On Fri, 26 Nov 1999, Mr. SISAVATH Sourith wrote:
> Thanks for the advice.
> What I meant about the least square methods is as follows:
> If I calculate the mean and the variance of y=log(x)
> using the "standard" equations I mentioned in the previous mail
> mean value m = sum [ log x(i)*probability(x(i))]
> variance = sum (log x(i) - m)^2
> how will I know that the distribution function I obtain is accurate
> or not?
By "accurate" you can only mean, I suppose, accurate enough for your
purposes. Knowing neither those purposes nor how otherwise to
operationalize "accurate", I cannot address this question.
> As you said, I can plot the function and compare it to the histogram.
> But is there a way to quantify the error induced by this approximation?
I suppose you could use one of the standard methods for comparing two
distributions; but these are reputed to be not very sensitive, and one
would still inquire "To what end?". (Suppose you did find evidence that
your initial hunch was wrong, and it isn't lognormal. Then what would
you do? Perhaps more to the point, how badly wrong -- how discrepant
from lognormal -- can you tolerate?)
> I thought methods, such as the least square method, would maybe be
> useful in these cases...
Seems to me you have this backwards. I know of no way to invoke any
least-squares method without having first defined a quantification of
"error": it is the squares of those errors that are minimized in any
such method.
> I hope it makes sense.
Not altogether, I'm afraid: at least not to me. Perhaps someone else on
the list will see more clearly what you're trying to do.
-- DFB.
------------------------------------------------------------------------
Donald F. Burrill [EMAIL PROTECTED]
348 Hyde Hall, Plymouth State College, [EMAIL PROTECTED]
MSC #29, Plymouth, NH 03264 603-535-2597
184 Nashua Road, Bedford, NH 03110 603-471-7128