[R] My dream ...

Greg Snow 538280 @end|ng |rom gm@||@com
Mon May 11 20:11:03 CEST 2020


It is a nice dream, but it is really abdicating ethical responsibility
to the computer instead of the researcher.  And I personally don't
trust computers over people for this.

What could go wrong?

First, how do you guarantee that the statistical plan was locked in
place before the data was collected?  With your proposed system some
people will still run a plan on their data, make changes, run again,
etc. until they get what they want, then claim that the statistical
plan came before the data that was used to tune it (I consider this
unethical, but see no way for R or computers in general to prevent
this without human oversight).

Second, what if the data shows something that you did not anticipate?
This methodology would prevent you doing Exploratory Data Analysis and
adapting accordingly.  If people start using this as a black box that
is "blessed" by the package as purely "objective", then many will not
even do any EDA and take the results as "True" when they are not even
appropriate.

Better would be a regular system for submitting your code to a
clinical trial registry or some other pre-study registry so that other
can compare what you did to what you claimed you were going to do.  If
you don't want to submit the entire code, you could submit an md5 hash
to the registry, then others could check to see if the code ran (put
into an online supplement) differs from the registered version.  If
the data show something unanticipated then you can show the original
code and the modified code along with your reasoning and let the
consumer decide if the changes were justified.

This could still be worked around if someone was really motivated, but
it is probably the best we will see.

In my opinion the advantage of computers is not Artificial
Intelligence, but rather Artificial Patience (most AI that I have seen
is really doing a bunch of what I would consider to be boring, really
fast so people don't have to).  Leave the Intelligence to the people.

On Mon, May 11, 2020 at 9:10 AM karl adenener <adenener using hotmail.com> wrote:
>
> It would be a dream, there would be a R-based software, which I configure according to my study (type of data, limits for meaningful measurements, handling of outliers and missing measurements, test method etc.), which then reads my original measurement data and after some computing time the software provides me with the statistical analysis. All steps of the evaluation have to be defined before the start of the study and cannot be changed after the start of the study.
>
> Where could problems arise?
> Does anyone know of a suitable R-Package or software?
> Does anyone have the time and inclination to create a flexibly customizable package?
>
> greetings
> Adenener
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> 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.



-- 
Gregory (Greg) L. Snow Ph.D.
538280 using gmail.com



More information about the R-help mailing list