Wednesday 21 October 2009

Offsetting behaviour and KPI measures

The first 5 minutes of this BBC video give wonderful examples of the perils of target based assessment in the public sector.

Set the target as being time on the hospital waiting list and they'll prioritize the easiest surgeries. Set the target as time to being seen in the Emergency Department and they'll assign a "Hello Nurse" (and not the good kind from Animaniacs either). Patients don't hurt targets about numbers left on gurneys if you take the wheels off the gurneys and call them beds. And so on.

Reminds me of the old stories of socialist planning bureaus. If the target were tons of widgets, you make one MASSIVE widget; if the target is number of widgets, you make all kinds of itty bitty ones. If it's number of shoes, why bother making both left and right?

The video goes on later about how this is all the fault of the market and of governments moving too much towards a market model, but I've never had my grocery store mess me about in this way. It only happens in the state sector where there is no performance measure OTHER than whatever the target is. In the competitive sector, profit and loss provide discipline. So if you hurt productivity trying to skew performance along one measure, the customers will make sure you pay for it. The video talks about problems in the UK's league tabling of schools where rich folks would move into good schools' zones, keeping poor kids from getting into good schools. But isn't the problem then less the league table and more the restrictions on zoning? If they'd have combined league tables with vouchers and zero zoning, wouldn't outcomes have been just a little different?

The lesson the BBC should have drawn was that giving customers vouchers and letting them choose their own providers, whenever possible (hospitals, schools yes; police, maybe not) makes an awful lot more sense than assessing bureaucratic performance against KPI measures when principal agent problems are rather severe.

No comments:

Post a Comment