virus: alpha and beta errors

Eric Boyd (6ceb3@qlink.queensu.ca)
Sun, 14 Mar 1999 23:00:20 -0500

Hi,

I said:
<<
Type one errors (alpha) result when you reject a hypothesis even though it is true (as above). A type two error (beta) occurs when you do not reject a hypothesis even though it is false.

The problem is that these two error are related -- if you decrease the probability of an alpha error, the probability of a beta error goes up, and vice-versa.
>>

KMO <kmo@c-realm.com> writes:
<<
> Now, depending on your outlook, you can decide which of > the two errors would be worse, and tip the scales accordingly.

Wow! You can do that? Right on. Do you have some managable algorithm for calculating the expected utility of an alpha mistake and a beta mistake to determine when and which way one should tip the scales? Maggs may recall from logic class some methods for calculating expected utility, but they aren't particularly managable, especially if you don't have a pen and paper and a few minutes to perform the operations.
>>

Well, as you said below, a good guide is your intuition... with friends you trust, you can afford to tip the scales more towards beta errors, in fact, I suspect most people do this naturally. As humans, we also have a very natural tendency to tip towards alpha errors in hostile situations, perhaps more so than is good for us.

However, I'm sure that it would be possible to construct formal models based on numerical schemes to decide when to tip and which way. As with all such models, the major weakness is in "numerifying" the data for analysis.

<<
Do you make the decission to tip the scales intuitively? Do you perceive a comfortable semantic distance between "having confidence in the reliablity of your intuitive judgements," "trusting your intuition," and "having faith that you will intuitively know when to tip the scales and which way to tip them?"
>>

Sure. I can trust myself without invoking any kind of "evidence of things unseen". Generally, I closely monitor my own intuitions to see if they pan out, and work at training them to be more accurate. That's what engineering is all about.

<<
And gullible or dogmatic, you're easily manipulated.
>>

Exactly.

ERiC