This article critiques two 2026 academic papers that argue that the Covid-19 SRD grant led to reductions in household income and expenditure and food security. We show that a failure to understand the design of the grant led to flawed methods and incorrect interpretation of results. The authors conclude that cash transfer programmes can have negative “unintended consequences”, when in fact their results show that the grant is overwhelmingly received by income-insecure households, but that its value is too low to adequately alleviate food insecurity.

This article responds to two recent papers by economists Lateef Bello and Dorah Dubihlela on the welfare effects of South Africa’s Social Relief of Distress (SRD) grant. These papers grabbed our attention, for two reasons.

First, because there isn’t a wealth of empirical research that focuses specifically on the SRD grant. The messy nature of the grant’s administration makes it difficult to cleanly identify and study its effects. Quantitative evidence on the impact of the SRD grant, in particular, is relatively scarce, though one important study found that it improves labour market participation and outcomes. Another study estimated that it prevented 2- to 2.8-million people from descending into food poverty during the Covid-19 lockdown period.

Second, Bello and Dubihlela’s findings are counterintuitive, and contrary to what has been documented in a vast body of research on cash transfers, including the papers mentioned above. They find that the SRD grant is causally linked to reductions in incomes and expenditure, and that the grant propelled increases in food insecurity. In other words, receiving the SRD grant makes people poorer, and more vulnerable to hunger.

These findings are in line with neoclassical economic theory but defy decades of empirical research. Such theory predicts that giving people cash transfers would make them more likely to opt for leisure over work, even if it results in reduced incomes. Empirical research, however, shows that cash transfers increase household incomes and expenditure, including over and above the value of the transfer (i.e. they have a multiplier effect).

While we should always welcome contributions that challenge conventional wisdom, these unusual findings prompted us to scrutinise the papers’ methods. What we found was that a failure to understand the design of the grant led to flawed methodology and distorted results. This is of serious concern, since the authors use these findings to caution South African policy makers that expanding social protection can have “unintended consequences”. A close reading of their results, however, reveals that they should have drawn the opposite conclusion: that the SRD grant reaches people in poverty and should be increased and expanded to support food security and other welfare outcomes.

The core problem with the authors’ approach is a methodologically unsound presumption of causation based on the statistical correlation between receipt of the SRD grant and poverty. Although the authors note in passing that “The results of this study should be taken with caution, as our findings are correlational, not causal, based on our empirical specification,” they make causal inferences throughout their papers. We explore this contradiction in detail below.

Bello and Dubihlela apply the same methodology across their two papers, varying the outcomes of interest in each. They use a fixed effects regression, complemented by a quantile regression, using a pseudo panel constructed from the General Household Survey (GHS) (2020 - 2023). They find in their first paper that the SRD grant reduces household income by a whopping 55.1%. The authors interpret this to mean that the grant pushes people further into precarious economic conditions as households become overly reliant on it, to the detriment of other livelihood strategies (the classic dependency argument). They also look at the impact of the SRD on household expenditure, and conclude, strangely—for reasons we will return to—that the grant increases total spending on basic necessities while reducing expenditure on non-essential goods.

In their second paper, they find that the grant increases food insecurity. This time, they are a little more cautious in their interpretation of the results, saying explicitly that they might be explained in part by selection issues (though they do not do anything to mitigate this). However, they still go on to argue that their results “underscore the complexities of cash transfer programs, suggesting that immediate financial relief may not translate into long-term stability”.

To be fair, there is a plausible argument that SRD grant-holders experience welfare deterioration over time, although not through the work-reduction channel. As many social protection advocates would argue, the current design of the grant does potentially institute a poverty trap. Specifically, the extremely low means test penalises any attempts by grant holders to generate additional income, while the very low grant value allows only for bare subsistence.

But regardless of whether we could reasonably hypothesize that the SRD grant could contribute to declining expenditure over time, we nevertheless find the authors’ results unreliable. This is because they have fundamentally failed to understand the design and purpose of the grant, which leads them to select flawed methods, namely a fixed effects specification. Fixed effects is a panel data method that looks at the evolution of a variable over time, while removing the influence of all other variables – both observable and unobservable – that do not vary with time. Let’s say you wanted to understand if joining a labour union gave workers a wage premium. Fixed effects would consider only the workers who have transitioned from not being unionised to being unionised and track their incomes over this time. You could be fairly certain that any large increase in income is not due to any innate individual characteristics such as work ethic, ambition, or talent as those don’t vary with time. So, after controlling for easily observable characteristics like education or sector, you could look at the difference in wages and conclude the increase was caused by their unionisation.

The authors attempt to employ a similar logic to understand the relationship between the SRD grant and household welfare. While the GHS does not track the same households over time (i.e. it is not a panel survey), they construct a pseudo panel by matching households with similar characteristics but different SRD grant statuses at different points in the reference period. They can then treat these combinations of different households at different times as if they were one household experiencing changes over time. This allows them to construct a group of households that either went from not receiving to receiving a grant between 2020 and 2023, or vice versa, to study the causal effect of the SRD grant on household income, expenditure, and food security.

But this approach is entirely inappropriate for the SRD grant. This is because a change in SRD grant status is not neutral. Unlike the decision to join a union, which can occur for a range of different reasons in different circumstances, a change in the status of a household’s SRD grant receipt always happens because the household’s economic circumstances changed. The grant is poverty targeted. Let’s look at it in both directions. Assuming a perfect targeting system (which we know is not the case in practice), a household that goes from not receiving a grant to receiving a grant, has experienced a reduction in their income in order to qualify – for instance, someone lost their job. In the case of those who have aged into the SRD grant (i.e. turned 18), this would entail a graduation from the child support grant which is set at a higher value—again denoting a fall in income. Similarly, going from qualifying for the grant to not qualifying means someone from that household found employment or some other source of income that raised them above the threshold.

Grant status always, and only, changes because of a change in the underlying variables determining a person’s income. These, too, vary with time and so are not removed by fixed effects. The grant does not cause people’s incomes to fall or cause them to become food insecure. Instead, they are receiving the grant precisely because their incomes fell.

There are other issues with the papers. These should be instructive for anyone looking to conduct impact analysis on the SRD grant. First, there is a very peculiar treatment of expenditure. The GHS calculates a household’s total expenditure and places them into various categories accordingly. For example, a household that spends R800 every month would fall into category 5, while a household that spends R6 700 would fall into category 9. The authors simplify these categories a little but retain the basic structure. They then interpret these categories to indicate expenditure on specific groups of goods. To them, being in the first category is taken to mean greater expenditure on basic goods, as opposed to lower overall expenditure. So, their regressions show that the SRD increases expenditure in the first category and lowers it in the others, which they take to mean that SRD households increase their expenditure on basic goods. But in their data it is not baskets of goods being measured but total expenditure—so they should be able to conclude only that SRD grant recipients’ total expenditure is lower than that of non-SRD-grant recipients.

This confirms our point: that people receiving the grant are just poorer. They were, of course, still less poor than they would otherwise have been had they not received the SRD grant.

Second, the panel merging method they use is interesting, and could potentially improve the usefulness of the GHS if employed more often. But the concern here—although it is not clear how much it affects their results—is the categories used to match similar households, i.e. to construct a sample of households whose SRD grant status changed and treat them as directly comparable. The authors match households based on age, province, and household size. But they neglect many other factors that have been shown to influence a household’s likelihood to receive an SRD grant. We know that the SRD grant is plagued with erroneous exclusion, and that factors such as gender, digital and financial exclusion, and rural location can increase this risk. It is standard in research that employs matching to at least assess the quality of matches. But the authors neglect to do so and as it stands there is a significant risk that they are treating households facing vastly different risks of exclusion as if they were the same across the surveys.

Finally, caution should be taken when using the GHS to analyse the SRD grant. This is because of the phrasing of the question on grants. It simply asks whether one receives each specific social grant. It’s a straightforward question when looking at the longer-standing grants such as the child support grant, because receipt of these grants is regular and consistent. The SRD grant on the other hand has much more irregular payments because income-eligibility is assessed on a month-to-month basis—this makes it more difficult to answer the question of whether one receives it with a binary yes or no. Do you answer yes if you received the grant in the current month but were rejected for the four months prior? Do you answer yes if you typically receive it but were rejected in the month of the survey? The result of this confusing formulation in the GHS is that its data on SRD grant receipt widely diverges from SASSA’s administrative data, and much more so than for the other grants, as shown in Figure 1.

Figure 1

Using GHS data for SRD grant analysis, especially in panel studies, therefore carries a high risk of bias and inaccuracy. With regard to the Bello and Dubihlela papers, a particular concern is the potential misclassification of non-recipients: households identified as non-recipients may actually be SRD recipients with irregular access patterns. For example, a household might receive the grant, use it to improve their financial situation or access other support in one month, and subsequently lose eligibility in a later month because their income no longer meets the threshold. In a panel analysis, this would create a false comparison where the grant’s welfare benefits are overlooked entirely.

Despite these methodological shortcomings, the findings of the papers are still useful on some level. Although they do not show what the authors claim, they do demonstrate that the SRD grant clearly reaches households that are much poorer, spend less, and are more at risk of food insecurity than the average South African household. This is of interest particularly in the context of recent government claims that the social grant system is beset by widespread inclusion errors and fraud, which have been used to justify heightened verification requirements and funding cuts for social protection. The papers show that the grant is targeted to impoverished households. They also show that it is at too low a level to meaningfully improve people’s already precarious economic conditions.

The conclusion and policy recommendation should therefore be that the SRD grant needs to be made permanent, and that its value and coverage should increase, because far too many are still being left behind.

Written by Siyanda Baduza and Kelle Howson, Econ3x3

*Econ3x3 has offered Bello and Dubihlela a right of response to this critique of their work, but so far, they have not responded to us.