No disagreement here. I said, and meant, that no P-value of this kind
_is_ 0, but as Roger says, it might be reported as zero to the
precision possible.
(It is also likely that if pressed beyond its limits Stata will report
a P-value as missing. This appears to be what -sktest- does for very
high skewness and kurtosis and large sample size, as reported last
week.)
Nick
njcoxstata@gmail.com
On 8 April 2013 10:28, Roger B. Newson <r.newson@imperial.ac.uk> wrote:
> Strictly speaking, it is entirely possible for Stata (or other software) to
> evaluate a P-value to zero, when the true P-value is smaller than the system
> precision limits. However, it is a valid point that P-values in general
> ought to be formatted as x.ye-z, where x and y are digits and z is a natiral
> number. In an age of megaSNP genome scans, "P<0.0005" is not as good as
> "P=0".
>
>
> On 08/04/2013 10:09, Nick Cox wrote:
>>
>> Massimiliano wrote
>>
>> "the Prob > F is usually smaller than 0.05 (actually being equal to
>> 0.00 several times)"
>>
>> No; this is a misconception. If Stata reports a P-value as 0 to the
>> number of decimal places given, that is what is meant. 0.000, for
>> example, means only <0.0005. No P-value from a test of this kind is
>> exactly zero.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
