Let's say I have an estimated parameter $\beta$ from a general linear model for 2 subject groups (with the $\beta$s corresponding to their means).

If I wanted to do a two sample t test in a particular way I can define a contrast with which I can formulate the t score as:

$$T = \frac{contrast\;of\;estimated\;parameters}{sqrt(variance\;estimate)} = \frac{c'\beta}{sqrt(variance\;estimate)}$$

where c' is a row vector (for example $[-1\;1] $), and $\beta$ is a column vector $[ \beta_1\;\beta_2]^T$. So for the t score one gets a scalar number.

My question is what happens when I have 3 groups and I would like to test multiple contrasts. I gather I should use an ANOVA which uses the F test rather than the t-test. But I don't quite understand how a statistical software would visualize an F score if one is testing for the overall differences between the groups using an F contrast. For example:

$$c' = \begin{bmatrix} 1 &-1 &0 \\ 0 & 1 & -1 \end{bmatrix}$$

Is there a way to formulate the F score for 3 groups like in the case of the t score for 2 groups.(the estimated parameters being $[ \beta_1\;\beta_2\;\beta_3]^T$ ) I would expect the F score to be a vector in this case. Am I wrong in assuming this?

Many thanks


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.