[R] How R calculate p-value ?

(Ted Harding) Ted.Harding at manchester.ac.uk
Sat Nov 1 22:56:17 CET 2008


On 01-Nov-08 20:59:29, RON70 wrote:
> Still no reply. Is my question not understandable at all?
> RON70 wrote:
>> I am wondering how R calculate p-value for a test. Does R do some
>> "Approximate" integration on the p.d.f of null distribution?
>> How I can see the code for this particular calculation?
>> 
>> Your help will be highly appreciated.
>> Regards,

How R calculates a P-value depends on the analysis being done,
but it will use a standard method.

For example, a t-test (standard frm, not Welch)will assume a normal
diretribution for the deviations from the mean (or means, for a
two-sample test, with the same variance for each group); will calculate
the T-statistic, and will calculate the exact (to within numerical
accuracy) for this T-value from the distribution of t on the correct
number of degrees of freedom.

In the case of fitting a linear model with normally distributed
errors (using lm()), the P-value for each coefficient will be
based on the "z" value (value of coefficient divided by the
standard error), referred to a t distribution with appropriate
degrees of freedom. Again, an exact P-value.

Similarly, for an analysis of variance, the F-statistic will be
calculated in the standard way, and the P-value will be obtained
by reference to the F distribution with appropriate degrees of
freedom. Again exact, to within numerical accuracy.

For some other kinds of analysis, standard (often large-sample
chi-squared approximations, in some cases to the distribution
of a likelihood ratio) methods will be used, and the resulting
P-value will be approximate to within th accuracy of the
approximating distribution. This kind of thing occurs for
example in fitting generalised linear models using glm().
These P-values are approximate, but the approximations
are standard.

There is as much variety in all this as in the variety of standard
methods for obtaining a test statistic and in using tractable
approximations to the distribution of the test statistic.

Hoping this helps,
Ted.

--------------------------------------------------------------------
E-Mail: (Ted Harding) <Ted.Harding at manchester.ac.uk>
Fax-to-email: +44 (0)870 094 0861
Date: 01-Nov-08                                       Time: 21:56:14
------------------------------ XFMail ------------------------------



More information about the R-help mailing list