Tuesday, March 17, 2009

Type I and II Errors

Every time I read about Type I and Type II errors, I have to look up the difference. I know what they are, but can't keep straight which is which. I think this is because the errors are sometimes defined in terms of whether an effect has been correctly identified, but elsewhere the errors are defined in terms of whether you reject the null hypothesis (which is quite the opposite). Thus, a Type I error is a "false positive" in the sense of finding an effect where none exists, but it could also be called a "false negative" in the sense of falsely finding that the null hypothesis is wrong. (Put another way, when I hear "Type I error," I tend to think, "Right, that's a false positive . . . but wait, does that mean falsely accepting the null hypothesis? Dang, I have to look it up again.").

Clearly, the two types of errors need to be renamed, as "Type I" and "Type II" are absurdly non-descriptive. And the names need to make clear that they're referring to how you've assessed the effect itself, not to how you've assessed the null hypothesis. A suggestion: "False Effect Error" and "False No-Effect Error." Any others?

5 Comments:

Blogger Paul Gowder said...

You're not alone. I'm technically some kind of social scientist, and I still can never remember which is which.

12:10 PM  
Blogger Dave said...

I can't remember which is which either, but I have always read about them in terms of rejecting the null hypothesis or not rejecting it.

6:55 PM  
Blogger Joseph said...

Type I: What you know that ain't so.

Type II: What you don't know.

1:38 AM  
Blogger Shalmirane said...

I keep them apart by referring to Type 1 as the seller's (part is rejected but not defective) error while Type 2 is a buyer's error (defective part is not detected and part is sold.)

1:30 PM  
Blogger Shalmirane said...

I like how Joseph put it. That should be taught in the classroom.

3:15 PM  

Post a Comment

Subscribe to Post Comments [Atom]

<< Home