Since there was no crossposting of my article in sci.stat.consult by default, I'm forwarding my posting this way. Sorry for inconv. ---------------------------------------------------------------------
Jay Weedon wrote: (in sci.stat.consult) > > > So does CR allow one to distinguish cause from effect, as claimed by > WC? > If you *know* that the more uniform variable is your cause, then I think you may conclude that, yes. But when I was in research we usually dealt with the assumption of normal distribution in population, and we usually had an error-margin, that I would not have trusted such a result. One time last year Steve Simon (I think he was the sender) provided empirical data and when I applied CR blindly on them, the results where not much convincing, as I saw them; I even builded some causality pathes reflecting the indicated hierarchical causal order - some looked reasonable, others looked plainly wrong -somehow randomly. I think if one uses measures of physical or electronical processes, where the distance from the uniform causes and the actual measurements may be shorter than in social sciences and the measure is relatively error-free, I would try to explore the power of the CR-idea in more depth. In my opinion the mixture of general questions involving that of cause-effect and of the question of distribution-sensitivity should be discussed separately to prevent wild, opportune and biased magical mystery-tours. If you *know* your causes are uniform distributed, and the best refinement of the CR-approach detects the most uniform variable -fine, so you say you can separate cause&effect. But in that special case, that you can reliably assume this unifomeness, you even need not two measures x and y, but you can quantify the empirical deviation from uniformeness even on the base of a single variable - just draw a histogram. This thought does not intend to discard the approach completely- as you see, I'm still finding it a somehow interesting approach, which should be studied in more detail. It's just my current valuation ... 0.02$ ... Gottfried Helms . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
