Dear all,

As per title, how to infer the direction of significance of a test, i.e. regression coefficients? I cannot find anything like that in the output variables.


Thank you in advance,
Ramtin

Hi Ramtin, 

this is quite a general question. 

Some considerations: For the t-test, the test is always one-sided; so the direction can be chosen based on the sign (positive or negative) of the contrast. For example, [-1 1] would test the opposite direction to [1 -1]. The F-test is two-sided and thus directionality cannot be tested. 

If you have a more specific example, I can provide more detailed advice. 

Andrew
Originally posted by ramtin:

Dear Andrew,

Many thanks for your response.

My question relates to any F-test I might perform with NBS. For instance, difference between two groups, contrast [1 1]. Or correlation with a covariate, contrast [0 1], with the first column of the design being the intercept.

At the moment, after a F-test I infer the direction by directly looking at the obtained result, e.g. boxplot or scatterplot of the average connectivity strength within the significant component for all subjects. Is not there any way to instead extract the fitted model and based on the coefficients infer the direction?


Thank you,
Ramtin

Hi Ramtin, 

the F-test is not a signed test. This is not specific to NBS but a general property of the F-test. 

You could do post hoc t-test to determine direction or look at the boxplots, as you are doing (that's perfectly fine).

Andrew 
Originally posted by ramtin:

I understand, many thanks for the clarification.


Kind regards,
Ramtin


Originally posted by Andrew Zalesky: