Income Mobility and Needs-Based Targeting

So much for writing regularly again. No sooner did I commit to it, I got loaded with a whole load of extra responsibilities. As a result, the last 2-3 months have been a blur, and I only now feel like I can sit down and reflect a little. That and because I have something to say.

There's a very quiet public policy debate going on about the extent of social assistance the public sector should provide. The overwhelming majority, in part driven by distaste of the policies of the previous BN government, is to roll back cash transfers and institute tighter targeting of benefits, with a particular focus on the poor. My view on this is:

1. Don't throw the baby out with the bath water; and
2. Targeting the poor suffers the same problems as targeting any income-based subset of the population (like the B40)

To be fair, the whole concept of aiming benefits primarily at the B40 has always been half-baked. Since the B40 describes the bottom 40% of households, no amount of policy tinkering would ever change the fact that there will always be a bottom 40%. As a result, it was never clear what these policies (like BR1M) were intended to achieve, which made them rightly open to criticism. By contrast, a focus on eradicating poverty, whether you use an absolute or relative measure (again, a subject of policy debate), has a very clear outcome in view. More importantly, it's an outcome than can be measured, which is a necessary precondition to any successful policy intervention.

However, as I've highlighted before, income based targeting is subject to serious error rates, of anything between 40%-95% of their intended beneficiaries. But let me be more specific. Targeting is subject to two types of errors:

1. Inclusion error - someone receives a public benefit that they ARE NOT entitled to;
2. Exclusion error - someone doesn't get a public benefit that ARE entitled to.

From an outcome perspective, inclusion errors primarily have a political cost because they undermine public support for social assistance programs (see for example: Welfare Queen). Exclusion errors are more serious however, because they essentially mean policy failure - you are not reaching the people you are trying to help. And globally, poverty targeting programs are rife with exclusion errors. As a general rule of thumb, the more restrictive your definition of a target group, the greater the likelihood of exclusion error. The reason for this is pretty straightforward: when we speak of the "poor", it will not be the same group of people at any given point in time.

Income at the bottom of the distribution tends to be highly volatile. The sample data I use for analysis at work, from a set of about 30k individuals over a 10 year period, shows income persistence (defined as the probability of remaining in the same income bracket) dropping sharply as you move down the income distribution, from over 30% for the top 1%, to just 10% for people at the 20th percentile. The figure is slightly higher for the B10, but only up to 25%. In other words, for every 4 people you might tag as "poor", only one was "poor" 10 years later, implying that 3 people fell into "poverty" in the interim. I won't claim these are robust results, but I do take them as indicative. The degree of turnover implied here would be challenging even if you had reasonably efficient means of determining income levels (for example: full compliance with income tax returns), and the complexity of the problem rises if you don't.

Moreover, delivery of social assistance based on arbitrary income measures subjects people to threshold effects and causes perverse incentives. If you receive RM1,000 every month (or equivalent non-cash benefits) if you're below the poverty line, and get nothing if you're above it, you're actually better off remaining "poor". The threshold effect can be addressed by making benefits variable to the level of income e.g. Friedman's negative income tax proposal, but such a system would again be conditional on good data on incomes. Another issue is that any such system implicitly presumes that incomes are relatively stable and predictable, which is simply not true at the bottom of the income distribution.

Given all the above, my conviction is that a social assistance strategy targeted primarily at the poor will have the twin paradoxical results of: not actually helping most of the people it is intended to help, and keeping those it does help in poverty. The main impact is to save money, and salve the conscience of the rest of us.