Dear NBS experts,

I was wondering if I could get some help with the creation of the design matrix for a regression analysis with the NBS. I have been trying many ways but I don't know which is correct. I have one group and 2 variables (A, B); at first I modelled the global mean and the two variables together, in this way (Mean A  B):
1 0 10
1 0 10
1 0 8
1 0 10
1 0 9,25
1 0 10
1 3 9,25
1 1 8,50
1 0 9,50
1 0 10
1 0 9
1 0 9
1 0 10
1 2 10
1 0 8,50
1 0 8,25
1 0 7
1 0 10
1 0 10
1 0 8
1 0 10
1 0 9,25
1 0 10
1 3 9,25
1 1 8,50
1 0 9,50
1 0 10
1 0 9
1 0 9
1 0 10
1 2 10
1 0 8,50
1 0 8,25
1 0 7
1 0 10
1 0 7,75
1 2 9
1 5 9
1 6 8,75
1 6 7,25
1 5 7,25
1 8 9
1 14 8
1 9 8,25
1 4 8
1 10 7
1 3 7,25
1 3 9
1 3 8
1 8 9
1 4 7
1 3 7
1 5 9
1 3 6,75
1 5 7
1 5 7,75
1 6 10
I couldn't get any significant correlation but if I remove the mean something became significant. Then I also modelled both variables separated because I presume that they are correlated. I just want to know which is the correct design to test correlations in the NBS toolbox and if I can set contrasts for both the negative and the positive correlation, n this way for the example before:
[0 -1 0] [0 1 0]
[0 0 -1] [0 0 1]
And the last question, sorry, Do I have to set in the 'Statistical Test': One Sample?

Thanks in advance for your help,

Lorna.

Hi Lorna,

Thanks for your interest. Here are a few points to consider:

1. The column of ones should be included to account for the global mean. The significant result you found when removing the column of ones is most likely the mean effect (unless you have explicitly demeaned your measure of connectivity).

2. A contrast of [0 0 1] with a t-test will test for a positive correlation between your measure of connectivity and your third predictor variable. A contrast of [0 0 -1] with a t-test will test for a negative correlation. Finally, a contrast of [0 0 1] with a F-test will test for any correlation, both positive and negative (Based on your email, it seems this is what you want). So you have three options: [0 0 1] with t-test, [0 0 -1] with t-test or [0 0 1] with F-test.

3. The "One Sample" test is not relevant for your statistical design.

4. If your two predictor variables are highly correlated, you may want to consider removing one of the variables. There are other options to deal with correlated predictors.

Andrew

Thanks a lot, Zalesky, your response was extremely helpful, now I can keep forward.

Regards,
lorna.

Hello Professor,

For your second point here about positive and negative correlations with a t-test - would this be considered a multiple regression analysis or analysis of covariance analysis if you are adding covariates such as sex and age? Or simply a partial correlation?

Thank you in advance. 
Originally posted by Andrew Zalesky:

Hi Lorna, 

It is definitely not a partial correlation. Partial correlation would involve first regressing out the age and sex effects and then performing correlation on the residuals. 

"Regression controlling for the effects of age and sex" would seem like a good description to me. 

best,
Andrew
Originally posted by jules_ozark: