33 percent and in Ethiopia it is 73 percent.

Second, the average learning profile is surprisingly shallow. Only half of young adult women
(aged 25-34) in the 51 DHS countries who had completed grade 6 could read all of a single
simple sentence.

Given the common conflation of “illiteracy” with “unschooled” these high levels of adult female
illiteracy among those who report attending school—roughly half those with grade 6 completed
cannot read--are perhaps stunningly (and to the sceptic perhaps unbelievably) high. The recent
learning profiles of current youth from ASER in India and Pakistan and UWEZO in Southern
Africa show similarly flat learning profiles across grades and similar lack of reading skills at
grade 6 completion. A recent set of nationally representative surveys about financial inclusion in
10 developing countries also included a question about whether the respondent (in this case
both men and women aged 25-65) could read a sentence (Kaffenberger and Pritchett 2016).
Table 1 shows the comparison between the DHS and Financial Inclusion Insights (FII) surveys
for the overlapping countries. These two entirely independent sources on the same concept of
“ability to read a sentence” show striking similarities—the average across the common countries
is 51.4 percent for DHS and 48.8 percent in the FII data--and a high correlation across countries
(.77) but some striking differences (e.g. Ghana). The broad pattern of our results is unlikely an
artefact of the DHS approach to assessing reading.

8
 Figure 1a: The learning profile varies widely across countries—in Rwanda
all women with Grade 6 can read while in Nigeria only 12 percent can
100.0%
90.0%
80.0%
70.0% Rwanda
Percent of women
aged 25-34 who can 60.0% Ethiopia
read all of a single
50.0%
sentence in their Average 51
chosen langauge, 40.0% countries
latest DHS data Peru
30.0%
20.0% Bangladesh
10.0% Nigeria
0.0%
0 2 4 6
Highest grade completed
Source: Oye, Pritchett and Sandefur (2016) based on DHS data.

Figure 1b: The grade attainment profile also varies widely across countries
100.0%

90.0%

80.0%

70.0% Rwanda

60.0% Ethiopia
Percent of women
aged 25-34 above 50.0% Average 51
each grade level countries
40.0%
Peru
30.0%
Bangladesh
20.0%
Nigeria
10.0%

0.0%
0 2 4 6
Highest grade completed
Source: Authors’ calculations based on DHS data.

9
 Table 1: Assessed ability of women with just primary education to read a
simple sentence or passage is similar between the DHS and FII data
DHS, women aged 25-34, FII survey, women aged 18-37,
Country highest grade was 6th, percent completed primary, able to read a
able to read all of a sentence sentence
Nigeria 12.0% 15.4%
Uganda 54.4% 23.2%
Bangladesh 32.6% 29.5%
Pakistan 50.7% 44.2%
India 34.6% 49.0%
Kenya 65.3% 69.7%
Indonesia 75.2% 76.7%
Tanzania 86.2% 82.5%
Ghana 7.7% 47.9%
Rwanda 97.1% 77.7%
Average 51.4% 48.8%
Source: Kaffenberger and Pritchett (2017) based on FII data.

2. How much would reaching UPC(6) help reach an SDG

of universal literacy?
The implicit assumption behind the targets aiming for universal primary completion was that all,
or at least most, countries had learning profiles like Rwanda’s so that if the country reached
universal primary schooling completion then nearly every child would also have reached at least
a minimal standard of universal literacy as in the SDG goal 4.6. But Rwanda is rather the
exception than the rule. Of the 51 countries only 51 percent of women who had completed
grade 6 could read a single sentence. Hence, nearly half of those meeting an MDG-like
schooling target (based on grade 6 as “primary”) would not meet the new SDG on literacy.

We can do a simple counter-factual calculation. Suppose in the cohort 25-34 all women who did
not reach grade 6 had instead completed grade 6. At the observed learning profile how much
would that have reduced the fraction of women who were illiterate?

2.1 Scenario of UPC(6) explained and illustrated

The arithmetic behind this calculation is simple. The actual literacy is just the schooling level
attainment weighted sum of the likelihood a woman at each level of schooling is literate
(equation 1) assuming women with higher grade completion are literate (even though this
assumption is almost certainly wrong as there is no particular reason to suppose there is a
discrete jump from grade 8 levels of literacy, which are well less than 100 percent, to exactly
100 percent at grade 9 and above). Hence all of the reported “literacy” rates are overstated by
this amount, but this doesn’t affect our estimates of changes across scenarios within a country.

10
 8

1) 𝐿𝑖𝑡𝑒𝑟𝑎𝑐𝑦 = ∑ 𝛼𝑔 ∗ 𝑠𝑔 + 𝛼𝑔>8
𝑔=0

Where αg is the share of women 25-34 with grade g as their highest grade attained (and no
schooling is g=0) and sg is the share of women with grade g who can read a complete sentence.

The arithmetic of the counter-factual of universal completion of grade 6 at a fixed learning profile
is easy, just assumed that all women with attainment less than grade 6 had grade six literacy
levels.

5 8

2) 𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑡𝑖𝑐𝑎𝑙 𝐿𝑖𝑡𝑒𝑟𝑎𝑐𝑦 = ∑ 𝛼𝑔 ∗ 𝑠6 + ∑ 𝛼𝑔 ∗ 𝑠𝑔 + 𝛼𝑔>8

𝑔=0 𝑔=6

The gain in literacy is the simple difference between the two.

3) 𝐺𝑎𝑖𝑛 𝐿𝑖𝑡𝑒𝑟𝑎𝑐𝑦 = ∑ 𝛼𝑔 ∗ (𝑠6 − 𝑠𝑔 )

𝑔=0

Equation 3 is very intuitive. The gain in literacy from the counter-factual of all women having
completed grade 6 is bigger: (a) the larger the shares of women who did not complete grade 6
(if all women had completed grade 6 then there would be no gain at all), and (b) the steeper the
learning profile and hence the larger the gap in literacy between women who completed grade 6
and those who completed less than no schooling (if the learning profile were completely flat then
there would be no gain at all).

The learning and grade attainment profiles in Figure 1a and 1b use four countries to illustrate
graphically the simple intuition of how this calculation works.

Ethiopia illustrates the case of a country where schooling expansion significantly reduces
illiteracy. In Ethiopia 85.9 percent of women aged 25-34 in 2011 had completed less than grade
6, with 69.7 percent having attended no schooling at all. In Ethiopia the learning profile is quite
steep (Figure 1a above) and 72.7 percent of those who did complete grade 6 could read. This
combination means that the potential gains in reducing illiteracy from expanding schooling at a
constant learning profile are massive. Actual female illiteracy in 2011 was 81.6 percent, but had
all women achieved grade 6 and achieved the observed grade 6 level of literacy of 72.7 percent
then illiteracy would have fallen by 56.7 percentage points. Ethiopia represented the scenario
that we believe many people promoting universal primary completion have in mind—since
schooling in fact appears to be reliably producing literacy (we are still bracketing the question of
causality and whether the learning profile can remain unchanged at massive changes of scale
of enrollment) the universalization of schooling eliminates most illiteracy.

11
Nigeria illustrates a different case. In Nigeria in 2014 43.2 of women aged 25-34 had not
completed grade 6 (37.8 percent with no schooling at all), less than Ethiopia but still very high
and far from universal primary completion. However, of women who had completed grade 6 in
Nigeria only 12 percent were literate. This implies that the gap in predicted literacy between
women with no schooling and those with grade 6 was only 12 percent (s 6-s0=.12) so that even
though there was a large share of women with no schooling the gain in literacy from expansion
of schooling completion would be small. Moving all the women from no schooling to grade 6
would have only reduced illiteracy by about 4.5 percentage points (.045≈.378*.12).

The hypothetical calculations in Nigeria illustrate the danger of a shallow learning curve. Even if
Nigeria had achieved universal completion of grade 6--compared to the actual large out-of-
school population--illiteracy would have fallen from 58 percent to only 53 percent. Achieving the
UPC(6) in Nigeria would have still left half of all women short of the SDG for literacy.

Peru illustrates the situation for many lower-middle and middle income countries. In Peru in
2013 only 2.4 percent of women had never enrolled in school and only 15.3 percent did not
finish grade 6. In this case even though the learning profile is the same as Ethiopia the absolute
magnitude of the gains are small simply because the scope for raising the share with at least
grade 6 complete is small—because those gains in schooling attainment have already been
made. In Peru this means the elimination of non-enrollment and drop-out before grade 6 would
have reduced illiteracy by only 5.9 percent—but from an already small base of 13.5 percent.

A fourth country, Bangladesh in 2015, illustrates an intermediate case. There were fewer
women who had no schooling than Ethiopia or Nigeria but more than Peru (33 percent) but
higher drop-out so 63 percent of women had less than grade 6 complete. Bangladesh’s learning
profile was steeper than Nigeria’s but much shallower than Ethiopia or Peru with only 32.6
percent of women able to read after completing grade 6. This means the gains were bigger than
Nigeria or Peru but much smaller than Ethiopia, as universal completion of grade 6 would have
reduced illiteracy by 16 percent. But again, in Bangladesh universal completion of grade 6
would still have left half of all women illiterate.

There are two obvious limitations of these calculations. First, it assumes that the descriptive
learning profile represents a causal relationship that is relevant to the incremental student. This
assumes that the difference between those who can read and who completed grade 4 versus
grade 3 is something like a Local Average Treatment Effect (LATE) and if a child who dropped
out in grade 3 instead had persisted to grade 4 their likelihood of learning to read in that year
was equal to the average gain. This assumption is almost certainly false and almost certainly
leads to the descriptive profile being steeper than the causal profile. This is because of the
positive self-selection of students into further enrollment. The drop-out of students in early
grades is almost certainly associated with students who are weaker students, both in the sense
of lower cumulative achievement and less likely to learn in the coming year. This positive self-
selection of students implies that at least some of the gain of the descriptive learning profile
does not reflect causal learning but just that those who could read (or were about to read)

12
persisted in school. Hence, all of our counter-factual simulations below are optimistic and
overstate the likely gains for this reason.

The second obvious limitation is that our simulations are not just over marginal or incremental
expansions of the schooling system but often imply massive increases in student populations.
Our assumption of a constant learning profile implies that the massive expansion would not
have caused the learning profile to deteriorate. Certainly this is a possible outcome but by no
means a foregone conclusion. Again this almost certainly makes our calculations optimistic in
that they overstate the likely gains.

2.2 Results for all countries

We break the reporting of the results of our simulations into three groups based on their initial
level of illiteracy: high illiteracy (above 60 percent), medium illiteracy (between 40 and 60
percent), and low illiteracy (below 40 percent) as these countries represent very different
conditions and, as we will see, produce strikingly different results.

2.2.1 Initially high illiteracy countries

There are 16 countries where initial illiteracy was over 60 percent. In most of these countries a
large fraction (typically more than 2/3) of women in this cohort never attended school. Hence
one of the conditions for a large impact on literacy of expanding primary schooling is met. What
Table 2a illustrates is that the impact of getting these under-schooled into and completing
school depends heavily on the steepness of the learning profile.

In 12 of the 16 countries (including a number of large countries like India, Bangladesh, and DR
Congo) less than half of women who completed grade 6 could read—ranging from shockingly
low levels of 20 percent or less in DRC, Chad, Liberia to only around a third in India,
Bangladesh and Benin. In all 12 of these countries even after UPC(6) illiteracy would still afflict
over half of all women.

In contrast, as we saw above with Ethiopia, a combination of high initial out-of-school population
combined with a reasonably steep learning profile implies UPC(6) would lead to enormous
reductions in illiteracy. In Mozambique, for instance, illiteracy would have fallen in 74 percent to
30 percent in the UPC(6) scenario.

2.2.2 Initially medium illiteracy countries

The countries with moderate levels of illiteracy (between 40 and 60 percent) show the same
feature—the gains in reducing illiteracy depend heavily on the learning profile and the learning
profiles vary widely across countries.

In Egypt the fraction of women who never attended school is 36 percent—high, but much lower
than in the high illiteracy group—but the learning profile is strikingly flat as only 18 percent of
women completing grade 6 were able to read. This implies the gains from additional enrollment

13
 are limited both because the “out-of-school” population is modest and because the learning
profile is shallow. Hence UPC(6) would reduce illiteracy by only 5 percentage points.

In contrast, in Morocco 70 percent of those completing grade 6 can read but 70 percent of
women in this cohort never attended school leading to illiteracy of 57.6 percent. But if all women
had completed grade 6, illiteracy would have fallen from 57.6 to only 10.2 percent.

Table 2a: Among high illiteracy countries (over 60 percent) universal primary completion would
have left female illiteracy over 50 percent in the 12 countries with a flat learning profile—but
produces massive gains in steep learning profile countries
Scenario:
Assuming that all women who never
Data on young women (25-34)
enrolled instead completed grade 6 and
from the most recent DHS
had the average literacy of women who
completed grade 6
Percent Percent
Percentage
Percent who who Percent of
Percent point
of cohort completed cannot illiteracy
who reduction
who grade 6 read eliminated
Country Year cannot in illiteracy
never who can (illiteracy) relative to base
read relative to
attended read a even in case by UPC
(illiteracy) base case
school simple UPC(6) (sorted)
in UPC
sentence scenario
Flatter learning profile countries (Percent grade 6 completers who can read<50 percent)
Sierra Leone 2014 87.7% 79.5% 3.5% 85.0% 2.7% 3.0%
Gambia 2013 69.3% 60.7% 4.1% 67.1% 2.3% 3.2%
Guinea 2013 91.6% 88.2% 4.5% 87.9% 3.6% 4.0%
Liberia 2014 74.6% 63.9% 7.4% 70.2% 4.4% 5.9%
Togo 2014 75.6% 61.2% 14.9% 68.2% 7.4% 9.8%
DRC 2014 64.4% 49.9% 20.0% 56.7% 7.7% 12.0%
Chad 2005 95.4% 93.8% 16.5% 80.6% 14.8% 15.5%
Bangladesh 2015 66.7% 63.1% 32.6% 50.7% 16.0% 24.0%
Niger 2013 95.0% 93.4% 25.2% 72.1% 22.9% 24.1%
India 2007 66.9% 62.7% 34.6% 48.3% 18.6% 27.9%
Benin 2012 87.4% 85.2% 32.3% 61.4% 26.0% 29.7%
Senegal 2015 82.0% 81.8% 40.9% 51.6% 30.5% 37.1%
Steep(er) learning profile countries (percent grade 6 completers who can read>50 percent)
Pakistan 2013 65.7% 70.0% 50.7% 37.3% 28.4% 43.3%
Cote d’Ivoire 2012 73.3% 73.2% 57.3% 36.0% 37.3% 50.9%
Mozambique 2012 74.2% 79.9% 66.3% 29.6% 44.5% 60.0%
Ethiopia 2011 81.4% 85.9% 72.7% 24.8% 56.7% 69.6%
Source: Authors calculations with DHS data.

14
 Table 2b: Among moderate illiteracy countries (40 to 60 percent) where the learning profile
is flat (<50 percent of grade 6 completers can read) the gains from UPC(6) are small to
moderate while in steeper learning profile countries UPC(6) nearly eliminates illiteracy
Scenario:
Assuming that all women who
Data on young women (25-34) never enrolled instead completed
from the most recent DHS grade 6 and had the average
literacy of women who completed
grade 6
Percent Percent
Percentage Percent of
Percent who who
Percent point illiteracy
of cohort completed cannot
who reduction eliminated
who grade 6 read
Country Year cannot in illiteracy relative to
never who can (illiteracy)
read relative to base case
attended read a even in
(illiteracy) base case by UPC
school simple UPC
in UPC (sorted)
sentence scenario
Flatter learning profile countries (Percent grade 6 completers who can read<50 percent)
Congo 2012 49.8% 24.0% 10.3% 47.9% 1.9% 3.7%
Ghana 2015 50.1% 38.8% 7.7% 47.6% 2.5% 5.1%
Nigeria 2014 58.0% 43.2% 12.0% 53.1% 4.9% 8.4%
Egypt 2014 42.0% 36.2% 18.3% 36.8% 5.2% 12.4%
Zambia 2014 54.0% 36.8% 26.5% 47.0% 7.0% 12.9%
Comoros 2013 56.1% 50.2% 26.8% 44.7% 11.4% 20.3%
Sao Tome
2009 52.2% 57.7% 49.1% 40.8% 11.5% 22.0%
Principe
Timor-Leste 2010 46.8% 40.6% 42.6% 32.6% 14.2% 30.4%
Cameroon 2012 51.8% 42.2% 49.8% 34.4% 17.3% 33.5%
Steep(er) learning profile countries (percent grade 6 completers who can read>50 percent)
Uganda 2012 58.0% 57.0% 54.4% 35.6% 22.4% 38.7%
Haiti 2012 49.1% 50.1% 55.6% 28.0% 21.1% 42.9%
Cambodia 2015 55.0% 64.8% 65.4% 27.0% 28.0% 50.9%
Madagascar 2010 53.1% 73.2% 76.9% 19.0% 34.1% 64.2%
Malawi 2011 47.0% 58.1% 79.9% 15.4% 31.6% 67.1%
Nepal 2011 59.4% 73.4% 76.9% 18.4% 41.0% 69.1%
Burundi 2011 47.9% 72.6% 90.2% 8.8% 39.0% 81.5%
Morocco 2004 57.6% 70.4% 86.5% 10.2% 47.3% 82.2%
Source: Authors’ calculations with DHS data.

2.2.3 Initially low illiteracy countries

In the initially low illiteracy countries there are two features that play an important role in whether
illiteracy is reduced by expansion in schooling. As in the other groups of countries the steepness

15
of the learning profile matters. But in this group there are a number of countries for which the
out-of-school population is so low that there just isn’t much gain left to be made—which is good
news.

In the Philippines, for example, illiteracy is already less than 10 percent and only 8.5 percent of
women attended no schooling. Therefore, even though the learning profile is steep (69 percent
can read at grade 6) the gain from UPC(6) is only 2.5 percentage points (but which is 25.6
percent of initial illiteracy).

In Rwanda, in spite of a steep observed learning profile (97 of grade 6 completers can read)
there is substantial illiteracy (33 percent) because a large fraction of the cohort did not attend
school (62 percent). This is a case where UPC(6) effectively eliminates illiteracy, reducing it to
only 2.6 percent.

16
 Table 2c: The gains in low illiteracy countries are those with both significant out-of-school
populations and steep learning profile
Scenario:
Assuming that all women who never
Data on young women (25-34)
enrolled instead completed grade 6 and
from the most recent DHS
had the average literacy of women who
completed grade 6
Percent Percent
Percentage Percent of
Percent who who
Percent point illiteracy
of cohort completed cannot
who reduction eliminated
who grade 6 read
Country Year cannot in illiteracy relative to
never who can (illiteracy)
read relative to base case by
attended read a even in
(illiteracy) base case UPC
school simple UPC
in UPC (sorted)
sentence scenario
Moldova 2006 2.6% 1.2% 57.9% 2.4% 0.2% 7.0%
Zimbabwe 2011 19.4% 8.7% 49.4% 17.2% 2.2% 11.5%
Guyana 2010 15.6% 12.5% 54.6% 13.3% 2.4% 15.1%
Gabon 2012 33.9% 24.7% 44.5% 28.1% 5.9% 17.3%
Philippines 2009 9.8% 8.5% 69.5% 7.3% 2.5% 25.6%
Indonesia 2013 16.2% 12.7% 75.2% 11.0% 5.2% 31.9%
Lesotho 2010 10.7% 13.6% 86.2% 7.2% 3.5% 32.8%
Kenya 2015 22.2% 16.9% 65.3% 14.3% 7.9% 35.6%
Namibia 2014 16.3% 16.6% 70.6% 10.0% 6.3% 38.7%
Swaziland 2007 15.4% 17.5% 74.7% 8.8% 6.6% 42.7%
Peru 2013 13.5% 15.3% 74.7% 7.5% 5.9% 44.1%
Dominican
2014 13.4% 19.7% 84.5% 5.7% 7.7% 57.3%
Republic
Tanzania 2010 32.8% 30.9% 86.2% 11.4% 21.4% 65.3%
Nicaragua 2002 21.2% 41.0% 93.5% 3.8% 17.4% 82.3%
Rwanda 2011 33.4% 62.1% 97.1% 2.6% 30.8% 92.3%
Bolivia 2008 9.8% 37.2% 100.0% 0.0% 9.8% 100.0%
Honduras 2013 12.7% 31.5% 100.0% 0.0% 12.7% 100.0%
Source: Authors’ calculations with DHS data.

17
 2.3 Aggregate calculations
We cannot do “global” calculations as the DHS data are neither comprehensive across all
countries nor a random sample of developing countries nor are the data all in the same year.
But we can do a population weighted average of the countries we have, which include 62
percent of the UN’s “less developed countries, excluding China” population. We use the
population in 2010 of females aged 25 to 34.8 Population weights are important because a
simple average of these countries would give equal weight to India with 95,407 thousand
women aged 25-34 in 2010 and Sao Tome and Principe with just 13 thousand.

Table 3 shows that even had UPC(6) been reached at the observed learning profiles there
would have remained a massive issue of women’s illiteracy. The three basic results are that in
the data there is a tremendous challenge of female illiteracy—57 percent of women could not
read a single sentence. This is the result of two big facts. First, in this age group 54 percent of
women did not complete grade 6. Second, 44 percent of those who had completed grade 6 (but
no higher) could not read all of a single sentence.

Table 3: Even with universal completion of grade 6 female adult illiteracy would have been 39
percent—reaching the SDG of universal literacy requires much more than just getting all girls in
school
Scenario:
Assuming that all women who never
Data on young women (25-34)
enrolled instead completed grade 6 and
from the most recent DHS
had the average literacy of women who
completed grade 6
Percent Percent
Percent Percentage
who who Percent of
Percent of point
completed cannot illiteracy
who cohort reduction
grade 6 read eliminated
Country Year cannot with in illiteracy
who can (illiteracy) relative to base
read less relative to
read a even in case by UPC
(illiteracy) than base case
simple UPC (sorted)
grade 6 in UPC
sentence scenario
Average across
all 51 countries,
Various 56.7% 53.8% 43.9% 39.0% 17.7% 31.9%
population
weighted
Average across
all 51 countries, Various 48.0% 48.1% 51.3% 32.1% 15.9% 36.5%
unweighted
Source: Authors’ calculations with DHS data.

8 https://esa.un.org/unpd/wpp/

18
 These simple facts imply that reaching a goal for universal literacy (like the SDG target 4.6) will
require making it possible for all girls to get into school and stay there. But, there is more to
“letting girls learn” than just “letting girls get to school.” Even had all girls completed grade 6 at
the observed learning profiles women’s illiteracy would have still been 39 percent. Eliminating
the deficit form grade 6 completion (the result of both never enrollment and early drop-out
entirely) would have a substantial impact on illiteracy—it would be reduced by almost a third (32
percent). However, a one-third reduction is far from elimination.

Figure 2 shows the basic result. Using a population weighted aggregate of the 51 countries with
DHS data an achievement of universal grade 6 completion at existing learning profiles would
have still left 39 percent of women in the world without even the most basic measure of literacy.

Figure 2: Reaching universal completion of grade 6 at existing learning profiles would reduce
illiteracy by about 17 percentage points—but leave 39 percent of women illiterate
60.0% 56.7%

50.0%

39.0%
40.0%
Percent of
women 25-34
30.0%
who cannot read
a single sentence
20.0%

10.0%

0.0%
Actual illiteracy (DHS method) Scenario of universal completion
of grade 6 at existing learning

Source: Authors’ calculations with DHS data (Table 3).

Figure 3 shows the impact of universal grade 6 completion on illiteracy in four countries we used
as illustrative above. Clearly the available gains from just “letting girls go to school” vary
massively by whether or not most girls already complete primary school (Peru) and by how well
girls on average learn once in school (e.g. contrast Ethiopia and Nigeria).

19
Figure 3: Reduction in total illiteracy (%) from universal grade 6 attainment with existing
learning profiles would still leave wide gaps in illiteracy among countries

7.6
Peru
13.5

53.1
Nigeria
58.0

50.7
Bangladesh
66.7

24.7
Ethiopia
81.4

0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0

Hypothetical illiteracy Actual illiteracy

Source: Authors’ calculations with DHS data.

Conclusion
Many of the global statements about global goals have conflated “schooling” and “education” by
treating them as synonyms or equating them by definition so that someone who attended school
is, by definition, educated. The Universal Declaration of Human Rights, article 26(1) spoke only
of education:

Everyone has the right to education. Education shall be free, at least in the elementary
and fundamental stages. Elementary education shall be compulsory. Technical and
professional education shall be made generally available and higher education shall be
equally accessible to all on the basis of merit.

The MDG targets spoke only of schooling (and within that, only of primary schooling). The SDG
has targets directly for “relevant and effective learning outcomes” (target 4.1) and that all youth
have literacy and numeracy (Target 4.6).

This paper uses DHS data on a very simple and very low standard of literacy—the ability to read
all of a simple, short, and concrete sentence--and shows that reaching the SDG will require
much more than simply reaching the MDG of all children completing primary schooling. It will
require a very sharp steepening of the learning profile—children with need to learn much more
in each year of schooling. This is necessary so that children will emerge from schooling not just
schooled but truly educated.

20
References

Asadullah, Mohmmad Niaz; Chaudhury, Nazmul. 2013. "Primary Schooling, Student

Learning, and School Quality in Rural Bangladesh." Center for Global Development,
Working Paper.
ASER. 2014. Annual Status of Education Report (Rural): 2013. New Delhi India: ASER
Centre.
Kaffenberger, Michelle and Lant Pritchett. 2016 (forthcoming). "Schooling Completion
and Learning: New Evidence from the Financial Inclusion Insights Data."
Kenny, Charles and Lant Pritchett. 2013. "Promoting Millennium Development Ideals:
The Risks of Defining Development Down " Center for Global Development, Working
Paper, (338).
Rose, Pauline. 2014. "Teaching and Learning: Achieving Quality for All " UNESCO, EFA
Global Monitoring Report.
Spaull, Nicholas and Stephen Taylor. 2015. "Access to What? Creating a Composite
Measure of Educational Quantity and Quality for 11 African Countries." Comparative
Education Review, 59(1), 133-59.
Statistics, UNESCO Institute for. 2013. "Adult and Youth Literacy: National, Regional
and Global Trends, 1985-2015," UIS Information Paper. Quebec, Canada:
____. 2012. International Standard Classification of Education. Quebec, Canada:
UNESCO Institute for Statistics.
UNICEF, UNESCO; World Bank;. 2014. Education Sector Analysis Methodological
Guidelines: Sector-Wide Analysis, with Emphasis on Primary and Secondary
Education.

21