Reuters/Ipsos: (Most) States

[Ozymandias]

This is pretty much exactly what I thought as well. It seems like these polls are actually very consistent, if you HEAVILY undersample the Latino vote. That would account for NM (the most Hispanic state) being where it is, as well as NV, CO, and even FL.

Would also explain Trump being up 7 in Arizona and 22 in Texas...

[Ozymandias]

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).

[Ozymandias]

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...

[Ozymandias]

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.

I can see the argument that 1-week cycles provide a valuable snapshot, but once you get up to two weeks or more, I think a poll loses any value as a "snapshot", which is why I think the much larger Survey Monkey samples are far more relevant for most states.

Also, Reuters/Ipsos only had 1-week windows for 8 states: IL, CA, TX, FL, VA, NY, OH, PA, two of which are clear outliers (CA: C+38, TX: T+23) relative to all the other recent polling.

So I'm VERY dubious that these Reuters/Ipsos polls are providing any useful info at all (though they're not as bad as the Google polls).

[Ozymandias]

Apologies if someone has already posted this message about Ipsos Polls from electoral-vote.com:

electoral-vote.com/evp2016/Pres/Maps/Sep21.html#item-11

some of the more interesting quotes...

"We Are Removing the Ipsos Polls from the Database"

"In nine of the 44 states Ipsos has polled, the Ipsos and non-Ipsos results differ by 10 or more points. That is way outside the margin of error (usually about 4%). In 21 states, the difference between Ipsos and non-Ipsos was 5 points or more."

"Interestingly enough, Ipsos is not biased. Averaged across all states, Ipsos favors the Democrats by a very small margin: Only 0.70%. That is fine. Except that it gets this small margin by having large errors cancel out."