I attended an event in central Tokyo yesterday.

Many IT workers attend events of this sort seeking information on
promising fields.  I notice that many of them are looking for a quick
and easy way in.  One with this attitude never gets far.  One may recall
that when asked by the king for an easy way to master geometry, Euclid
replied there is none.

I noticed that the man who had in the past represented a
crypto-currency organization was manning a booth of users of
generative AI.  I asked, and he told me that he is still is in
crypto-currency.  Those who profess to be specialists in subjects that
require advanced math are respected.  On the other hand the
intellectual demands places these fields beyond the reach of many an
IT engineer.  Perhaps one receives more respect when he declares he is
in ChatGPT compared to Bitcoin.

I spoke to the person at the sales booth of one of Japan's major
computer textbook publishers.  I suggested that a book to help older
people re-learn math might be a good idea.

---

I brought up a statistics issue that has recently got my attention
with some people.

  Note: My concern here is the variance, not the popularity figures.
  The variance (or standard deviation, sigma) tells us how much
  confidence the surveying agency has.

Three surveys were quoted in the following article on US presidential
election polls:

https://www.yahoo.com/news/presidential-election-polls-2024-latest-160812649.html

  The survey of 1,500 registered voters, conducted from Oct. 19 until
  Oct. 22, showed Trump leading 47% to 45%, within the poll's margin of
  error of 2.5 percentage points.

  ...

  The poll of 1,000 voters nationally showed Trump with a 48% to 46%
  lead over Harris, within the poll's margin of error of 3.1%. The
  survey took place from Oct. 15 to Oct. 19.

  ...

  The poll of 1,260 likely voters conducted between Oct. 21 and Oct. 23
  shows Harris leading 50% to 47% with a 2.8% margin of error.

- Summary -

1. 1500 samples, margin of error (MOE) 2.5%

2. 1000 samples, MOE 3.1%

3. 1260 samples, MOE 2.8%

In an ordinary survey, the above MOE figures make sense.  However, the
above pollss are not ordinary: to cope with differences in the
response rates among demographic groups, pollsters apply adjustments
in the form of "weights."

For groups with a small number of samples the variance will be large.
There are many such groups.  This should affect the aggregate MOE: the
figures reported above don't look correct.

I think that if correctly calculated, the MOE would be so large that
people would start wondering whether these polls are worth attention.

However the mainstream media makes loud headlines out of small shifts
which are only to be expected from sample surveys.  This is going on
in the US on a daily basis reently.

---

I believe it is a good idea to occasionally discuss math issues with
co-developers.  It tells people how well a grasp one has upon certain
concepts.

Statistics is widely used in programs and quality evaluation thereof,
but it is widely known that as with accounting, you can make things
up.  We'd like to assign matters to people who understand the math and
can be trusted for their integrity.


Thank you for reading.

Akira Urushibata

_______________________________________________
libreplanet-discuss mailing list
[email protected]
https://lists.libreplanet.org/mailman/listinfo/libreplanet-discuss

Reply via email to