POSTING RULES & NOTES
#1 YOU MUST clip all extraneous text when replying to a message.
#2 This mail-list, like most, is publicly & permanently archived.
#3 Subscribe and post under an alias if #2 is a concern.
*
On 8/11/2017 1:15 PM, Erik Toren via Marxism wrote:
Hispanic is a big category. I wonder which subgroup of Hispanics has
surged.
Erik
Public opinion surveys claiming things about Latinos generally would not
be worth the paper they're printed on if someone printed them. But the
entire field of public opinion polling is in crisis, for the techniques
that worked in the 20th Century no longer do. Most of all, people aren't
cooperating with surveys
But apart from that, Zogby surveys are systematically biased (in a
statistical sense) due to their design. It is a self-selected population
skewed towards heavier internet and especially email usage.
Zogby says they're surveying "Likely voters". Likely voters in what
election? You may say, well, the next one is November 2018 but that is
wrong. My county board of elections had 12 elections scheduled for the
current year, one a month. In reality, from among the self-selected
participants, Zogby is deciding who merits being counted and whose
answers should be discarded as coming from a member of the great unwashed.
The traditional screens for "likely voter" include questions like do you
know where your voting place is? Who is running in this election? When
did you register to vote? Those really only become significant when the
election is very close.
Zogby did not detail its methodology for this survey, but we know they
survey by email invitation. A greater proportion of Latinos are Internet
users but a higher percentage (than among the general population) have
access mostly or exclusively through smartphone, so they use email less.
Although "Hispanic" is now the largest racial/ethnic minority (that the
government counts), surveys rarely break out numbers for Latinos, even
though they do so for Blacks. That's because they have so few Latinos
that actually respond. The response rate among Latinos is lower than for
anglos and Blacks even in bilingual phone surveys.
But the truth is the OVERALL response rate is now so low for the
population as a whole as to call into question the validity of polling,
period. That's why the largest and most prestigious polling
organization, Gallup, has stopped doing election polls.
When I first became involved in the field at CNN in 1990, the response
rate for phone polls was close to 50%. For the last few years it has
been under 10%, according to Pew. For email-invitation surveys, the
response rate is even lower.
Zogby says they had 1300 people in their new poll. They do these
presidential approval surveys frequently and in July they had 856 and in
June, 1031 and at the end of April 876. This raises the obvious
question: did the methodology change? Were different numbers of
invitations to participate sent out? Or did the response rate vary
wildly, so that half again as many people responded now as three or four
weeks ago? If so, THAT is probably a more significant data point worth
exploring than what they think of Trump.
But we don't know because Zogby only says they send out "thousands" of
emails. Nor do they publish a detailed, question-by-question breakdown
of results like reputable pollsters do.
And then consider the Latino sub-sample. If they actually got the same
percentage as among the 2016 voters (10%) that would have been 130 of
the 1300 total respondents. But my experience tells me it was about
2/3rds that number, let's say 90. The MoE with that sample is more than
10%, which means the numbers this survey gives for Latinos are
meaningless. Even at 130 they're junk (MoE=8.7%). And the previous
surveys that the new Hispanic number is being compared to was 856, so
there was no way you could have gotten meaningful percents out of that
poll.
BTW under representation of a huge problem because it is not just
Latinos. Experience shows polling respondents systematically under
represent young people, minorities, and people without a college
education. So you take your raw results, and then you look at the
demographics you SHOULD have had, based on, for example, census bureau
numbers, and weigh the results accordingly. So if you should have had
120 Blacks but actually only had 100 Blacks respondents, you give a
"weight" of 1.2 to each Black respondent.
But when it comes to "likely" voters, you define what the target
universe looks like, how many Blacks, how many high school graduates.
So, are the demographics those of the 2016 presidential or the 2014
off-year electorate? If we're talking about the 2020 electorate, who is
the Democrat? Clinton gives you one electorate, Bernie, a very different
one.
(This was the big reason Gallup abandoned