Looking for Patten Makers Photography? Just click the name and you will be transported accross!

Thursday, November 7, 2013

One tailed F-test - Possible?

The F distribution is asymmetrical.
Some people believe that this means you cannot have a one tailed F test.
This is incorrect and is the result of what an F statistic represents and how a one tailed test works.
Here I will explain why you can have a one tailed F test despite the F distribution being asymmetrical.

1) The F distribution

The F distribution is asymmetrical because it is the distirbution of squared values. In the case of a distribution where F(1,X) it is simply the square of a t distribution. Those who are familiar with a t distribution now have all the informatoin they need to know that one tailed tests should be possible with an F statistic.


2) Calculating an F statistic

One area of misunderstanding comes from the fact that F statistic calculation involves squaring values. This means all values become positive leading some people to believe that the F statistic no longer conveys information about the direction of an effect. As one tailed tests rely on information about the direction of an effect these same people conclude that F statistics cannot be used for one tailed tests.

This is not correct.

While the Statistic itself contains no information about the direction of the difference the original data does.

This post is a stub and will be updated in the future.
If it appears to be useful to people the update will occur sooner.
Feel free to add questions and comments below to encourage attention and updates.

0 comments:

Post a Comment