Thank you a lot Donald for the answer!Regarding using con_images vs. t_images:1. My idea is that I not interested in comparing the magnitude of the response between two regions, but rather their selectivity. So, isn't use of t_images more appropriate? I also concern not to compare absolute differences, but rather relative change - otherwise in the more activated region the difference by definition will be larger (see related second question)2. I do not quite understand what are the units of con_images / t_images and whether the values are absolute. For example, if for a region A the signal in cond1=100, cond2=90, while for a region B the signal in cond1=10, cond2=9, assuming that std scales similarly, would values in con* and t* images be the same? I thought that t-images is in units of t-distribution, so they are by definition not-absolute. If, so, then what about con*?On Thu, Dec 5, 2013 at 10:33 PM, MCLAREN, Donald <[log in to unmask]> wrote:
See below.On Mon, Nov 25, 2013 at 5:47 AM, John Gelburg <[log in to unmask]> wrote:
Hi,I have a design with three conditions (facesA, facesB, non-faces) and the baseline. Here is my analysis pipe-line:1. Select the most active region based on a-priori anatomical knowledge using contrast facesA+facesB+non-faces> baseline. Based on anatomical coordinated I split this most active region to two subRegions: "subRegion a" and "subRegion b".If there are two subregions, could you increase the threshold to split this region?2. I want to test face selectivity profile in my subRegions. So, I make t-contrast facesA + facesB > non-faces. To compare the profiles between subregions I extract t-values for subRegion a & b and run a paired t-test between them (in matlab).
Two issues here: (1) Comparing t-values does not allow you to say one region is more active than another region as the t-values formed by the magnitude of the response and its standard deviation; and (2) t-values don't follow a normal distribution and thus should not be compared using a t-test. I would suggest repeating the analysis with the contrast values instead.
3. One of the possible interpretations for the difference in selectivity between subRegions is that one of subRegions was just more active vs. baseline. To test this, for both subRegions I extract t-values for facesA+facesB+non-faces> baseline contrast. Similarly to (3), I also run a t-test.As above, you should use the contrast values as the contrast values represent the amount of activity. I'm not sure how All>baseline necessitates faces>non-faces being greater.
Question: How valid it would be to run 2-way repeated measure ANOVA for (3) and (4) in order to show interaction?I'm not sure what the interaction tells you in this case. The difference between faces>non-faces is different from all>baseline. What does this mean?
Thanks for help,John