Exporting PCA factor scores (in microvolts)

  • Isaac

    Isaac - 2012-06-03

    I figured out how to do the sequential temporospatial PCA (thanks!).  Now, I want to export the factor scores (in microvolts) from the temporospatial PCA to characterize each subject by one amplitude for every combination of condition, temporal factor, and spatial factor.

    Looking through the tutorial, I see a couple options:
    1. The factor score matrix is in the PCA tab of the Edit window, and I could export the factor scores.  A couple problems here, though.  Are the factor scores in microvolts?  Also, there is no column for subject ID and condition, so I would have to guess which row corresponds to which subject and condition.
    2. I could use the AutoPCA function to generate the voltage for every subject and condition for each of the temporal and spatial factors.  This, however, appears to be the voltage accounted for at the peak time point and peak channel for a given temporospatial factor, and not a representation of the entire factor structure (all of the time points and channels) for that temporospatial factor.

    Is it possible to export the factor scores from a temporospatial PCA that are represented in microvolts and represent the entire factor structure for a given temporospatial factor (not just the peak time point and peak channel)?

    I really love the toolkit, and thanks so much for your help through the learning process.  I really hope to use it as my go-to tool for data reduction.

  • Joe Dien

    Joe Dien - 2012-06-06

    Thanks for your interest!  The PCA files are the entire factor structure so you just window as desired if you want the results in microvolts as recommended.  See my chapter (Dien & Frishkoff, 2005) for how this all works conceptually and the tutorial for the directions on what to do.

  • Isaac

    Isaac - 2012-06-07

    Reading the tutorial (p. 74), it appears that the AutoPCA function selects the amplitude of a factor by the amplitude at the peak time point and peak channel for that factor.  Thus, amplitudes from the windowing procedure do not take into account information from the the other time points and channels that compose the factor, correct?  I want to make sure that my amplitudes reflect all of the time points and channels composing the factor, and not just the amplitude of the peak time point and peak channel.  Thanks for clarifying.

  • Joe Dien

    Joe Dien - 2012-06-07

    I'm sorry, your question suggests you're not understanding how this works.  The whole point of a temporal-spatial PCA is that it makes the rest of the time points and channels irrelevant.  There is a single factor score number that accounts for the subject and condition variance and this is what you care about for the ANOVAs.  The temporal factor loadings accounts for all the time points and the spatial factor loadings accounts for all the channels.  Try doing an ANOVA with different time points for example.  You'll get the same ANOVA result.  Consult my 2005 chapter to understand how this works.  I don't really have the time to give you a full PCA tutorial.  I need to get this new update finished.  Sorry!  Joe


Log in to post a comment.