2014 Boyle County Exit Poll: How accurate are the results?

Here is a comparison of the 2014 BCEP results for various races and the actual results from the Kentucky Secretary of State‘s website:

Race Actual results Survey results Difference Survey results with sample weighting Difference
McConnell 56.5% 50.2% 6.3% 51.8% 4.7%
Grimes 40.5% 47.3% 6.8% 46.1% 5.6%
Patterson 3.0% 2.4% 0.6% 2.1% 0.9%
Guthrie 62.5% 58.4% 4.1% 59.2% 3.3%
Leach 37.5% 41.4% 3.9% 40.6% 3.1%
McKinney 55.0% 59.9% 4.9% 61.4% 6.4%
Harmon 45.0% 40.1% 4.9% 38.6% 6.4%
Mike Perros 51.0% 43.8% 7.2% 45.4% 5.6%
Paige Stevens 49.0% 56.2% 7.2% 54.6% 5.6%
Steve Becker 20.3% 18.8% 1.5% 19.6% 0.7%
Lowery Anderson 15.9% 17.7% 1.8% 18.4% 2.5%
Susan Matherly 25.7% 24.4% 1.3% 23.8% 1.9%
Paige Matthews 23.6% 24.7% 1.1% 23.9% 0.3%
Elaine Wilson-Reddy 14.6% 14.4% 0.2% 14.3% 0.3%
Rick Serres 17.0% 17.7% 0.7% 17.7% 0.7%
Denise Terry 17.1% 15.9% 1.2% 15.5% 1.6%
Kevin Caudill 20.1% 20.4% 0.3% 20.7% 0.6%
Kent Mann 13.2% 12.3% 1.0% 12.3% 1.0%
J. H. Atkins 19.8% 20.4% 0.6% 19.7% 0.1%
Buck Graham 12.3% 13.4% 1.1% 14.2% 1.9%
AVERAGE DIFFERENCE 2.8%   2.7%

A reminder that a sample weighting was applied to figures reported on this blog, which is a standard procedure used to correct for differences in how different demographic groups respond to the initial invitation to take the survey.

In general, it seems that the exit poll sample tended to overstate support for Democratic candidates in the partisan races as well as the implicitly liberal-leaning candidate in the non-partisan Danville mayoral election by anywhere from 3%-6%.[1] When it comes to the Danville city commission and school board, however, the exit poll sample was only about 1% off from the actual final results. The average difference between the exit poll sample and the actual sample for all races is 2.8% which is reduced slightly to 2.7% once a sample weighting procedure is applied.

Given that the results vary by about an average of 2.5%-3%, we can confidently assume that the other responses from the survey questions are likely somewhere in the same ballpark. This range is well within the standard accepted margin of error of 3% for most national surveys and professional polling firms.

[FN1] There are a number of explanations that can account for this, including the possibility that people were slightly less likely to accurately report their voting patterns for these higher-profile offices on the survey form. However, it is ultimately impossible to definitely verify this one way or the other.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s