On 2019-10-16 19:43, duncan smith wrote:
On 16/10/2019 04:41, DL Neil wrote:
On 16/10/19 1:55 PM, duncan smith wrote:
On 15/10/2019 21:36, DL Neil wrote:
On 16/10/19 12:38 AM, Rhodri James wrote:
On 14/10/2019 21:55, DL Neil via Python-list wrote:
...
So, yes, the "label" is unimportant - except to politicians and
statisticians, who want precise answers from vague collections of
data... (sigh!)
[snip]
No not (real) statisticians. People often want us to provide precise
answers, but they don't often get them.
"It ain’t what you don’t know that gets you into trouble. It’s what you
know for sure that just ain’t so." (Mark Twain - perhaps)
+1
Although, you've undoubtedly heard people attempt to make claims of
having 'accurate figures' (even, "that came from Stats") when you told
them that the limitations and variations rendered the exercise laughable...
My favorite (of the moment) is a local computer store who regularly
offer such gems as: (underneath the sales (web-) page for an upmarket
*desktop* computer) "people who bought this also bought" followed by at
least two portable PC carry cases. They must be rather large carry-bags!
(along with such surprises as keyboard, mouse, ...)
This morning I turned-down a study for a political group. One study has
already been completed and presented. The antagonist wanted an A/B
comparison (backing his 'side', of course). I mildly suggested that I
would do it, if he'd also pay me to do an A/B/C study, where 'C' was a
costing - the economic opportunity cost of 'the people' waiting for 'the
government' to make a decision - (and delaying that decision by waiting
for "study" after "study" - The UK and their (MPs') inability to decide
"Brexit" a particularly disastrous illustration of such)
Sorry, don't want to incur the anger of the list-gods - such
calculations would be performed in Python (of course)
Clearly, all such analyses should be done in Python. Thank God for rpy2,
otherwise I'd have to write R code. It's bad enough having to read it
occasionally to figure out what's going on under the hood (I like
everything about R - except the syntax).
> I have too many examples of people ignoring random variation, testing
hypotheses on the data that generated the hypotheses, shifting the
goalposts, using cum / post hoc ergo propter hoc reasoning, assuming
monocausality etc. In some areas these things have become almost
standard practice (and they don't really hinder publication as long as
they are even moderately well hidden). Of course, it's often about
policy promotion, and the economic analyses can be just as bad (e.g.
comparing the negative impacts of a policy on the individual with the
positive impacts aggregated over a very large population). And if it's
about policy promotion a press release is inevitable. So we just need to
survey the news media for specific examples. Unfortunately there's no
reliable service for telling us what's crap and what isn't. (Go on,
somebody pay me, all my data processing / re-analysis will be in Python
;-).)
Even when using Python, you have to be careful:
Researchers find bug in Python script may have affected hundreds of studies
https://arstechnica.com/information-technology/2019/10/chemists-discover-cross-platform-python-scripts-not-so-cross-platform/
--
https://mail.python.org/mailman/listinfo/python-list
