NY Times polls of four southern states (AR, KY, LA, NC)

[Tender Branson]

The polls were done by Kaiser Family Foundation.

I know that they are usually polling healthcare-related stuff.

Have they ever polled political races before ?

If not, I would be somewhat sceptical of these results.

NC and KY are certainly in line with other polls, AR maybe a bit too high for Pryor (but he has definitely improved).

Cassidy seems to be too low in LA, but why didn't they poll the run-off (unlikely that Landrieu gets 50+ in the 1st round) ?

[Tender Branson]

2) There are questions whether this sample polled all adults, even those who aren't registered to vote, which could explain the fishy presidential vote sample.

They polled all adults in the states, but for the election questions they only polled RV.

The "presidential vote" question is of all adults, that's why it includes a high share of "I didn't vote/can't remember".

You need to remember that voter turnout in KY, AR and LA was very low, therefore more than 1/3 who say they didn't vote ...

[Tender Branson]

Simfan, take a look at my "MEGA election 2012 turnout chart":

https://uselectionatlas.org/FORUM/index.php?topic=181934.0

Or if you prefer, Prof. McDonald's blog of turnout data (he uses 2011 data for his calculations, whereas I already used the 2012 data from the Census Bureau):

http://elections.gmu.edu/Turnout_2012G.html

...

AR only had 48% turnout in 2012, relative to it's adult population (VAP).

KY had 53%, LA 57% and NC had 60% turnout.

Which means that if you have a sample of 1000 adults, where about half has not voted and 60% of those who voted, then you only get some 30% overall for Romney.

Of course, Obama's percentage is higher than it should be. Why is that ?

Maybe because Southern people are simply liers in these polls: Remember the polls in AR that showed Obama down by only 10-15 points ahead of the election ? Then, in the secrecy of the voting booth, those Southerners are returning to their redneck-voting-behaviour and voted for Romney by over 20 points ...

Or people simply don't remember and give false answers, or don't care. And these subsamples usually have high MoE too.

SurveyUSA often has very strange crosstabs too, but the overall topline is often very accurate.