Substack

Friday, May 11, 2007

Statistical Analysis in Policy Making

I am a big admirer of Steve Levitt and I have found "Freaknomics" an extremely insightful piece of work. It is a must read for not only all economists, but also policy makers and administrators.
Taking cue from Levitt, I have indulged in two experiments over the past year-and-a-half. One on malaria control and another on road safety. The results have been nothing short of extraordinary.

Vijayawada, where I work as Municipal Commissioner, is one of the more Malaria endemic cities, with more than 6000 positive cases every year.
Government agencies like the Health department have excellent and readily available database, which if analysed can throw up interesting conclusions. We took the Malaria information for the last five years for all the 53 malaria sections in the City, and found certain striking similarities and trends. A significant proportion of the positive cases were from 10 sections. However, with no regard to this, the anti-malaria activities were carried out with the same intensity across all the 53 divisions. Further analysis of the individual cases in these 10 sections over the five year period showed certain colonies and streets as endemic areas. It was decided to carry out intensive anti-malaria operations in these areas, including an Information Education Campaign (IEC). The focus was to be on anti-larval activities and we discontinued the highly unscientific malathion fogging activity, hitherto the center-piece of anti-malaria activities of the Department.

The results over the past 15 months have been no less than stunning. The number of positive cases fell by over 60% from 6271 cases in 2005 to 2921 cases in 2006. (There were 5912 cases in 2004 and 4769 cases in 2003) In the ten high incidence sections, the number of positive cases fell by more than half from 2871 cases in 2005 to just 1131 cases in 2006. (The corresponding numbers for 2003 and 2004 were 2009 and 2604) The downward slide has continued unabated this year too. In the first four months, the figure for positive cases has fallen from 577 in 2006 to 242, a 58% drop.
And all this has been achieved with a much smaller consumption of anti-larval chemicals and minimal use of malathion.
It is now proposed to eliminate Malaria from the City by end-2010, and an action plan has been formulated.

The second experiment was on road safety. Vijayawada City is serviced by two very heavy density National Highway (NH) Corridors which are prone to a large number of fatal accidents. An analysis of the accident profile by the District Road Safety Committee revealed that a majority of the fatal accidents took place in two stretches of the NH, and between 9.00 PM and 01.00 AM. It also emerged that most of the accidents resulted from drunken driving. Absence of visible signboards at certain divider crossings and road intersections, also contributed to some of these accidents. The regular highway traffic patrolling used to be done with the same intensity (or lack of it!) throughout the day. In light of the analysed information, it was decided to intensify patrolling in these two stretches during the most accident prone periods.
Breath analyser tests were introduced at check points in these stretches, strict vigil was kept on highway side bars and wine shops, and blinkers and fluorescent sign boards were erected at select locations.
These efforts saw a drop of over 75% in Highway accidents in the subsequent months.

As the afore-mentioned two examples indicate, the role of statistical analysis and data mining in policy and decision-making is very important. Government Departments are typically excellent at data collection, given the large network of ground level staff. The challenge is to use this valuable data by sifting the wheat from the chaff and drawing causative conclusions. The conclusions should be rigorously tested with respect to the technical analysis and other determining socio-economic and political factors, before being exposed to policy or decision-making.

The major problem with such analysis is that of confusing co-relation for causality. If the diagnosis is misplaced or incorrect, the prescription is invariably wrong. To take a small example, why is the standard of education very poor in a typical Government Elementary school in a village? The analyses range from lack of interest among children and parents, teachers not showing interest, unqualified and badly motivated teachers, irregular attendance of teachers and students, to a host of other factors. The correct answer is probably a mixture of all the afore-mentioned. But the challenge for policy makers is to identify the most important determinant. This determinant may vary from place to place, and can be arrived at only by careful analysis, chief among which being the intelligent mining of the existing information base.

No comments: