On 21 Mar 2003 08:41:53 -0800, [EMAIL PROTECTED] (Robert J. MacG. Dawson) wrote:
[ snip, some] > Do you mean "probability density" here? (They are in > a sense equal, but in the context of "the curve" probability > density sounds more appropriate. Yes, that is what I mean -- in the other context of generalizing to *discrete* distributions, the height seems useful. > > The problem here is surely that probability densities > are difficult to compare. A "critical density" that would be reasonable > for a N(0,1�) would be far too high for an N(0,10�). Likelihood testing uses the ratios, not the raw densities, which indeed become tiny. > > Scaling this out is essentially dividing by the density of the mean. > And why should we care about values less discrepant than what we've > observed <grin>? More- or less- discrepant -- It bothers me, that this seems to bother anyone. If we are doing some testing, we are assuming that there is at least an order to results, are we not? And it seems to me that the Bayesians have to assume even more than the the rest of us do, concerning what is not immediately observed. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
