For many children around the world, access to higher education and the labour market depends
on becoming fluent in a second language. This presents a challenge to education policy: when and
how in the school programme should a transition to the second language occur? While a large
theoretical literature exists, empirical evidence is limited by the difficulties inherent to measuring
the causal effect of language of instruction. In South Africa, the majority of children do not speak
English as their first language but are required to undertake their final school-leaving examinations
in English. Most schools offer mother-tongue instruction in the first three grades of school and
then transition to English as the language of instruction in the fourth grade. Some schools use
English as the language of instruction from the first grade. In recent years a number of schools have
changed their policy, thus creating within-school, cross-grade variation in the language of instruction
received in the early grades. We use longitudinal data on school characteristics including language
of instruction by grade, and student test score data for the population of South African primary
schools. Simple OLS estimates suggest a positive correlation between English instruction in the
first three grades and English performance in grades 4, 5 and 6. After including school fixed effects,
which removes the confounding effects of selection into schools with different language policies, we
find that mother tongue instruction in the early grades significantly improves English acquisition,
as measured in grades 4, 5 and 6. The significance of this study is twofold. Firstly, it illustrates the
power of school-fixed effects to estimate causal impacts of educational interventions. Secondly, it is
the first South African study (and one of a very few international studies) to bring robust empirical
evidence to the policy debate around language of instruction.