on automatic compensation
utk1 at 013.net
Sat Nov 5 11:21:18 EST 2005
One point I am a little confused about is which compensation is "right" (or both?): "automatic" compensation (using a spillover matrix and its inverse) vs manual compensation (which lacks some of the arguments the matrix uses). Lets assume that we have perfect stained and unstained controls and we use the exactly the same gates for any determination, all in software. In the manual case we'd line up the medians (after a couple of iterations) reaching a "perfect" compensation correction. In the automatic case we'd get a matrix of corrections that includes arguments not existent in the manual case. 2 sets of compensations, both done "right". which one is correct? Should the automatic compensation bring us to the same outcome where the medians of the compensated parameters line up? or will it be slightly off but "the most correct" overall? which one to use if one wants to measure expression values in contrast to just determining population boundaries (where imperfect-but-close compensation is enough)?
By the way, I am assuming that the median is the statistic to use, but here and there the mean has been mentioned (in past posts), and I would be grateful if someone more statistically literate than me could explain a little on that (or point a suitable reference). Intuitively the mean is better because it is more robust as a representation of the central tendency of the distribution (in asymmetrical cases), but I don't know how does the median fare in log-lin-log conversions. The mean on the other side, although skewed by outliers, is a more important parameter of "general" statistics.
As always your comments are highly appreciated!
The Laboratory for Cellular and Molecular Immunology
The Hebrew University - Hadassah Medical Organization
Jerusalem - ISRAEL
"God does not care about our mathematical difficulties. He integrates empirically."
-------------- next part --------------
HTML attachment scrubbed and removed
More information about the Cytometry