The effect of the Universal Primary Education program on consumption and on the employment sector: Evidence from Tanzania

Highlights

• The massive primary education program implemented in Tanzania from 1974 to 1978 improved access to basic education.

• I adopt a two-sample estimation approach to predict household consumption for the entire population.

• The returns to education are higher in nonfarm self-employed activities, but are still positive in agriculture.

• Education raises the probability of working in agriculture for women.

• This switch in favor of agriculture is probably explained by the schooling curriculum, composed of agricultural classes.

Abstract

This paper uses the Tanzanian Universal Primary Education (UPE) program implemented between 1974 and 1978 to study the effect of education on household consumption and on labor market participation in a rural environment. Combining regional disparities of access to school with the timing of the program, I adopt a difference-in-difference approach. To estimate the returns to education for the entire population and not only for wage workers, I use a two-sample estimation approach to predict consumption for every household and find that education increases predicted consumption for household heads working in every sector. I also provide evidence that education increases the probability of working in agriculture for women. These results, at first surprising, suggest that education may influence the structural transformation and that returns to education are positive in agriculture, provided that skills taught at school are consistent with agriculture.

Introduction

Over the last decades, policy-makers have put education at the top of their agenda, and several governments of developing countries have implemented policies to universalize primary education. Yet, while the literature on the returns to human capital is extensive, little evidence exists on developing countries where the agricultural sector prevails (see Card (2001) for a literature review).

So far, Duflo, 2001, Maluccio, 1998 have examined the causal effect of education on earnings in developing countries, but both authors restrict their analysis to wage earners who constitute a very selected sample, preventing the extrapolation of the results to a broader population. To get around this issue, Duflo (2001) imputes a wage for self-employed workers, but this approach cannot be applied in agriculture-based countries, characterized by a small formal economic sector. In the nigerian context, Oyelere (2010) deals with the endogenous nature of education, but the author also limits her analysis to income earners.

Another strand of the literature estimates the returns to education among agricultural households by considering the agricultural production. Lockheed, Jamison, and Lau (1980) review papers estimating the impact of education on agricultural production and find very mixed results depending on the country and the specification of education. However, these studies do not consider the endogeneity of education and the estimations were prone to bias.

Using the extensive change in the access to education induced by the Tanzanian Universal Primary Education (UPE) program, this paper contributes to this existing literature by investigating the returns to education for the entire population. To include farmers and non-agricultural self-employed workers in the analysis, and to be able to compare the returns to education across the different sectors of activity, I estimate the effect of education of household heads on the predicted level of consumption of households. This predicted consumption is constructed by adopting a two-sample procedure from census data and household survey matching (Elbers et al., 2003, Tarozzi and Deaton, 2009).

The second contribution of this paper is to address the effect of education on the labor market participation, and more precisely, on the choice of the sector of activity. Education might not only increase earnings but also provide access to better-paid activities. Duflo et al., 2017, Ozier, 2015 investigate this question in African contexts, by focusing on secondary education. Duflo et al. (2017) find that an increase in secondary education in Ghana raises the probability of having a formal employment contract. Similarly, Ozier (2015) shows that in Kenya, children who complete secondary school are more likely to have a formal employment, but the effect is not always robust depending on the specification.

These questions are of public interest because even though the structural transformation is underway in rural Africa, agriculture remains the main source of employment, while agricultural productivity is still low and exposes population to poverty. The majority of field experiments in agriculture show that this low productivity is explained by a lack of information which prevents farmers from exploiting new technologies (De Janvry, Sadoulet, & Suri, 2017). Even when the information is available, farmers may fail to capture every aspect of the information (Duflo and Pande, 2007, Hanna et al., 2014). Social networks turn out to be useful in transmitting knowledge, but the available information is often incomplete (Cai, De Janvry, & Sadoulet, 2015), and the effectiveness is very context specific. Thus, this raises question of the relevance of investing in education. Is education, whose very essence is to transmit knowledge, effective in transmitting skills and in increasing household earnings in a rural environment ?

To assess the causal effect of education, I exploit the UPE program as a natural experiment. This program, implemented in Tanzania from 1974 to 1978, aimed at reducing disparities in access to education. In 1974, educational levels were low at the national level, with wide variations across regions arising from the activity during the colonial period. To lower these inequalities, the Tanzanian socialist government gave priority to disadvantaged areas, which led the latter to experience higher schooling expansion. The results were substantial: 3.3 million children aged 7 to 13 were enrolled in 1980, compared to 1.2 million in 1974 (Bonini, 2003). This achievement was made possible by the construction of new schools and by the villagization process which removed regional disparities of access to school by bringing together scattered homestead into community villages. The particularity of this UPE program was also to stand out against the elitist colonial education by providing a curriculum geared towards skills that are relevant for rural communities. In this respect, practical classes such as agricultural classes were incorporated.

The exposure to the UPE program of an individual depends on his age and on the regional educational attainment before the program was launched. On one side, the age informs whether the individual was young enough to go to school at the time of the program, and delineates the pre-treatment and the treatment age-cohorts. On the other side, the regional schooling supply illustrates the efforts that the government has made to remove disparities in access to schools. After controlling for region of birth and year of birth, the identification strategy combines information on the initial regional education level with information on the age of individuals to create a measure of the exposure to UPE program and to instrument education. A large literature exploits a similar identification strategy to study the effects of Free Primary Education programs on the education attainment (Al-Samarrai and Zaman, 2007, Hoogeveen and Rossi, 2013, Grogan, 2009), on the quality of education (Lucas & Mbiti, 2012), and on financial outcomes (Ajayi & Ross, 2018).

The first-stage results show that the education expansion increases with the exposure to the UPE program. This difference-in-difference estimate represents the effect of the UPE program on schooling, if there is no omitted regional time-varying characteristics correlated with the intensity of the UPE program.

To test the validity of the instrument, I attempt to identify the potential sources that violate the exclusion restriction assumption and I provide robustness checks. A major concern is that regions exposed to the program have specific characteristics that influence the education expansion. To account for these structural differences that can interfere with education, I control by a set of regional characteristics interacted with a time trend. The estimates may also be explained by a convergence phenomenon. To address this issue, I check the pre-trend to test whether the education expansion was higher in less educated regions than in the more educated ones before and after the introduction of the UPE program. Finally, I examine and discuss whether other programs occurring at the same time that the UPE program may be responsible of the results.

The 2SLS estimates show that the returns to education for the entire Tanzanian population are about 6 percent, which is close to what it is found for wage-workers in developing countries (Duflo, 2001, Maluccio, 1998). When I compare the returns to education across sectors of activity, I find that they are positive in every sector, including agriculture. I also find evidence that education increases the probability of working in farm activities for women. I explain this finding by the design of the program, whose aim was to provide agricultural classes and other practical classes in the curriculum. This paper demonstrates the importance of education and of the choice of the curriculum in helping households to increase their level of consumption in predominantly rural countries.

The remainder of the paper is organized as follows: Section 2 provides a broad picture of the Tanzanian context and describes the data as the main variables of the analysis. Section 3 outlines the estimation strategy and the effect of the UPE program on education. The results are introduced in Section 4 while Section 5 discusses the robustness of the results and Section 6 concludes.