Design for outcomes

Designing a device to save lives is not enough. People may not use it, or may not use it correctly. Or be unable to maintain it. Or …

I’ve seen analogous problems with statistical methods. People will not necessarily adopt a new statistical method just because it is better. And if they do use it, they may use it wrongly, just like medical devices.

(“Better” in the previous paragraph is a loaded term. Statistical methods are evaluated by many criteria: power, robustness, bias, etc. When someone says his new method is better, he means better by the criteria he cares most about. But even when there is agreement on statistical criteria, a superior statistical method may be rejected for non-statistical reasons.)

Related posts

2 thoughts on “Design for outcomes

  1. Looks to me like Google reader doesn’t like object tags within an XML CDATA. What does the non-Feedburner’d version look like?

Leave a Reply

Your email address will not be published.