Google Consumer Surveys Presidential Poll Fielded 8/18-8/19 Results, Crosstabs, and Technical Appendix 1 This document contains the full crosstab results for Red Oak Strategic's Google Consumer Surveys Presidential Poll, elded August 18th through 19th, 2016. Question 1 is a screening question, so we report the simple unweighted distribution of answers. For the following questions, all crosstabs are subsetted to respondents who are "somewhat" or "very" likely to vote, with weighting based on age, gender, and region so the sample better reects the demographic distributions of the United States population of Internet users. Following the crosstabs, a technical appendix contains information on weighting, margin of error and other aspects of the survey. Q1 [Screening Question]: "How likely are you to vote in the November US Presidential Election?" Very likely 67.5% Somewhat Likely 7.5% Not Likely 9.4% Not Registered 6.7% Other / Don't Know 8.8% Results for questions 2-7 reported from sample of respondents who are "Very Likely" or "Somewhat Likely" to vote, with weighting based on age, region, and gender Q2: "If the US Presidential election were held today, who would you vote for?" Results from previous survey (elded 8/4-8/9) displayed in middle column, with dierence from current poll results in right column. Hillary Clinton 37.4% 36.6% +0.8% Donald Trump 30.4% 29.6% +0.8% Gary Johnson 10.4% 9.8% +0.6% Other / Don't Know 21.8% 23.9% -2.1% Margin of error ± 3.3% N=900 1 Max Samels, Data Analyst
2016 preferences among voters who prefer a candidate Hillary Clinton 47.9% 48.2% - 0.3% Donald Trump 38.9% 38.9% +0% Gary Johnson 13.3% 12.9% +0.4% Margin of error ± 3.9% N=701 2016 preferences among voters who prefer a major party candidate Hillary Clinton 55.2% 55.4% -0.2% Donald Trump 44.8% 44.7% +0.1% Margin of error ± 4.1% N=620 Results for questions 3-7 reported from sample of respondents who are "Very Likely" or "Somewhat Likely" to vote, with weighting based on age, region, and gender. Q3: "Who did you vote for in the 2012 US Presidential election?" 2016 preferences by 2012 vote Barack Obama 44.0% Mitt Romney 34.3% Don't recall / Other / Didn't vote 21.7% Margin of error ± 3.4% N=900 2012 Vote Clinton Support Trump Support Johnson Support Other / Don't know Barack Obama 70.9% 5.4% 8.7% 15.0% Mitt Romney 6.2% 66.0% 9.3% 18.5% Other / Don't Know 19.1% 24.9% 15.4% 40.6% Q4: "Thinking about politics these days, how would you describe your own political viewpoint?" Conservative 28.5% Liberal 22.9% Moderate 31.5% Don't recall / Other 17.2% Margin of error ± 3.2% N=900
2016 preferences by ideological identication Ideology Clinton Support Trump Support Johnson Support Other / Don't know Liberal 78.1% 3.3% 6.7% 12.0% Conservative 5.9% 69.2% 7.4% 17.5% Moderate 46.5% 21.3% 13.2% 19.0% Other / Don't Know 19.3% 18.8% 15.0% 46.9% Q5: "Generally speaking, do you think of yourself as...?" 2016 preferences by party aliation Democrat 26.8% Republican 26.3% Independent 33.1% Don't recall / Other 13.8% Margin of error ± 3.2 % N=900 Party Identication Clinton Support Trump Support Johnson Support Other / Don't know Democrat 80.7% 6.0% 4.1% 9.3% Republican 5.7% 68.3% 8.1% 17.9% Independent 35.6% 24.2% 18.3% 21.9% Other / Don't Know 18.5% 20.4% 8.1% 53.1% Q6: "What is the highest level of education that you have completed?" 2016 preferences by education Did not graduate High School 1.8% High School Graduate 11.2% Some college or 2-year college degree 29.8% 4 years college degree / postgraduate degree 57.2% Margin of error ± 3.9% N=900 Level of Education Clinton Support Trump Support Johnson Support Other / Don't know Did not graduate 29.5% 39.8% 6.1% 14.6% High School High School Graduate 29.5% 44.9% 10.1% 15.5% Some college or 2-year 29.1% 33.3% 9.1% 28.5% college degree 4 years college degree 43.6% 25.8% 11.2% 19.4% / postgraduate degree
Q7: "What is your race or ethnic group?" 2016 preferences by race / ethnicity Black 4.2% White 73.5% Hispanic 6.2% Other / Don't wish to answer 16.2% Margin of error ± 3.0% N=900 Race Clinton Support Trump Support Johnson Support Other Don't know Black 77.4% 2.4% 8.6% 11.5% White 35.1% 35.7% 11.1% 18.1% Hispanic 48.9% 16.7% 12.1% 22.3% Other 33.4% 18.8% 6.9% 40.9% / Don't wish to answer
Technical Appendix Weighting and Coverage Error The survey is weighted over age, gender, and region in order to approximate the typical Internet using population of the United States. Google provided these demographic characteristics for respondents based on their IP address and online behavior, and Google also constructed the weightings based on these characteristics. Any respondents for whom gender, age or region could not be imputed were not included in our nal sample. This population without demographic data amounted to 224 out of 1500 respondents to our survey. This survey's main source of coverage error is that it excludes members of the population without access to the Internet or a smartphone. Statistical Error and Sample Size The margin of error is reported on a question-by-question basis, in line with suggestions from both the American Association for Public Opinion Researchers and the practices of established polling rms. 2 To determine the margin of error for each table, we take the maximum size of one half of the 95% condence interval for each cell in the table with a substantive response (i.e. not Don't know / Other). We then multiply this number by the design bias, which in this case is 1.02. 3 In other words, we believe that there is 95% chance that the true value of these estimates for the general population is within the margin of error of our reported number. Total surveyed population of likely voters with demographic data N = 900. However, since all of these tables subset the population in various ways to produce the estimates for each cell, we report sample sizes on a per table basis as well. For tables which break down 2016 preference by demographic characteristic and therefore use smaller samples to estimate cells, the error for those cells grows accordingly. Response Rate The response rate for this survey was approximately 8.1%. After running our survey, we were able to compute the percent of people presented with the survey who answer the rst question, and then the number of people who continued on to each subsequent question for the survey. We could then multiply these values to determine the overall response rate. 2 "AAPOR Guidance on Reporting Precision for Nonprobability Samples", American Association for Public Opinion Researchers, accessed 8/9/16 3 Design bias dened as 1 + ( SD(W eights) Mean(W eights) )2