Thread by @dogemperor

I am STILL. NOT. DONE in pointing out the methodology problems with these polls, and again this is ANOTHER poll (the ABC/Ipsos poll) that is based, again, on *paid survey pools* and which have the SAME methodolology issues as most online polling

Here we go

1/x

So we have to, again, dive for the actual PDF and I should note a bit more on Ipsos as a polling group.

Ipsos gets its polling pools from two sources: Its own paid polling, and redirects from various "paid survey" portals like Qmee

2/xlangerresearch.com/wp-content/upl…

As I noted back with the CBS/YouGov poll, paid-survey polls are *opt in* (as in, you actually have to join these), and they are PRIMARILY done by students and elderly seeking "beer money" and occasionally WAHM sorts, many of which trend R.

3/x

Many younger folks won't participate in online polling of this sort (knowing it's generally a scam and they are generally the "product"), meaning that pretty much ALL "paid survey" polls will self-select for "easy marks" which trend R

4/xthreadreaderapp.com/thread/1842984…

And the reason I'm posting info on this, and the small pools that these survey groups use, and HOW they're self-selecting, is because I am pointing out some effectively *uncorrectable* issues with polling (that are in fact being used to drive a "horse race" narrative)

5/x

So with that said: The Ipsos poll only uses a poll of 2631 nationwide, and (yet again) this is a *registered voter* instead of a *likely voter* poll

The *likely voter* poll is even smaller: 1714 polled *nationwide*.

Which is already pretty fucky, but anyways

6/x

So this whole "dead heat" narrative was...based on an attempt to "balance" likely vs registered voters, out of *likely voters* Harris is ahead 50/48 (but we haven't even gotten into separation by age, etc.)

And as we dig in, we have hints of another right-wing pool

7/x

Ipsos includes "forcing questions", aka "if your choices were ONLY the two candidates, who would you pick?" for people who indicate they won't vote for Harris OR Trump (and this was a SMALL percentage, including 1 Stein and 1 Libertarian voter and one "Other")

8/x

Out of this data set of literally 2% of voters, you got mostly "Would not vote" (and more skips), and more Trump than Harris, which indicates again a more rightwardly skewed poll AND also sharing some methodology issues with the NYT poll.

9/x

(One methodology problem that's been pointed out with NYT is that *even if someone skips* and *even if someone hangs up* on a phone survey their response is still recorded; Ipsos also appears to record "skips" (refusal to answer) as a response as well.)

10/x

That said--even among THOSE sets, there's really not a whole lot of budging, though 11% of Republicans would consider Harris vs 8% of Dems for Trump, and HARD bars of 90% No (for Rs re Harris) and 94% No (for D's re TFG) for RVs

11/x

Also, mote notably, there is actually a higher enthusiasm for Harris (48-50 Very Enthusiastic and 34-36% Somewhat Enthusiastic to vote) vs TFG (43-45% Very Enthusiastic and 32-34% Somewhat Enthusiastic)

And this is a MAJOR dropoff for TFG from 2020

12/x

Some of the specifics on questions (on specifics re crime, the economy, and particularly immigration) also point to a conservative pool having been self-selected for (and if anything, this is not anything that's radically changed in the entire course of the survey)

13/x

And again--showing clear signs of a Dobbs Effect--more people trust Harris on abortion, by a LOT (47% strongly or somewhat, vs 32% strongly or somewhat for TFG). And that 32% STRONGLY points to "self-selected conservative pool"

14/x

Other signs that strongly point to "conservative self-selected poll" include at least one polling question related to tariffs (57% opposed to tarriffs, 42% approving) with fairly low margins on reducing corporate income taxes (42% opposed, 56% in favor).

15/x

Another indication of self-selected conservative poll--56% of recipients support forced deportation to home countries for undocumented immigrants (vs 36% in 2016)

And this is where I'm gonna note that the WORDING of questions can actually impact answers.

16/x

This year, the specific question is on "should the federal government forcibly deport undocumented immigrants?"

In 2016, the question is "Should undocumented immigrants be allowed to stay or deported?"

17/x

And THAT question resulted in 50% for people agreeing they should be allowed to stay, vs 44% calling for deportation.

Again: How a question is asked can in and of itself inject significant bias, and also unconsciously goad towards a specific answer.

18/x

So another sign of the Dobbs Effect--63% (and rising) explicitly oppose the Dobbs decision that destroyed Roe v Wade, and only 34% support

And the question immediately thereafter shows what I mean by "Bias in questioning can in fact impact results".

19/x

This year, the next question was worded "Would you rather have each state make its own laws on abortion, or have the federal government restore abortion access as it was before the Supreme Court ruling?

That resulted in 41% "leave to states", vs 56% "restore Roe".

20/x

The *Pre-Dobbs* wording--last done in *2013*--was "Do you think each state should make its own laws on abortion, or do you think this should be decided for all states on the basis of the US Constitution?"

That resulted in 30% "leave to states", vs 66% "keep Roe".

21/x

It is literally a subtle difference in phrasing, but it also has resulted in a difference of literally *ten points*.

That's what I mean by "Phrasing of questions has to be CAREFULLY done to avoid even accidental bias".

Let's continue.

22/x

Now, interestingly, there is a stronger general favorability rating for Harris (44%) vs TFG (35%)...and a STRONG unfavorable rating for Trump (58%) vs Harris (47%), with again a 10 point difference in "strongly unfavorable" (35% for Harris, vs 45% for TFG).

23/x

And ANOTHER indication we got a conservative polling group was that almost every Ipsos poll has had Harris as "too liberal" (48%) vs TFG being "too conservative" (at 43%) and both Harris (at 41%) and TFG (at 43%) close to "about right"

24/x

Now, those of you who've followed my writings on polls and methodology may have noted that, along with specific states being polled not being mentioned, there is LITERALLY one thing that DOES NOT SHOW UP IN THE IPSOS POLL AT ALL THAT SHOWS UP IN ALL OTHER POLLS.

25/x

And it's an INCREDIBLY important factor, because there are specific demographics that explicitly vote STRONGLY towards progressive causes that are *very hard* to capture, that only average around anywhere from 9% to 15% in political surveys

26/x

If you guessed "There is literally not a single tabulation in this entire fucking poll that shows ANY kind of breakdown by age demographic and how many people in the pool were in specific age cohorts" as being the thing missing, come down and collect your prize

27/x

Which means we actually HAVE NO IDEA OF THE ACTUAL POOL other than it's literally a pool of 2226 registered voters of which 1714 of those are likely voters

They could almost ALL be 45+ for all we know, but they didn't include ANY info on demographics by age

28/x

And THAT is sus as hell IMHO, especially since the colour text DOES indicate that they are able to track specific trends by race and by age cohort as well as gender, but again *no actual data regarding number of people who are of different age or racial groups is included*.

29/x

And honestly if they're not willing to put the breakdown of how people answered a particular question by gender, by race, or by age we can pretty much safely treat it as a lower-quality poll until this info is presented

Or, better yet, considered a distraction.

30/x

The lesson I'm trying to drive home is that there are functional *limits* to what we can derive from polls with shrinking pools of "pollable" individuals, which are *heavily* self-selecting for older conservatives as a *general trend*.

31/x

The one thing that we can use--and this particular poll's data is not even that particularly useful for what I'll describe--is seeing specific trends in populations and what THEY consider Hot Topics to use so that those can be used to GOTV for us.

32/x

Generalized polls of "X number of total folks like candidate A" are less useful for GOTV efforts than "X percentage of a large up-and-coming cohort that isn't showing up in the polls well at all DO consider abortion their primary driving issue in the election".

33/x

And the point I'm driving home today in pointing out the flaws in these polls is that there ARE considerable populations that just plain aren't showing UP in teeny-tiny polling pools, and that the polling pools themselves *heavily self-select towards likely R voters*.

34/x

And the reasons are honestly things that it is hard, if not functionally *impossible*, to correct for.

You can't really work *around* people under 45 being unwilling to participate in online or phone polls for the most part; not without some MASSIVE changes in privacy laws

35/x

You can't really work around the fact that *as an aspect of basic human psychology* the people more likely to do paid-survey polling and phone polling are ALSO "natural marks" that are *significantly more likely to vote R and fall to MAGAt conspiracy theory peddling*.

36/x

And when disregard for privacy, telemarketing, scams, spams, phishing, selling of metadata to the highest bidder (where it'll probably be exfiltrated and end up on a Russian black market) are the NORM, people protecting themselves means polling will INEVITABLY get worse

37/x

And that is WELL before I get into the concept of "polling as disinformation", which is to say, far-right narrative polls (and even some of the supposed "high quality" polls) being used to drive specific narratives or generate FUD

REMEMBER. THE. ASSIGNMENT KIDS

38/x

THE ASSIGNMENT IS ALL GAS NO BRAKES TILL NOVEMBER.

Check reg! vote.gov
Learn your options! iwillvote.com
Make plans to vote! threadreaderapp.com/thread/1763057…
Make a voting guide: threadreaderapp.com/thread/1800169…

AND VOTE IN THE ONLY POLL THAT MATTERS, Y'ALL.

39/end

@threadreaderapp unroll