The impact of learning first in mother tongue: evidence from a natural experiment in Ethiopia

ABSTRACT

This study explores the impact of mother-tongue instruction in early grades on the performance of students later after they switch to English instruction. Students in Ethiopia switch to English-instruction classrooms either in grade 5, 7, or 9 depending on the state in which they attend school. Typically, this switch is from mother-tongue to English instruction for language-majority students and from second-language to English instruction for language-minority students. As a result, the intensity of the impact of the switch to English instruction varies by language group. Exploiting these two plausibly exogenous sources of variations across states and language groups and using data from a school survey, we estimate triple-differences model. The estimate from our preferred specification suggests that learning first in mother tongue (in grades $1-4$) improves mathematics tests scores later (in grade 5) by 0.159 standard deviations, suggesting students taught first in their mother tongue learn in English better after they switch to English-instruction classrooms.

KEYWORDS:

• Medium of instruction
• primary school education
• triple-differences model
• Ethiopia

JEL CLASSIFICATION:

• I25
• O10
• O12

Acknowledgments

I thank Getachew Abegaz, Chanie Ejigu, Seid Hassen, the editor, and the anonymous referees for their helpful comments. In addition, I would like to thank the Center for the Study of African Economies (CSAE), Research on Improving Systems of Education (RISE) Program, and the Working Group in African Political Economy for their financial support to present this article at their respective annual conferences. Finally, I thank Young Lives (www.younglives.org.uk) for making the data used in this article available to me. Any errors are the responsibility of the author.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ The 1994 education reform also prescribes the introduction of teaching English as a subject starting from grade 1 and mandates English to be the official medium of instruction from grade 9 onwards in all schools in Ethiopia. On the other hand, students learn Amharic – the official language of the federal government – as a subject starting from either grade 1, 3, or 5 depending on the state in which the student attends school. It is compulsory for students to continue learning Amharic and English as subjects up to grade 10 and 12, respectively. Amharic is offered as an elective subject for grades 11 and 12 students.

² Ethiopia is a federal country with three levels of government: federal, state (or regional), and local. The country has nine states and two chartered cities.

³ Amharic has been the only official language of the federal government of Ethiopia since the Ethiopian history has been recorded.

⁴ It has been estimated that more than 90 languages are being spoken in Ethiopia (Bamgbose 1991).

⁵ The 2007 Ethiopian population census shows that SNNP state is the third largest state in Ethiopia in terms of population size, with a total population of about 14 million and accounts for about 19% of the population in Ethiopia.

⁶ Even if a number of languages are spoken in SNNP state, the five most widely spoken languages are the mother-tongue languages for more than half of the residents of SNNP state. Data from the most recent Ethiopian population census, which was administered in 2007, reveal that Sidamigna is the mother-tongue language for about 20% of the population in SNNP state. On the other hand, the proportion of population in SNNP state who are native speakers of Wolayitigna, Hadiyigna, Gamogna, and Guragiegna are 10, 8, 7, 7%, respectively.

⁷ Primary education in Ethiopia covers 8 years of schooling which are equally divided into two: the first (i.e. grades $1-4$) and second (i.e. grades $5-8$) cycles of primary education.

⁸ Gambella state also mandates students to switch to English instruction in grade 5. In this article, however, we do not focus on Gambella state since the state is not surveyed in YL 2012–2013 Ethiopia school survey which is the primary source of data in this article.

⁹ In Ethiopia, Oromiffa speakers are the numerical majority. For the purpose of the present study, however, the term language majority/minority is defined locally at community/school district level. For instance, if an Oromiffa-mother-tongue student lives outside Oromiya state and attends primary school where the medium of instruction is different from Oromiffa, then we consider the student as language-minority student.

¹⁰ Even if our analysis focuses on a switch from mother-tongue to English instruction in primary schools in Ethiopia, it can equally be relevant to other developing countries with similar contexts where the change is from mother-tongue to other foreign-language instruction.

¹¹ Alternatively, we can think of the treatment as having two components in SNNP state. The first one is exposure to mother-tongue instruction for 4 years, where only language-majority students are exposed to it. The second treatment is English instruction in grade 5 where both language-majority and language-minority students are exposed to it, but its intensity varies by language group.

¹² Even if YL has administered both mathematics and literacy tests, we exclusively focus on mathematics tests scores – see the discussion in Section 3 for our justification for this.

¹³ In Ethiopia, the school year begins in September and ends in June.

¹⁴ In bilingual education programme, students with limited English proficiency learn first in their native language whereas in English-immersion programme they are expected to learn in English from the beginning.

¹⁵ Using data from YL’s 2012–2013 school survey has an important advantage since information on both students’ mother tongue and the languages schools use as media of instruction are collected. This enables us to develop an identification strategy that exploits the differential impact of a switch to English instruction on students from different language groups.

¹⁶ Only a fraction of YL’s younger cohort children surveyed in the longitudinal household surveys were also surveyed in the school survey. This is partly because some of the YL’s younger cohort children were attending neither grade 4 nor grade 5 in the 2012–2013 school year and partly because some of the YL’s younger cohort children attend schools located outside of the YL’s sentinel sites.

¹⁷ This, along with the fact that Addis Ababa is the seat of the federal government, explains why Addis Ababa has adopted Amharic as its official language as well as the medium of instruction in primary schools located within its boundary.

¹⁸ Our data show that about 7 and 10% of students in SNNP and other states have, respectively, changed schools, but this difference is not statistically significant.

¹⁹ The lagged mathematics test score, i.e. mathematics test scores from the test administered at the beginning of the 2012–2013 school year (i.e. Wave 1) is included as one of the control variables in the regression we present and discuss in the following sections. Thus, in effect, we estimate lagged value-added model – see Section 4 for further discussion on this.

²⁰ On the contrary, it is difficult to extend this argument to literacy tests, where reading comprehension is part of the literacy tests and also where grade 5 students are being taught the majority language, Amharic, and English as subjects. Generally speaking, the language in which the literacy tests are administered may interact significantly with the contents of the tests themselves, making it difficult to interpret the results from the econometric analysis (which is discussed in Section 4) where the outcome variable is literacy tests scores in particular. Thus, we have decided to focus on and report results from the econometric analysis where the dependent variable is mathematics tests scores.

²¹ If, in fact, the language in which the mathematics tests was administered plays a major role in explaining differences in test scores of students who come from different language groups, then this will bias our estimates upward.

²² See Table A1 in Appendix A for descriptive statistics of the outcome variable, i.e. mathematics tests scores, by experimental state, language group, and grade.

²³ It is not surprising that test scores vary by language group since it has been documented that students in Ethiopia who are taught in their mother tongue perform better (see, e.g. Seid 2016). This should not, however, affect our identification strategy since we are here interested in exploring whether this difference in test scores continues in grade 5 after students switch to English instruction.

²⁴ The falsification test analysis, which is presented below in Section 5 and compares students in other states (excluding those in SNNP state) among themselves, shows that the coefficient estimate of the third-level interaction term in the triple-differences model is statistically insignificant. This is comforting as it indirectly suggests that there is no differential growth trend in test scores by language group in the absence of the treatment – see Section 5 for further detail on this.

²⁵ It is worth mentioning that the adoption of mother-tongue instruction in primary school as well as the legislation passed by SNNP state that mandates students in the state to switch to English instruction in grade 5 were implemented long before 2012–2013, the year when the data used in this article were collected. Thus, it is fair to argue that the possibility that the temporary disruption associated with implementing the 1994 education reform may bias our estimates is not a concern here.

²⁶ Even if YL has also administered separate longitudinal household surveys that can provide a wide range of information on household-level characteristics in Ethiopia, we have chosen not to use information gathered in the YL’s household survey. This is because only 493 students were surveyed both in the YL’s Ethiopian household and school surveys, implying that the small sample size does not allow precise estimation. Thus, we employ information collected on household-level characteristics through the school survey. However, as in many other school surveys, the YL’s Ethiopian school survey primarily focuses on collecting information on school inputs, and, hence, it has only collected limited information on the demographic characteristics of students’ families by directly asking students about their families.

²⁷ On the other hand, the coefficient estimate of the lagged mathematics test score itself is positive and statistically significant (see column 2 of Table 2), with estimated magnitude of 0.482. The fact that its estimate is less than 1 provides an evidence in support of imperfect persistence of achievement among Ethiopian students.

²⁸ This concern assumes that grade 5 mathematics teachers in SNNP state can speak the majority language (i.e. the language the school uses as medium of instruction in grades $1-4$). This, however, is not a necessarily accurate assumption. The language requirement to be qualified as grade 5 mathematics teacher in SNNP state is ability to teach mathematics concepts in English, not in a language the school uses as medium of instruction in grades $1-4$.

²⁹ In our data, we observe both the mother tongue of mathematics teachers and the languages schools use as media of instruction in grades $1-4$. Even if the mother-tongue language of the majority of grade 4 mathematics teachers is the same as the language the school has adopted as medium of instruction, this is not the case for grade 5 mathematics teachers.

³⁰ In these regressions, however, the coefficients are not precisely estimated due to small sample size.

³¹ This leads to assigning Amhara, Somali, and Tigray states into placebo experimental states and Afar and Oromiya states into non-experimental states.

³² Since the only thing that is different in the falsification tests analysis (relative to the main analysis presented earlier) is the absence of exposure to the treatment, the results from the falsification tests analysis indirectly provide evidence that the statistically significant positive effect that is reported in the main analysis (where students in SNNP state are included in the sample) is driven by exposure to the treatment.