[R] Comprehensive power analysis/sample size package in R?

[ukoenig at med.uni-marburg.de]

Many thanks for your helpful suggestions and
the detailed feedback!

We will have a look at your suggestions before buying
the (quite expensive) PASS software.

-Udo


Quoting Tobias Verbeke <tobias.verbeke at openanalytics.be>:

> Frank E Harrell Jr wrote:
>> Greg Snow wrote:
>>> I don't know of a single package that is comparable to PASS, but   
>>> the R system itself is the most comprehensive tool available for   
>>> power and sample size computations.
>>>
>>> For the simple cases you already found the pwr package, there are   
>>> also some power functions in the stats package and in some other   
>>> packages and these will be comparable to the equivalent (or   
>>> possibly better) than the simple ones in PASS.
>
> FYI, Russ Lenth is porting his piface package
>
> http://www.cs.uiowa.edu/~rlenth/Power/
>
> to R
>
> http://r-forge.r-project.org/projects/piface/
>
> Best,
> Tobias
>
>>> When things get a bit more complicated then there are a few   
>>> different options for what to do next:
>>>
>>> 1. Don't provide anything for the more complicated cases.
>>> 2. Provide a minimal set of routines for more complicated cases   
>>> based on programmer assumptions rather than information from   
>>> someone familiar with the source of the data (assumptions often   
>>> hidden).
>>> 3. Provide many different routines encompassing every alternative   
>>> set of assumptions that the programmer can think of forcing the   
>>> user to sort through all the options to find the one that is   
>>> closest (and maybe the same) as what they want to do.
>>> 4  Provide a full programming language so that the people familiar  
>>>  with the question(s) of interest and the source of the data can   
>>> explicitly spell out the desired analysis and assumptions.
>>> 5. possible others, but I can't think of any.
>>>
>>> It looks like PASS uses option 3, giving many different routines   
>>> that any one user in only likely to use a few of.
>>>
>>> R is option 4.  You can decide what assumptions you want to make   
>>> about the data (and later change any of those assumptions), decide  
>>>  how you plan to analyze the data, then by simulation you can work  
>>>  out the power/sample size/etc. knowing exactly what assumptions   
>>> went into the analysis.
>>>
>>
>> As one example of what Greg is talking about see   
>> http://bm2.genes.nig.ac.jp/RGM2/R_current/library/Hmisc/man/spower.html
>>