by Theodore de Macedo Soares.
Republican Party 2016 primaries. This table, Republican Party Table 2, shows the discrepancies between exit poll results and the unverified computer vote counts applying total survey error (TSE) calculations. Discrepancies that exceed the total survey error of the exit polls are shown in the last column. Compared to Table 1, applying TSE instead of the standard MOE results in decreasing the discrepancies by which the states exceed error calculations. Additionally, the discrepancy in Ohio, exceeding the MOE in Table 1, does not, in this table, exceed the TSE applied.
Surveys of all kinds measure a small sample of the population at interest. Survey related errors can accumulate in all phases of a survey: Coverage error (not able to sample selection of the population), nonresponse error (not able to poll all persons in the chosen sample), sampling error (when the survey sample differs from the population being measured), and measurement error (from inaccurate responses).* When the true values of the population are known, the total survey error can be easily calculated.
The approach taken in the following tables takes into account all the possible sources of error inherent in exit polls. It does this by comparing 106 exit poll results for the main candidates in the Republican Party primaries (Carson, Bush, Kasich, Rubio, Cruz, and Trump) with their actual vote totals. In this way the actual total survey error (TSE) can be measured. With the exception of the primaries in two states, West Virginia and Texas, with very large discrepancies (18% and 10.6% respectively) disfavoring Trump, the exit polls for Republican Party primaries have been fairly accurate.
By increasing the standard statistical margin of error (MOE) by 32%, and excluding the outliers in West Virginia and Texas, the percentage of exit polls exceeding this increased MOE goes down to 4.76% which is just under the expected 5% in an exit poll with a confidence interval of 95%. The tables below take into account the total survey error by increasing the MOE by 32%.
Coincidentally, the total survey error calculated for the 2016 primaries that increases the standard MOE by 32% is very close to the 30% increased MOE applied by other analysts to these and other exit polls. The significant difference is that whereas their applied margin of error only addresses the statistical sampling error associated with the size and configuration of the sample, the almost identical TSE applied here takes into account all the possible sources of error common in surveys such as the exit polls being examined.
*Groves, R. (1989). Survey errors and survey costs. New York. Wiley.
[1] Exit polls (EP) conducted by Edison Research and published by CNN shortly after the closing of polls for the state and downloaded by TdMS. Copies are available. Exit poll results are derived from the gender category–the proportion of men and women voting for each candidate. As these first published exit polls were altered/adjusted to conform to the unverified computer vote counts, the discrepancies noted above were adjusted to near zero in the final EPs.
[2] New York Times reported 100% vote count (99% for NY, CT, MD, PA Primaries 6/12/2016). http://www.nytimes.com/elections/results
[3] This table compares candidate Trump’s results with the combined results of the competing candidates.
[4] The margin of error calculated in this column represents the total survey error (TSE) derived by comparing 106 exit poll results for the various candidates in the 2016 Republican Party primaries to their vote count. As the completed exit polls are compared to the vote totals the TSE measures all errors associated with an exit poll–sampling, coverage, nonresponse, measurement, and post-survey errors–and not just the sampling error associated with the standard margin of error. Calculations by TdMS. The found TSE results in increasing the standard margin of error (MOE) by 32%. MOE calculated according to Franklin, C. The ‘Margin of Error’ for Differences in Polls. University of Wisconsin, Madison, Wisconsin. October 2002, revised February 2007. Available at: https://abcnews.go.com/images/PollingUnit/MOEFranklin.pdf ).
[5] In contrast to the Democratic Party primaries results, where 22 of 25 EP/VC discrepancies favored Clinton the discrepancies in the Republican Party primaries were more balanced–for and against Trump–which is the expected result of an impartial vote count and random errors in the exit poll.
[6] Two states, West Virginia and Texas exhibited very large discrepancies, far exceeding the TSE of the exit polls. In West Virginia Trump’s vote count at 9% less than predicted by the exit poll plus a 5.3% increase in the vote totals of minor candidates plus a 2.7% increase in the vote totals for Kasich compared to the exit poll predictions formed the bulk of the large 18% discrepancy. In Texas, Cruz’s home state, Trump’s vote count at 5.8% less than predicted by the exit polls plus Cruz’s 4.7% increase in the vote count than predicted by the exit polls comprised the bulk of the discrepancies. These outliers were not included in the TSE calculations noted above.
* According to personal communication on May 23, 2016 from Mr. Lenski, Executive Vice President of Edison Research, as the National Election Pool (NEP) consisting of ABC, CBS, FOX, NBC, and the Associated Press did not provide funding to conduct exit polls for the remaining primaries, they did not occur. The remaining primaries for which no exit polls were conducted:
May 24 – Republican Party primary: Washington. June 5 – Democratic Party primary: Puerto Rico
June 7 – Democratic and Republican Party primaries: California, New Jersey, New Mexico, South Dakota, and Montana
June 14 – Democratic Party primary: District of Columbia
Revisions:
July 13, 2016: Revised to include an introduction to the methods applied in calculation survey errors.
July 17, 2016: Minor edits to table and notes.
July 24, 2016: Table simplified by the removal of the column showing number of exit poll respondents for each state
[…] https://tdmsresearch.com/2016/06/20/republican-party-table-2016-primaries/ (Table and notes will open in new Tab) […]
[…] of the primaries in two states, West Virginia and Texas, with very large discrepancies (18% and 10.6% respectively) going against Trump, the exit polls for Republican Party primaries have been fairly […]
[…] The explanatory notes for this table can be seen at: https://tdmsresearch.com/2016/06/20/republican-party-table-2016-primaries/ […]
[…] the 2016 Republican Party primaries, however, the computerized vote counts matched the exit polls within the margin of error for all […]