 |
 |
All three methods should produce the same results. However, the approach mentioned by Chris will be more straightforward. At the first level, you can have a 2x2 factorial design and form the interaction at the first level with the contrast [1 -1 -1 1] or [-1 1 1 -1]. Then you can use that contrast in a one-sample t-test. Best Regards, Donald McLaren ================= D.G. McLaren, Ph.D. Research Fellow, Department of Neurology, Massachusetts General Hospital and Harvard Medical School Postdoctoral Research Fellow, GRECC, Bedford VA Website: http://www.martinos.org/~mclaren Office: (773) 406-2464 ===================== This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is intended only for the use of the individual or entity named above. If the reader of the e-mail is not the intended recipient or the employee or agent responsible for delivering it to the intended recipient, you are hereby notified that you are in possession of confidential and privileged information. Any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited and may be unlawful. If you have received this e-mail unintentionally, please immediately notify the sender via telephone at (773) 406-2464 or email. On Fri, Dec 5, 2014 at 2:41 AM, Amy Frithsen <[log in to unmask]> wrote: > Sorry for the rather naive question, but I just want to make sure I am > doing things properly, so any feedback will be greatly appreciated. > > I use SPM for memory-based fMRI experiments mostly. So many of my > contrasts of interest involve looking at the difference in activation > between 'old' items and 'new' items (usually old > new). How I usually set > up my design is in the the following way: > > at the first (subject) level, I create a beta weight for 'old' items and > also one for 'new' items. I then create a con image for each subject using > 1 for the 'old' beta weight and -1 for the 'new' beta weight. This leaves > me with a con image that shows the differential activity between 'old' > items and 'new items' for each subject (old > new). > > then I take these con images up to the second (group) level of analysis > and run a one-sample t-test against zero to see if this 'old > new' > contrast is significant at the group level. > > All is fine and good. > > However, recently I've included a different factor in the experimental > design. Now there are 2 conditions that each subject goes through - a > 'High Probability' condition and a 'Low Probability' condition. I do the > same analysis that I described above, separately for the High and Low > conditions. Now I have two one-sample t-tests that show me the 'old>new' > contrast at the second level for each condition separately. > > But what I want to do is compare these two 'old>new' contrasts between the > two conditions. Basically I want to see: > > Is the 'old > new' activity during the Low Probability condition > significantly different than the 'old > new' activity during the High > Probability condition? > > As far as I can see, there are a few ways to get at this, but I'm not sure > which (if any) are statistically the most correct. Here's what I was > thinking as options: > > 1. Run a paired-samples t-test at the second level of analysis. To do > this, I would pair each subject's 'old>new' Low Probability con image with > their 'old > new' High Probability con image. This seems to be the most > straight-forward way to go about this > > 2. Use ImCalc to create a 'difference image' for each subject. Simply > subtract each subject's 'old >new' High Probability Condition con image > from their 'old > new' Low Probability con image. Then bring those > 'difference images' up to the second level and run a one-sample t-test > against zero on those. > > 3. Use a flexible factorial to run a one-way ANOVA (setting the > independence parameter to 'No') and put Condition as a factor with two > levels (High and Low) and then input the 'old > new' con images for each > subject for both conditions. I would then create a t contrast that > directly compared the two levels to each other. > > Are these the right way to go about this? Is there one method that is > statistically better than the other? I get slightly different results > depending on what method I use (due to differences in degrees of freedom > which lead to differences in thresholding), so I just wanted to reach out > and see what you thought > > Thanks in advance for any advice! > > ~ Amy >