9 Mar
2011
9 Mar
'11
2:29 p.m.
Hi Katie, sorry for the delay. This is actually a pretty complicated question. It probably does mean that you are getting a better fit, though you have to be careful if you attempt to keep playing with your model and only keep the one that gives you the most activation. Eventually, you will pick one that just fits the noise. Technically, what you should do is to Bonferroni correct across all the models that you use (which would probably quickly kill your activation). Or you could take a subset of subjects, find the best model, then apply that model to a new set of subjects. Neither option is great, so don't try too many models.
doug
Katie Bettencourt wrote:
Still looking for an answer to this question:
So in doing a more complicated version of the analysis with paradigm weighting, I had another experiment where I wanted to test out two different models, 1) that BOLD activity was related to my behavioral measure alone or 2) that BOLD activity was related to my behavioral activity plus a factor that represented the number of distractors present (which is constant across set size) (had a high and low distractor condition, so that this factor was effectively 0 for the low distractor condition and something like 1 for the high distractor condition). I modelled the second my simply adding the behavior measure and the distractor factor together (which means that in practicality, nothing changed for the low distractor conditions between model 1 and model 2, but that the weight was increased by a constant factor in the high distractor conditions between model 1 and model 2).
When I ran this analysis, what I got (shown in the attached picture in a group average) was an increase in the amount of activity and area of activity for model 2 (targets & distractors) compared with model 1 (targets only). Am I right in interpreting this as model 2 being a better fit? Or is can these maps be compared to show areas that are a better fit for model 1 and other areas that are a better fit for model 2?
I have ROI data on the unweighted analyses that goes along with the idea of model 2 being better, but I wanted to make sure I was intrepreting the activity in this weighted analysis correctly.
Thanks,
Katie
On Tue, Feb 22, 2011 at 1:35 PM, Douglas N Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> wrote:
Yes, your understanding is correct. People might choose Version 1 because they strongly feel that it accurately represents the underlying mechanism. It saves one DoF, but that's not going to change much. I think the problem with Version 1 is that, for most designs, it forces you to extrapolate outside of your data range. Few things are truly linear, but they are often well modeled by straight lines over a limited range. For your data, you might really expect the response to be 0 at x=0, but it become a very non-linear curve in that region. doug Katie Bettencourt wrote: Ok, so just so that I understand (I realize I'm being a bit dense, and I appreciate you walking me through this). If I am weighting my paradigm by my behavioral measure, then the activity I see with this weighted analysis, is areas where we are seeing a linear relationship between BOLD activity and my behavioral measure. In version 1:1vs0, this assumes that my linear fit passes through xy=0. However, in version 2:2vs0, we put an offset (or a sort of y intercept?) into the model so that the data doesn't have to pass through xy=0. And you are saying that the larger amount of activated area that I am getting in Version 1:1vs0, is because I'm seeing areas activated for the fact that there's an increase from baseline but not exactly the linear fit between BOLD and behavioral measures, but for Version 2:2vs0, I see only the areas that show this linear correlation? Does that sound about right? And if I am right, why would anyone do Verison1? Is it just because if your fit does pass through xy=0, you get more power this way compared with Version 2? Katie On Tue, Feb 22, 2011 at 1:00 PM, Douglas N Greve <greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>>> wrote: Katie Bettencourt wrote: You should also look at Version2:1v0. I bet a lot of the areas from Version1:1v0 will also show up. You can also create a Version3 in which you divide your presentations into a low-weight and a high-weight (but set the weight=1). Then create contrasts of low+high and high-low. The low+high should look like Version2:1v0 and the high-low should look like Version2:2v0. So, what would this say? That the extra activity I'm getting in Version 1 is just the slope/offset, and isn't actually load related data that I am trying to get at (ie. isn't activity that is increasing as the number of items the subjects is remembering increases, but is just baseline visual stimulation activity or something?) Yes. It's easier to explain if I can draw it out. But imagine an xy plot with your weight on the x axis and the fMRI response on the y axis. If there is an offset but no change in fMRI with weight, then the data points will be on a horizontal line. Now what happens if you try to fit that data with a line that is forced to pass through xy=0? You'll get some positive slope, but overall it won't fit very well. And What exactly to do you mean by divide my presentations into low and high weight? What exactly would this be comparing? Sorry if I'm being dense. I mean create a new paradigm file with two non-null conditions. Condition 1 would be all presentations with weight < 0.5 (low weight); Condition 2 would be all those with weight>0.5. You should pick your own weight threshold of course. doug Katie doug Katie Bettencourt wrote: Yes, those are the maps I"ve been comparing. I've been comparing it to BV sort of, but that analysis is not surface based and I"m not used to it, so I can't quite tell which is more accurate, though Version 2 gives a much smaller area of activity, which fits with the description of what I've been given about what to expect in BV. Attached is two pictures of the difference I get for Version 1:1v0 (labeled with "single" in the image name) and Version 2:2v0 (labeled with "double" in the name). As you can see, Version1 activates a much larger area than Version 2. I guess part of my problem is that I'm having trouble understanding exactly what these two versions are telling me about the data and what the differences is. Can you try to give me a sort of layman's explanation? Katie On Tue, Feb 22, 2011 at 11:45 AM, Douglas N Greve <greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>>>>> wrote: I assume that you are comparing maps of Version1:1v0 and Version2:2v0 ? I could imagine it going either way. If the true slope is 0 but the offset is non-0, then Version1 will give you an artificially high slope (and Verion2 will give you the correct slope at 0, and so no activation). Are you comparing this to a BV analysis? doug Katie Bettencourt wrote: So I created a weighted regression analysis to look at the effect of memory load in a particular brain region. Basically, I weighted the paradigms by a behavioral measure that reflected the number of items actually remembered (as set size was increased). As far as Doug told me there are basically 2 ways to weight your paradigm files. Version 1: Have 2 conditions, baseline (condition 0) and all the set sizes (condition 1). Condition 1 would then be weighted by the behavioral measure. Version 2: Have 3 conditions, baseline (condition 0), and then I represented each presentation as two different conditions, one with a weight that is always 1 (condition 1), the other weighted according to the behavioral measure (condition 2). The difference, as far as I understand it, in version 1, it is assumed that the response amplitude is ) when the weight is 0 (ie. that when you are attending to 0 items, brain activity = 0). Whereas, version 2, tests the slope of the HRF amplitude vs weight without the assumption above. However, I'm a bit confused as to the results I got. When I looked at the data from both versions, version 1 provided a much higher amount of activation and more areas activated than version 2. However, I believe version 2 better fits with the multiple regression analysis that is done in Brain Voyager. Can anyone give me a better explanation of what the difference between these analysis models is? Katie -- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>>>> Phone Number: 617-724-2358 Fax: 617-726-7422 Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail. ------------------------------------------------------------------------ ------------------------------------------------------------------------ -- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>>> Phone Number: 617-724-2358 Fax: 617-726-7422 Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> -- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> <mailto:greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu>> Phone Number: 617-724-2358 Fax: 617-726-7422 Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html> -- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu <mailto:greve@nmr.mgh.harvard.edu> Phone Number: 617-724-2358 Fax: 617-726-7422 Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting <http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting> FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html <http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html>
--
Douglas N. Greve, Ph.D.
MGH-NMR Center
greve@nmr.mgh.harvard.edu
Phone Number: 617-724-2358
Fax: 617-726-7422
Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html