[R] Code modification for post-hoc power

[peter dalgaard]

That doesn't work. In caricature, post-hoc power is

- I observe a difference of nearly zero 
- However, to find a significant difference of that size I'd need 200000 observations
- I only used 100 observations
- Therefore my study is useless and can be discarded

(or: I calculate the probability of Type II error if the true difference is the observed and get 0.999... Therefore, etc.)

Best way out is a confidence interval. Second best (but in principle wrong) is to redo the pre-study power calculation and say that the study was designed to find a difference of delta, which it clearly didn't, so the true difference is probably less than delta.

-pd

> On 26 Aug 2019, at 18:29 , Michael Dewey <lists using dewey.myzen.co.uk> wrote:
> 
> Dear Anne
> 
> In addition to Marc's comments if you are forced to do this then, assuming your package computes sample size from power then just feed it a range of powers and find the one for which it calculates the sample size you had. There is a more elegant way to do this using uniroot but brute force should work.
> 
> Michael
> 
> On 26/08/2019 13:42, Marc Schwartz via R-help wrote:
>>> On Aug 26, 2019, at 6:24 AM, CHATTON Anne via R-help <r-help using r-project.org> wrote:
>>> 
>>> Hello everybody,
>>> 
>>> I am trying to accommodate the R codes provided by Donohue for sample size calculation in the package "longpower" with lmmpower function to estimate the post-hoc power (asked by a reviewer) of a binary GEE model with a three-way interaction (time x condition x continuous predictor) given a fixed sample size. In other words instead of the sample size I would like to estimate the power of my study.
>>> 
>>> Could anyone please help me to modify these codes as to obtain the power I'm looking for.
>>> 
>>> I would really appreciate receiving any feedback on this subject.
>>> 
>>> Yours sincerely,
>>> 
>>> Anne
>> Hi,
>> Three comments:
>> 1. Don't calculate post hoc power. Do a Google search and you will find a plethora of papers and discussions on why not, including these:
>>   The Abuse of Power: The Pervasive Fallacy of Power Calculations for Data Analysis
>>   The American Statistician, February 2001, Vol. 55, No. 1
>>   https://www.vims.edu/people/hoenig_jm/pubs/hoenig2.pdf
>>   Post Hoc Power: Tables and Commentary
>>   https://stat.uiowa.edu/sites/stat.uiowa.edu/files/techrep/tr378.pdf
>>   Observed power, and what to do if your editor asks for post-hoc power analyses
>>   http://daniellakens.blogspot.com/2014/12/observed-power-and-what-to-do-if-your.html
>>   Retraction Watch:
>>   Statisticians clamor for retraction of paper by Harvard researchers they say uses a “nonsense statistic”
>>   https://retractionwatch.com/2019/06/19/statisticians-clamor-for-retraction-of-paper-by-harvard-researchers-they-say-uses-a-nonsense-statistic/
>>   PubPeer Comments on the paper cited in the above RW post:
>>   https://pubpeer.com/publications/4399282A80691D9421B497E8316CF6
>>   A discussion on Frank's Data Methods forum also related to the same paper cited above:
>>   "Observed Power" and other "Power" Issues
>>   https://discourse.datamethods.org/t/observed-power-and-other-power-issues/731/30
>> 2. If you are still compelled (voluntarily or involuntarily), you may want to review the vignette for the longpower package which may have some insights, and/or contact the package maintainer for additional guidance on how to structure the code. See the vignette here:
>>   https://cran.r-project.org/web/packages/longpower/vignettes/longpower.pdf
>> 3. Don't calculate post hoc power.
>> Regards,
>> Marc Schwartz
>> ______________________________________________
>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
>> 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.
>> ---
>> This email has been checked for viruses by AVG.
>> https://www.avg.com
> 
> -- 
> Michael
> http://www.dewey.myzen.co.uk/home.html
> 
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.

-- 
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Office: A 4.23
Email: pd.mes using cbs.dk  Priv: PDalgd using gmail.com