David wrote:
> Greetings and thanks in advance for any help
> For interest, if I sample values from a perfectly Normal
distribution
> and then 'discretise' the values by applying class boundaries...
When
> I recalculate the standard deviation of the discrete values, the
> estimate of sd will have increased compared to the estimate of sd
for
> the sample values. Is there a correction formula that will adjust
the
> sd in line with a continuous variable?

"Sheppard's correction" does this. See for example old Kendall &
Stuart vol 1, or via Google. The correction is more usually stated as
an adjustment to the estimated variance. There are corresponding
adjustments for other cumulants. If you "know" there is an exact
underlying Normal distribution, then you would potentially be better
off estimating via maximum likelihood, but I don't know if the
difference is great.

David Jones


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to