Thursday, September 29, 2011

Type I and II

From time to time, it's worth revisiting our old friends Type I and Type II:
In general, a Type 1 error makes you depart from the status quo and take some action that is wrong, while committing a Type 2 error means that you stick with the status quo when you should have acted.

Maintain control
Decision theory is always a trade-off.

In hypothesis testing, you make sure that the Type 1 error is the severe error, so you can control the probability of making it.

As a result, the Type 2 error may in fact have a larger probability of occurring (but it is the lesser of the 2 errors).

A pharmaceutical company is testing the safety of a new drug:
Type 1 error: concluding it is safe when it’s not (kill people)
Type 2 error: concluding it is not safe when it is (lose some money)

The criminal justice system:
Type 1 error: convicting an innocent person (terminology: the defendant is guilty)
Type 2 error: not convicting a guilty person (terminology: the defendant is not guilty. It’s important to note that you never say the defendant is innocent).

A scientific researcher
Type 1 error: publishing your research when it’s wrong (embarrassment)
Type 2 error: not publishing your research when it’s correct (usually this means continuing your research until you can show progress)

Important terminology:
  • There is sufficient evidence to reject the null hypothesis in favor of the alternative.
  • The is insufficient evidence to reject the null hypothesis (NEVER ACCEPT THE NULL HYPOTHESIS – no one is ever declared innocent; they are declared not guilty)

I didn't make this stuff up. I credit a good friend, Dr. Pat Bond, a professor at Florida Tech, for these insights

 Bookmark this on Delicious  
Are you on LinkedIn?    Share this article with your network by clicking on the link.