SOCIAL SECURITY OP-ED
Proposed Family Poverty Grant is excellent in theory — but there are problems with its implementation
The burning question which has seized the government and civil society in the past few months is: what, if anything, will replace the R350 Social Relief of Distress (SRD) grant in March 2022? The SRD has been extended three times following protests, most recently after the July riots.
Some civil society formations and the Department of Social Development have been advocating for a new Basic Income Grant (BIG), while the National Treasury is reported to be in favour of a new household-targeted “Family Poverty Grant”, and others are interested in a further extension of the SRD grant.
We were part of a Southern Africa Labour and Development Research Unit (Saldru) team commissioned by the National Treasury to model and write a report on various options. The findings of this report have been the subject of some debate, with a particular focus on the report’s conclusions about the proposed Family Poverty Grant.
Our view is that the evidence suggests it would be premature to favour the Family Poverty Grant based on how it would work in theory, when actually implementing the grant is likely to reduce its advantages and pose additional problems.
Excellent in theory, but…
To interpret the report’s results, one needs to understand how the grant options were modelled. At its core, the method is quite simple: use nationally representative household survey data, identify which individuals in the data should be eligible for a particular grant, and assign them additional income equivalent to the grant’s value.
One of the main metrics in the report was a grant’s theoretical “efficiency”: the percentage of the grant spending that directly goes to increasing incomes below a particular poverty line. A higher efficiency score suggests greater “value for money” in terms of poverty reduction.
The Family Poverty Grant performed extraordinarily well on this metric. Because it is targeted directly at the poorest households with a household means test, it performs far better than the next most efficient options.
This seems to be the primary basis upon which the National Treasury has been advocating for the grant. But choosing the grant on this basis would ignore two key issues discussed in the report:
- The Family Poverty Grant would be by far the most difficult grant to implement; and
- The efficiency advantage of the Family Poverty Grant is especially sensitive to implementation failures.
Why would the Family Poverty Grant be so difficult to implement?
Unlike all of the existing social grants in South Africa, which go to individuals, the Family Poverty Grant would go to households, with household eligibility and grant size depending on the number of people in the household and their income.
This means that the government would have to implement a means test for every person in an applying household, and they would need to know precisely which individuals live in which households.
And yet the South African government does not have any registry which links individuals to specific households. Indeed, it is worth asking how a credible up-to-date registry could even be developed. In a country with very high rates of household adjustment, dissolution and re-formation, widespread informal housing arrangements, and the complex household structures associated with migrant labour and large extended families, how would the government even begin to verify who is a particular member of a particular household?
There is no existing administrative infrastructure and capacity to take on such a task in South Africa. For Brazil’s Bolsa Familia, which the Family Poverty Grant draws inspiration from, this kind of work is done by local municipalities. We have a long way to go before we have that kind of local capability here.
Special sensitivity to implementation problems
A further problem is that the extraordinary theoretical efficiency of the Family Poverty Grant comes directly from its very precise targeting. If you start to allow implementation errors in the targeting mechanism, the grant’s efficiency drops rapidly.
Economists usually worry about two types of targeting errors: false exclusion errors, where eligible people erroneously don’t receive the grant, and false inclusion errors, where ineligible people mistakenly do receive the grant. We are usually more concerned about exclusion errors, and indeed the report has a substantial and important discussion about severe exclusion errors associated with the current SRD roll-out. But when it comes to the efficiency of a grant, we are more worried about inclusion errors (though exclusion can be important here too).
The idea is that if the grant goes to the wrong people, that is money going to the non-poor, which means the efficiency score will drop. The graph below shows what the efficiency score of the Family Poverty Grant and modified SRD grant would be if we randomly allocated ineligible people to receive the grant (ie, if we create false inclusions). We also show the efficiency score of a targeted R624 BIG (discussed in the report), where we assume no inclusion errors as the targeting is automatically administered through the tax system.
The graph shows that with no false inclusions (the far left of the graph), the proportion of the total budget that goes directly towards reducing poverty (our measure of efficiency) for the Family Poverty Grant is 94%, while the SRD and R350 BIG have scores of 69% and 53% respectively. However, as the percentage of false inclusions increases (moving along to the right of the graph), the efficiency of the grants becomes more similar. With a false inclusion rate of 40%, the Family Poverty Grant has an efficiency score of 58%, the SRD has 50%, and the R624 BIG is still at 53% (by assumption).
As inclusion errors increase further, the theoretical efficiency advantage of the Family Poverty Grant begins to disappear — and in fact it can be worse than the other options at very high levels of error.
What is a reasonable false inclusion rate to expect? It’s hard to know, but a recent global review suggests that for means-tested household-targeted grants, inclusion error rates in the range of roughly 35% to 55% are plausible, though these are likely to be overestimates to some degree.
This exercise is more about illustrating a general concept, rather than a serious model of inclusion errors. We don’t know how implementation errors will actually occur, and the inclusion errors we model above are all about the means test rather than additional household registry issues. For this exercise we also assume no false exclusion errors.
But the general principle shown in the graph will hold regardless of these issues: not only is the Family Poverty Grant more susceptible to implementation errors than the other grants, but its efficiency score is also especially sensitive to these errors.
One other reason some may prefer the Family Poverty Grant is because it is cheaper than some of the other options: R59-billion per year rather than R71-billion for a modified SRD or R206-billion for a targeted R624 BIG (R114-billion for a similar R350 BIG, which also has slightly higher efficiency). But unless there are compensating exclusion errors, inclusion errors will also dramatically blow up the Family Poverty Grant budget, much more than the SRD.
Using the same model as in the graph above, 40% inclusion errors imply a budget of R120-billion and R118-billion for the Family Poverty Grant and SRD respectively, or R270-billion and R144-billion at a 70% inclusion error rate.
Modelling is not a crystal ball
The modelling approach used in the report obviously comes with some weaknesses. For example, it can’t account for how grant receipt might affect people’s behaviour or macroeconomic conditions. Most importantly for our discussion here, it also doesn’t include implementation errors.
One can always build a more complicated model. But economists, for the most part, don’t have a very precise and settled idea of how to model these things.
The explicit trade-off is a simpler model where you can more easily understand where the results come from and what caveats you need to keep in mind. But when using the output of such a model, you have to keep thinking about what the model ignores.
In addition, shifting to a household-targeted grant presents serious sociological questions, to which traditional economic analysis is often not very well suited.
Sociologists know that the household is not a neutral institution for the distribution of government money. Households and families can be places of love and mutual care, but they can also be sites of domination and exploitation, which are often highly gendered.
How much money will intended beneficiaries of the grant actually receive if the total grant is sent to their household patriarch or matriarch, rather than them having an individual entitlement? How will this affect intra-household relations? What are the implications, in a society with high levels of domestic violence, if the household registry is not updated regularly, and people are unable to leave households when they want, because the grant follows the household rather than them individually? The Saldru report calls for further sociological research before the Family Poverty Grant is implemented.
The report also notes that the narrow criterion of “efficiency of poverty reduction” is perhaps an inadequate ground for deciding social policy. In particular, other goals could include broadening the social grant safety net to include those vulnerable individuals not currently covered by an existing grant. This would mean recognising that it is not only children, the elderly, and those with disabilities who require state assistance, but a much wider range of unemployed people who are not served by South Africa’s exclusionary labour market. The SRD and BIG would do a much better job in this respect than the Family Poverty Grant.
The bottom line
We believe that it is premature to prefer the Family Poverty Grant over other options purely on the basis of its theoretical efficiency ranking. The model results need to be interpreted alongside implementation and other issues which are excluded from the model.
Indeed, this is what the report directly concludes: that the Family Poverty Grant is theoretically the most efficient but entails major implementation concerns, while the targeted BIG is relatively inefficient and easier to implement, and the modified SRD presents a kind of middle ground when it comes to efficiency and implementability. What we’ve shown here is that the theoretical efficiency of the Family Poverty Grant is dramatically reduced by expected implementation errors.
With these implementation issues in mind, there is currently talk of the National Treasury proposing a “pilot” of the Family Poverty Grant for next year, starting with one million households. Given the severe administrative challenges, it is unclear how even a pilot will be implemented, let alone in a way that truly informs us about the likely issues of a national roll-out. A rigorously evaluated pilot, however, is probably not a bad idea. If the Family Poverty Grant can actually be implemented well, and doesn’t have adverse sociological implications, it is a very attractive extreme poverty alleviation programme.
However, it seems strange to cancel an actually existing national programme, the SRD, to replace it with a temporary pilot which will reach far fewer people. What will the millions of current SRD recipients do while the trial runs its course? DM
Ihsaan Bassier and Joshua Budlender are research affiliates at the Southern Africa Labour and Development Research Unit (Saldru) at the University of Cape Town. They recently contributed to a report which evaluated social grant options to replace the SRD in March 2022, commissioned by the National Treasury. They write in their personal capacities.