MT-AL: Rob Quist (D) vs. Greg Gianforte (R) vs. Mark Wicks (L), May 25

[cinyc]

Are there any new Google polls coming out before the 25th?

As of right now, I'm planning on putting one into the field this weekend.  I might not if we get a credible poll between now and then.  But, that's doubtful, given we haven't seen a good poll yet.

Are you still planning on doing that poll cinyc?

Yes.  It will go into the field very late tonight or tomorrow afternoon.  The sample size will likely be 566, which hopefully gets us 400 useable respondents.  And I'm going to tweak the question a little because some people may have already voted absentee.

New wording (assuming it fits the character limit):

Many Montanans will go to the polls on May 25 to vote for a new U.S. Congressman. Others may have already voted or will vote absentee.  For whom would or did you vote?

[cinyc]

Are there any new Google polls coming out before the 25th?

As of right now, I'm planning on putting one into the field this weekend.  I might not if we get a credible poll between now and then.  But, that's doubtful, given we haven't seen a good poll yet.

Are you still planning on doing that poll cinyc?

Yes.  It will go into the field very late tonight or tomorrow afternoon.  The sample size will likely be 566, which hopefully gets us 400 useable respondents.  And I'm going to tweak the question a little because some people may have already voted absentee.

New wording (assuming it fits the character limit):

Many Montanans will go to the polls on May 25 to vote for a new U.S. Congressman. Others may have already voted or will vote absentee.  For whom would or did you vote?

Cool! Do you think you can give us some updates as the responses start to come in? Like at 100 responses/300 responses etc

I usually try to wait until the poll is completed before giving updates.  The problem with a very early update, like at 100 responses, is that the poll isn't reliable with 100 responses.  I've seen leads change before with that few responses.  I might drop a few hints if someone's lead seems stable.

[cinyc]

For example, I purchased a very small 60 sample size Google poll for the Montana race, there's only 26 responses in, but Quist leads Gianforte beyond the margin of error that google displays.

With 26 responses out of 60 in, the result so far is:

Rob Quist 46.2%

Greg Gianforte 11.5%

Undecided 30.8%

Won't vote 7.7%

When I look at the floor/ceiling that google shows me of Gianforte and Quist's support, Quist leads so much that his floor based on the margin of error is basically tied with the ceiling of Gianforte's vote share.

The ceiling of Gianforte's vote share is 28.9%, the floor of Quist's vote share is 28.8%. So would that indicate that the worst outcome for Quist is a 0.1% Gianforte lead?

Can you please educate me about this? I'm really interested

The link for the poll (it's still in progress): surveys.google.com/reporting/survey?survey=jaxfdpd3rxotdiygpxubxdp4c4 (can't post links yet)

The sample size is going to be too small for any meaningful analysis, unfortunately.  Quist likely will lead it in the end, but the MoE is going to be huge.  Raw, it's currently 12-3-1 Quist-Gianforte-Wicks.  There's no way that the poll would end up that lopsided with more respondents.

Your poll also currently faces the Google App vs. Website Ad polling sample issue - Usually, Google Surveys draw from both types, but with a small sample, you're getting all pro-candidate responses from websites, instead of the App.  The two types can poll differently. For example, if the first 25 voters are from the App (where respondents get paid for answering the poll and, IIRC, can't skip the question), you might end up with different results if the next 35 are from website Google Ads. At a quick glance, your poll is also slightly Western-Montana heavy - and we've seen a huge regional divide in all Google polls.

I usually track the different types in my spreadsheets.  (Here's the spreadsheet for Castro's recent poll, for example).

[cinyc]

I see what you're saying.

So why are the Google polls being done on this race all showing Quist in the lead, even with weighting, while all the Gravis polls and the Dem super pac poll are showing Gianforte in the lead? Is it to do with the Google polls being biased towards internet users or something?

Not all the Google polls have shown Quist in the lead.  My last poll back in mid-April found Gianforte with a 10 point lead (weighted).  That poll was taken around the time of the first Gravis poll and the Emerson poll.

Quite frankly, I'm not sure whether these Google polls are good for anything but entertainment value.  They have been all over the place.  Part of the reason is because most people other than me have had relatively small sample sizes - 333, which maybe yields 240 non-Undecideds/Not Likely to votes).  Even my polls have only gotten 300-355 usable responses. My first Google Survey back in March had Quist up by something unbelieveable, like 14 or 17 points.  

The one-question methodology we are using is extremely suspect.  The fact that we don't have a registration based sample or a way to exclude non-voters in a one-question poll is suspect.  The fact that we can only weight by age, sex and (maybe) region is suspect.  And the Google polls that various Atlas users commissioned before the November election were off more than they were on.  So I always take these with a huge grain of salt.  This may be my last attempt at a Google Survey if it isn't close to reality this time.

[cinyc]

Have you started the google poll yet, any early returns?

The survey was put into the field around midnight, Mountain Time.  It's too early to say anything about the returns.  We're only up to 165/567 respondents.  The poll should be completed by tomorrow or (most likely) Tuesday.

[cinyc]

When is the pending Gravis poll supposed to be released?

I don't think Gravis necessarily absolutely promised to release a new poll.  Their Tweet only says that they're going to TRY to release another one. 

The blog on their website that usually lists poll releases is currenly down.

[cinyc]

Can you just say whether Quist or Gianforte is in the lead so far? Pleeease? (I know it doesn't mean much I'm just really impatient lol)

It's way too early to responsibly do that.  There are too few respondents.  A pro-Quist or pro-Gianforte sample as lopsided as your Google Poll can easily change the bottom line.  Once you get more respondents, that becomes less possible.

Patience is a virtue.

[cinyc]

cinyc, how many responses so far? Any clues?

We're up to 355/567 responses.  Of those, 236 are usable - i.e. didn't choose the "I am not likely to vote in this election" option.  We're on track to get about 375 or so usable responses, if current rates continue.

[cinyc]

5PM EST = 6PM EDT.

Gravis is such a joke.

Let's wait to see if they polled the non-existent Green candidate again before definitively calling them a joke.

They're just late.  But they said around 5 Est, not 5PM on the dot.  It takes time to write up their analysis.  Note that the server for their website's Blog and Poll results has been down since at least yesterday.

[cinyc]

Can you give us some hints from your google poll, is it close so far or is one candidate far ahead?

Sorry.  I don't want to say much about the poll until it is done.  I don't think I've commented much on the last 5 Google Surveys I've done before before they were done - and would like to keep that policy.  Early results aren't necessarily representative (though we're probably getting to that stage now).  It's at 390/567, so we're probably looking at it being complete some time tomorrow unless the speed picks up.  And I've never been sure that these Google Surveys are worth anything, anyway, even when complete.

[cinyc]

Gravis just Tweeted their poll.  It's something like Gianforte +14.

Gianforte 49%
Quist 35%
Wicks 8%
Undecided 9%

https://twitter.com/GravisMarketing/status/866846140150423552/photo/1
https://twitter.com/GravisMarketing/status/866846545483894784/photo/1

[cinyc]

Gravis Crosstabs:
https://www.scribd.com/document/349161124/CROSSTABS-1

Gravis Poll:
https://www.scribd.com/document/349160889/Montana-May-22-2017-Gravis-Marketing

[cinyc]

Do my eyes deceive me, or does that poll give Trump an 80% approval rating with Hispanics, and a 58% approval rating with postgraduates?

There are very few Hispanics in Montana - about 2% of the VAP, or roughly 10,000.  So the MoE is likely very high.  There are even fewer Asians.

[cinyc]

Gravis Crosstabs:
https://www.scribd.com/document/349161124/CROSSTABS-1

Hispanics approve of Trump 80%-18%, Asians 56%-44%, Others 63%-21%, but Whites only approve of him 47%-45%?!?

Gianforte is also leading among Hispanics 75%-19%, Asians 56%-44%, Others 59%-20%, but only 47%-38% among whites.

Again, low sample sizes lead to strange results.  2% of the VAP is Hispanic, according to the 2016 CPS.  Less than 1% are Asian or Black.

You must be logged in to read this quote.

You should see the East vs. West regional gap in the Google Surveys we've done so far - they're also large.  You're always going to end up with some wonky cross-tabs.

You must be logged in to read this quote.

Probably.

[cinyc]

Lake county is generally a good bellwether in the state. If one just wants a rough approximation that's probably right on election night, this is probably the county to watch.

Other than Obama in 2008, I can't find a statewide race since 2000 where it picked the wrong winner.  

Lake County is kind of a bizarre bellwether.  Most of it is on the Flathead Indian Reservation, and it's 22% Native American.  Its demographics are not reflective of the rest of Montana.

[cinyc]

Lake county is generally a good bellwether in the state. If one just wants a rough approximation that's probably right on election night, this is probably the county to watch.

Other than Obama in 2008, I can't find a statewide race since 2000 where it picked the wrong winner.  

Lake County is kind of a bizarre bellwether.  Most of it is on the Flathead Indian Reservation, and it's 22% Native American.  Its demographics are not reflective of the rest of Montana.

Whatever it's demographics are, it's still a decent bellwether as Miles said.

I'm not doubting it's a decent bellwether - but it isn't one because it reflects the demographic makeup of the state.

[cinyc]

Gravis polled five Hispanics and four of them were Republicans.

I'm aware this may not have been meant seriously, but the crosstabs indicate that 2% of Hispanics are undecided on Trump - disapproval is at 18% and approval is at 80%. To get those numbers they would have needed to survey 50 Hispanics or more.

Wrong.  Gravis is likely weighting their crosstabs, as most pollsters do.  Any number of respondents could get you an 80-18 split.  Even 3 repondents, in theory.  

Suppose Gravis overpolled female seniors, and they were weighted very low because there are so many of them.  She was undecided.  Next, they polled a 50 year old, and he was pro-Quist and somewhat low weighted, since there were a lot of 50-year old males in the sample.  Then, they polled a 18 year old female, who picked Gianforte.  Because there were so few millenial females in the sample, the weight was high.  Put the three together, and you could theoretically get to 80-18-2.

They probably polled more than 3 Hispanics, but it doesn't need to be 50.

[cinyc]

On a bigger note, why didn't any respectable pollsters get a read on this race? All we got was Gravis and a bunch of Google surveys.

We also got an Emerson College poll showing Gianforte up 15 in mid-April.  Not that that improves things much.

As KingSweden said, Montana requires live dialers, which increases polling costs.

[cinyc]

We're at 482/567 on my Google Survey.  It should be finished tomorrow.  So far, it doesn't confirm Gravis' poll.  

[cinyc]

We're at 482/567 on my Google Survey.  It should be finished tomorrow.  So far, it doesn't confirm Gravis' poll. 

Do you do these polls just to help satiate your curiosity? Regardless of the reason, thank you for sharing them here!

It's part curiosity, part hope that they're actually accurate and actually tell us something about the race.  Amdcpus got me hooked back during the 2016 election.  Once you've commissioned one Google Survey, it's somewhat addictive.  But I've been increasingly skeptical of their reliability.  Had I not done the first poll back in March, and the second poll in April, I wouldn't have done this one.   This poll is probably my last.  It's a huge waste of money.

[cinyc]

We're at 482/576 on my poll.  It should be finished tomorrow.  So far, it doesn't confirm Gravis' poll. 

But does it confirm my 79 sample size "poll" with Quist 40 points ahead?

It's not done yet, but I'm pretty confident that nobody will end up 40 points ahead.  These Google polls can be pretty wonky sometimes from results release to release, though.

[cinyc]

My Google Survey is complete.   The poll is available here, along with my weighting spreadsheet with weighted crosstabs.

573 total respondents were polled from May 21-23.  Of the 573, 385 chose a candidate.  The question asked was:

You must be logged in to read this quote.

Weighted by Age and Sex (to the November 2014 CPS for actual voters) and Region (West vs. East), the Results were:

Quist 54%
Gianforte 39%
Wicks 7%

Or Quist +15.

Raw, it was Quist+14.  Simple Weighted by only sex and age, like we've done for other polls this cycle, it was Quist +16.

As usual for our Google Surveys, the poll showed a big East/West Divide.  Western Montanans (n=181) preferred Quist by 32 points (weighted by age, sex and region and raw).  Eastern Montanans (n=134) picked Gianforte by 4 (5 raw).

Recall that I've split Eastern and Western Montana this way in our Google Polling:


There is a gender gap, with men supporting Quist by 12 (weighted)/7 (raw), and women supporting Quist by 23 (weighted)/26 (raw).  Every age group except 45-54 year olds backed Quist.  And, as in other Google Polls, there is a huge Suburban/Rural gap, with suburban voters backing Quist by 24/21, and rurals only going Quist by 1/3.  Rural voters made up 37% of the weighted and raw samples.

More, perhaps, to come.  But this poll looks like as big an outlier as Gravis', unfortunately.  Average the two polls together, and we might be on to something.

[cinyc]

Here's a chart on how the results came in over time (cumulative weighted):



As you can see, there was an early huge spike for Quist, and Gianforte never recovered.  The poll almost reverted to single-digit respectability on late Monday, but then started veering into outlier territory again.  Quist's lead actually grew to 20 points overnight before reverting downward to 15 in the final tally rather quickly.  

This is why I don't like releasing poll results early.  They tend to be swingy.  All of this is probably moot, though - unless Quist wins by double-digits (unlikely), I'm never commissioning another Google Survey.  They're crap.

[cinyc]

Oh, and needless to say, there is zero chance that Quist will win by 15 points either. These Google polls are obviously biased in favor of Ds.

I'm not sure these Google polls are biased in favor of anyone in particular.  They're just bad.  Remember - I got a plausible result in line with Emerson in April (Gianforte +10 weighted, compared to Emerson's G+15), after getting a ludicrous Quist+17 in March.  The two recent 60ish sample polls had Quist romping in one and Gianforte in a squeaker in another.  And who knows?  If someone else commissions a Google Survey in between now and the election, it could show Gianforte up big.

I just think they're trash.  People who are reading the newspaper articles are randomly clicking on anything to get to the article.

And then there are ways that they make sense.  For instance, bellwether Lake County is polling pretty much on par with the whole state in this poll.  Granted, the sample size is small, and it all may be happenstance, but it's interesting.

[cinyc]

What does the relatively High turnout mean?

Absentee turnout so far is higher than 2014, but lower than 2016.