Mother-tongue instruction and later labour market outcomes: evidence from a natural experiment in Ethiopia

¹ Given the data we have, however, we are not able to disentangle how much of its effect is through its positive effect on students’ school performance and through its negative effect on language proficiency.

² Amharic has been the only official language of the federal government of Ethiopia since the Ethiopian history has been recorded.

³ 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. Zones are lower-level/local governments that can be considered as equivalent to US counties.

⁴ Here, it is worth mentioning that the 1994 education reform has other aspects as well. However, the only aspect of the reform that has affected schools in treated and control states differently is the introduction of mother-tongue instruction in primary school. Moreover, no other education policy change that affects schools in treated and control states differently has been introduced during the period of analysis.

⁵ See Seid (Citation2016) for further discussion on the 1994 Ethiopian education reform and its effect on educational outcomes in primary school in Ethiopia.

⁶ See Seid (Citation2019) for a discussion on the impact of mother-tongue instruction in early grades on the performance of students later in primary school after they switched to English-instruction classrooms.

⁷ It is worth mentioning that Southern Nations, Nationalities, and Peoples’ (SNNP) and Harari states also fall in the first and second group, respectively. Besides, there are additional states/cities (i.e. Addis Ababa, Afar, Benishangul-Gumuz, and Dire Dawa) that can potentially be grouped into a separate fourth group where students switch to English instruction in grade 7. But we do not focus on these states since we exclude them from our sample of analysis – see Section III for a discussion on sample restriction.

⁸ Since some children delay primary school enrolment and/or repeat grade, some individuals who were older than 14 years in 1994 might be partially exposed to mother-tongue instruction in primary school. We checked the sensitivity of our results to potential bias due to delayed enrolment and grade retention by using an additional specification which uses a more disaggregated age group. The results from this specification (not reported here) are qualitatively similar to those reported in our main analysis in Section IV, suggesting that bias due to delayed enrolment and grade retention is not a serious concern in our study.

⁹ More specifically, individuals aged $18-25$ and $34-41$ in 2013 are assigned into treated and control birth cohorts, respectively. Individuals younger than 18 years in 2013 are excluded since they might still be in school when the 2013 Labour Force Survey was conducted. Besides, this ensures that our analysis exclusively focuses on the labour market outcomes of adults by adopting a stricter cut-off age for child labour. On the other hand, we exclude individuals older than 41 years in 2013 so that the age range for individuals included in both treated and control birth cohorts is equal.

¹⁰ Of the total 9 states and 2 federal cities in Ethiopia, our final sample includes observations from 5 states. See Section III for a discussion on our sample restrictions and the reasoning behind the sample restrictions.

¹¹ The 2013 LFS data are supplemented by data from the 2-percent, public-use microdata samples of the 2007 Ethiopian population census when we check the robustness of our results. See Section III in the supplementary online appendix for a brief description of the 2007 Ethiopian population census.

¹² As mentioned earlier, the official school starting age in Ethiopia is 7, implying that students who were too young to attend primary school when the 1994 education reform was signed into law were younger than 12 and 18 years in 1999 and 2005, respectively. This suggests that data from the 1999 and 2005 LFS are not best suited to explore the labour market outcomes of individuals who have attended primary school after the 1994 education reform.

¹³ See Section II for further discussion on how individuals are assigned into treated and control birth cohorts.

¹⁴ See Table A1 in the supplementary online appendix for the patterns of internal migration by birth cohort and states’ treatment status.

¹⁵ Since individuals in our sample are a `selected’ group who have chosen not to migrate, however, our findings may not be generalized to individuals who have chosen to migrate out of their home state at some point in their life.

¹⁶ The ethno-linguistic diversity of Addis Ababa and Dire Dawa partly explains why primary schools in these cities have continued to use Amharic as medium of instruction even after the 1994 education reform.

¹⁷ The population in the states that are included in our final sample are relatively ethno-linguistically homogenous. Data from the 2007 Ethiopian population census reveal that the faction of population who are native speakers of Amharic in Amhara state (excluding Awi, Oromiya, and Wag Hemra zones) is 97%. The proportion of native speakers of Oromiffa in Oromiya state, Somali in Somali state, and Tigrigna in Tigray state are 89, 97, and 95%, respectively. Similarly, the same census data show that about 76% of the population in Gambella state are native speakers of the three dominant languages in the state (i.e. Agnuak, Nuer, and Mezhenger) which are also adopted as the only media of instruction in primary school in the state. See Table A2 in the supplementary online appendix for the population characteristics of relevant states in Ethiopia.

¹⁸ The selected four dependent variables are closely related to each other, but they capture slightly different aspects of individual’s labour market performance. Probability of employment, for instance, captures the likelihood of employment whereas permanent employment is a good proxy to the quality of employment since permanent employment provides job security and other non-monetary benefits to workers (at least compared to temporary employment). Hourly earning, on the other hand, is a good proxy to labour productivity while self-reported employee satisfaction captures the overall quality of employment as perceived by the worker.

¹⁹ See Table A3 in the supplementary online appendix for descriptive statistics of the labour market outcomes by birth cohort and the states’ treatment status.

²⁰ As mentioned earlier, we employ four labour market outcomes as our dependent variables. That is, probability of employment, and among those who are employed, we consider probabilities of permanent employment and whether an employee is satisfied with her/his current job and hourly earning as our dependent variables.

²¹ These fixed effects capture any secular changes occurring across Ethiopia in any given year.

²² Note that Equation (1) is estimated for three of our labour market outcomes that are defined as binary indicators (i.e. probabilities of employment, permanent employment, and whether an employee is satisfied with her/his job). For the other outcome variable, i.e. hourly earning, we estimate the following equation.

(2) $log(y_{is})={\alpha }_0+{\eta }_{0}T{C}_{is}+{\tau }_{0}T{S}_s+{\gamma }_0(T{C}_{is}\ast T{S}_s)+{\beta }_1{\mathbf{X}}_{is}+\psi +{\in }_{is},$(2)

where ${\in }_{is}$ is the idiosyncratic error term and all the other notations are as defined in Equation (1)

²³ The average marginal effects of the explanatory variables on the likelihoods of labour market outcomes are estimated by averaging the underlying partial effects over the distributions of the explanatory variables and the unobserved effects. The computation of the average marginal effects of the interaction term, in particular, follows the suggestion of Ai and Norton (Citation2003) which is shown to be a correct way of estimating average marginal effects of interaction terms in a general class of nonlinear models such as the one used in this paper.

²⁴ For the purpose of this study, an employee is considered as a private-sector employee if she/he works in a private firm/organization or non-governmental organization (including international organization). On the other hand, an employee is considered as a public-sector employee if she/he works in the government or government parastatal.

²⁵ In the public sector, it seems both Amharic, the official language of the federal government, and other local languages, mostly the official language of the state, are more valued. In the private sector, on the other hand, English (and to some extent Amharic and the official language of the state) seem to be more valued.

²⁶ It is worth mentioning that the magnitude of the negative labour market effect of longer exposure to mother-tongue instruction is not uniform across the four dependent variables employed in this paper. Table 2, for instance, suggests that the adverse effect of longer exposure to mother-tongue instruction is lower in employment than earning, with 28% decrease (4.3% to 3.1%) and 32% decrease (7.2% to 4.9%) for probability of employment and earning, respectively, when exposure to mother-tongue instruction increases from 4 to 6 years.

²⁷ Note that we also control for state-level unemployment rate and cohort-of-birth fixed effects in the regression framework, which presumably control for differences in macroeconomic trends between treated and control states. Thus, our results are less likely to be confounded by differential macroeconomic trends between treated and control states during the period of analysis.

²⁸ For uniformity, we exclude individuals older than 49 years from the placebo treated birth cohort group so that the age range in both the control birth cohort group and the placebo treated birth cohort group is equal.

²⁹ As mentioned earlier, for the purpose of this study, primary sector includes agriculture and mining; secondary sector includes manufacturing, utilities, and construction; and tertiary sector includes trade, transport, financial services, and community services.

³⁰ In the interest of space, we do not present results from the heterogeneity analysis here, but they are available upon request.

³¹ See Section III in the supplementary online appendix for a brief description of the 2007 Ethiopian population census.

³² It must be noted that, however, small sample size is not an issue here since we use census data.