As featured in the New York Times, the Civis Analytics Research Team conducted a survey of the Republican Primary of 757 self-identified Republican or Republican-leaning adults across the country. You can read more about our findings in Why Donald Trump Won’t Fold: Polls and People Speak by Michael Barbaro, Nate Cohn, and Jeremy W. Peters from Sunday August 23, 2015 and our methodology in There’s evidence that Trump’s polling support is overstated by Nate Cohn on August 26, 2015.
We wanted to share a bit more on the survey we conducted, the topline findings, and the methodology used to conduct this poll.
Question and Toplines
Self-identified Republicans or self-identified Independents that lean Republican were asked the following question:
I’m going to read a list of potential candidates for the 2016 Republican nomination. Please tell me which one you would like to see as the Republican presidential nominee.
[IF NOT SURE, ASK: Well, if you had to decide today, which one would you choose?]
|Registered Republican||Subsample General Population|
|Don’t know/No answer||17||16|
|None of them||5||5|
Civis Analytics conducted 3,007 live telephone interviews of adults in the United States contacted on landline telephones from August 10th to August 19th. Among respondents of this survey there were 757 self-identified Republican or lean Republican adults. These respondents were asked their candidate preference for the GOP primary.1
Sample for this survey was pulled from nationally representative voter and consumer files provided by TargetSmart Communications.
The results were weighted to two universes for purposes of comparison. The first method weights respondents to the Republican Primary question to be representative of registered or modeled republicans in the United States according to voter files provided publicly by Secretaries of State. This method can only be achieved when sampling from voter and consumer lists. The second method is closer to the way most public polls using random digit dialing weight their surveys. Using this method, the the entire sample (all 3,007 respondents) is weighted to be representative of the adult population and respondents who qualified for the Republican Primary question are simply analyzed as a subset of the broader adult population.
The margin of error for this survey is 4.2 percent based upon the standard-error for a sample of 757 respondents plus additional error introduced through weighting.
You can read more about the methodology and findings in Nate Cohn’s article in The Upshot.
The Upshot article highlighted a simple regression analysis conducted by Civis Analytics Data Scientists about the relationship between candidate support and past voting history. We’ve provided the original analysis and charts for reference below.
This first chart shows average vote share (% Trump, % Bush, etc.) by historical election participation at the individual level, as measured by the voter file. The dashed line in the graph indicates the average number of elections that adults in our respondent pool have voted in. The dotted line in the graph indicates the average number of elections that 2012 Republican Presidential Primary voters have voted in.
The following chart indicates the difference in vote share for each candidate between ‘average adult’ respondents and ‘average primary’ respondents with standard-error bars for reference.
Keep on the lookout for more posts from the Civis Analytics Research Team about interesting work we are doing and our approach to research methods!
This is in-line with public polling, where 26% of people identify as Republican (and 31.7% identify as Independent): http://elections.huffingtonpost.com/pollster/party-identification. ↩
Dotted lines indicate the average number of elections for a primary voter and an adult. ↩