This doesn't exactly answer your question, but what if you grouped together your stimuli into, say, 3 levels of complexity. i.e. If your complexity scale is 1-10, you could group together all of 1-4, 5-7, 8-10, or something like that. Do this for both stimulus types and set up a factorial design. Then you can see interaction effects related to stimulus complexity.
________________________________________
From: SPM (Statistical Parametric Mapping) [[log in to unmask]] On Behalf Of Donghoon Lee [[log in to unmask]]
Sent: Thursday, December 09, 2010 2:59 AM
To: [log in to unmask]
Subject: [SPM] Methods for removing confounding effects from intrinsic stimulus charateristics
Dear SPM experts,
I would like to ask experts in this group about statistical methods for removing confounding effects from stimulus characteristics.
The confounding effects from stimulus characteristics which I meant are like unexpected influences of intrinsic stimulus chracteristics on the experimental factors.
For example, visual complexity of stimulus is sometimes correlated with semantic factors of visual stimuli.
If I want to compare a neural response from man-made objects(e.g. tools) with that from natural objects (e.g. animals), there might be more activation in the primary visual cortex for animal pictures because visual complexity of animal pictures is generally higher than that of tool pictures. So, in this case, general visual complexity is a confounding factor.
Surely, such confounding factors should be very carefully controlled when an experiment is designed. However, some variables are not easily controlled if they are naturally correlated with experimental factors.
Even though I controlled the influence from such factors to be equal on average across experimental conditions, the effect could be hetrogenous across trials.
If I have a measurement of the confounding factor, I could enter the score as nuisance variable in the GLM. Yes, I mean ANCOVA.
My question is how this procedure is applicapable in SPM. I've looked at PPI analysis and the parametric modulation option in the first level of model setup.
However, in my understanding, these two methods seem not proper for my purpose because they are actually treating such factors as experimental factors. So, the results are if the effect from such variable is significant or not (e.g., Buchel et al, 1998). Rather than this, I just want to test the main effects and interaction effects from the experimental factors on the data after removing the unexpected effects from the confounding variable.
How can I do this with SPM? The covariate option seems only available in the second level of analysis.
Simply, I want some measurements about stimulus charateristics(such as visual complexity, word frequency and so on) as covariates in the trial level (Not in the subject level, Nor in the scan(or time) level)
Many thanks in advance,
==============================
Donghoon Lee, Ph.D.
Post-doc fellow
Department of Psychology
Pusan National University
Geumjeong-Gu, Busan 609-735 South Korea
Phone: 82-51-510-3933
Fax: 82-51-581-1457
E-mail:[log in to unmask]<mailto:[log in to unmask]>
=============================
"I prefer keeping in mind even the possibility that existence has its own reason for being."
- Wislawa Szymborska, "Possibilities"
|