Sometimes in life there are things that you know,
instinctively, to be true; but you lack the proof with which to convince
others. It is then particularly gratifying when the necessary proof emerges.
I have experienced just such a moment of gratification with
the appearance of the Development Pathways’ paper on Targeting Effectiveness, "Hit and Miss"! I
have always felt that poverty-targeting, in all its forms, is a fundamentally
flawed approach to deliver social protection – see my other blogs on this site. The new paper
provides incontrovertible proof.
In an ideal world, poverty-targeting would be sensible,
which is why it intuitively appeals to those who are new to social protection, or
who don’t understand the true complexities of poverty. I would even go so far
as to accept that, if all of five conditions were met, there are circumstances
in which poverty-targeting might be the optimal choice. But we don’t live in an
ideal world, and the Development Pathways’ paper clearly demonstrates how very
unlikely it is – in the real world – to fulfil each one of those five
conditions, let alone to meet all five simultaneously!
For poverty-targeting to be the best option, all of the following five assumptions need to hold true:
- The poor represent a small residual group
- The poor will remain poor (and the non-poor will remain non-poor)
- Inequitable outcomes are acceptable
- The resource envelope is fixed
- You can accurately identify the poor
Let’s look at each of these in turn, in the light of the
evidence presented in the Development Pathways’ paper.
1) The poor represent
a small residual group
It only makes sense to undertake the expensive and
complicated attempt to target the poor if they represent a relatively small
portion of the population. And yet, paradoxically, it is in exactly these
circumstances that poverty-targeting is most difficult, as the paper shows. In
any case, it is absolutely clear from the paper that in all lower- and most
middle-income countries, the number of those who can be considered poor by any comprehensive
metric is high: usually in the region of 80 percent. It illustrates this with
graphs of the wealth distribution in Brazil and South Africa, and with a telling
figure showing incomes in purchasing power parity for five such countries:
In the USA, anyone living on less that US$10 a day would be
considered extremely poor. If some 80 percent of the population is poor, it
makes no sense to waste money on targeting: the cost of doing so would probably
exceed the cost of including the extra 20 percent in the programme, and you
would lose many of the political economy benefits of an inclusive universal
approach. Conversely, if a much smaller proportion of the overall population is
poor, then you still have a serious problem, because the paper shows clearly
that poverty-targeting is less and less accurate the lower the coverage – see
point 5 below.
2) The poor will
remain poor (and the non-poor will remain non-poor)
Poverty is extremely dynamic. The paper gives two examples of such churn in and out of poverty, from Uganda and Indonesia:
These graphs show that in Uganda, only 46 percent of households that were in the poorest quintile in 2013 had been in the poorest quintile in 2011; and, in Indonesia, 38 percent of households in the poorest quintile in 2015 had not been in that quintile just one year earlier. As the paper makes clear, all forms of poverty-targeting would miss this: even the very best such programmes only undertake retargeting every three or four years, and in the vast majority the interval is between five and ten years (for instance, neither Pakistan nor the Philippines has updated its PMT since 2009). There is simply no extant example of a poverty-targeted approach that could capture such volatility in wealth.
3) Inequitable
outcomes are acceptable
Even if it could, that would create another challenge. The
main reason for such volatility is that, in most low- and middle-income
countries, there is very little difference in the income (or consumption) of
the poorest 80 percent of the population[1],
so even a small change in income, caused by only a minor shock, can knock a
household a long way down the wealth distribution. Conversely – and this is
where the next challenge comes – making a cash grant to a selected poor
household will inevitably propel it far up the wealth distribution,
leap-frogging a number of almost equally poor households who are now poorer
than the beneficiary household. And yet the beneficiary household is likely to
go on receiving the same benefit for a number of years, while the now-poorer
households get nothing. This is inequitable, and frequently causes the resentment,
jealousy and social tension that are the inevitable by-products of
poverty-targeting.
4) The resource
envelope is fixed
If you assume that the fiscal space for social protection is
fixed and immutable, then there is some justification in arguing that the limited
resources should be focussed on the poor. As Devereux[2]
argues when presenting the case for targeting, “Given the reality of budget
constraints, scarce public resources must be used optimally and allocated
efficiently, where they can achieve the maximum impact. If poverty reduction is
an objective of public policy, social spending should be directed towards the
poor who need income support, not spread thinly over the entire population
including to those who do not need it”. But it is wrong to assume that poverty
reduction is the only – or even the main – objective of social assistance, and
it is naïve to think that budgets are immutable: investment in social
protection is a political choice. If politicians spy electoral benefit, the funding
will inevitably follow. As a result, the value of transfers is likely to be
much higher with universal programmes than with poverty-targeted ones, meaning –
ironically – that the poorest do better out of such inclusive programmes than
they do when they are specifically targeted.
5) You can accurately
identify the poor
Finally, we come to what the Development Pathways’ paper demonstrates
to be the killer assumption: that you can accurately identify the poor. It is
abundantly clear that you cannot…or at least that current approaches cannot.
The paper systematically analyses national household survey datasets from 23
countries, and examines the targeting accuracy of 38 social protection schemes or
targeting registries in those countries. It does this using two metrics of
targeting effectiveness: (i) the proportion of households incorrectly excluded,
when measured against the scheme’s intended coverage; and (ii) the degree of exclusion
of the poorest 20 percent of households from the scheme. Both give results that
can only be described as lamentable, especially when considering schemes that
aim to reach 25 percent of the population or less (representing two-thirds of
the programmes studied). In terms of exclusion against intended coverage, the very
best performer amongst these has errors of 47 percent, while 12 such schemes
have exclusion errors above 70 percent, 8 above 80 percent and 5 above 90
percent (meaning that fewer than one in ten of intended beneficiaries is
actually selected). In terms of exclusion of the poorest 20 percent of households
(which should logically all be included in a programme aimed at 25 percent of
the population), the results are similar: the very best programme excludes 46
percent of the poorest quintile of households, while 12 schemes exclude more
that 70 percent of the poorest wealth quintile, 9 exclude more than 80 percent
and 4 exclude more than 90 percent – indeed, the worst-performing manages to
correctly identify fewer than one in twenty of these poorest households!
The purpose of the paper is not to shame the worst
performers, but to demonstrate that the whole approach is fatally flawed. Yet
there is a remedy: to increase programme coverage and – ultimately – move to
universality. The paper summarises this on a very instructive chart, showing
the relationship between coverage and exclusion error:
The best-targeted programmes, using both metrics applied in
the paper, are those which have higher coverage of their target populations.
Thus, the best results using a PMT, though still not very good, are for the two
such programmes where coverage exceeds 40 percent of the intended population: the
old age component of Ecuador’s Bono
Desarrollo Humano and Uruguay’s Asignaciones
Familiares, both of which exclude “only” some 30 percent of their intended
recipients and 17 percent of the poorest quintile of households. Similarly, the
best example of community targeting, admittedly out of an alarmingly poor set
of results, is Rwanda’s Ubudehe
classification, covering 30 percent of the population, although this still
excludes over half of those intended and half of the poorest quintile of
households.
As coverage increases further, exclusion error decreases. South
Africa’s Child Support and Old Age Grant, which are means-tested to exclude the
most affluent, have over 70 percent coverage of their target populations:
exclusion of the intended beneficiaries is only 13 percent and 8 percent
respectively; and exclusion of the poorest quintile of households is zero in
both cases – in other words every household in the poorest quintile is included,
a remarkable achievement. As we move to the four universal programmes analysed
in the paper, this finding is further reinforced. Bolivia’s Renta Dignidad and Bono Juacinto Pinto, Georgia’s Old Age Pension and Mongolia’s Child
Money Programme all exclude fewer than one in ten of their intended
beneficiaries and fewer than one in ten of the poorest households.
The bottom line is that poverty-targeting does not work. The
best way to help the poor is not to target them, but to move to higher-coverage
programmes, and ideally to universal life-course approaches! I know I have said
this before…but now I feel I have the evidence to prove it! Thank you,
Development Pathways!
[1] It
is much easier to distinguish the wealthiest 20 percent in such cases, which is
probably why the high-coverage, affluence-tested programmes studied in the
paper – in particular South Africa’s Old Age and Child Support Grants – perform
so well.
[2] Devereux, Stephen, 2016. Is targeting ethical? Global Social Policy, I-16.
No comments:
Post a Comment