SD: Google Consumer Surveys - Trump +17

[cinyc]

Following amdcpus' lead, I conducted a one-question poll of South Dakota using Google Consumer Surveys.  First the results, then caveats about the methodology:

cinyc/Google Consumer Surveys South Dakota Poll
Trump 50%
Clinton 33%
Johnson 14%
Castle 3%

October 15-17; 341 (probably) Registered Voters.  This was weighted for the percentage of South Dakota registered voters by age and sex in 2012 according to the U.S. Census Bureau.

In Google's weighted results of Internet Users (i.e. weighted for the universe of Internet viewers, which skews younger than RVs), results were similar:
Trump 49%
Clinton 33%
Johnson 15%
Castle 3%

About methodology:
In this one-question survey, I asked "If the presidential election were held today, who would you vote for?"  The answers were only those candidates on the ballot - Donald Trump of the Republican Party, Hillary Clinton of the Democratic Party, Gary Johnson of the Libertarian Party and Darrell Castle of the Constitution Party, randomized in position.  I added an option for "I am not registered to vote in South Dakota" (always at the end) to try to filter out out-of-staters and those who were not registered.  I did not include an option for undecided or other because I couldn't follow up with the undecideds to push leaners.  The reported results are for those who didn't choose "Not registered".  506 total voters were surveyed.

The links above go to my weighting spreadsheet (thanks to amdcpus for his help), and the Google results.

[cinyc]

I think this is waaaay too friendly for the Trumpster but South Dakota is anyway safe R.

The South Dakota two-party Democratic (Obama) percentage was 40.8% in 2012.  Here, in my weighted numbers, it's 39.8%.  So, assuming a uniform swing (a big assumption that's likely not true), this poll says we're more or less on par with 2012 election results nationwide - or at least in South Dakota.  That result is pretty much is in line with the national two-way polling average right now, too.

[cinyc]

Should have led with the questions about registration and likelihood to vote...but nice going with the poll!

To add 2 more questions would have cost $150.

Is it $150 or $750 total?  I couldn't tell and didn't try.  In any event, that's out of my budget. 

GCS will let you do a poll with an actual screen so that you get only RVs or LVs - but it costs something like $3 per respondent and up.

[cinyc]

You can apparently submit this to Nate Silver and he'll include it in his model. He just added a Utah poll made by someone using Google Consumer Surveys. If you email polls@fivethirtyeight.com I think they'll add it.

I wouldn't want 538 to add this poll.  The one-question methodology is non-standard, as is me putting not registered as a choice and not including an undecided option.  I'm not sending it in.  If it ends up on 538, either someone else did or we know 538 is lurking here or at RRHelections.  They can get whatever info they want from the links I provided, and weight it however.

Anyway, I downloaded the raw data and tried to calculate the East River/West River divide.  Excluding unregistereds and without weighting the results,  East River is Trump +9, while West River is Trump +28.  240 respondents are East River.  53 are West River.  101 don't have specific geographic info.  Those people are Trump +23.  Small sample sizes, but the poll makes geographic sense.

You can see most other crosstabs at the link.  The poll probably has a gender issue - women were underpolled and actually prefer Trump slightly more than men.

[cinyc]

How much does commissioning the poll cost? Curious.

A one-question state poll costs 15 cents per respondent.  So a one-question 500-respondent poll costs $75.  I can tell when I'm not logged into Google that GCS is offering some people a coupon for $50 off your first poll if you click on my link to the Google weighting of my poll.  So, in theory, a new "pollster" can do a 500-respondent state poll for $25.

Adding questions adds cost.  amdcpus thinks it's linear (15 cents per additional question), but I think it's $1.50 for a 2-10 question state poll, no matter how many questions are asked.  Putting screening questions on the poll to get only RV/LV up front costs even more than that, but the pricing is poll specific.