Re: Languages don't matter. A mathematical refutation

["Jean François Martinez" <darkquark99@gmail.com>]

On Thursday, April 2, 2015 at 6:39:11 PM UTC+2, Dmitry A. Kazakov wrote:
> On Thu, 2 Apr 2015 08:58:41 -0700 (PDT), Jean François Martinez wrote:
> 
> > On Thursday, April 2, 2015 at 3:36:02 PM UTC+2, Dmitry A. Kazakov wrote:
> > 
> >>> To change the subject a bit, is there any possibility of using
> >>> the probabilistic approach to estimating the long-term cost of
> >>> maintenance given a history of errors with a particular language?
> >> 
> >> I don't think so. Errors do not occur, they are present or introduced. The
> >> only probabilistic model could be random selection of a given error from
> >> the pool of errors. This is not related to the language choice. All
> >> difference here is in the pools, since errors made in C are different from
> >> errors made in Ada. You change the language, you get *other* errors, no
> >> statistics across to compare.
> > 
> > Really?  Errors occur because of progarmmer quality, physical shape (had a
> > bad night), distracting events, use of alcohol and drugs and plenty,
> > plenty of factors.   One of them being that a bird has just landed  in
> > front of your window.  Random.  
> 
> You are describing the process of programming. This is another model, which
> is certainly not random. If that were random, the same programmer asked to
> design the same program several times would randomly introduce any possible
> errors. That is not the case. People have memory, they learn, decision
> making process is nowhere random. In general, only typo errors could be
> considered random. Anything above simple motoric functions is not random.
>  
> -- 
> Regards,
> Dmitry A. Kazakov
> http://www.dmitry-kazakov.de

The problem is:  None of what you have said from the beginning is relevant
about how statitical tests work.

Let's say we have white balls and black balls.  The reasons they are white or black is definitely not random since it depends on the operator filling the machine with white or black paint.  :-).  Then from a conceptually infinite-sized urn  you take two samples at random.  The averages of the samples will differ from one another and from the expected value http://en.wikipedia.org/wiki/Expected_value (that is the propoortion of white balls in our urn) but usually not by that much and anyway you can calculate probability of the difference and probability of difference being above a certain threshold.   More exactly we approximate to a Student law or to a gaussian law depending on the size of the sample.  

Now we have two samples of programmers  and in each sample there are white balls (programmers who succeeded) and black balls (programmers who failed). 
Just like for balls being black or white due to deterministic reasons, they
may succeed or fail for deterministic reasons   We <em>don't care</em> and <em>don't care</em> why.  We make the <em>hypothesis</b>  these are samples of the same urn (ie from an homogeneous population so same proportion of white and black balls, of succeeding and failing programmers) calculate the averages and the probability we would have of getting a such difference or higher. Then we compare that probability with a threshold (usually 5% or 1%) and if it is superior we accept the hypothesis (two samples from the same urn) if it
is inferior we reject the hypothesis and assumme these are not two samples from the same urn but from different urns with one of them having a higher mix of white balls (succeeding programers).  In our case probability was inferior to one in one million so the null hypothesis (same urn) is rejected.

That is math, <em>it is correctly applied</em> and that is all what we can deduce with math.  To find the reasons of why one urn that happens to be labelled Ada has a higher proportion of white balls than the urn labelled C
requires applying field knowledge. 


Regards

Jean-François Martinez