Wednesday, July 29, 2009

RCTs as electoral strategy

Governments are naturally reluctant to subject their programs, especially those involving individual welfare handouts, to rigorous impact evaluation. What happens if the program is found to have not delivered on what it promised? This has confined academic researchers and program evaluators to analysing programs funded by Non-Governmental Organizations and multi-lateral institutions and thereby seriously limited their canvas of study. However, I am inclined to believe that it may be an electorally prudent strategy for governments and ruling party representatives to embrace program evaluations like randomized control trials (RCT). Here is why.

Program evaluation by way of RCTs can be beneficial for politicians facing elections besides providing valuable learnings about the impacts of various development program components. RCTs perforce divides the target group into two randomly distributed treatment and control groups. In other words, one half is administered the welfare benefit while the other half is denied the same.

The situation is tailor-made for the ruling party politician to offer the incentive of extending the program to the other half in return for re-electing him. This is likely to over-ride any ill-effects of them having been denied the first chance of sharing in the benefits of the program. They can also be incentivized with the prospect of a revamped program that is more effectively administered and whose punch for the beneficiary is greater, a result of the lessons learnt from the RCT evaluations. Presumably, the first half should be happy in two ways - at benefitting from the program and more so when his neighbour has not!

Interestingly, by the same logic, randomized phasing of welfare programs will also contribute towards the sustainability of the program even in the case of a government change. The overall success of the program in delivering welfare benefits and the fact that half the group have not got their share of the benefits will maintain the pressure on the new government to continue the program, if only to deliver the benefits to those hitherto denied their share.

Randomization into two (or more) groups is also beneficial to the ruling party in so far as they can now more optimally utilize the scarce resources available by covering only one-half of the population without losing the loyalty of the other half. They can therefore use the same resources to now cover two or more programs. Further, since the beneficiaries are randomly and transparently selected, instead of the usual partisan manner, the government may find it easier to rationalize away any discontent amongst those denied the benefits.

However, it is important that on the net, these programs deliver substantial benefits to those in the treatment group. This is rarely a problem since any new welfare program would deliver some benefit or the other to its beneficiaries, in its own ineffcient and poorly targeted manner. The challenge is only to design it to deliver the greatest bang for the buck and to the specific target group.

This approach was adopted in Mexico when the PROGRESA Conditional Cash Transfer scheme was first launched in 1998 in only half of the 506 targeted communities. It has been claimed that the randomized phase-in of the program (and the resultant ability to incrementally make changes based on the results of the RCT results) has played a major role in ensuring the continuity of the program despite the governmental change in 2000.

5 comments:

Anonymous said...

You said:
"has played a major role in ensuring the continuity of the program despite the governmental change in 2000. "
What is the design that will ensure continuity of the party in power? This would interest the policy makers a lot indeed.

Anonymous said...

You said:
"since the beneficiaries are randomly and transparently selected, instead of the usual partisan manner, the government may find it easier to rationalize away any discontent amongst those denied the benefits."

How is this done. Has it been practically possible anywhere. How much would such design ensure positive voter behaviour towards the ruling party.

How does this compare with the saturation approach. Any studies?

Anonymous said...

You said:
"The situation is tailor-made for the ruling party politician to offer the incentive of extending the program to the other half in return for re-electing him."

This might also engender animosity towards the ruling party too. Loss of credibility in terms of perception of public mindedness of the ruling party.

M Asgar said...

This kind of cut and paste approach should be discouraged
1)A ruling party should work for the benfits of all the sections of the society and not for only the people it got elected through.
2)Enblock voting for a particular party from a region is bygone, now that the vote difference between the elected and defeated candidates is meager and some time running into just hundreds of votes spread across the constituency and in any case selection of the region will become cumbersome thus driving the ruling party sympathetic voters of 'other' region into the opposition camp.

It is apt to repeat here the famous words of Tagore:
Where the world has not been broken up into fragments by narrow domestic walls ....
Where the clear stream of reason has not lost its way
Into the dreary desert sand of dead habit ....

Let us implement the policies based on the economic background and for the pure intentions it was carried out.

gulzar said...

1. the incentive to implement the program for the other half if the ruling party is elected to power may offer compelling attraction for those standing to benefit from it. As I had argued, those who benefit from the program are more likley than not to vote for the ruling party.

Interestingly, the Indiramma program in Andhra Pradesh, which sought to cover one-third of villages with all basic facilities at a time, could have been a good example, if the full program was spread over two sides of an election.

Maybe we could look at the results of the Panchayat elections, which were held after the first phase of villages were selected. Is it possible to control for other factors and find out whether it improved or reduced the prospects of the ruling party in those elections?

2&3. I have not been able to locate any studies on the impact of saturation approach followed by the Andhra Pradesh Government. But the experiment offers a natural setting for RCT researchers to compare the relative impacts of various welfare programs across phases. Were the programs more effective in the second and third phases, due to learnings from the first phase?

However, the fact that the Government could phase-in the program by randomly (not fully randomly though!) without any political or other opposition from those denied in the first and second phases, is enough to prove that such randomized phasing is possible. By all accounts, the ruling party did not attract any animosity due to this sequential phasing in of the program. Even a spiritedly obstructionist opposition did not make this an issue. It helped that those not selected in the first and second phases knew that they would get atleast in the third phase.