[R] significance level (p) for t-value in package zelig

Rune Haubo rhbc at imm.dtu.dk
Tue Jun 26 14:41:40 CEST 2012


My point was just that the situation in a cumulative link model is not
much different from a binomial glm - the binomial glm is even a
special case of the clm with only two response categories. And just
like summary(glm(...., family=binomial)) reports z-values and computes
p-values by using the normal distribution as reference, one can do the
same in a cumulative link model by applying the same asymptotic
arguments.

In both models the variance is determined implicitly by the mean, so a
t-distribution is never involved.

Cheers,
Rune

On 25 June 2012 11:05, Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:
> On 25/06/2012 09:32, Rune Haubo wrote:
>>
>> According to standard likelihood theory these are actually not
>> t-values, but z-values, i.e., they asymptotically follow a standard
>> normal distribution under the null hypothesis. This means that you
>
>
> Whose 'standard'?
>
> It is conventional to call a value of t-like statistic (i.e. a ratio of the
> form value/standard error) a 't-value'.  And that is nothing to do with
> 'likelihood theory' (t statistics predate the term 'likelihood'!).
>
> The separate issue is whether a t statistic is even approximately
> t-distributed (and if so, on what df?), and another is if it is
> asymptotically normal.  For the latter you have to say what you mean by
> 'asymptotic': we have lost a lot of the context, but as this does not appear
> to be IID univariate observations:
>
> - 'standard likelihood theory' is unlikely to apply.
>
> - standard asymptotics may well not be a good approximation (in regression
> modelling, people tend to fit more complex models to large datasets, which
> is often why a large dataset was collected).
>
> - even for IID observations the derivation of the t distribution assumes
> normality.
>
> The difference between a t distribution and a normal distribution is
> practically insignificant unless the df is small.   And if the df is small,
> one can rarely rely on the CLT for approximate normality ....
>
>
>> could use pnorm instead of pt to get the p-values, but an easier
>> solution is probably to use the clm-function (for Cumulative Link
>> Models) from the ordinal package - here you get the p-values
>> automatically.
>>
>> Cheers,
>> Rune
>>
>> On 23 June 2012 07:02, Bert Gunter <gunter.berton at gene.com> wrote:
>>>
>>> This advice is almost certainly false!
>>>
>>> A "t-statistic" can be calculated, but the distribution will not
>>> necessarily be student's t nor will the "df" be those of the rse.  See,
>>> for
>>> example, rlm() in MASS, where values of the t-statistic are given without
>>> p
>>> values. If Brian Ripley says that p values cannot be straightforwardly
>>> calculated by pt(), then believe it!
>>>
>>> -- Bert
>>>
>>> On Fri, Jun 22, 2012 at 9:30 PM, Özgür Asar <oasar at metu.edu.tr> wrote:
>>>
>>>> Michael,
>>>>
>>>> Try
>>>>
>>>> ?pt
>>>>
>>>> Best
>>>> Ozgur
>>>>
>>>> --
>>>> View this message in context:
>>>>
>>>> http://r.789695.n4.nabble.com/significance-level-p-for-t-value-in-package-zelig-tp4634252p4634271.html
>>>> Sent from the R help mailing list archive at Nabble.com.
>>>>
>>>> ______________________________________________
>>>> R-help at r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>>> PLEASE do read the posting guide
>>>> http://www.R-project.org/posting-guide.html
>>>> and provide commented, minimal, self-contained, reproducible code.
>>>>
>>>
>>>
>>>
>>> --
>>>
>>> Bert Gunter
>>> Genentech Nonclinical Biostatistics
>>>
>>> Internal Contact Info:
>>> Phone: 467-7374
>>> Website:
>>>
>>> http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm
>>>
>>>        [[alternative HTML version deleted]]
>>>
>>>
>>> ______________________________________________
>>> R-help at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>> PLEASE do read the posting guide
>>> http://www.R-project.org/posting-guide.html
>>> and provide commented, minimal, self-contained, reproducible code.
>>>
>>
>>
>>
>
>
> --
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
>
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.



-- 
Rune Haubo Bojesen Christensen

Ph.D. Student, M.Sc. Eng.
Phone: (+45) 45 25 33 63
Mobile: (+45) 30 26 45 54

DTU Informatics, Section for Statistics
Technical University of Denmark, Build. 305, Room 122,
DK-2800 Kgs. Lyngby, Denmark



More information about the R-help mailing list