On Wed, 06 Aug 2003 20:01:26 +0200, agentsmart <[EMAIL PROTECTED]> wrote:
> [ ... ] > Is skewness of variable distriubtion a hint in investigation of its > nature? ... How could it be otherwise? Yes, if the variable is log-normal, over a large range, then it is definitely, without a doubt, going to be skewed when you plot it without taking logs. It is certainly nice to see variables that are (a) Normal, with (b) linear relationships, and (c) homogeneous variances all along the prediction. - It is certainly convenient and encouraging that, at times, we observe variables are log-normal: We see heterogeneity, skew, and nonlinearity, which all go away when we subsequently take the logs. But there are not guarantees. Having skewness is certainly not "proof" -- more like a clue that raises our hopes -- that when we take the log, we will get normality, or improve some aspect of linearity, or make variances homogeneous. Those are what I think of, as "its nature"; I hope that is enough answer. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
