Dear Donald

thank you very much for the great reference!

In order to chase any doubt, please correct me if I am wrong.

In order to test wether distinct treatments have a significantly different impact on my brain outcome I have to valid options:

1. either test for an interaction between group and time in my flexible factorial design

2. use a two sample test in order to test if the "difference" (pre-post) significantly differs among groups. E.g. (Jbefore-Jafter) vs (Obefore - Oafter).

Thank you a lot for the great feedback and for your time!

Best

Lorenzo

On Mon, Mar 7, 2016 at 10:45 PM, MCLAREN, Donald <[log in to unmask]> wrote:

See below.

Best Regards, 

Donald McLaren, PhD

On Mon, Mar 7, 2016 at 7:57 PM, lorenzo pasquini <[log in to unmask]> wrote:

Dear all

thank you very much for the kind responses.

I would greatly appreciate if you could help me with following questions, in order to me understand what is going on with my data.

1. dear donald, what do you mean exactly with the contrast is not valid? why are the error terms incorrect? is there any alternative way of performing a main effect of group?

It's not statistically correct. The T/F-value is typically inflated due to incorrect df and using the wrong error term.

The error term of the GLM with repeated-measures represents the within-subject error. The within-subject error in inappropriate for between-subject comparisons.

If you want the main effect of group, average the conditions for each subject, then use a one-way anova with the average images.

 

2. as previously stated, I have a highly significant main effect of group. the same region appears, when performing the t test J group after treatment bigger O group after treatment, though at a very liberal threshold. This region is obviously highly significant, when I reduce my t tests by using a inclusive region of interest as mask. No group differences are present for the contrast J before vs O before. Can I draw the conclusion that my data indicating that there is a difference between groups treated with distinct treatments?

No. See Nieuwenhuis et al. 2011 (full reference below) as to why you can't make this conclusion.

Nieuwenhuis, S., Forstmann, B. U., & Wagenmakers, E.-J. (2011). Erroneous analyses of interactions in neuroscience: a problem of significance. Nature Publishing Group, 14(9), 1105–1109. http://doi.org/10.1038/nn.2886

Thanks a lot

LOrenzo

On Mon, Mar 7, 2016 at 12:13 PM, MCLAREN, Donald <[log in to unmask]> wrote:

As Helmut pointed out, it should be a two-sample t-test. 

My apologies for the typo.

Best Regards, 

Donald McLaren, PhD

On Mon, Mar 7, 2016 at 8:47 AM, Cutter Lindbergh <[log in to unmask]> wrote:

Hi Dr. McLaren,

Thank you for responding to Mr. Pasquini's questions. I am currently working through analyses using a very similar design and had some similar questions. It would be greatly appreciated if you could clarify the following:

-----Is there a way to code a contrast for J-Before vs O-Before? and also J-After vs O-After?

--Use a one-sample t-test.

Specifically, why do you recommend a one-sample t-test instead of a two-sample t-test? Given that each contrast (i.e., J-Before vs. O-Before and J-After vs. O-After) involves a comparison of two groups, it is unclear to me how a one-sample t-test would be appropriate.

Many thanks,

Cutter

 

On Sun, Mar 6, 2016 at 8:12 PM, MCLAREN, Donald <[log in to unmask]> wrote:

See below.

Best Regards, 

Donald McLaren, PhD

On Fri, Mar 4, 2016 at 1:22 PM, lorenzo pasquini <[log in to unmask]> wrote:

Dear SPM users,

I have found several posts regarding mixed repeated measures, bit some doubts still exist.

I have fMRI data of two groups. J is getting a substance, O is getting a different substance. Both J and O are measured before and after substance delivery. 

I order to statistically evaluate my data, I am using a flexible factorial design, where I model the factor subject (1), the factor treatment group (2), the factor before-after substance (3), and the interaction between treatment group and before-after substance (2 3).

I used the Glaetscher tutorial in order to define F contrasts for the main effect of group, main effect of "time"and the interaction between both.

1. First question:

In the code suggested by the Glaetscher tutorial, e.g. the interaction in my sample would be coded as: 

0 0 0 0 -0.5 0.5 0.5 -0.5 

I have seen in the posts some people using 1 instead of 0.5. Does it make any difference?

It does not make a difference. 

2. Second question:

I find significant effects only for the contrast main effect of group, though very significant.

This is not a valid contrast because the error term is wrong. 

Is there a way to code a contrast for J-Before vs O-Before? and also J-After vs O-After?

Use a one-sample t-test. 

Is it statistically correct to use the cluster derived from the main effect of group in order to test the contrast J-After vs O-After and J-Before vs O-before in separate two sample t tests? 

No. The main effect of group isn't valid. 

Thank you very much and enjoy your weekend

Lorenzo Pasquini