Martin Wittenberg, 1962 - 2024
A version of this post was published in the South African Journal of Economics in December 2024.
Martin Wittenberg was born in Bethel, Germany in 1962; he died in Cape Town in 2024. He made major contributions to the intellectual life of South Africa through his scholarship, mentorship, and public service. The three of us have known him as students, as coauthors, and colleagues, and we write here to express gratitude for that privilege.
Below, we discuss some of his intellectual contributions in four areas: (1) building public goods that serve the research community and securing the foundations of empirical research for public policy; (2) data quality and measurement issues; (3) the economics of households and the labour market in South Africa; and (4) mentorship and teaching. Our discussion is not exhaustive. Nor do we aim to minimise his achievements as an anti-apartheid activist (having served as a secretary of the United Democratic Front in KwaZulu-Natal in the 1980s), or his personal qualities as a friend and as a father - but those aspects of his life are best described by others.
Public Service and Public Goods
Much of Martin’s life and work was dedicated to shoring up the foundations of empirical knowledge in service of the public good. An early example of his willingness to work towards that vision is his involvement with the Financial and Fiscal Commission’s work on intergovernmental allocations in the late 1990s and early 2000s. Martin helped to develop the “equitable share” formula that still guides how the National Treasury allocates funds to provincial and local government. Importantly, that formula requires information on economic conditions at the local level; for example, one of several variables that enter the formula is the fraction of the population eligible for social grants.1 But implementing the equitable share formula would then require accurate data - a mission that would come to define much of the last part of his career.
These same motives also drove his leadership of DataFirst, where his investments in the public good of data quality and accessibility became institutionalised. As DataFirst’s director from 2012 to 2024, Martin led a series of national workshops where empirical researchers worked to resolve many of the data quality issues mentioned below. Under his leadership, DataFirst conducted a capacity building programme, jointly with Statistics South Africa and SALDRU, in which public-sector workers could come to UCT and acquire new skills in statistics, coding and in certain relevant parts of economic theory (such as the theory of consumption or that of price indices). Over the ten years in which this postgraduate diploma was offered, over 350 students took at least one of its constituent courses. The administration of the resulting diploma programme was essentially unpaid additional work; Martin simply did it because it was important.
Measurement and Data Quality
Martin worked extensively to make post-apartheid survey data more reliable and more easily comparable across time. Often, this involved reweighting existing datasets (Branson and Wittenberg (2007), Branson and Wittenberg (2014), Thornton and Wittenberg (2022)), detecting when data had been imputed or altered in misleading ways (Wittenberg (2017a), Wittenberg (2017b)), detecting when changes in questionnaires or fieldwork practices had implicitly altered the design of a survey or - as a last resort - documenting the inconsistencies as a cautionary tale for others (Wittenberg (2004), Kerr and Wittenberg (2015)).
This line of work led to the creation, with Andrew Kerr and David Lam, of the Post-Apartheid Labour Market Series (PALMS), a consistently documented and largely “clean” dataset consisting of many cross-sectional surveys harmonised across time. PALMS currently covers the period 1993 - 2019 and is freely available from DataFirst (Kerr, Lam, and Wittenberg (2023)). In Wittenberg (2017a) and Wittenberg (2017b), Martin used these painstakingly constructed data to document an increase in wage inequality over the post-apartheid period, but with an important subtlety - while the upper tail of the wage distribution (say, the 90th percentile) has “pulled away” from the median, there has been much less of an increase in wage dispersion below the median (and even some compression at very low percentiles).2 He also showed, by comparing tax data to data in publicly available surveys, that very high incomes are substantially understated in surveys, while the earnings of the self-employed are relatively better captured by surveys; unfortunately, this work was never published (Wittenberg (2017c)).
Martin’s work on improving the quality of publicly available data will continue to pay dividends for many years not only to those interested in the South African labour market, but also to education researchers. The harmonisation and reweighting of the many surveys in PALMS, and the GHS series of surveys (Thornton and Wittenberg (2022)), are an easily accessible and transparently documented source for researchers wanting to study how educational outcomes have changed in the South African population over the last three decades.
Households and the Labour Market
Much of Martin’s research on the supply side of the labour market featured a sensitivity to the heterogeneity in people’s fortunes, and in particular to how gender, geography, education, and labour market institutions like unions could affect those outcomes. For example, work with his former student Miracle Benhura (Ntuli and Wittenberg (2013)) showed how, for black women, the cross-sectional association between labour force participation and characteristics such as education and urban residence strengthened over the first decade of democracy. Or, in work with former student Nicola Branson (Branson and Wittenberg (2007)), they showed how post-apartheid increases in unemployment were partially explained by increases in labour supply - that while “young people are leaving school earlier, while being better educated than their elders”, the labour market was unable to absorb them into employment.
That people live together in households is obvious, but the implications of that observation for economic questions can be subtle. Households are sites both of consumption (relevant for policy questions around the provision of services like electricity and water) and of production (relevant for labour supply questions, as well as many decisions related to investment in health and education). Recognising this, Martin pursued fieldwork over several decades at the Agincourt Demographic Surveillance site in Mpumalanga, together with public health researchers like Mark Collinson. Their joint work (eg Wittenberg and Collinson (2007)) showed that while national survey data appeared to show a decrease in average household size, driven in particular by the growth of single-person households, this was not true in the higher-frequency Agincourt panel. Kerr and Wittenberg (2015) later confirmed that the large increase in the number of single-person households was due to changes in sampling and fieldwork practices in the October Household Surveys. Despite being overstated by the national surveys, the decrease in average household size was a real trend, and Martin showed how this had resulted in a reduction in electricity access (Wittenberg, Collinson, and Harris (2017), Wittenberg and Collinson (2020)).
Martin’s career-long collaborations with colleagues in Agincourt demonstrated his ability and willingness to work across fields to improve data quality and to uncover important patterns in the economy. His work with public health researchers influenced how economic variables were included into ongoing surveys at a time when most demographic surveillance sites were not collecting any economic data. And his collaborations bringing together and the MRC/Wits Agincourt Unit resulted in the creation of the Agincourt Household Energy Panel dataset (DataFirst and MRC/Wits Agincourt Unit (2023)), which now allows social scientists a unique opportunity to study how access to modern energy spreads and impacts rural communities over two decades (Dinkelman et al. (2024)).
Because consumption data is often collected at the household level, and households consist of multiple people, it is difficult to make inferences about individual welfare. This disaggregation problem is one that the profession continues to struggle with; for his part, Martin was driven to innovation both in econometric methods and in the use of unusual data to try and learn about the distribution of individual welfare. In Lubotsky and Wittenberg (2006) and Wittenberg (2011), he developed methods for the use of multiple proxies as a substitute for an unobserved variable. The leading application in the South African case, as in many other developing-country contexts, is to proxy for the effects of wealth or income, often by the use of “asset indices” (Filmer and Pritchett (2001), Filmer and Scott (2012)).
One measure of welfare that is collected at the individual level, though, are anthropometric ones - i.e. height and weight. In Wittenberg (2013), Martin used these measures to attack the question of whether unemployment in South Africa was largely “voluntary” - as sometimes claimed by some economists. In that paper, he shows that body mass is indeed monotonically related to both wealth and income. As Martin succinctly put it, “The fact that the unemployed are lighter than the employed… suggests that they are not choosing this state.”
Though much of his work concerned households, Martin did not ignore the demand side of the labour market. Amongst local labour economists, he was an early adopter of the “flow” approach to labour markets that emphasises the dynamics of quits and firing (on the worker side) and of vacancy creation and hiring (on the firm side). This perspective is explicit in Wittenberg (2002), although it would be more than a decade before firm-level data, in the form of the Quarterly Employment Statistics survey, became available for research. Their estimates of job creation and destruction rates for South Africa were published in Kerr, Wittenberg, and Arrow (2014), showing for the first time that gross job flow rates in South Africa were of a similar order of magnitude to those found in OECD countries. While not decisive about the question of whether South Africa’s labour regulations deter firms from creating jobs - a commonly held belief amongst the public and policy commentators - these estimates represented a large jump in the quality of that debate, constraining the beliefs one could reasonably hold.
Martin as a Colleague and Mentor
Most people know that Martin taught quantitative classes in econometrics and statistics at the postgraduate level for many years. Anyone who has taken classes with Martin has felt the thrill and the terror of learning from someone who always seemed to know everything, and could always point out our mistakes! Those of us who gravitated towards him for advising, despite this terror, were willing to feel stupid around him because we knew that we would be better for it; that interacting with him would make us grow; that understanding how to seek truth mattered.
Martin always challenged his students, pushed them, and cared about their work and about them as human beings. We knew that getting critical feedback from Martin was like “tough love”; we needed his wise insights to improve. And to a person, every former student we have talked with holds the utmost respect for Martin’s intellect and his role as a mentor.
Many of his turns of phrase will stay with us forever: “garbage in, garbage out”; “do not conflate fancy methods with correct empirics”; “don’t beat the data until it confesses”; and “bloody” everything: things were “bloody amazing,” “bloody ridiculous,” “bloody bizarre,” “bloody mind-boggling” or sometimes “bloody boring”.
A distinguishing feature about Martin as a teacher is that he never stopped being a mentor to any of us. If you were taught by him once, you could always count on him to have sensible advice at any point in your life. This lifelong mentorship is just part of what made Martin special, and human. In personal correspondence, he noted that “Inequality will only be broken when the next generation is better taught,” and his life as a teacher was always with this aim in mind.
Martin was always at home in academia, coming as he did from a long line of Wittenberg missionaries, preachers, and academics. His intellect, his wit, his struggle background and his varied life experience meant that discussions with Martin were opportunities to learn and to gain a different perspective on the world. South Africa is better off for his life of fearless, tireless, dedicated service, and we will miss him.
Taryn Dinkelman, University of Notre Dame
Andrew Kerr, University of Cape Town
Jesse Naidoo, University of Pretoria
References
Footnotes
Wittenberg (2006) is an account of the political tensions that, in the aftermath of South Africa’s negotiated transition to democracy, led to the development of that formula.↩︎
The statistical relationship between education and earnings - the “returns to education” - has become stronger over the post-apartheid period (see, e.g. Branson and Lam (2021)). These patterns, taken together with the changes in the wage distribution described above, may be consistent with a model of skill-biased technical change, as extensively studied in other countries (Katz and Autor (1999), Acemoglu and Autor (2011)). Local economists are yet to fully exploit PALMS to study the extent to which trends in wage and employment patterns are explained by such theories.↩︎