Reuters/Ipsos: (Most) States

[Seriously?]

Haven't heard a peep about this from the media, but ol' Nate has it...

http://projects.fivethirtyeight.com/2016-election-forecast/kentucky

Of all places, Kentucky is actually in play.

This same updated pile of junk now has Trump ahead by 14 in New Hampshire. Trump also leads in Wisconsin and Michigan (unlikely). Hillary leads in Nebraska.

They somehow have "insufficient data" to give Hillary DC and Hawaii. And Trump Montana, North Dakota, Wyoming and Alaska.

Until the kinks are all ironed out or they have more respondents, these polls are probably as bad as the YouGov effort of every state four years ago.

[Seriously?]

I refuse to format this junk, but here you go. The latest data, including battleground Kentucky, Reuters doesn't know who wins DC and the 14-point New Hampshire Trump surge!

State     Clinton   Trump
Alabama   42%   48%
Alaska   —   —
Arizona   43%   46%
Arkansas   39%   48%
California   61%   25%
Colorado   43%   40%
Connecticut   50%   37%
Delaware   —   —
Florida   49%   42%
Georgia   43%   46%
Hawaii   —   —
Idaho   29%   49%
Illinois   53%   26%
Indiana   34%   55%
Iowa   43%   40%
Kansas   37%   48%
Kentucky   43%   45%
Louisiana   39%   50%
Maine   39%   40%
Maryland   53%   30%
Massachusetts   46%   33%
Michigan   43%   44%
Minnesota   44%   35%
Mississippi   37%   53%
Missouri   40%   46%
Montana   —   —
Nebraska   45%   41%
Nevada   41%   39%
New Hampshire   34%   48%
New Jersey   48%   34%
New Mexico   —   —
New York   53%   31%
North Carolina   48%   44%
North Dakota   —   —
Ohio   47%   41%
Oklahoma   32%   54%
Oregon   45%   37%
Pennsylvania   50%   43%
Rhode Island   —   —
South Carolina   46%   46%
South Dakota   —   —
Tennessee   33%   46%
Texas   32%   45%
The District of Columbia   —   —
Utah   34%   39%
Vermont   —   —
Virginia   48%   37%
Washington   49%   33%
West Virginia   39%   47%
Wisconsin   36%   39%
Wyoming   —   —

[Seriously?]

There Nebraska statewide has Clinton leading Trump by four points (538). I'll just put that out there.

Also, still not sure how 538 is getting the sample sizes. And why would a polling company even put out polls with samples under 200?

I remember You Gov doing something similar in either 2014 or 2012. The sample sizes for some of the states were putrid. This thing has an interesting methodology where some states are 1, 2 or 3 week samples depending on the sample size.

Reuters/Ipsos must be supplying the information to 538. I don't see anything on either Reuters' or Ispos' websites.

[Seriously?]

Can we stop cherry picking results that we like from this poll, and just toss it all in the trash?

Actually, I don't think that's the right answer.

So long as you understand that out of the 50 states+DC, there are going to be an average of 2.5 polls that fail as a matter of science, the model they are using for turnout and know the MOE in the thinly polled states, these polls won't be terrible.

Like anything else, they are simply data points at the end of the day, so long as they are put in the right context. Hopefully, the thinly polled states will get some heft to them in the upcoming weeks.

If you put it into that context, we may laugh at a number of the results, but they are not absolutely terrible as a matter of science to be completely discounted as a whole. The ones where they have a decent sample should be somewhat decent polls.

[Seriously?]

What in god's name is this pile of junk?!

I really think people should stop trying to poll all 50 states. It never seems to go well. Though these make the Morning Consult ones look excellent by comparison.

It's the MOE fluctuations from a small sample size that's the issue here. If they can increase the number of people polled, they may have the start of something decent here. I don't think any internet polling outfit, including YouGov has quite gotten there yet though. The smaller states just have outrageous MOEs, which make the poll useless.

There should, however, be some heft to the larger states with bigger sample sizes.

[Seriously?]

Complete List

This assumes Hillary +3 (Hillary 44, Trump 41)

NAME   CLINTON   TRUMP
Alabama   39%   52%
Alaska   —   —
Arizona   41%   45%
Arkansas   42%   48%
California   63%   24%
Colorado   45%   39%
Connecticut   47%   39%
Delaware   —   —
Florida   48%   45%
Georgia   41%   47%
Hawaii   —   —
Idaho   28%   58%
Illinois   50%   37%
Indiana   32%   56%
Iowa   41%   44%
Kansas   37%   52%
Kentucky   42%   46%
Louisiana   37%   57%
Maine   42%   42%
Maryland   52%   32%
Massachusetts   48%   32%
Michigan   41%   42%
Minnesota   42%   33%
Mississippi   30%   59%
Missouri   35%   51%
Montana   —   —
Nebraska   38%   45%
Nevada   43%   35%
New Hampshire   44%   45%
New Jersey   47%   36%
New Mexico   —   —
New York   50%   28%
North Carolina   49%   44%
North Dakota   —   —
Ohio   43%   46%
Oklahoma   37%   48%
Oregon   44%   39%
Pennsylvania   48%   42%
Rhode Island   —   —
South Carolina   45%   48%
South Dakota   —   —
Tennessee   31%   49%
Texas   32%   49%
The District of Columbia   —   —
Utah   34%   35%
Vermont   —   —
Virginia   50%   37%
Washington   45%   35%
West Virginia   38%   55%
Wisconsin   38%   38%
Wyoming   —   —

[Seriously?]

And of course 538 adds them.

So can anyone explain why Ipsos/Reuters have an A- ranking from 538?

These statewide polls leave a bit of something for everyone when it comes to cherrypicking individual state polling numbers, but their results are so far off the map when it comes to not only comparing states in terms of national PVI. but additionally the numbers from even large states like California and Texas, let alone random inexplicable numbers between Arkansas/Kentucky and Louisiana/Mississippi for example.

Differences between Nebraska/Kansas don't appear intuitively appear to make sense....

So, we'll see but when you do the state by state comparisons these results don't make much logical sense.

A lot of this is just polling around the MOE. You also have to keep in mind that about 2 1/2 of each 51 polls will fail as a matter of science.

The methodology is also a little unique based on sample sizes recorded in any given state over a period of time.

Each state sample must have 500 likely voters to get a one week result. If that number is not met, the poll goes out to a two week or three week sample. If they can't get a sample with 70% of the simulations has one candidate winning over another over two or  three-weeks, the state goes to "insufficient data." It also goes to insufficient data is the sample size is less than 60.

As a result, there's a bit of a lag in real time with some of these states, hence the Clinton +3 sample with the "State of Polls" result while it's Trump +1 in the nationwide Reuters "Polling Explorer."

If you keep all of those variables in mind, you get a better idea of how this poll works and how you can have some of the variances we are seeing.

Source: http://www.reuters.com/investigates/special-report/states-of-the-nation-explainer/#sidebar-methodology

[Seriously?]

Thanks Seriously!

That's the best explanation I have seen thus far....

Just because you have a "Blue" avatar and I have a "Red" avatar doesn't mean that there isn't at least a common area of legitimate discussion.

If your man loses.... don't blame me I voted for Bernie.

For me there's a difference between the academics of polling on this board and the partisanship that goes on in the 2016 elections board. A lot of folks can't grasp that there are differences between these two boards and think that if I am hackish when I support my candidate, that I can't be reasonable when it comes to numbers.

Numbers are numbers as far as I am concerned. The only real difference is your interpretation of them at the end of the day.

With that said, like You Gov four years ago, these Reuters samples will be hit or miss, but provide data points that may be otherwise lacking in smaller states. I just wish I didn't have to go to 538 to get the relevant sample sizes.

[Seriously?]

Here's an analysis of the Reuters States polls.

As you can see, the biggest seven states (CA, TX, FL, NY, IL, PA, OH) generally get 1 week surveys.

Thirteen states from 10 to 16 EVs (Georgia to Minnesota), with the exception of Maryland are on a two-week cycle.

Maryland and states generally from 5 to 9 EVs are on a three-week polling cycle.

When you get to about 6 EVs, the polling gets a little unreliable, where sample sizes will be in the 100s, hence the MOEs for these states increases significantly.

Most states with 3 or 4 EVs do not get polled, however, IA, ME and NH are exceptions. NM is the only 5 EV state without polling.



Sourcing for LV and length of poll is 538. Population totals are from the Census 2015 estimate. MOE calculated at 50% interval at 95% confidence based on number of persons surveyed. EVs and polling data are from the Reuters Data for 1, 2 or 3 week period ending 9/1.

[Seriously?]

Donnie is really using these garbage polls to promote himself?

Statistically speaking, the Ohio poll is not that terrible. It's 1 week poll with the MOE ~ +/- 5%

[Seriously?]

My take from all of this are that the polls in the smaller states <6 EV are unreliable and that's why we're seeing weird results in New Hampshire and Utah, for example.

The other polls are generally OK if you take trends into account.

[Seriously?]

9/8 numbers. Only NM seems off significantly as far as I can tell from a quick look.

State           EV   Clinton   Trump
Alabama   9   42%   51%
Alaska   3   —   —
Arizona   11   40%   46%
Arkansas   6   41%   50%
California   55   60%   32%
Colorado   9   40%   43%
Connecticut   7   46%   40%
Delaware   3   46%   24%
Florida   29   47%   47%
Georgia   16   38%   48%
Hawaii   4   —   —
Idaho   4   30%   56%
Illinois   20   52%   34%
Indiana   11   32%   56%
Iowa   6   41%   44%
Kansas   6   38%   52%
Kentucky   8   37%   53%
Louisiana   8   33%   60%
Maine   4   52%   33%
Maryland   10   54%   30%
Massachusetts   11   53%   30%
Michigan   16   43%   41%
Minnesota   10   42%   31%
Mississippi   6   35%   58%
Missouri   10   34%   53%
Montana   3   30%   60%
Nebraska   5   28%   56%
Nevada   6   43%   41%
New Hampshire   4   49%   36%
New Jersey   14   47%   35%
New Mexico   5   38%   48%
New York   29   52%   30%
North Carolina   15   46%   45%
North Dakota   3   —   —
Ohio   18   47%   46%
Oklahoma   7   33%   52%
Oregon   7   47%   35%
Pennsylvania   20   49%   44%
Rhode Island   4   —   —
South Carolina   9   45%   49%
South Dakota   3   —   —
Tennessee   11   28%   49%
Texas   38   31%   48%
The District of Columbia   3   —   —
Utah   6   29%   46%
Vermont   3   —   —
Virginia   13   50%   37%
Washington   12   45%   37%
West Virginia   5   39%   54%
Wisconsin   10   38%   40%
Wyoming   3   —   —

[Seriously?]

Colorado is a little bizarre too.  Virginia is extreme given how close the other swing states are.

I don't have sample sizes, but CO and VA have some heft to them, so on balance, there's some science there. NM is likely a junk sample.

[Seriously?]

"Only NM,"

-looks at WI and CO-

lol

From a MOE standpoint, WI and CO are possible if you are on the heavy Trump side of recent samples (assuming MOE ~ +/- 5%).

Wi has been consistent pro-Trump with Reuters for whatever reason.

[Seriously?]

Is it head-to-head numbers?

Yes. 2-way, LV. no 4-way numbers.

[Seriously?]

On FiveThirtyEight, an Ipsos poll has Trump leading here. Is this legit or no?

I wouldn't dismiss the possibility of Trump having a lead there or drawing things close in Colorado as nonsense. The Ipsos polls of the state actually have a statistically significant sample and a two-week frame of reference.

Given the state of the national race, it could be closer than some of the older data suggests.

I am not saying Trump is ahead in the state, but it could be closer than you think. There hasn't been a legit landline/cell poll of the state in a while.

[Seriously?]

Reading this correctly, it's Battleground Michigan for all the marbles.

[Seriously?]

Wow@that CA margin.

I'd say that's very very bad for Clinton and the Democrats. Trump can run up the score elsewhere.

[Seriously?]

Go to the turnout model and reduce everyone to 0 but Hispanics to 100. I can't do that right now but last time ir showed that there was not enough in the sample to measure for states like NM

Yes, and your point is?

The point is child, that in the past two weeks out of 4,899 respondents, only 162 respondents are Hispanic. That's 3%. That sample isn't accurate enough to give you an accurate national poll, never mind states ones.

Not if you reweigh those Hispanics to the equivalent of your suggested turnout model, which it appears that Reuters has done. So Reuters in essence takes the 162 Hispanics and extrapolates them to the real number which is likely about 400 or so.

Whether they did it on a micro level for each state is the only real question that you have when attempting to reskew these polls.

These numbers fundamentally are fine and in line with most state polls within the margins. Of the four companies doing these 50-state polls (Morning Consult, Google, Survey Monkey and Reuters), the way Reuters is doing it makes the most sense to give you the most up-to-date polls possible.

It's the states with 6 or fewer EVs where you start to run into issues methodologically.

[Seriously?]

These numbers fundamentally are fine and in line with most state polls within the margins. Of the four companies doing these 50-state polls (Morning Consult, Google, Survey Monkey and Reuters), the way Reuters is doing it makes the most sense to give you the most up-to-date polls possible.

How so? I thought Survey Monkey is the one that actually came closest to conducting 50 separate state polls (which is why it has the highest weights at 538)?

And for what it's worth, Survey Monkey is more strongly correlated with the Morning Consult MRP analysis (R=0.94) than with Reuters/Ipsos (R=0.88) or Google (R=0.79).

It's the staggered releases, which to me makes them more relevant. If the poll is big enough, Reuters releases them over a 1, 2 or 3 week cycle.

Google's samples are 1-week and generally too small in smaller states. Google just balances their samples based on age and gender. They don't even ask any other questions.

Survey Monkey and Morning Consult go over a longer period. (a month), so your snapshot is not as accurate as the race changes over the course of a month.

[Seriously?]

Google's samples are 1-week and generally too small in smaller states. Google just balances their samples based on age and gender. They don't even ask any other questions.

Google doesn't balance by racial demographics?  

Wow, no wonder their polls are so useless...

It's literally like a 3-4 question poll.. Likelihood to vote, Voter Intention, Gender and Age.

[Seriously?]

Latest update 9/16-9/25, 1 or 2 or 3-week polls, depending.

259-191 Clinton. Lots of close states though. OH and NC in the Clinton camp.



http://www.reuters.com/statesofthenation/

[Seriously?]

9/29 update. Some pre-debate and some post-debate numbers. Roughly 50/50.



Reuters Link: http://www.reuters.com/statesofthenation/

[Seriously?]

Through 10/6. 1, 2 and 3 week surveys depending on state.