[Development] The place of QML

[André Somers]

Op 16-5-2012 1:31, André Pönitz schreef:
> On Tue, May 15, 2012 at 07:13:19AM -0700, BRM wrote:
>>> From: Donald Carr<sirspudd at gmail.com>
>>> [...]  This is out of a sample set of 110 people, which is
>>> infinitesimally small in comparison to the Qt user base. It would
>>> be a stretch to call this a statistically significant poll, [...]
>> [...] But the sample sizes are no better than yours - 114/139/19
>> respectively for each poll.  Again, statically insignificant.
> I wonder how the two of you came to the conclusion of statistical
> insignificance, and how you explain the apparent discrepancy of this
> result with, say, the section titled "Estimating proportions and means"
> on http://en.wikipedia.org/wiki/Sample_size_determination. [1]
>
> I trust the audience would be delighted if you gave some insight
> into the calculation.
>
> Andre'
>
> [1] That link is admittedly a random pick, but Wikipedia seems to
> be an acceptable replacement for lecture notes nowadays.
>
If we're throwing wikipedia links back and forth anyway, also consult 
http://en.wikipedia.org/wiki/Sampling_%28statistics%29#Errors_in_sample_surveys 
please. I agree with you that perhaps the sample _size_ is not an issue, 
but the sample suffers from *many* of the errors mentioned on the URL 
above, including selection bias, undercoverage and measurement error. 
Doing good surveys is just plain hard to do, and the polls discussed so 
far certainly don't qualify as "good" or even "acceptable". There are 
people getting a PhD on this kind of thing...

However, I think the whole point is moot. I don't think the people who 
get to decide - that is, those doing or funding the actual work - care 
much about any such statistics, no matter if they are sound or not. And 
more generally: facts don't seem to matter much in this discussion 
either for some participants, or at least not those facts that tell a 
different story than they want to hear.

André