I've conducted a temporal PCA with your toolbox using some data collected from a classical oddball paradigm. Everything's been working quite well for the primary analysis in which I'm looking at the peaks/latencies of the millivolt-scaled factor waveforms generated from all available trials.
As a secondary analysis, I'm trying to look at how the P3a and P3b components are changing during the course of the experiment. The hypothesis is that there won't be much latency jitter, but the amplitudes will change. To start with, I just want to examine the degree to which the first and second halves of the trials each contribute to the variance in the overall grand means.
My thinking is that since the factors are not expected to have much latency variation between the two trial subsets, that it doesn't make sense to run a PCA on each. Instead, I plan to use the factor loadings from the "all trials" PCA and generate factor scores and a standard deviation vector for the first and second halves of the trials separately.
In your experience, is this a legitimate approach? If so, is there a straightforward way of accomplishing this using the toolbox? If not, do you think running a PCA for the two halves of the experiment would make more sense?
Thanks for putting this toolbox together. It's been a tremendous benefit to my research.
Regards,
Brian
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Sure, sounds good! Couple things I’m not quite clear about. I imagine you could just look at the grand averages for the two halves so you don’t need to guess whether there is latency variation. What do you see? Second, I’m not quite clear what the question is about. So if you have averages for standards and targets for the first and second half for each subject (four total cells), then you run the PCA and then you’re done. What’s the problem?
I've conducted a temporal PCA with your toolbox using some data collected from a classical oddball paradigm. Everything's been working quite well for the primary analysis in which I'm looking at the peaks/latencies of the millivolt-scaled factor waveforms generated from all available trials.
As a secondary analysis, I'm trying to look at how the P3a and P3b components are changing during the course of the experiment. The hypothesis is that there won't be much latency jitter, but the amplitudes will change. To start with, I just want to examine the degree to which the first and second halves of the trials each contribute to the variance in the overall grand means.
My thinking is that since the factors are not expected to have much latency variation between the two trial subsets, that it doesn't make sense to run a PCA on each. Instead, I plan to use the factor loadings from the "all trials" PCA and generate factor scores and a standard deviation vector for the first and second halves of the trials separately.
In your experience, is this a legitimate approach? If so, is there a straightforward way of accomplishing this using the toolbox? If not, do you think running a PCA for the two halves of the experiment would make more sense?
Thanks for putting this toolbox together. It's been a tremendous benefit to my research.
Hello Dr. Dien,
I've conducted a temporal PCA with your toolbox using some data collected from a classical oddball paradigm. Everything's been working quite well for the primary analysis in which I'm looking at the peaks/latencies of the millivolt-scaled factor waveforms generated from all available trials.
As a secondary analysis, I'm trying to look at how the P3a and P3b components are changing during the course of the experiment. The hypothesis is that there won't be much latency jitter, but the amplitudes will change. To start with, I just want to examine the degree to which the first and second halves of the trials each contribute to the variance in the overall grand means.
My thinking is that since the factors are not expected to have much latency variation between the two trial subsets, that it doesn't make sense to run a PCA on each. Instead, I plan to use the factor loadings from the "all trials" PCA and generate factor scores and a standard deviation vector for the first and second halves of the trials separately.
In your experience, is this a legitimate approach? If so, is there a straightforward way of accomplishing this using the toolbox? If not, do you think running a PCA for the two halves of the experiment would make more sense?
Thanks for putting this toolbox together. It's been a tremendous benefit to my research.
Regards,
Brian
Sure, sounds good! Couple things I’m not quite clear about. I imagine you could just look at the grand averages for the two halves so you don’t need to guess whether there is latency variation. What do you see? Second, I’m not quite clear what the question is about. So if you have averages for standards and targets for the first and second half for each subject (four total cells), then you run the PCA and then you’re done. What’s the problem?
Joe
On Aug 20, 2014, at 1:48 AM, Brian Silverstein briansi@users.sf.net wrote: