1. A Republican edge in terms of self-described interest in the election. 2. Lower levels of self-described interest among younger and Latino

Transcription

1

2 2

3 Academics use political polling as a measure about the viability of survey research can it accurately predict the result of a national election? The answer continues to be yes. There is compelling evidence though, that modest shifts in the composition of the electorate, including a higher percentage of 18 to 29 year old voters, Latinos, and African Americans in some swing states produced a result which tipped a narrow Obama advantage to a more comfortable four point win. It is clear a traditional Likely Voter Model based only on self-described interest and selfdescribed likelihood to vote missed the scope of the turnout of year olds and Latinos in With what we know now, it is important to look again at the survey research completed in We need to understand what it was telling us about level of interest and likelihood to vote. We need to think through how what we learned impacts what we do in terms of developing and strengthening a Likely Voter Model for future election cycles. 3

4 1. A Republican edge in terms of self-described interest in the election. 2. Lower levels of self-described interest among younger and Latino respondents. 3. The ballot being tighter or favoring Romney among the most interested voters. 4. Romney carrying White voters by the largest margin since Obama carrying Latino voters by the highest margin since Lower overall election interest compared to the last two election cycles suggested a lower percentage of turnout of eligible voters, but an increase in the total votes cast because of the increase in the number of Citizens of Voting Age (CVA) Population. 4

5 Over one billion dollars in advertising was spent in a relatively small number of targeted states. Half of Americans in mid-october reported being contacted by one the campaigns. This is up sharply from the 30% who recalled being contacted when asked the same question during the 2004 election cycle. A stunning 70% plus of respondents in swing states in mid-october reported being contacted by one of the presidential campaigns with three weeks left to go in the election! Vote by mail and early voting has changed the equation significantly reducing the interest and involvement required to cast a vote. The extensive use of and social media and the direct appeal that follows from family and friends to vote may have had a differential impact on turnout among younger voters. The increase in Citizens of Voting Age (CVA) Population is largely because of the surge of young voters entering voting age and the growth of the Latino electorate. This creates the confounding problem that even with lower self-described interest total turnout can increase because of the larger number of younger and Latino voters. 5

6 Expectations of the cycle were based on national surveys conducted by multiple different organizations. Category Expectation Outcome Age Latinos Voters Party ID Voter Turnout Increasing the younger and Latino percent of the electorate would require total votes cast to increase. Younger and Latino voters were a larger percent of the electorate, but total votes cast dropped. 6

7 Step One: The CDC estimate of cellphone only households is the likely percentage of voters who are cellphone only in a presidential election. Surveys have to keep pace in terms of whom they contact. Step Two: There is a known base about turnout and that s the relatively narrow band of difference in terms of the percentage of Citizens of Voting Age Population (CVA) who vote. A turnout model has to be generous enough in terms of whom it includes to approximate this known percentage. Step Three: Self-described interest is a powerful predictor of the likelihood to vote, but it cannot be the only indicator used. Step Four: It is possible to help compensate in a turnout model to allow some target populations, like year olds and Latinos, to more easily qualify. 7

8 Step Five: Step Six: But additional weighting will probably still be required to help compensate for the continued likelihood of missing potential likely voters. In presidential years we should move as a default to gender being 47% men/53% women. Step Seven: We may not be asking the wrong questions, but we may be missing some additional right questions. Step Eight: We need large scale, living laboratories to further test and refine any Likely Voter Model. 8

9

10 10

11 There were seven telephone surveys incorporating live interviews with cellphone respondents released right before the election. The overall results across most of these polls generally mirrored the final results within the margin of error: Four surveys showed Obama winning by a range of one to four points. This was within the margin of error of the final election result. Two surveys had the race tied. Gallup, though, showed Romney ahead by one point among their pool of likely voters. 11

12 Surveys Released on Sunday and Monday Before the Election Final Result % White Total Romney Total Obama 12

13 13

14 Among the changes in politics has been the use of poll aggregation to help monitor campaigns. It is striking how close the poll average the day of the election in 2004 and 2008 matched the final result. Compellingly, this is not as clear in An Obama advantage of one point in the Real Clear Politics (RCP) average on election day became a 3.9 point margin in actual votes cast. This gap between the RCP average and the actual vote is also reflected in eight of eleven key states, where the difference between the last RCP average and the actual votes cast was three points or more. 14

15 Comparing the Final Results to the Real Clear Politics Average the Morning of the Election RCP Average On Election Day Final Results Republican Democrat Net Difference Republican Democrat Net Difference

16 Public polling showed movement in President Obama s direction over the final week of the campaign. State (# of Electoral Votes) Romney October 29 November 6 Obama Net Difference Romney Obama Net Difference Wisconsin (10) Nevada (6) Iowa (6) Ohio (18) New Hampshire (4) Colorado (9) Virginia (13) Florida (29) North Carolina (15) Michigan (16) Pennsylvania (20)

17 Even with this late momentum, presidential result by state compared to the RCP average for election day still shows significant movement for Obama. State (# of Electoral Votes) Romney RCP Average November 6 Obama Net Difference Romney Results Obama Net Difference Wisconsin (10) Nevada (6) Iowa (6) Ohio (18) New Hampshire (4) Colorado (9) Virginia (13) Florida (29) North Carolina (15) Michigan (16) Pennsylvania (20)

18 Averaging polls together increases their sample size making them much more powerful statistically than any one poll taken alone. But the errors in the polls are sometimes correlated, meaning there are years when most of them miss in the same direction. Mr. Romney remains close enough to Mr. Obama that he could fairly easily win the popular vote if there is such an error in Mr. Obama s favor this year. - Nate Silver November 6 th, 2012 Election Day Predictions Obama Chance of Winning Projected Obama Popular Vote Margin 18

19 These findings help reinforce our impression that the composition of the electorate in 2012, especially in the swing states, helped shift the election an additional two to three points toward President Obama. 19

20 20

21 The following slides illustrate Romney s startling margin among White respondents, with President Obama absolutely maximizing his vote among African Americans and Latinos. We built a model before the election which simulated the results based on each candidate s margin and the composition of the electorate s ethnicity. It was clear if the percentage of White voters dropped to 72% as the exit poll says it did that Romney could not win this election. The most critical part of this model was demonstrating that even if White voters were 74% of the electorate, as they were in 2008, that in some scenarios Romney still could have lost the election by 1.5 points. 21

22 Margins for Romney among White voters and margins for Obama among Latino voters were startlingly high. Presidential Ballot By Ethnicity Among Whites Net Republican Margin % % % Mid-October 2012* +23% Among African Americans Net Republican Margin % % % Mid-October 2012* -86% Among Hispanics/Latinos Net Republican Margin % % % Mid-October 2012* -45% *Data from 2000, 2004 and 2008 Exit Poll data from the presidential elections. Data from 2012 comes from NBC/WSJ October 2012 Survey. 22

23 Obama s Margin Among Non-Whites Romney s Margin Among Whites Final modeling before the election What turned out to be the 2012 Exit Poll Romney Best Possible Outcome IF... IF... Percent of White Vote Popular Vote Margin Popular Vote Margin 23

24 24

25 Likely Voter Models have been deployed for decades to better understand the gap between adults, registered voters, and likely voters. It is worth remembering even in the highest intensity elections, like the one held in 2008, almost four out of ten American adults did not vote. The models were created before the scope of the 2012 campaign could be fully understood and calibrated. It is possible a person who expressed lower interest or likelihood to vote and did not qualify in a Likely Voter Model did end up voting because of the sheer reach and frequency of contact by the presidential campaigns. 25

26 The following data may be an intriguing clue about how changing technology impacted turnout. Look at Obama s margin on /social media contact, especially among year old voters. Recall of Contact Net Difference All Voters Likely Voters Swing States Telephone/personal contact Mail /social media /social media among age

27 27

28 28

29 The CDC estimates of adults with a cellphone only (CPO) is a critical way to approximate the percentage of a presidential electorate who do not have a landline. % Cellphone Only CDC/NHIS Report National Exit Poll

30 Indeed, on the initial exit poll, the president carried cellphone only voters by 11 points, while losing narrowly among the balance of respondents. Exit Poll Ballot by Phone Type Cellphone Only (33%) Balance of Voters 30

31 The data is very clear: Every telephone poll has to account for what is being done to make sure they are interviewing a sufficient quantity of people who are cellphone only (CPO) respondents. Cellphone only adults are more likely to be younger, Latino, and African American. If you miss these voters, you are missing three core Democratic sub-groups, and the poll is too Republican. Percentages here matter. In the past election, if you only interviewed 20% CPO, you were almost a net two points more favorable to Romney. 31

32 32

33 Roughly 80% of adults interviewed in a survey say they are registered to vote. Other academic research has suggested that respondents willing to complete interviews are, in fact, already more likely to be civic participants and more likely to vote. Here is the important step: In a presidential year, if you allow roughly 80% of registered voters to qualify, that equates to roughly 64% turnout (you interview 100 adults, 80 are registered and 80% vote, that s 64 voters or 64% of adult turnout). This provides a known base on which to build a model in terms of what percentage of registered voters should be allowed to qualify to be included as likely voters. Here is the point though the 80% of registered voters you select for inclusion in the Likely Voter Model is the likely voter pool. When you start cutting down further based on most likely to vote based on self-described interest or other factors, you are narrowing the sample you are looking too low to really project to a large enough electorate. Repeating our example: If you interview 100 adults, 80 are registered, 80% are likely to vote, that s 64 people or 64% of adults. But, if you take these 64 adults and only look at the 75% who say they are very interested in the election, 75% of 64 adults is 49 people or 49% of adults and a turnout that is a good deal lower than reality in a presidential cycle. 33

34 34

35 Public Opinion Strategies has done its own studies on the relationship between self-described interest in an election and actual turnout. Some states, like California, have terrific voter files which contain information about who has voted. The following slide highlights a study conducted using data and turnout information from California after the 2010 election. There is no question the people who rated their interest the highest were the most likely to actually vote 71% of people who rated their interest as a 10 voted, compared to only 40% who rated their interest as a 1. 35

36 In a California post-election study in 2010, the people who rated their interest the highest were the most likely to actually vote. Self-described Election Interest (1-10) % Actually Voted * Findings based on 3,465 interviews conducted in California between Labor Day and Election Day Of these 3,465 interviews, 1,763 of the voters said they were very interested in the November 2010 elections as a 10 on a scale of one to ten, 349 rated their interest as a 9, and another 505 rater their interest as an 8. 36

37 Looking only at people who say they were a 10 or very interested, there is still a striking difference by age and ethnicity. Only 37% of registered voters age who were 10s in interest actually voted, compared to 89% of seniors who also rated their interest as a 10. Similarly, 77% of White 10s voted compared to the mid-50% range of other registered voters who actually voted who are non-white. This data helps illustrate while self-described interest is an important variable, it cannot be used as the only variable when predicting turnout. Even in an off-year election, 40% of those who rated their interest a 1 still voted. (Critical caveats: These interviews were done off of a California voter file, not a sample using random digit dialing. This means the list was already screened to include people who had voted in both the last two federal elections, or had voted in 2008, or who had registered to vote in the last year. In other words, even a 1 on this file had already demonstrated some proclivity to vote or had registered more recently.) 37

38 There was a striking difference though in turnout by age and ethnicity, even among those who rated their interest as a 10. Among Election Interest 10 % Actually Voted CA Post-Election Study White Hispanic/Latino African American Asian 38

39 39

40 Survey research has become more difficult as cooperation rates on telephone surveys have dropped. In addition, we also know there are some audiences more difficult to reach and involve in completing telephone survey research. These audiences include the youngest respondents, the poor, African Americans, Latinos, and younger single women. Once again this list is comprised of core Democratic sub-groups. Given these factors, it is appropriate and needed that any Likely Voter Model be modestly adjusted to allow a higher percentage of younger and Latino respondents interviewed to qualify as likely voters. (In the Obama era of 2008 and 2012, African American interest, intensity, and selfdescribed vote behavior were so high that no additional adjustment was required. Depending on the nominees in 2016, this may have to be reviewed, and also allowing African Americans to qualify more easily through a likely voter screen may be considered or needed.) 40

41 Here is another approach to the fast growing Latino vote that allowed us with some confidence to say their share of the electorate would continue to increase in In our final surveys, we had the Latino percentage of the electorate at either 9% or 10% of the vote. This was based on projecting their percent of the 2012 electorate based on the Latino percent of the electorate in the past three cycles. This projection technique based on the increasing Latino percentage of the electorate needs to be carried forward into future elections. 41

42 Projecting the Latino vote as a percentage of the electorate. Percentage of Latinos in the Adult Population % of Adults* 10.98% 12.79% 12.84% 14.53% % of Exit Poll 6% 8% 9% 10% Projected % of Latino Vote in 2012 Based on 2008 alone 10.05% Based on % Based on % *Data is from the Census Bureau data is from the 2005 ACS, as the 2004 ACS was unavailable. 42

43 43

44 Even with all of this caution and work, the odds are it will still be necessary to weight the data of cellphone only respondents, younger respondents, Latinos, and African American respondents. This is what makes polling difficult. You cannot prejudge a cycle based on past trends as sooner or later, trends can change. But as the following page makes clear, there are certain minimums over past cycles for each of these sub-groups as a percent of the electorate that is difficult to ignore. Again, for example, in our polling year olds had to be a minimum of 16% of the likely voter sample. Based not on just the past two cycles, but the past four cycles, it is clear the exit polls suggest an electorate that is younger and less White than most traditional telephone polls. Keeping these minimum weights in place for these key sub-groups is required to get closer to the final ballot result. 44

45 Interest in Election and Composition of the Electorate Exit Poll % of Electorate Mid-Oct. %9-10 Interest Exit Poll % of Electorate Mid-Oct. %9-10 Interest Exit Poll % of Electorate Mid-Oct. %9-10 Interest Exit Poll % of Electorate Overall 100% 82% 100% 87% 100% 76% 100% Age % 71% 17% 82% 18% 52% 19% Age % 77% 29% 84% 29% 74% 27% Age % 87% 38% 89% 37% 84% 38% Age % 89% 16% 88% 16% 88% 16% White 81% 81% 77% 86% 74% 77% 72% Af. Am. 10% 83% 11% 93% 13% 83% 13% Latino 6% n/a 8% 88% 9% 68% 10% 45

46 46

47 The data is not unanimous, but there is enough here to suggest in a presidential year to move to a 47% men/53% women as the gender quota or weight. Exit Poll Results '92-'12 Men Women % 53% % 52% % 52% % 54% % 53% % 53% 47

48 48

49 Self-described interest in the election and self-described likelihood to vote continue to have great value, but, they seem too constricted and not adequate. Here are some other thoughts about measures to consider: 1. Over the last two election cycles, NBC/WSJ has also been incorporating self-described previous voting behavior which has been a successful addition to the Likely Voter Model. 2. We have been experimenting with a few other questions to possibly use in a Likely Voter Model, including whether you are more or less enthusiastic about the election compared to past elections, and whether you believe the outcome will make a difference in your life. 3. Recall of voter contact, including social media and personal contact. Every voter program has demonstrated a direct ask from someone you know has consequence. We should be more carefully tracking people who say they have been contacted by a family member or friend asking them to vote for a candidate. 4. We went back and looked at the pool of NBC/WSJ non-likely voters from August forward and looked at three measures of intensity: very favorable for one candidate, very unfavorable toward the other; definitely voting for a candidate; and a strong partisan. Adding the non-likely voter respondents who qualified on all three measures of intensity would have added one percent to our pool of likely voters. This addition alone though would have moved the ballot 0.6% toward Obama. In general, we have to commit to a wider battery of questions on each survey as we work to determine which one(s) work best in more accurately predicting turnout. 49

50 50

51 Many states have very good voter files that allow you to know for sure who did and did not vote. Public Opinion Strategies will be merging and using its extensive files to compare our survey data so we can match back to a respondent to identify whether or not he or she actually voted. We will highlight those likely voters who did not vote and those non-likely voters who did vote. We will work to see what, if anything, better predicted people actually voting or not voting. Moving forward, we have designed a series of follow up interviews and other studies with targeted respondents to learn more about how best to refine a Likely Voter Model. Finally, we will be using the 2013 elections to further refine and calibrate a Likely Voter Model. 51

52 PUBLIC OPINION STRATEGIES 214 N. Fayette St. Alexandria, VA Bill McInturff, Partner