Switching to English in Rwanda’s Educational

Curriculum: Estimating the Return on Investment∗

David D. Laitin† and Rajesh Ramachandran‡

May 2024

Abstract

In 2008, Rwanda elevated English as the medium of instruction for all schools, re-
placing Kinyarwanda (the universally shared indigenous language) at the primary and
French at higher levels. The primary justification for this policy change was centered
on economic development, as English is seen as a requirement for active participation
in the global economy. Additionally, English was seen as key to Rwanda’s regional
and global integration, including its joining of the East African Community and the
Commonwealth, with the hope that an English-literate population would bolster trade,
development, and investment. However, these benefits require that the population can
effectively acquire human capital in English despite having little knowledge and ex-
posure to the language in daily life. Our analysis of the educational effects of this
policy shift compares Burundi and Rwanda by combining differences across countries
in the language of schooling with differences across cohorts induced by the timing of
the change in the medium of instruction. Our results suggest that this change resulted
in reducing literacy, the probability of entering secondary schooling, and consequently
the average years of completed schooling for the generation most affected by the change.

Keywords:

JEL classification codes:

∗
To add.
†
Stanford University, Department of Political Science, Encina Hall West, Room 423, California 93405,
[email protected].
‡
Monash University Malaysia, Department of Economics, School of Business, 47500 Subang Jaya, Selan-
gor, Malaysia, [email protected].

1
1 The Linguistic Choice for New Nations: Indigenous
vs. Colonial

The medium of instruction in education is a crucial policy area, with important implications
for human capital, health, income and productivity (Tollefson and Tsui, 2003; Laitin and
Ramachandran, 2016). In 2008, Rwanda instituted a change in its language-use-in-education,
where English became the primary medium of instruction in schools, replacing Kinyarwanda
at the lower levels, and French at the higher levels (Samuelson, 2012). This shift, though
intended to promote English proficiency and international integration, sparked debates about
its effectiveness and potential drawbacks. Given the importance of education in shaping a
nation’s future, understanding the consequences of such a policy change is essential.

This study aims to assess the impact of Rwanda’s language policy change, which in 2008
introduced English as the primary medium of instruction, on human capital development.
Specifically, we investigate how this policy shift affected educational attainment, literacy
levels, and the transition to secondary schooling among the cohort affected by the change in
the policy.

To be able to assess the impact of the policy change, we employ nationally representative
data from the Demographic and Health Surveys (DHS), and compare Rwanda and Burundi
using a difference-in-differences (DID) estimator. Burundi, the neighbor of Rwanda has a
very similar demographic makeup with almost the entire population having the same mother
tongue as in Rwanda, as well as a very similar language-use-in-education policy with their
shared indigenous language being employed for the first 3-5 years of primary schooling before
transitioning to using French as the medium of instruction. In the time period of our study
Burundi did not change its language-use-in-education policy and forms the control group
of interest. By comparing cohorts affected by the language policy change in Rwanda with
similar aged cohorts in Burundi, which did not undergo such a policy change, we can isolate
the effect of the policy on human capital outcomes.

Our analysis reveals several key findings. First, we find that the introduction of English
as the primary medium of instruction in Rwanda led to a significant decrease in years of
education for the affected cohort compared to what would have been the case without the
reform. Specifically, cohorts exposed to the policy change experienced, on average, a re-
duction of 0.54 years in schooling, indicating a negative impact on educational attainment.
Second, the policy change also had a detrimental effect on literacy levels. We observe a

2
7.2-percentage-point decline in the probability of individuals being able to read an entire
sentence following the policy implementation. This suggests that the language shift may
have hindered literacy development among students in Rwanda. Third, we find evidence of
a decline in the transition rate to secondary schooling among cohorts affected by the policy
change. The probability of entering secondary schooling decreased by 4.7 percentage points,
indicating potential barriers created by the language policy shift. We find qualitatively sim-
ilar patterns, though the economic magnitudes are larger, when we restrict the comparison
to individuals who reside close to the border, enabling us to account for the role of other
unobservables.

Our study is directly related to the small but growing literature that tries to estimate
the impact of language-use-in-education polices on human capital in postcolonial states. The
large bulk of the evidence finds that reliance on the former colonial language whose knowledge
is restricted to a small section of the society, and not used for day-to-day communication, has
negative effects on educational attainment and learning outcomes; see Eriksson (2014) and
Taylor and Coetzee (2016) in the context of South Africa; Ramachandran (2017) in the con-
text of Ethiopia; Laitin et al. (2019) in the context of Cameroon; Laitin and Ramachandran
(2022) for evidence from sub-Saharan Africa. The switch in the medium of instruction that
we analyze herein, although attributed to several factors, would not have occurred without
the genocide in 1994 that brought in a political regime hostile to the linguistic status quo.
Since the genocide and the subsequent civil war was not fought over language choice, we can
consider it somewhat exogenous and useful for a case involving language shift and human
development, amenable to empirical analysis.

The most closely related papers Eriksson, 2014; Laitin et al., 2019 show that English
acquisition in fact improved when indigenous languages are employed as mediums of in-
struction and English is taught as a subject. These papers, and our findings, are consistent
with the work of authors such as Phillipson (1992) who highlights the fact that it is a oft
repeated fallacy that English acquisition is best promoted through using it as a medium of
instruction, whereas evidence suggests that competence in the first language can in fact help
acquisition of a second language (Cummins, 1979, 1991). Our findings also offer insights into
the trade-offs associated with language policies aimed at promoting international integration
and proficiency in global languages like English. While such policies may have strategic ben-
efits, they can also have unintended consequences for educational outcomes, particularly in
settings where the existing medium of instruction is the principal language of communication
at home, and the society at large.

3
 The policy change, while advertised by the incumbent regime and its Ministry of Educa-
tion and validated by international donors as bringing Rwanda into the modern technocratic
age, the switch from Kinyarwanda and French to English, as many regime critics foresaw,
has been an educational setback, one that has placed the growth of Rwanda’s production of
human capital behind Burundi, its nearly impoverished neighbor. Rwanda serves as a poster
child for Africa’s development future; but now it must recover from its less than successful
language policy reform. The rest of the paper is organized as follows: Section 2 provides
the background to language-use-in-education policy in Rwanda and the political economy
surrounding the change in 2008. Section 3 introduces the data and the empirical strategy
to estimate the impact of the policy change. Section 4 presents the results. Section 5 dis-
cusses the implications of our findings for policy and Section 6 provides concluding remarks
in reference to the concept of institutional coherence.

2 Media of Instruction in Rwanda

The official languages of Rwanda (Constitution of the Republic of Rwanda, 2003, p.5) are
Kinyarwanda (an indigenous Bantu language), French (the colonial language after the colony
was awarded to the Belgians as a League of Nations Mandate after World War I), and
English (the language of international business and that of the country’s president who as
an exile grew up in Uganda, whose colonial language was English). An estimated 99% of
the population can speak Kinyarwanda, and 90% speak only Kinyarwanda. Estimates of
the total number of English speakers range from 1.9% - 5%. Approximately 5% to 15% of
the population speaks French (Samuelson, 2012). Educational policies in regard to media
of instruction in Rwanda have switched among these three languages (see Table 1), moving
between Kinyarwanda, French, and English.

This paper focuses on the switch in 2008. Point 11 of the Council of Ministers announce-
ment (author translation) “asked the Minister of Education to set up an urgent program of
teaching in English at all levels of public and subsidized free education. He has also asked
the Minister of the Civil Service to set up a program aid allowing all State agents, first and
foremost those in high positions, to learn English” (Tuvuzimpundu, 2014). Thus, for the first
time, the government gave sole status to English the medium of instruction for both private
and public education in Rwanda’s Primary 1-3, as well as from primary 4 through secondary
education to English, while giving only secondary status to French. As of 2008, French
was still compulsory in public education, but the courses are not subject to examination for

4
promotion, and therefore not taken seriously by most students. Furthermore, (author trans-
lation) “the majority of higher education institutions no longer provide French departments.
The National University of Rwanda, for example, was forced to abolish the French Language
and Literature department due to a lack of candidates. The Kigali Higher Educational In-
stitute (known as KIE: Kigali Institute of Education) is the only public institution where a
French affiliate can still be found” (Tuvuzimpundu, 2014) .

There are two theories accounting for this historic switch. Most often heard was the
political story. During the post-war period (1994-present), French has suffered from negative
attitudes due to the alleged involvement of the French army in lending support to the Hutu
genocidaires in the mass killing of Tutsis and those alleged to support them (Prunier, 1997).
After the victory of the Tutsi-led exiled forces, many of these English-speaking returnees,
including President Paul Kagame, have shown little interest in learning French, which they
view as the language of the French allies and supporters of the Francophone genocidaires.
Meanwhile, the Hutu population, only recently facing exile, has had minimal exposure to
English. It is difficult to know their preferences, however, as due to the genocide of 1994,
the Rwandan Senate passed legislation prohibiting “genocidal ideology” or “divisionism”
(Republic of Rwanda, 2006), thereby suppressing public dialogue about language preferences
which have been considered ethnic and scorned by authorities.

The French press interpreted this switch as a frontal attack on francophonie, a core policy
goal of their country’s foreign policy, one that would bring international status and trade
to France. In a typical analysis in Le Monde, the switch is interpreted as a foreign policy
decision (author translation): “The questioning of French in Rwanda, a country belonging
to the traditional French-speaking area, is a consequence of the political crisis which has
been affecting relations between Paris and Kigali for years. The Rwandan authorities accuse
France of having helped the Hutu militias responsible for the 1994 genocide (800,000 deaths
in three months among Tutsis and moderate Hutus)”1 . But francophonie faced deeper
problems; data from Tuvuzimpundu (2014) before the switch to English reveal that the
spread of French in Rwanda was shallow, with most domains (novels, news, theatre, and
poetry) rendered in Kinyarwanda, with English surpassing French in the diffusion of news
on radio and TV. And few Rwandans heard any French outside the classroom. Similarly,
Albaugh (2014) drawing on various sources estimates that around 12% of the Rwandas could
speak French.

With critics of the government, the political story has had a more personal touch. Pres-
1
See here for further details.

5
ident Kagame’s once ally, David Himbara, analyzes this switch not as politically motivated,
but done out of personal gain or hubris. He writes on social media “Kagame said adieu to
the French language in 2009. Fast forward to 2022 – he sang bienvenue to le français. What
made the notoriously truth-allergic strongman eat his words?” He continues, “General Paul
Kagame’s flip-flipping on education proves that he does doesn’t give a damn about human
capital and the future of Rwanda’s young people. It is stating the most obvious to say that it
is human capital that drives individual self-fulfillment and national socioeconomic develop-
ment. To repeat, without human capital – meaning education, knowledge, skills, and health
– neither can a country’s citizens realize their own potential, nor can informal economies
transform into productive and prosperous nation states. The Rwandan strongman’s legacy
is the reverse – destroying the already dismal education, thereby creating a generation of
broken youths in their millions.” A similar indictment of the policy from the point of view
of the francophone teaching corps by a former Secretary of State for Education, Théoneste
Mutsindashyaka, called the policy shift a fiasco that will hinder the educational progress
of a generation. Still others claimed the decision was a politics of ”elite closure,” where
the move to English was a means of disenfranchising French-speaking former Hutu elites
(Tuvuzimpundu, 2014).

The official explanation was different, with the notion that “adopting English as the
official language can promote better communication for business, foreign investment, devel-
opment, and technology transfer” (Samuelson). English was portrayed as key to Rwanda’s
regional and global integration, including its joining of the East African Community (July
2007) and the Commonwealth (November 2009), with the hope that an English-literate popu-
lation and business community would bolster trade, development, and investment. However,
these benefits require that the population can effectively acquire human capital in English
despite having little knowledge and exposure to the language in daily life.

This concern came out clearly in an evaluation report on Rwanda’s educational reforms
that were supported by USAID. In a collaboration with the Rwandan Ministry of Education
(MINEDUC), USAID and technical partners, the Literacy, Language and Learning (L3)
Initiative worked with pre-service and in-service facilitators to introduce proven reading
and mathematics teaching strategies. Among the five intermediate results, the third (IR-3)
offered support for English that involved inter alia a revision of the existing ESL curriculum
(Rurangirwa, 2012).

This initiative revealed some basic issues in implementing IR-3. It reported, “based
on meetings and training sessions organized recently with teachers...from rural schools in

6
Bugesera, teachers’ ability to follow training sessions facilitated only in English is a concern.
During the recently organized sessions, the vast majority of teachers had difficulty follow-
ing English language explanations or instructions, even when the explanations/instructions
were offered in simplified English. Although Rwandan [School-based Mentors] SBMs will
be able to fall back on Kinyarwanda in cases where teachers do not understand, SBMs re-
cruited from Kenya or Uganda will not. It will be necessary to monitor the use of English
and Kinyarwanda in school-based mentoring/training activities to Bugesera to identify the
appropriate mix of language usage in video (and non-video) based trainings” (Rurangirwa,
2012).

Reaching the level of English language skills students need before making the transition
to English as a language of has proven to be a challenge. A consultant at the Global
Partnership for Education (GPE) Conference stressed the need for students to have acquired
a vocabulary of between 5000 and 7000 words before transitioning to English as a language
of instruction. Unfortunately, she stressed, “the current P1 to P3 English curriculum does
not allow students to build anywhere near this level of vocabulary.” For the Curricular and
Pedagogical Materials Development department in the MINEDUC, “her presentation was a
revelation and brought home the need to revisit the primary English curriculum in order to
ensure that students have the language skills necessary to begin learning in English at the
start of P[rimary]4” (Rurangirwa, 2012).

Similar issues were brought to the fore in a seminal Master’s thesis submitted by a leading
official in the MINEDUC (Umulisa, 2022). In this study, several concerns were registered.
A 2012 assessment of teachers’ proficiency in English revealed that a mere 2.9% of the corps
had reached an intermediate level of English proficiency. The government invested heavily
in teacher training and yet in a World Bank (2020) assessment of the training program, less
than half of teachers scored intermediate level in English (Kraay, 2019).2 . This exacerbated
Rwanda’s rural disadvantage, as there was a dearth of print media and opportunities outside
the capital for students to hear and speak English.

In light of the rapid shift, educational results were hardly surprising. Summary statistics
show that repetition rates in early grade primary were higher during the period of English
language of instruction (59% of students) than during the period of Kinyarwanda language
of instruction (37% of students). Using OLS with a battery of controls, early grade students
taught in English were approximately 17 percentage points more likely to repeat than early
grade students taught in Kinyarwanda and results are statistically significant at all levels.
2
Find the Rwanda specific report here.

7
Surprisingly however (and this may be due to selection in which only the best equipped
students survived the English medium in Primary 1-3, and therefore were most able to
complete secondary education in English), students who were taught strictly in English in
earlier and upper grades were approximately 21 percentage points less likely to repeat than
those taught in Kinyarwanda in early grades and taught in English in upper grades (Umulisa,
2022).

This overview of Rwanda’s educational policy in regard to media of instruction reveals

experimentation without clear metrics for success. Here we are able to estimate the returns
on human capital for one of the reforms, viz. the switch from Kinyarwanda to English as
the medium of instruction in P1-3 instituted in 2008. We can thus measure the effects of a
switch from indigenous to an international language for educational success.

3 Data and Identification Strategy

3.1 Data

To be able to examine the impact of the switch in the medium of instruction on human
capital, we employ data from the Demographic and Health Surveys (DHS). The DHS are
nationally representative datasets providing information on education, health and population
in developing countries. We draw on data from the men and women’s round of the DHS for
Rwanda from the years 2014-15 and 2019-20 and from Burundi from 2016-17. The Rwanda
data has information on 13,497 women and 6217 men aged 15-49 in 2014-15, and 14,634
women and 6513 men aged 15-49 in 2019-20. For Burundi, there is data on 17,269 women
and 7552 men aged 15-49 in 2016-17.

The children attend primary school from the ages of 6-12 in Rwanda and Burundi. The
language of instruction in Rwanda and Burundi was characterized by the use of the same
indigenous language for the first 5 grades of primary schooling (Albaugh, 2014). However,
in 2008 the change in language altered the language of instruction in primary schooling in
Rwanda from Kinyarwanda to English: initially right from Grade 1, and then from 2011
from Grade 3 onwards. We thus consider the cohorts aged less than 12 years old in 2008 in
Rwanda as the treated cohort as regards the change in the medium of instruction in primary
schooling. The individuals who were aged older than 19 were unaffected by the language
policy change. We thus consider the cohorts aged 19-25, 26-32, 33-39 and 40-46 as the

8
untreated cohorts. The number of observation by cohort and country is shown in Table 2.

The cohort aged 13-18 was already in secondary schooling at the time of the language
policy change in 2008. Delayed school entry could imply that some of these students were
also affected by the change in the medium of instruction. In addition, the switch to English
meant that they were also affected by the change in the medium of instruction in secondary
schooling. We thus omit this cohort when analysing the implications of the change in the
language of instruction in primary schooling.

3.2 The Dependent Variables

We are interested in estimating the impact of the language policy change on human capital.
With regards to the change that occurs at the primary schooling level, we focus on two key
quantity based proxies for human capital: (i) a dummy for entering secondary schooling; and
(ii) completed years of education. In addition, the DHS also has information available on a
quality based indicator, that is whether an individual can read an entire sentence based on
an actual test of reading comprehension rather than being self-reported. More specifically,
the DHS interviewer coded the respondents as: (1) cannot read at all; (2) able to read only
parts of the sentence; (3) able to read the whole sentence; (4) no card with required language.
We code individuals who are able to read a whole sentence in any language chosen by the
interviewee as literate and those who cannot read at all or only able to read parts of the
sentence as illiterate. We thus also consider the ability to read an entire sentence as the
third proxy for human capital.

Table 3 shows the mean values for the three dependent variables, as well as for some key
characteristics: rurality, distance to the capital, demographic composition and ownership
of assets. We see that on all indicators Rwanda does better than Burundi, and that these
difefrenes are statistically significant.

3.3 The Difference-in-Differences Estimator

To be able to estimate the effect of the policy change we compare individuals in Rwanda and
Burundi. Burundi, which is a neighbor of Rwanda, has not only a very similar demographic
composition with almost everyone being a Kirundi speaker, virtually the same language
as Kinyarwanda. Moreover, Burundi and Rwanda had similar language use-in-education

9
policies prior to 2008: use of Kirundi as the medium of instruction up till the upper primary
grades and then a switch to French for the rest of the educational system. They thus
constitute our control group of interest.

In Rwanda, the language policy change affects all individuals aged 12 and younger in
2008 in Rwanda, and they form our treated cohort. On the other hand, the cohorts aged
19-46 had already finished secondary schooling and were unaffected by the language policy
change. They form our control cohorts. In other words, the treatment is defined by the
interaction of the country dummy with the cohort dummy.

More specifically, we compare the cohorts aged 6-12, 19-25, 26-32, 33-39 and 40-46 across
Burundi and Rwanda using a difference-in-differences estimator. In particular, we estimate
the following estimating equation:

k=4
X
Hikrp = (Rwandai ∗ Ck )δk + ζk Cohortk + Xikrp + Xkrp + Φr + ϵikrp (1)
k=1

where Hikrp is the human capital outcome for individual i from cohort k, region r and
cluster p. The coefficients ζk are the one associated with the cohort dummy for the cohort
aged 40-46, 33-39, 26-32 and 6-12 in 2008 and captures the time trend in the dependent
variable, and where the omitted cohort is the one aged 19-25 in 2008. The cluster level
controls are denoted Xrcep and account for geographical features – growing season length,
proximity to water, slope, proximity to national borders, latitude and longitude – that affect
economic opportunities (Nunn and Puga, 2012), and thus have a bearing on human capital
outcomes. Finally, Xircp represents a set of individual-level controls for age, distance from
the border and urban residence. The standard errors, ϵikrp , are clustered at the level of the
primary sampling unit to account for spatial correlation. Φr is a set of administrative unit
1 fixed effects.

The principal coefficients of interest are δk , the ones associated with the interaction of
the dummy Rwandai that takes the value 1 when the individual i is from Rwanda and 0
when is from Burundi, and the the cohort dummy Ck . The use of a D-I-D estimator assumes
that the groups being compared have parallel trends in absence of the change in policy. The
omitted cohort is the individuals aged 19-25 in 2008 . This should imply that δk = 0 for
k = {2, 3, 4}, that is the cohorts aged 26-32, 33-39 and 40-46 in 2008. On the other hand,

10
δ1 the coefficient associated with the interaction between the dummy for the cohort aged
6-12 in 2008 and the dummy for the individual being from Rwanda captures the effect of
the change in language policy on the proxy for human capital.

We estimate Equation 1 for the entire country, as well as by restricting the comparison
to individuals who are residents with a certain distance from the border. This allows us to
restrict our comparison to contiguous areas ensuring maximum comparability between the
treated and control groups.

4 Results

4.1 Descriptive

Figure 1 shows the evolution on the three proxies for human capital by cohort and country.
We observe that there is an increase in educational achievement in both countries over the
34 year period the data spans. For the older cohorts, for all three proxies, Rwanda has
a substantial lead over their Burundian counterparts. These gaps tend to remain largely
constant, or slightly even increase, before the youngest cohort aged 6-12 in 2008 in Burundi
catches up with their Rwandan counterparts, or in fact on one indicator even overtakes
Rwanda. For instance, for the cohort aged 40-46 the average years of schooling for Burundi
and Rwanda was 2.26 and 4.27, respectively. However, for the youngest cohort aged 6-12
the average years of schooling stands at 5.97 and 6.38 for Burundi and Rwanda, respectively.
Turning to the category of an individual having at least entered secondary schooling, for
the cohort aged 40-46 in 2008, the shares for Burundi and Rwanda stood at 0.08 and 0.095,
respectively. These for the youngest cohort aged 6-12 in 2008 increases to 0.47 and 0.45 in
Burundi and Rwanda, respectively; so in fact, Burundi overtakes Rwanda on this indicator.
The descriptive evidence thus suggests that Rwanda had better human capital outcomes
compared to Burundi and maintained more or less a constant lead till for the youngest
cohort aged 6-12 in 2008, Burundi closes the gap. Thus, the evidence suggests that the
change in language policy had a negative effect on educational attainment in Rwanda.

11
4.2 Difference-in-Differences Estimates

Table 4 shows the results of estimating Equation 1. The dependent variables in columns
(1)-(2), (3)-(4) and (5)-(6) are years of education, a dummy for literacy and dummy for
entering secondary schooling, respectively. Column (1), (3) and (5) includes administrative
level 1 fixed effects and a dummy for being rural, whereas columns (2), (4) and (6) in
addition include geographical controls at the level of the primary sampling unit: distance
to the country’s capital, location equipped for irrigation, growing season length and global
human footprint, which is “created from nine global data layers covering human population
pressure (population density), human land use and infrastructure (built-up areas, nighttime
lights, land use/land cover), and human access (coastlines, roads, railroads, navigable rivers)”
(Mayala et al., 2018, 18).

On all three indicators, the cohort in Rwanda is negatively affected by the change in
language policy. It reduces years of education by around 0.54 years, the probability of being
able to read an entire sentence by 7.2% points, and the probability of entering secondary
schooling by 4.7% points relative to the cohort aged 19-25 in 2008. These are relative to the
average for the years of education, literacy and a dummy for entering secondary schooling
for the cohort aged 19-25 in Rwanda being 5.04; 0.69 and 0.19, respectively.

These estimates can be interpreted as the effect of the language policy change conditional
on the assumption that Burundi and Rwanda exhibit parallel trends prior to the change in
language policy. The coefficients on the interaction between the Rwanda and the cohorts
aged 26-32, 33-39 and 40-46 provide a test of this assumption. The cohort aged 26-32 is
seen to show parallel trends relative to the cohort aged 19-25 for all three proxies for human
capital. In fact, for the category for entering secondary schooling all 3 cohorts in the range
of 26-46 show parallel trends relative to the cohort aged 19-25 in 2008. For the other two
indicators, the cohorts aged 33-39 and 40-46 in 2008 from Rwanda do not exhibit parallel
trends, and in fact, are positive and significant. These suggest that Rwanda was in fact
outpacing Rwanda till the cohort aged 33-39, after which they exhibit parallel trends till
they fall behind.

Table 5 again estimates Equation 1 but now restricts individuals who live within 50kms
of the border from Burundi/Rwanda; where 50kms is the median distance from the border to
Burundi/Rwanda in the data. Qualitatively the results are similar to Table 4, however, the
negative effects of the language policy change are between 1.5 to 2 times in magnitude. The
key coefficients of interest, that is, the ones associated with the interaction of the Rwanda

12
dummy with the cohort dummy, accounting for all fixed effects and control is shown in
Figure 2, and correspond to columns (2), (4) and (6) of Table 5. The switch to English
is seen to reduce years of schooling by 1.08 years, the probability of being able to read an
entire sentence by 9.1% points, and the probability of entering secondary schooling by 6.5%
points.

5 Discussion

The results presented in Tables 4 and 5 provide evidence that is consistent with the change
in language policy in Rwanda having had negative effects on human capital formation. It
is difficult to say whether the policy shift was decreed for the acquisition of English for
economic advance or motivated by ideological sentiment. The fact that only twelve years
have passed since the implementation also implies that we cannot ascertain whether the
results we have outlined signify a short-term setback during the transitional phase before all
the issues are resolved within the policy, or a long-term descent into linguistic incoherence.
But we do know that at least a generation of Rwandan children have suffered, not realizing
the level of human capital gains that earlier generations were realizing.

If English acquisition was indeed one of the key motivations behind the current policy, one
can ask if there a better way to promote English fluency among the population? One of the
important binding constraints for the the ministry of education was teachers with adequate
language skills to be able to use English as the medium of instruction (Rurangirwa, 2011).
With that in mind, one potential route would be to have extended the use of Kinyarwanda
as the medium of instruction, at least through the entire span of primary schooling, and
focused on teaching English as a subject. The scant existing evidence from sub-Saharan
Africa seems to suggest that such a policy could have had higher returns both in promoting
generalized human capital, as well as better English language skills. For instance, Eriksson
(2014) employs the the Bantu Education Act, which intended to restrict knowledge of English
from the black population in apartheid South Africa, as a natural experiment and finds that
the he provision of an extra 2 years of local language instruction — instead of in English
or Afrikaans— had a positive effect on wages, the ability to read and write, educational
attainment, and the ability to speak English. Taylor and Coetzee (2016), again in the context
of South Africa, find that that provision of mother tongue instruction in the early grades

13
significantly improves English acquisition, as measured in grades 4–6. Laitin et al. (2019)
study an experimental program in the English speaking part of Cameroon that provided local
language instruction instead of English as the medium of instruction for the first three years
of schooling. They find that use of local languages as the medium of instruction had positive
effects not only on math test scores but also significantly improved English language skills.
This suggests that the objective of English language acquisition could have been better
achieved through using the available resources to train a small cadre of English-language
teachers, and the rest of resources bring devoted to improving pedagogical practices using
the language in which both teachers and students are fluent.

Here it is important to stress that there is no rigorous empirical evidence that allows
us to evaluate at which stage of schooling, and if at all, transition to English would be the
optimal policy from the point of view of maximizing human capital and English language
skills in the population. The existing practices from economically successful nations across
the globe suggest that investing into Kinyarwanda as the language-of-education across the
entire schooling system and teaching English as a subject might be the best way forward. In
this regard, we have in mind the equilibrium in the small states of northern Europe where
the indigenous languages (Dutch, Norwegian, Swedish, Danish) serve as media of instruction
through the schooling system, along with several disciplines of tertiary education, with En-
glish taught throughout as a subject. These also could have other benefits for postcolonial
states ranging from reclaiming cultural identity, empowering marginalized communities, chal-
lenging colonial legacies and fostering a sense of cultural pride and autonomy (Wa Thiong’o,
1986).

Finally, it is also important to note that one of the key constraints in promoting indige-
nous languages as official, that is, primary mediums for education, governance, and politics,
in contexts such as sub-Saharan Africa is the high levels of linguistic diversity. It has been
often claimed that high levels of diversity imply that the only way to assuage competing
group claims is by retaining an ethnolingusitically neutral language as the primary language
of education. In fact, in recent work, Laitin and Ramachandran (2024), provide a concep-
tual framework, and empirical evidence, consistent with higher levels of linguistic diversity
within a country resulting in a higher probability of retaining the colonial language. This so-
lution through ethnolingusitically neutral languages promotes the interest of a small section
of elites with knowledge of the former colonial language at the expense of the large majority
of the population (Laitin and Ramachandran, 2022). In this regard, Rwanda stands out
as a special case as it is characterized by the entirety of the population speaking a single
language, Kinyarwanda. This implies the political economy of which group’s language is to

14
be employed is a challenge that does not need to be surmounted. In fact, given that the
entire population of Burundi speaks the same language as in Rwanda, suggests there might
even be economies of scale for regional cooperation to invest into language standardization,
translation and teacher training.

6 Conclusion

The underlying idea of instrumental incoherence, and here summarizing J.P. Faguet’s state-
ment that informs this special issue and (Faguet and Shami, 2022), is that institutional
reform is often driven not by its main effects, but rather by its side effects, which are often
orthogonal to the main effects, but which solve some specific, short-term problem for those
pushing reform. So politicians propose a change in institutional structure not to enhance
public goods at lower cost, but because they want to serve short-term political goals such as
for pay-back to political enemies. This approach allows us to distinguish between unintended
consequences in institutional reform, which are unforeseen effects that are unsystematic, and
instrumental incoherence. In the latter, reformers’ incentives map onto the specifics of reform
design via their side effects, which in turn lead to the medium and long-term consequences
that are eventually realized.” In sum, Faguet’s instructions underline that “The goal [of sub-
mitted papers] is to highlight differences between the short-term incentives driving a reform
vs. its main effects.

In this paper, we have pointed to the mixed political motives that drove the 2008 language
reform in Rwanda from Kinyarwanda as the medium of instruction to English, and on the
side, marginalizing French at later points in the curriculum that was the principal language
of higher education of the teaching corps. In one fell swoop, President Kagame was able to
marginalize the Hutu professionals that had no foundation in English and French diplomats
who were charged with the promotion of francophonie. On these goals, he has been successful.

But as the data in this paper show, a generation of Rwanda students paid a cost in
human capital. The overall growth in Rwanda’s educational achievements flattened and a
generation of students faced a setback in their human capital. Moreover, Rwanda’s rural
students and those living close to the border fell behind with a lack of competent English
speakers in their schools and no social networks to provide them with exposure to English.
All Rwanda students of that period of transition found that their counterparts in Burundi
(a country with the same language but a weaker economy) were catching up on educational

15
achievement relative to Rwanda.

This pattern of results – political success hiding failure in the optimum provision of public
goods – is a telling example of institutional incoherence.

References
Albaugh, E. A. (2014). State-Building and Multilingual Education in Africa. Cambridge
University Press.

Constitution of the Republic of Rwanda (2003). Constitution of the Republic of Rwanda.

Impr. Scolaire.

Cummins, J. (1979). Linguistic interdependence and the educational development of bilin-

gual children. Review of educational research 49 (2), 222–251.

Cummins, J. (1991). Interdependence of first-and second-language proficiency in bilingual

children. Language processing in bilingual children 70, 89.

Eriksson, K. (2014). Does the language of instruction in primary school affect later labour
market outcomes? Evidence from South Africa. Economic History of Developing Re-
gions 29 (2), 311–335.

Faguet, J. and M. Shami (2022). The incoherence of institutional reform: Decentralization

as a structural solution to immediate political needs. St Comp Int Dev 57, 85–112.

Kraay, A. (2019). The world bank human capital index: a guide. The World Bank Research
Observer 34 (1), 1–33.

Laitin, D. D. and R. Ramachandran (2016). Language policy and human development.

American Political Science Review 110 (3), 457–480.

Laitin, D. D. and R. Ramachandran (2022). Linguistic diversity, official language choice and
human capital. Journal of Development Economics 156, 102811.

Laitin, D. D. and R. Ramachandran (2024). The historical sources of language policy.

Conditionally accepted at Journal of Politics.

Laitin, D. D., R. Ramachandran, and S. L. Walter (2019). The legacy of colonial language
policies and their impact on student learning: Evidence from an experimental program in
Cameroon. Economic Development and Cultural Change 68 (1), 239–72.

16
Mayala, B., T. D. Fish, D. Eitelberg, and T. Dontamsetti (2018). The dhs program geospatial
covariate datasets manual. Rockville: ICF .

Nunn, N. and D. Puga (2012). Ruggedness: The blessing of bad geography in Africa. Review
of Economics and Statistics 94 (1), 20–36.

Phillipson, R. (1992). Linguistic imperialism. Oxford University Press.

Prunier, G. (1997). The great lakes crisis. Current History 96 (610), 193.

Ramachandran, R. (2017). Language use in education and human capital formation: Evi-
dence from the Ethiopian educational reform. World Development (98), 195–213.

Republic of Rwanda (2006). Rwanda: Genocide ideology and strategies for its eradication.
Kigali: Rwandan Senate., Republic of Rwanda.

Rurangirwa, S. (2011). Literacy, language and learning initiative. Technical report, USAID,
Agreement No. AID-696-A-11-00006, FY12 - Q2 Quarterly Report.

Rurangirwa, S. (2012). The absence of kinyarwanda in the current language policy of rwanda.
The Journal of Pan African Studies 5 (8), 169–177.

Samuelson, B. L. (2012). Rwanda switches to english: Conflict, identity, and language-in-

education policy. In Language Policies in Education, pp. 210–232. Routledge.

Taylor, S. and M. Coetzee (2016). Estimating the impact of language of instruction in South
African primary schools: A fixed effects approach. Economics of Education Review 50,
75–89.

Tollefson, J. W. and A. B. Tsui (2003). Medium of instruction policies: Which agenda?

Whose agenda? Routledge.

Tuvuzimpundu, J. (2014). La place du français dans le système éducatif du rwanda, anglo-

phone depuis 2008. Le français et les langues partenaires : convivialité et compétitivité,
65–87.

Umulisa, A. (2022, August). Impact of language of instruction on repetition of primary

school children: Evidence from rwanda.

Wa Thiong’o, N. (1986). Decolonising the mind: The politics of language in African litera-
ture. Heinemann Educational Publishers.

17
 Table 1: The Changing Media of Instruction in Rwanda Public Education

Year Primary 1-3 Primary 4ff Note

1917 French French Belgium gets trusteeship
1967 Kinyarwanda French
2003 Kinyarwanda French or English English added by new Tutsi leadership as an official
language
2008 English French or English Teachers given crash course in English with implementation
in 2009
2011 Kinyarwanda English Failure to get teachers at a proper level of English and
only 2-5% of the population fluent in English
2019 English English Apparent input from private schools

18
 Table 2: Cohorts and sample sizes: Burundi and Rwanda

Age in 2008 Burundi Rwanda

(1) (2)
6-12 7074 11893
19-25 4903 8593
26-32 3784 6575
33-39 2621 4195
40-46 1007 1592
Total 19389 32848
Notes: The table shows the number of observations for each of the five cohorts included in our analysis for
Burundi and Rwanda, respectively. The data is drawn from the DHS: Burundi from the year 2016-17 and for
Rwanda from the years 2014-14 and 2019-20.

19
 Table 3: Key characteristics of the sample: Comparing Burundi and Rwanda

Burundi Rwanda
Mean Diff. P-value
(1) (2) (3) (4)
Completed years of education 4.496 5.480 -0.984 0.000
Dummy can read an entire sentence 0.670 0.743 -0.072 0.000
Dummy entered secondary education 0.270 0.262 0.008 0.034
Rural residence 0.784 0.762 0.022 0.000
Distance to capital (Kms.) 80.003 57.100 22.902 0.000
Age at time of survey 29.904 30.366 -0.462 0.000
Age of household head 44.029 44.361 -0.333 0.014
Household has: television 0.142 0.275 -0.133 0.000
Household has: radio 0.471 0.606 -0.136 0.000
Household has: television 0.142 0.275 -0.133 0.000
Household has: refrigerator 0.071 0.144 -0.072 0.000
Household has: bicycle 0.268 0.262 0.006 0.535
Household has: motorcycle/scooter 0.078 0.138 -0.060 0.000
Household has: car/truck 0.068 0.137 -0.069 0.000
Number of household members 6.137 5.348 0.789 0.000
Number of children under-5 in Household 1.097 0.857 0.240 0.000
Notes: The table shows the mean values on key variables for the sample of cohorts aged 6-46 in 2008. The data
is drawn from the DHS: Burundi from the year 2016-17 and for Rwanda from the years 2014-14 and 2019-20.
The sample sizes are shown in Table 2

20
 Figure 1: Evolution of three proxies of human capital by country and cohort
Coefficient on SC dummy*Low Quality Exam

7 6
Years of Schooling
4 3
2 5

40-46 33-39 26-32 19-25 6-12

Age in 2008
Coefficient on SC dummy*Low Quality Exam

Share entered at least secondary

.9

.5
.8

.4
Share literate
.7

.3
.6

.2
.5

.1

40-46 33-39 26-32 19-25 6-12 40-46 33-39 26-32 19-25 6-12
Age in 2008 Age in 2008

Rwanda Burundi

Notes: The sample consists of individuals aged 6-46 in 2008. The data is drawn from the DHS: Burundi from
the year 2016-17 and for Rwanda from the years 2014-14 and 2019-20. The sample sizes are shown in Table 2

21
Table 4: Effect of language of instruction on human capital: Difference-in-Differences esti-
mates comparing Burundi and Rwanda
Years of Literacy Entered Secondary
Education Dummy Schooling Dummy
(1) (2) (3) (4) (5) (6)

Rwanda*Age 6-12 in 2008 -0.54*** -0.54*** -0.072*** -0.072*** -0.047*** -0.047***

(0.15) (0.15) (0.014) (0.014) (0.016) (0.016)
Rwanda*Age 26-32 in 2008 0.10 0.087 0.0075 0.0062 -0.0046 -0.0051
(0.12) (0.12) (0.014) (0.014) (0.0099) (0.0099)
Rwanda*Age 33-39 in 2008 1.43*** 1.42*** 0.088*** 0.087*** 0.0065 0.0057
(0.14) (0.14) (0.017) (0.017) (0.011) (0.011)
Rwanda*Age 40-46 in 2008 1.09*** 1.10*** 0.032 0.033 0.012 0.012
(0.16) (0.16) (0.024) (0.024) (0.012) (0.012)
Age 6-12 in 2008 1.65*** 1.64*** 0.20*** 0.20*** 0.23*** 0.23***
(0.18) (0.18) (0.018) (0.018) (0.018) (0.018)
Age 26-32 in 2008 0.0072 0.025 0.0041 0.0056 -0.0077 -0.0069
(0.12) (0.12) (0.014) (0.014) (0.010) (0.010)
Age 33-39 in 2008 -0.86*** -0.82*** -0.099*** -0.097*** 0.023 0.025
(0.17) (0.17) (0.023) (0.023) (0.016) (0.016)
Age 40-46 in 2008 -0.85*** -0.82*** -0.11*** -0.11*** 0.052** 0.053**
(0.23) (0.23) (0.032) (0.032) (0.021) (0.021)
Age -0.048*** -0.049*** -0.0011 -0.0013 -0.0075*** -0.0075***
(0.011) (0.011) (0.0012) (0.0012) (0.0010) (0.0010)
Distance to Capital -6.4e-06*** -7.6e-07*** -8.8e-08
(2.0e-06) (2.0e-07) (1.8e-07)
Geographical Controls No Yes No Yes No Yes
Region Fixed Effects Yes Yes Yes Yes Yes Yes
Rural Dummy Yes Yes Yes Yes Yes Yes
Constant 6.07*** 5.79*** 0.69*** 0.72*** 0.41*** 0.36***
(0.34) (0.39) (0.039) (0.043) (0.032) (0.037)
Observations 42,768 42,768 42,709 42,709 42,777 42,777
R-squared 0.216 0.219 0.086 0.087 0.223 0.224
SE clustered by PSU
*p<.05; **p<.01; ***p<.001
Notes: Standard errors are clustered at the level of the primary sampling unit. *p<.05; **p<.01; ***p<.001.
The control group is Burundi and the omitted cohort is the individuals aged 19-25 in 2008. In columns (1)-(2),
(3)-(4) and (5)-(6) the dependent variables are years of completed schooling; a dummy for being able to read
an entire sentence; and a dummy for entering secondary schooling, respectively. The geographical controls refer
to controls for irrigation, growing season length and global human footprint. The region fixed effects refer to
administrative 1 level fixed effects.

22
Table 5: Effect of language of instruction on human capital: Difference-in-Differences esti-
mates comparing Burundi and Rwanda within 50kms from the Border
Years of Literacy Entered Secondary
Education Dummy Schooling Dummy
(1) (2) (3) (4) (5) (6)

Rwanda*Age 6-12 in 2008 -1.09*** -1.08*** -0.093*** -0.091*** -0.066*** -0.065***

(0.19) (0.19) (0.022) (0.022) (0.024) (0.024)
Rwanda*Age 26-32 in 2008 -0.18 -0.20 -0.0062 -0.0079 -0.015 -0.016
(0.15) (0.16) (0.020) (0.020) (0.012) (0.012)
Rwanda*Age 33-39 in 2008 1.18*** 1.18*** 0.11*** 0.11*** -0.014 -0.015
(0.20) (0.20) (0.027) (0.027) (0.014) (0.014)
Rwanda*Age 40-46 in 2008 0.86*** 0.86*** 0.018 0.020 -0.0096 -0.010
(0.23) (0.22) (0.035) (0.035) (0.015) (0.015)
Age 6-12 in 2008 1.99*** 1.96*** 0.23*** 0.22*** 0.22*** 0.22***
(0.22) (0.21) (0.026) (0.026) (0.025) (0.025)
Age 26-32 in 2008 0.35** 0.38** 0.016 0.020 0.00087 0.0020
(0.15) (0.15) (0.022) (0.022) (0.012) (0.012)
Age 33-39 in 2008 -0.41 -0.37 -0.095*** -0.088** 0.043** 0.045**
(0.25) (0.25) (0.034) (0.034) (0.022) (0.021)
Age 40-46 in 2008 -0.32 -0.29 -0.073 -0.067 0.084*** 0.086***
(0.32) (0.32) (0.044) (0.044) (0.028) (0.028)
Age -0.054*** -0.057*** -0.00092 -0.0014 -0.0081*** -0.0082***
(0.014) (0.014) (0.0017) (0.0016) (0.0014) (0.0013)
Distance to Capital -0.000010*** -1.5e-06*** -5.2e-07**
(2.4e-06) (2.6e-07) (2.0e-07)
(0.011) (0.011) (0.0012) (0.0012) (0.0010) (0.0010)
Geographical Controls No Yes No Yes No Yes
Region Fixed Effects Yes Yes Yes Yes Yes Yes
Rural Dummy Yes Yes Yes Yes Yes Yes
Constant 6.19*** 4.89*** 0.67*** 0.44*** 0.44*** 0.44***
(0.43) (1.25) (0.051) (0.14) (0.043) (0.11)

Observations 21,245 21,245 21,214 21,214 21,249 21,249

R-squared 0.243 0.247 0.097 0.103 0.227 0.228
SE clustered by PSU
*p<.05; **p<.01; ***p<.001
Notes: Standard errors are clustered at the level of the primary sampling unit. *p<.05; **p<.01; ***p<.001.
The control group is Burundi and the omitted cohort is the individuals aged 19-25 in 2008. In columns (1)-(2),
(3)-(4) and (5)-(6) the dependent variables are years of completed schooling; a dummy for being able to read
an entire sentence; and a dummy for entering secondary schooling, respectively. The geographical controls refer
to controls for irrigation, growing season length and global human footprint. The region fixed effects refer to
administrative 1 level fixed effects.

23
 Change in MOI policy

Figure 2: Difference-in-differences estimates- Comparing Burundi and Rwanda for sample

living within 50kms from the border
Coefficient on SC dummy*Low Quality Exam

Rwanda Dummy*Cohort Dummies

DV- Yrs of education
2

Change in MOI policy

Point estimate/CI
-1 0 -2 1

40-46 33-39 26-32 6-12

Age in 2008

Coefficient on SC dummy*Low Quality Exam

Rwanda Dummy*Cohort Dummies Rwanda Dummy*Cohort Dummies

DV- Literacy DV- Enters secondary schooling
.05
.2 .1
Point estimate/CI

Point estimate/CI
0
0

-.05
-.1

-.1
-.2

40-46 33-39 26-32 6-12 40-46 33-39 26-32 6-12

Age in 2008 Age in 2008
Change in MOI policy

Notes: The above plots the coefficients associated with the interaction of the cohort dummy with the Rwanda
dummy shown in columns (2), (4) and (6) of Table 5. The gray dashed line refers to when the policy change is
introduced in Rwanda and thus affects the cohort aged 6-12 in 2008.

24