Saturday, March 15, 2008

Measurement, Control and Negative Feedback

A few days ago I mentioned a problem of measurement, and I was delighted to get a bit of feedback in the comments there.

The "observer effect" is a term for when the act of measurement causes a change to the variable being measured. For example, using different examples to Wikipedia: a speed camera emits radar that strikes the object being measured, thus changing its speed, a flowmeter in a water pipe causes a restriction that impedes the flow.

But the effect is perhaps most significant in infinitesimal physics experiments, which do not concern me, and in behavioural economics, which does.

In sociology, in society generally, the knowledge that people are being observed causes those people to change their behaviour. Hence the need for double-blind studies and libraries of other control techniques. And when the actual thing that you want to measure cannot be measured directly, or if you want to predict future change, then you need to find a proxy measure.

It's amusing that Tim Harford in the FT has asked a very similar question today. Sometimes historical performance of the same variable will be the best measure, but for the "big economic measures" (interest rates, stock indexes, inflation, etc) we must constantly evaluate underlying variables looking for causal ones. There is much potential for statistical correlation techniques here.

Despite the practical difficulties with its implementation, the idea of an expenses-based evaluation of office need had theoretical appeal because the incentive works in the opposite direction to the benefit. This is what you would want from a measure designed to control anything, you want there to be "negative feedback". That phrase has new meanings in the user rated world of eBay and customer surveys, but I refer to it in the Engineering sense of Control Theory - negative feedback is an essential component of almost all systems to keep them operating within desired parameters. Ideally, the greater the force pushing something from the ideal, the greater the force of the automatic counterbalance. I've mentioned this before but my college tutors would kill me for the oversimplification and overgeneralisation.

I am conscious that I am combining two separate problems - how best to measure and how best to control. One day I'll try to elaborate on the fundamental differences, but I'll leave you with another quote from the same Financial Times article mentioned above: My guess is that it is just a matter of time before economists embrace methods from other disciplines in an effort to understand dynamic processes better than we do.

No comments: