Rich Ulrich <[EMAIL PROTECTED]> wrote in message
[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> Page 304:
> "One of the principal reasons for presenting the Lilliefors
> test is to show how the quantiles in Table 15 were obtained."
>
> " ... Lilliefors generated random normal deviates on a
> high speed computer."
>
> I wonder what tests are older than that, that use cutoffs
> obtained from simulations.  And whether this was the first
> table generated by a 'high speed computer'.
>
> By the way, the original references might be unreliable:
> Table 15 is "Adapted from Table 1 of Lilliefors (1967),
> with corrections." (I wonder, too, how many samples he
> generated.)

To my recollection, Lilliefors used n=1000.
It is too small a simulation, but probably all he could
manage at the time. A larer simulation is very easily performed
in a few moments in most stats packages for a given sample
size, so we hardly need the tables - it's a few minutes at worst
to get the p-value any time we need it. (Less time than it
takes to post a message here, I expect, and certainly less
time than it takes to get a reply!)

Glen


.
.
=================================================================
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