The 2021 Nobel Prize in Economics was awarded to David Card “for his empirical contributions to labour economics”, and to Josh Angrist and Guido Imbens “for their methodological contributions to the analysis of causal relationships”. In this article, we describe some of their main contributions to economics as an academic discipline, and then highlight some of the ways in which these insights and methods have been reflected in SALDRU’s work.
There are a few reasons why these three economists won the most prestigious prize in economics. On one level, their work has shaped our understanding of several extremely important topics in the world that we live in. These include labour market policy and regulations, minimum wages, returns to education, and the effects of schooling resources on educational outcomes. These are also important spheres where policy makers might intervene, and policies on these issues have the potential to impact on hundreds of millions of people. In addition to the welfare-improving impact of their work, they also popularized a set of empirical methods that improved the way that economists use microeconomic data to identify causal effects. And these methodological contributions, in turn, have shaped the evolution of the discipline for generations to come.
What is so difficult about empirically identifying causal effects? To begin, we need to first understand what most applied micro-economists mean when they talk about causal effects. This framework was derived from Donald Rubin’s work on causality and empirically identifying it, and is often referred to as Rubin’s Causal Model (RCM). In RCM, the ‘effect’ of a ‘treatment’ is always relative to ‘some other treatment’, which is also called the counterfactual. For example, suppose that we are interested in the efficacy of the first vaccine for COVID-19, the counterfactual would then be ‘no vaccine’. The average treatment effect of the new vaccine (relative to no vaccine) is given by the fraction of people who would get sick if everyone got the vaccine, less the fraction of people who would get sick if everyone did not get the vaccine. While this is reasonably clear intuitively, it raises two substantive challenges if one wants to use this framework empirically. The first challenge is immediate and obvious; it is literally impossible to find anyone who both received the vaccine and also did not receive the vaccine.
Naturally one would then think about looking at the incidence of infection amongst groups of people who were vaccinated, and comparing these infection rates with groups who were not vaccinated. We could thus use the unvaccinated people to obtain a measure of the counterfactual outcomes for the vaccinated group. In most cases, however, these would not provide an ideal measure of the counterfactual outcomes. The unvaccinated people may be different from the vaccinated people in multiple ways; they may be less well informed, they may have lower levels of access to vaccines, they might be in a different educational and income group, and they may be more risk-loving than the vaccinated group. For all of these possible reasons and many more, the comparison would be confounded and result in a biased estimate of the causal effect of the vaccine.
Thus, several econometric methods were developed to obtain unbiased measures of counterfactual outcomes. These methodological approaches are sometimes referred to as ‘natural experiments’, as they clarify particular sets of circumstances under which one can (sometimes) use real-world observational data to estimate causal effects.
A really neat illustrative example of one of these methods was presented in a seminal paper from 1994 on the effects of minimum wage increases in the United States by David Card and Alan Krueger. They were interested primarily in whether an increase in the minimum wage caused unemployment to rise, and if so, by how much? To answer the question, they studied the relative change in employment outcomes of fast food franchises that were close to the border between New Jersey and Pennsylvania, over a time period when New Jersey raised their minimum wage while Pennsylvania did not. Their approach measures the change in employment in the ‘treatment group’, i.e. New Jersey, from before the minimum wage was increased to after the law was passed; and compares this change to the corresponding change in the ‘untreated group’, i.e. Pennsylvania. This empirical approach is very widely used, and is intuitively called a ‘difference-in-differences’ (DiD) approach. Surprisingly, they found that the minimum wage did not cause unemployment to rise, at least not within the fast food establishments in their sample.
That study was one of the first to show that the relationship between minimum wages and unemployment may not be as clear cut as it was widely believed to be some thirty years ago, and led to what is now a vast international empirical literature on the effects of minimum wages. The current consensus has also shifted, and is much more nuanced than it used to be. Most labour economists at present would say that minimum wages sometimes cause unemployment, and sometimes do not, and that even when they do cause unemployment these effects are relatively small in magnitude.
Several other approaches have been developed to identify causal effects using observational data, and these have rather non-intuitive names such as instrumental variables (IV), regression discontinuity designs (RDD), regression kink designs (RKD), and propensity score matching (PSM) estimators. In the bullet points below, we list a selection of papers where we in SALDRU have used these methods in our South Africa focussed research. What is clear is that the 2021 Nobel Prize in Economics winners’ work has had a significant impact for applied empirical economics in SALDRU in particular, and in South Africa more generally.
- Pellicer, M., & Piraino, P. (2019). The Effect of Nonpersonnel Resources on Educational Outcomes: Evidence from South Africa. Economic Development and Cultural Change, 67(4), 907-934.
The authors make use of discontinuities in the government school funding formula to estimate the impact of non-personnel funding on school outcomes. Their results show a small but positive effect of resources on student throughput during the last years of high school, and on the number of students writing the matriculation exam. However, additional resources do not translate into a higher number of successful exams, leading to an overall negative effect on pass rates. (RDD)
- Ranchhod, V., Lam, D., Leibbrandt, M. and Marteleto, L., (2011). Estimating the effect of adolescent fertility on educational attainment in Cape Town using a propensity score weighted regression. Cape Town: SALDRU, UCT. (SALDRU Working Paper No. 59).
The authors estimate the effect of a teenage birth on the educational attainment of young mothers in Cape Town. They control for a number of early life and pre-fertility characteristics, and reweight their data using a propensity score matching process to generate a more appropriate counterfactual group. Accounting for respondent characteristics reduces estimates of the effect of a teen birth on dropping out of school, successfully completing secondary school, and years of schooling attained. (PSM)
- Branson, N., Hendry, J., Ranchhod, V. (2020). The effects of credit rationing on re-enrollment rates at a University in South Africa. Cape Town: SALDRU, UCT. (SALDRU Working Paper No. 274).
The authors use institutional data to measure the impact of credit rationing on re-enrollment rates at the University of Cape Town (UCT). Identifying variation is obtained from a policy change in the eligibility requirements for continued financial aid that occurred in 2015. The difference-in-differences point estimate is -0.074. They also estimate a difference-in-difference-in-differences model to identify whether the policy had heterogenous effects for relatively lower income students who received funding from the National Student Financial Aid Scheme (NSFAS). They find that the policy resulted in a 5.5 percentage point decrease in re-enrolment rates amongst students who were not previously on NSFAS funding, while the corresponding estimate amongst NSFAS students was approximately 13 percentage points. These findings suggest that credit constraints are binding on the decision to re-enroll, but only for a relatively small proportion of the students who were affected by the change in the policy. (DiD)
- Pellicer, M., Ranchhod, V. (2020). Estimating the effect of racial classification on labour market outcomes: A case study from Apartheid South Africa. Cape Town: SALDRU, UCT. (SALDRU Working Paper No. 259).
The authors use a change in the relative importance of appearance as compared to ancestry as an instrumental variable to identify the causal effect of racial classification in apartheid South Africa. Specifically, they identify the effects of being classified as White (instead of Coloured) on education, employment and income. The authors’ preferred estimates indicate that being classified as White instead of Coloured resulted in a more than threefold increase in income for men. This corresponds to approximately 65% of the difference in mean incomes between the two population groups. Their findings for women are inconclusive. (IV)
- Ranchhod, V., & Finn, A. (2016). Estimating the Short Run Effects of South Africa’s Employment Tax Incentive on Youth Employment Probabilities using A Difference‐in‐Differences Approach. South African Journal of Economics, 84(2), 199-216.
The authors utilise several waves of nationally representative data and implement a difference-in-differences methodology in order to identify the effects of South Africa’s Employment Tax Incentive (ETI) on youth employment probabilities in the short run. Their primary finding is that the ETI did not have any statistically significant and positive effects on youth employment probabilities. Any decrease in tax revenues that arise from the ETI are effectively accruing to firms which, collectively, would have employed as many youth even in the absence of the ETI. (DiD)
- Ardington, C., Wills, G., & Kotze, J. (2021). COVID-19 learning losses: Early grade reading in South Africa. International Journal of Educational Development, 86, 102480.
This paper establishes short-term COVID-19 learning losses in reading for grade 2 and 4 students from under-resourced school contexts. They find that in 2020 grade 2 students lost between 57% and 70% of a year of learning relative to their pre-pandemic peers. Among a grade 4 sample, learning losses are estimated at between 62% and 81% of a year of learning. Considering that in 2020 students in the samples lost between 56%–60% of contact teaching days due to school closures and rotational timetabling schedules compared to a pre-pandemic year, this implies learning to schooling loss ratios in the region of 1–1.4. There is some evidence from the grade 4 sample that the reading trajectories of children benefiting more from attending school pre-pandemic – namely girls and children with stronger initial reading proficiency – are more negatively impacted. Mitigating the long-run implications of these learning losses will require a significant pivoting of the education system to ensure that instructional practices are appropriately levelled to optimise learning. (DiD)
- Ranchhod, V., & Bassier, I. (2017). Estimating the wage and employment effects of a large increase in South Africa’s agricultural minimum wage. REDI3x3 Working Paper.
The authors estimate the short-run effects of a 52% increase in the agricultural minimum wage in South Africa in 2013. They find that the law had a substantial effect on the earnings of farmworkers who remained employed after the law came into effect, but that there was also a small and gradual decrease in agricultural employment. The descriptive evidence from the cross-sections indicate an increase in mean income per month of 17.9% about a year after the law came into effect. This coincided with a mean decrease in adult employment by this industry of about 8.2% over the same time period. Their difference-in-differences estimates indicate substantial increases in wages in this industry after the law, but this increase is not systematically related to an individual’s wage rate prior to the law. There is also only very limited evidence that employment losses were statistically significant after the law. (DiD)