Type 1 and Type 2 Errors

來源:Victor Niederhoffer
日期:2010/06/10

Errors in statistics are usefully classified as type 1 and type 2. A type 1 is a false positive or undue credulity and a type 2 error is a false negative or false skepticism. The greater you try to reduce the level of error in one the greater the likelihood of error in the other.

Don't reject reject

no effect hypothesis true correct type 1 error

no effect hypothesis false type 2 error correct

A useful way of considering the decision making is above. Consider for example the no effect hypothesis that a pill is not healthy. if it's not healthy and you say it's healthy you make a type 1. If it's healthy and you don't say it is healthy you make a type 2.

A certain agency that regulates drugs is famous for only considering the type 1 errors, making sure with endless and ruinous double blinds that type 1 errors are minimized to the excessive making of type 2 errors and keeping off magic bullets that would extend life span and health enormously.

There are many areas where these trade-offs between errors occur. For example in spam filters. You can reject good things, that's type 1. You can accept bad things– that's type 2.

Our own field often has trade-offs like this. The hypothesis that a system or set point for a trade is random is a good null hypothesis. If you accept the system, you're just incurring churning for a worthless randomness. If you don't accept the system, and it's good, why then you've lost some good money.

The decision to expand your business or trading is another area that crops up frequently. If you expand it you might get in over the head. If you dont expand it, you might miss the gold. The movement into a new field, or the engagement of an employee or employer is another frequent trade off of type 1 and type 2, gullible reaching versus excessive caution that frequently arises.

The usual way to trade off between the two types of errors is to consider the cost of both errors, and to balance your decisions based on the relative costs. Considerations relative to randomness, and variability must also be considered. Also, the myriad psychological biases that lead us to place too much reliance on avoiding the two types of errors that the cognitives have contrived with their silly experiments on college students et al.

What other trade-offs of type 1 versus type 2 do you see that mite be of use to market people or others and what better way to consider gullibility versus skepticism do you see?