R help - Jul 2016 - dependent p.values in R

Le 09/07/2016 ? 17:17, Fernando Marmolejo Ramos a ?crit
:
> hi all
>
>
> does any one know a method to combine dependent p.values?
>
>
First, it is a stats question and not a R question. So you could have 
better chance to ask this in stackexchange forum.
Second, your question is difficult to answer without context: why 
p.values are dependent ? Do they come from the same dataset ? Or are 
they linked by an external source ? For both these situations, combining 
dependent p.values seems strange for me.
When you will ask question in stackexchange, be more precise.
Sincerely,
Marc Girondot

-- 
__________________________________________________________
Marc Girondot, Pr

Laboratoire Ecologie, Syst?matique et Evolution
Equipe de Conservation des Populations et des Communaut?s
CNRS, AgroParisTech et Universit? Paris-Sud 11 , UMR 8079
B?timent 362
91405 Orsay Cedex, France

Tel:  33 1 (0)1.69.15.72.30   Fax: 33 1 (0)1.69.15.73.53
e-mail: marc.girondot at u-psud.fr
Web: http://www.ese.u-psud.fr/epc/conservation/Marc.html
Skype: girondot

Fernando Marmolejo Ramos <fernando.marmolejo.ramos <at>
psychology.su.se>
writes:
> 
> hi marc
> 
> say i have a vector with some x number of observations
> 
> x = c(23, 56, 123, ..... )
> 
> and i want to know how normal it is
> 
> as there are many normality tests, i want to combine their p.values
> 
> so, suppose i use shapiro.wilk, anderson darling and
>    jarque bera and each will give a pvalue
> 
> i could simply average those p,values but to my knowledge 
>    that approach is biased
> 
> so i thought, in the same way there is a method to combine
>    independent pvalues (e.g. stouffer method); is
> there a way to combine dependent pvalues?
> 
> best
> 
> f
> 
  Yikes.  There is extensive discussion, e.g. at
http://tinyurl.com/normtests , that suggests that much of the
time (if not always) formal statistical hypothesis tests for
normality are misguided.  Combining p-values from different tests
feels like compounding the issue.  In any case, I would definitely
say that this a question for CrossValidated 
(http://stats.stackexchange.com), rather than r-help ...

  Ben Bolker

Hello Fernando

First, ask yourself what Gosta Ekman would have said if you asked him 
this question.  He would have asked "does it make any difference to
your conclusion?"  He might also have asked you "Did you do a visual
test?"  Plot your data as a QQ plot or density plot?

If the test doesn't make a difference in conclusions, it is a waste of 
your time (and ours) to worry about how to cite a
'combined p.value' (if such an animal exists), presumably to
more decimal places than is worth worrying about.

If the test *does* make a difference about normality, then ask yourself
does the degree of non-normality impede my substantive conclusions.

HTH,
-ichael

On 7/10/16 3:39 AM, Fernando Marmolejo Ramos wrote:
> hi marc
>
> say i have a vector with some x number of observations
>
> x = c(23, 56, 123, ..... )
>
> and i want to know how normal it is
>
> as there are many normality tests, i want to combine their p.values
>
> so, suppose i use shapiro.wilk, anderson darling and jarque bera and each
will give a pvalue
>
> i could simply average those p,values but to my knowledge that approach is
biased
>
> so i thought, in the same way there is a method to combine independent
pvalues (e.g. stouffer method); is there a way to combine dependent pvalues?
>
> best
>
> f
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Fernando Marmolejo-Ramos
> Postdoctoral Fellow
> G?sta Ekman Laboratory
> Department of Psychology
> Stockholm University
> Frescati Hagv?g 9A, Stockholm 114 19
> Sweden
>
> ph = +46 08-16 46 07
> website = http://sites.google.com/site/fernandomarmolejoramos/
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> ________________________________________
> From: Marc Girondot <marc.girondot at u-psud.fr>
> Sent: Sunday, 10 July 2016 8:25 AM
> To: r-help at r-project.org; Fernando Marmolejo Ramos
> Subject: Re: [R] dependent p.values
>
> Le 09/07/2016 ? 17:17, Fernando Marmolejo Ramos a ?crit :
>> hi all
>>
>>
>> does any one know a method to combine dependent p.values?
>>
>>
> First, it is a stats question and not a R question. So you could have
> better chance to ask this in stackexchange forum.
> Second, your question is difficult to answer without context: why
> p.values are dependent ? Do they come from the same dataset ? Or are
> they linked by an external source ? For both these situations, combining
> dependent p.values seems strange for me.
> When you will ask question in stackexchange, be more precise.
> Sincerely,
> Marc Girondot
>
> --
> __________________________________________________________
> Marc Girondot, Pr
>
> Laboratoire Ecologie, Syst?matique et Evolution
> Equipe de Conservation des Populations et des Communaut?s
> CNRS, AgroParisTech et Universit? Paris-Sud 11 , UMR 8079
> B?timent 362
> 91405 Orsay Cedex, France
>
> Tel:  33 1 (0)1.69.15.72.30   Fax: 33 1 (0)1.69.15.73.53
> e-mail: marc.girondot at u-psud.fr
> Web: http://www.ese.u-psud.fr/epc/conservation/Marc.html
> Skype: girondot
>
>