Fox's awful scheme for selecting GOP debators

[Rich Ulrich]

about - FOX's awful scheme for selecting GOP debators -
How would you improve the selection?

This is a question of wise data collection, before it is one
of statistics. But statisticians have the experience that
ought to be relevant.

Here is the situation: The GOP has 15 or 20 candidates
for President. Fox TV and the Republican National Committee
announced that the first debate would have 10 persons;
and those 10 would be the candidates who were "doing
best in the polls." Everyone has presumed that "doing best"
will use the usual poll-result: Having the highest percentage
on, "Who is the one that I prefer over all the others?"

Oh, and the RNC promised to punish anyone who took
part in any "unsanctioned" debate by barring them from the
other, later officially sanctioned ones yet tol be scheduled.

Here is one feature that makes this an awful scheme: Even
ignoring whether 10 people can "debate", what promises to
be the lowest qualifying "score" is 3 or 4 percent, at most...
This is not at all a rousing endorsement; plus, it strikes out
the next 3 or 4 candidates who will rate only a fraction of a
point lower (and well within the limits of polling accuracy).

By the way, according to Rachel Maddow on MSNBC, the GOP in the
three earliest Primary states are apparently joining together to
protest, so the original plans might yet be altered.


I suggest that the original scheme migt be rescued -- at least,
improved -- by finding a more reliable and face-valid choice of
"top 10."

I'm pretty sure that results would be more replicable, and
probably more meaningful, if pollees were simply allowed
to name their top three ( or 4, 5, 6 ...) five choices, and all
those were tablulated; instead of counting just one. Or the
Aussie system could be used or modified: Rank your choices,
and let several ranks be included for each person after
striking out their least popular choices.

The only other poll result that has typically been presented
is "Who I would never vote for". Today, Donald Trump
announced he will run. By name recognition, it is apt to
end up in the top ten for the first debate, but he also has
rung up, presently, the *highest* rating for unfavorability
seen by the pollster that MSNBC cited, out of 98 candidates
in the last 15 years. He is presently at 57%, compared to
43% for the next highest.

- This example lets me point out that "name recognition"
is one confounding variable, which might or might not be
something to accommodate -- What can we do with "Missing",
which potentially should be recorded in place of an opinion?

Also, "Unacceptable" is another piece of information that *ought*
to be accommodated, since it certainly does speak to the
original question of "doing best in the polls." - I suspect that
there are a lot of people who should like a rule to exclude Trump.

Does anyone have a model for this sort of selection?

One thing that comes to mind - partly because I always
prefer "scores" over "ranks" - is something like the IMDB
(International Movie Data Base) scores: Let every pollee
rate the candidates that they *choose* to rate, on a scale
of 1 to 10. Warn the pollees about how a weighted scoring
will be done: that they do need to rate multiple candidates,
and to down-score some candidates, or their scores will have
little effect for their candidate. (Does a pollee's scores get
weighted by their variance? by how many candidates they
mark? or what? - This offers some possibilitiies for tuning.)

[Rich Ulrich]

On Thu, 18 Jun 2015 01:10:38 -0400, Rich Ulrich

Post by Rich Ulrich
about - FOX's awful scheme for selecting GOP debators -
How would you improve the selection?
This is a question of wise data collection, before it is one
of statistics. But statisticians have the experience that
ought to be relevant.
Here is the situation: The GOP has 15 or 20 candidates
for President. Fox TV and the Republican National Committee
announced that the first debate would have 10 persons;
and those 10 would be the candidates who were "doing
best in the polls." Everyone has presumed that "doing best"
will use the usual poll-result: Having the highest percentage
on, "Who is the one that I prefer over all the others?"
Oh, and the RNC promised to punish anyone who took
part in any "unsanctioned" debate by barring them from the
other, later officially sanctioned ones yet tol be scheduled.
Here is one feature that makes this an awful scheme: Even
ignoring whether 10 people can "debate", what promises to
be the lowest qualifying "score" is 3 or 4 percent, at most...
This is not at all a rousing endorsement; plus, it strikes out
the next 3 or 4 candidates who will rate only a fraction of a
point lower (and well within the limits of polling accuracy).
By the way, according to Rachel Maddow on MSNBC, the GOP in the
three earliest Primary states are apparently joining together to
protest, so the original plans might yet be altered.
I suggest that the original scheme migt be rescued -- at least,
improved -- by finding a more reliable and face-valid choice of
"top 10."

Okay, no one wants to offer? I still think that the pollsters
should be trying for better descriptions, even if that takes them
to foreign ground like cluster analysis.

I've seen one more Poll result offered recently, though
it was not presented for the whole set:
"Would you be willing to vote for this candidate for President?"

This is a little particular than Favorable/Unfavorable, and maybe
more generous.

One more option that would give higher and more reliable counts
is something that I borrow from the Pennsylvania ballot for
judges. "Select no more than 10 candidates. Ten will be elected."
I think the pollers for Fox would get better results if they asked
for only 5 or so.

The problem, both for polling and the elections, is how much
the results can be dominated by name recognition.

Post by Rich Ulrich
I'm pretty sure that results would be more replicable, and
probably more meaningful, if pollees were simply allowed
to name their top three ( or 4, 5, 6 ...) five choices, and all
those were tablulated; instead of counting just one. Or the
Aussie system could be used or modified: Rank your choices,
and let several ranks be included for each person after
striking out their least popular choices.
The only other poll result that has typically been presented
is "Who I would never vote for". Today, Donald Trump
announced he will run. By name recognition, it is apt to
end up in the top ten for the first debate, but he also has
rung up, presently, the *highest* rating for unfavorability
seen by the pollster that MSNBC cited, out of 98 candidates
in the last 15 years. He is presently at 57%, compared to
43% for the next highest.
- This example lets me point out that "name recognition"
is one confounding variable, which might or might not be
something to accommodate -- What can we do with "Missing",
which potentially should be recorded in place of an opinion?
Also, "Unacceptable" is another piece of information that *ought*
to be accommodated, since it certainly does speak to the
original question of "doing best in the polls." - I suspect that
there are a lot of people who should like a rule to exclude Trump.
Does anyone have a model for this sort of selection?
One thing that comes to mind - partly because I always
prefer "scores" over "ranks" - is something like the IMDB
(International Movie Data Base) scores: Let every pollee
rate the candidates that they *choose* to rate, on a scale
of 1 to 10. Warn the pollees about how a weighted scoring
will be done: that they do need to rate multiple candidates,
and to down-score some candidates, or their scores will have
little effect for their candidate. (Does a pollee's scores get
weighted by their variance? by how many candidates they
mark? or what? - This offers some possibilitiies for tuning.)

[Rich Ulrich]

Just a followup about the the leadin to the Fox debates.

On Wed, 24 Jun 2015 13:43:17 -0400, Rich Ulrich

Post by Rich Ulrich
One more option that would give higher and more reliable counts

Last week, Rachel Maddow happened to show a set of
bar charts that also mentioned "second choice", even though
only the first choice was mentioned in her commentary.

Rachel M. has regularly over-stated the statistical error
of these polls. This evening, she shows several polls with the
errors *specified* for each, as 3% or 4%. Now, this evening
she has a pollster on, and he just stated that the "3%" is
the error at 50%, whereas, in fact, it is different in the tails.
Right! but ...

Unfortunately (for good sense).
Either I missed some body language, or his full suggestion
actually was that there is MORE error at the tails. Which
is not how the binomial works.



Chris Matthews made an interesting summary this
evening, which made mention of "sets" of candidates.

He proposed that the candidates who are thoroughly
"Teaparty" share among themselves 50% support
as the first choice among likely Republican primary
voters; whereas the 3 candidates who make some
attempt to remain acceptable to main-line Republlicans
win only 30%. He concludes that when the fields
shrink, it is really likely that a Teapartyer will win.

Matthews also showed us poll results on national match-ups
between GOP leaders and the second Democratic candidate,
the self-proclaimed socialist Bernie Sanders. Sanders
led by about 20 points against Trump, and also led against
the other two.

Meanwhile Sanders continues to draw crowds larger
than any other candidates - the most recent crowds
being in strong Republican territory including Dallas
and Louisiana. - My own intuition about "Teaparty"
is that many of these folks supporting those candidates
may be willing to support Sanders, since a lot of their
motive is disaffection for the conventional.