Background on Wrong-Way Voting

Additional Findings and Patterns in Wrong-Way Voting

The Methodology Used to Measure the Number of Votes Affected by Wrong-Way Voting

Definition of Wrong-Way Voting

Wrong-way voting is when a person casts a vote that contradicts their beliefs; their vote is the opposite of their intention. On Prop 8, wrong-way voting occurred when someone who supported allowing gay and lesbian couples to marry voted Yes on 8, or when someone who wanted to ban gay and lesbian couples from marrying voted no.

Why We Can Measure Wrong-Way Voting on Prop 8

Wrong-way voting on Prop 8 is measurable thanks to a smart decision in questionnaire design by Lake Research. After asking the horse race question to find out how each voter was planning to vote (yes or no on 8), Lake asked in plain language whether each voter wanted to eliminate marriage for gay and lesbian couples, or to retain it.

The Lake polling revealed that both types of wrong-way voters existed: those whose mistake benefitted the No side, and those whose mistake benefitted the Yes.

Of course, until any voter actually casts a ballot, that voter is merely a potential wrong-way voter. But if their confusion persists through the time when they cast a ballot, they are no longer potential wrong-way voters. They become actual wrong-way voters, voting for a result that is the opposite of what they intend.

The Lake data is once again useful in determining whether and when potential wrong-way voters became actual wrong-way voters. This is because Lake also aimed to gauge when and whether each voter voted, starting when absentee voting began on October 6 for the November 4 election.

For a full discussion of the methodology used in this report to calculate the scale of wrong-way voting, see the section of this Appendix headed “Model Used to Estimate Wrong-way Voting” (below).

Magnitude of Wrong-Way Voting

Wrong-way voting was very high. An estimated 1,528,744 voters cast their ballot in a way that differed from the voter’s public policy intention. This equals 11.4% of the total number casting a ballot on Prop 8. In total, 876,987 voters who wanted to eliminate gay marriage cast no votes while 651,757 voters voted yes while trying to keep gay marriage.

Wrong-way voting reduced Yes on 8’s margin of victory. If wrong-way voting had been fully corrected during the Prop 8 campaign, the Yes margin of victory would have increased from 599,602 to 1,050,064. Yes on 8 would have received 53.9%, rather than 52.2%, of the vote.

The rate of wrong-way voting was even higher among early voters. An estimated 3,848,090 voters cast mail ballots before October 30. Of these voters, 615,427, or16.0%, cast their ballots the wrong way. Specifically, 310,988 early voters voted no despite trying to eliminate gay marriage by their vote. On the other hand, 304,439 early voters voted yes despite trying not to eliminate gay marriage by their vote. Each wrong-way vote is a net differential of two votes: it subtracts one from the voter’s intended side and adds one to the voter’s intended opposition. Thus, No benefited by approximately 13,099 net votes overall.

Among late voters, Yes on 8 lost hundreds of thousands of votes from voters who wanted to eliminate gay marriage. The rate of wrong-way voting decreased over the course of the campaign. But even as the rate of wrong-way voting was falling, more voters were voting, and the absolute number of wrong-way voters therefore increased. Among those who voted after October 30, 565,999 voters who wanted to eliminate gay marriage cast no votes. In late voting, 565,999 voters voted no while trying to eliminate gay marriage. On the other hand, 347,318 late voters voted yes while trying to keep gay marriage. Thus, No on 8 was the beneficiary of a further 437,363 net votes via wrong-way voting near and on Election Day.

Additional Findings Regarding Wrong-Way Voting

In addition to Finding 1 (near the beginning of this report), here are supplemental findings about the patterns of wrong-way voting that occurred in Prop 8.

Wrong-way voting diminishes as Election Day looms, and voters with a higher education vote the wrong-way at a much lower rate than others.

Some predictable conclusions emerge from the data. Voters voted the wrong way less often over time, and as the campaign went on and Prop 8 achieved higher and higher visibility, wrong-way voting fell among almost all groups. Voters with higher education levels voted the wrong way at a much lower rate. Among voters with postgraduate education, only 2% of pro–gay marriage (henceforth, "pro-gay") voters voted yes, and only 5% of anti–gay marriage (henceforth, "anti-gay") voters voted no—while 16% of anti-gay voters with a high school or less education voted no, and 10% of pro-gay voters with a high school or less education voted yes.

But there are other conclusions that emerge from the data that are less easy to foresee.

At every stage of the campaign, wrong-way voting was about 4% higher among voters against same-sex marriage than among those who favored same-sex marriage.

Overall, wrong-way voting was always about 4% higher among Yes voters. As of late September, among anti-gay voters, 22% of men and 19% of women were planning to vote the wrong way; among pro-gay voters, the numbers were 16% and 17%. By October, anti-gay voters had reduced their wrong-way voting to 12% for men and 14% for women, and pro-gay voters had reduced theirs to 11% and 8%. No on 8 maintained its advantage over time, although the particulars of the advantage changed substantially (see below).

Wrong-way voting tended to get corrected for voters on both sides belonging to a group that shared their views on same-sex marriage.

For example, pro-LGBT voters who are part of a demographic group that also factors the prop-LGBT position ‑ such as younger voters – are the most likely to correct their wrong-way most likely to correct their wrong-way voting before they vote. On the other hand, pro-LGBT voters who belong to a group which that was predominantly opposed to same-sex marriage – such as voters over age sixty-five – are the least likely to correct their wrong-way voting before they vote.

A voter who belonged to a pro-gay demographic and was pro-gay, tended to move toward voting no (the “correct” vote) over time.

Similarly, a voter who belonged to an anti-gay demographic and was anti-gay tended to move toward voting yes.

This scenario did not hold up, however, for voters who belonged to groups with whom they disagreed on the issue of same-sex marriage. These voters were less likely to learn the “correct” vote—in fact, in some cases, they appear to have unlearned the correct vote.

The clearest example looks at voters by age; age has substantial correlation with a voter’s support for same-sex marriage. According to Lake, voters under thirty supported same-sex marriage by a margin of 51 to 38 at the end of the campaign. Voters over sixty-five opposed same-sex marriage 55 to 29; other age groups fell in between. But the change in wrong-way voting between late September and late October is striking.

·         In the pro-gay marriage cohort (under thirty), voters who were pro–gay marriage reduced their wrong-way voting from 26% to 6%. Overwhelmingly, they learned that no was the “correct”" vote.

·         In the anti-gay marriage cohort (over sixty-five), voters who were anti–gay marriage reduced their wrong-way voting from 20% to 15%. They also learned that yes was the “correct” vote.

·         In the pro–gay marriage cohort (under thirty), voters who were anti–gay marriage—ie, that disagreed with most of their peers—increased their wrong-way voting from 6% to 16%.

·         Similarly, in the anti–gay marriage cohort (over sixty-five), voters who were pro–gay marriage increased their wrong-way voting from 21% to 28%.

·         This is the example that is most striking. But examples abound across the data. To take two:

·         Among the largest subgroups, by gender: men, who leaned toward Yes, anti-gay voters corrected their wrong-way voting (22% to 12%) more than pro-gay voters did (16% to 11%). Among women, the opposite was true: pro-gay voters reduced wrong-way voting from 17% to 8%, while anti-gay voters reduced theirs only from 19% to 14%.

·         Among voters intending to vote for McCain, anti-gay voters reduced their wrong-way voting from 16% to 7%, while pro-gay voters reduced it only from 24% to 23%. Voters intending to vote for Obama displayed the mirror image: pro-gay voters reduced theirs from 15% to 6%, while anti-gay voters reduced only from 26% to 22%.

There are two likely causes of this phenomenon: an affinity effect (people learn how to vote from their peers) and potential campaign effects. It’s hard to disentangle some sort of affinity effect from potential campaign effects (campaign communications are more salient to people who agree with them and are directed at potential supporters, and thus campaign communications tend to correct wrong-way voting amongst people who agree with the campaign). It is difficult to disentangle the two effects, and both likely operated in this election. But the data—very inconclusively—suggest the affinity effect was greater than the campaign effect. For example, Bay Area voters over fifty, who probably disproportionately received communication from the No campaign because of where they lived, nevertheless displayed wrong-way voting characteristics of a Yes group (anti-gay wrong-way voting fell over time; pro-gay wrong-way voting increased).

The principal exception to the above is that Latino pro-LGBT voters were less likely to be wrong-way voters.

Latino pro-gay voters wrong-way voted at a much lower rate than would be expected given other factors. The available data don’t explain why this is so. Latinos as a whole resembled whites when asked their public policy position on marriage by the end of the campaign; within both groups, the anti-marriage position led the pro-marriage position by 7 points. Latino anti-gay voters voted the wrong way at a slightly higher rate than white anti-gay voters (16% to 14%) by that time. But Latino pro-gay voters voted the wrong way at a much lower rate than whites (11% to 6%). Latinos similarly experienced a large drop throughout the campaign: anti-gay Latinos cut their wrong-way voting at about the average rate (22% to 16%), while pro-gay Latinos cut their wrong-way voting dramatically (from 28% to 6%).

There are several tempting explanations for this that are not borne out by the data. One possibility might be that pro-gay Latinos are overwhelmingly young. But pro-gay Latino voters over fifty cut their wrong-way voting from 22% to 0%, according to Lake, while anti-gay Latinos over fifty cut their wrong-way voting only from 27% to 21%. (Latinos under fifty showed a similar pattern.) It might be expected that Latino women account for the disparity, but pro-gay Latino men cut their wrong-way voting dramatically, from 30% to 6%. It might be hoped that Latino Democrats account for the disparity, but pro-gay Latinos who are Independents or Republicans cut their wrong-way voting from 48% to 6%!

It’s possible that highly educated Latinos account for the disparity. Lake’s published data do not allow a test of that supposition, although exposure to questionnaire-level data would allow it.

It’s also true that these subsets of Latino voters have very small sample sizes that make it difficult to draw conclusions. But the aggregate sample size of Latino voters is considerable, so it makes it harder to dismiss the conclusion about Latino voters as a whole. Perhaps the set of assumptions that do least violence to the data would be:

·         Latino voters as a whole display a higher “community effect,” and if all factors were equal would wrong-way vote at a lower rate because they would learn more from their peers about how to vote

·         Nevertheless, older Latino voters have lower education levels as a whole, and along with their language challenges this makes them more susceptible to wrong-way voting

·         In contrast, younger Latinos disproportionately are influenced by other pro-gay cohorts (eg, are more influenced by their peers of different races who are more pro-gay), and hence pro-gay young Latinos correct wrong-way voting more quickly.

Although these theories are suggested by, and not inconsistent with, the data, the data involved have very small sample sizes and do not definitively prove these conclusions. Regression analysis of Lake’s questionnaire-level data over time (including previous surveys, to increase the sample sizes) to isolate variables correlated with Latino identification could potentially permit closer examination of this finding.

The phenomenon of pro-gay Latinos having low wrong-way voting seems highly likely, however, even given that the data pertaining to subgroups is not fully conclusive.

Model Used to Estimate Wrong-Way Voting for This Report

Here, step-by-step, is the model used to estimate wrong-way voting so that all readers can independently examine the calculations and conclusions offered in this report.

1)    Determine the number of people who voted in the November 2008 election in California. The answer is 13,743,177 (see California Secretary of State).

2)    Estimate the percentage of people who had voted by October 30. The Lake Research Daily Tracking Polling projected as of October 30 that 28% of voters had voted.

3)    To estimate the number of early voters who voted the wrong way,

a)    Determine the set of tracking polls containing disjoint samples. Lake reported “three-day rolling tracks,” meaning that each daily tracking poll includes the results of three consecutive days of polling. The advantage of this approach is that it reduces the likelihood that a one-day anomaly will be mistaken by the campaign as significant. A series of Lake tracking poll results therefore counts each voter three times. For example, a voter polled on Monday will be included in tracking polls for Friday through Monday, Saturday through Tuesday, and Monday through Wednesday before being dropped from the survey. To aggregate the result from multiple tracking releases, we reduce the set of surveys to those that will not contain the same respondent more than once. This set includes the surveys ending: 10/9, 10/13, 10/16, 10/20, 10/23, 10/27, and 10/30.

b)    For each survey in the disjoint set, examine the top line for the Be Clear question and examine the crosstabs comparing the responses to the Be Clear question (which begins “Just to be clear”) with the responses to the horse race question (which begins “I realize you have already voted”).

c)    To estimate the behavior of early voters as a whole, average the results, on a percentage basis, of the seven surveys. To estimate the percentage of early voters who wanted a particular policy outcome, average the results of the seven top lines for the given answer to the Be Clear question. To estimate the voting behavior of the people who preferred a particular policy outcome, average the crosstabs for a given answer to the Be Clear question against the responses to the early-voting horse race question across the seven disjoint surveys.

To Be Clear: “Eliminate Gay Marriage”

Date

Horse Race

N (sample size)

Yes

No

Undecided

Refused

10/9

83

10

4

3

107

10/13

80

12

6

3

103

10/16

73

19

4

5

88

10/20

81

14

0

6

84

10/23

80

15

3

2

97

10/27

81

13

1

5

114

10/30

77

20

1

2

158

ESTIMATE

79.3

14.7

2.7

3.7

(N/A)


d)    To estimate the number of early voters who voted wrong-way No, multiply:

i)      the percentage of early voters who said they wanted to eliminate marriage;

ii)     the percentage of “be-clear-eliminate” voters who responded no to the horse race question, plus half the percentage of “be-clear-eliminate” voters who responded undecided to the horse race question;

iii)    the percentage of voters who voted early;

iv)   the number of ballots cast in the election.

e)    To estimate the number of early voters who voted wrong-way Yes, reverse the responses above.

4)    To estimate the number of late voters who voted the wrong way,

a)    Examine the last Lake tracking poll, which ended interviewing on October 30. Examine the top line for the Be Clear question and the crosstabs comparing the responses to the horse race question beginning “Proposition 8 on the November ballot.”

b)    To estimate the behavior of late voters as a whole, use the results of the last survey, which ended interviewing on October 30.

c)    To estimate the number of late voters who voted wrong-way No, multiply:

i)      the percentage of late voters who said they wanted to eliminate marriage;

ii)     the percentage of “be-clear-eliminate” voters who responded no to the horse race question, plus half the percentage of “be-clear-eliminate” voters who responded undecided to the horse race question;

iii)    the percentage of voters who voted late;

iv)   the number of ballots cast in the election.

v)    To estimate the number of late voters who voted wrong-way Yes, reverse the responses above.

Correcting for the Limitations of the Data, and Assumptions Made to Use the Data

Frequency of mail voting

It seems likely there is a deficiency in the Lake Research early voting model. The percentage of voters who have already voted in Lake’s sample is already 22% on October 7, which was the second day of early voting. Following that, the percent of Lake’s electorate that reports casting a ballot declines, eg, on October 9, it drops to 17%. According to Lake Research, this is because it changed the way it was asking whether someone had already voted on October 7. After that, the proportion of “already voted” voters essentially stays the same. It does not increase above 17% until October 25, varying between 13% and 17% for that period.

It is highly unlikely that no one voted between October 9 and October 25, although there is no hard evidence on this question.

We nevertheless use Lake's estimate of the number of early voters who had voted by October 30 because it is the best data available. We derive our estimates for the number of early votes by multiplying that number by the number of total ballots cast in the November election.

Measuring wrong-way early voting

We estimate the number of people who voted the wrong way by averaging the percentages across each disjoint sample available during the period in which Lake was tracking.

We could use the later numbers, as they should include early voters from the entire early voting period. But this would lead to problems: respondents might correct their wrong-way voting after the fact. Or they might move on the marriage question, though this is less likely.

However, one of Lake’s methodological choices makes this an easy decision—even with early voters the interviewer asks how a voter would vote if the election were held today. Wrong-way voters following the interviewer’s instructions should correct their wrong-way vote if they have since learned the meanings of Yes and No.

One could do other statistical manipulations to try to adjust for the time series available, but Lake’s data (implying that there was no early voting at all between October 9 and October 25) makes a blunter approach more realistic. Averaging the percentages is the approach chosen here.

Mail voting post-October 30

The step-by-step model described above assumes that a substantial percentage of voters voted by mail between October 30 and the end of the campaign. We further assume that these voters behaved the same as precinct voters (Election Day voters), and that both groups behaved the way that poll respondents who had not yet voted on October 30 were intending to behave.

Behavior of voters undecided on vote intention

A substantial number of voters indicated that they were undecided on the horse race question but were able to give an answer on the Be Clear question. The assumption used in the model above is that half of those voters voted the wrong way—that they voted essentially randomly.

Binder data is consistent with the findings above

Although the model described above analyzes the No on 8 polling conducted by Lake Research, it should be noted that the only other known data that bears on this question—the September 2–4 poll by David Binder Associates— is consistent with this data. Binder asked the standard horse race question, and then later asked a question designed to ascertain persuadability surrounding marriage. Voters were asked: “Generally speaking, now that marriage for gay and lesbian couples is the law in California, what best describes your opinion?”

·         It is acceptable to you (“acceptable”)

·         You do not like it, but it is acceptable (“dislike”)

·         You do not like it, and it is unacceptable (“unacceptable”)

13% of the “acceptable” voters were planning to vote yes, while 19% of the “unacceptable” voters were planning to vote no. This is very similar to the wrong-way voting estimate derived from the Lake data from the nearby period using the Be Clear question: in Lake’s September 22–24 poll, 19% of voters who said their vote was to eliminate gay marriage were voting no, while 16% of voters who said their vote was to not eliminate gay marriage were intending to vote yes.

Additional corroboration of wrong-way voting

Two independent analyses using data other than the Lake Research polling data have reached conclusions consistent with the findings in this report on wrong-way voting.

Prof. Gregory B. Lewis and Dr. Charles W. Gossett wrote a paper “Why Did Californians Pass Proposition 8?” that analyzed data from other sources, primarily the Field Research Corp. (which publishes the Field Poll) and the Public Policy Institute of California (PPIC). From the abstract:

Using data from 24 polls of Californians since 1985, we consider four hypotheses to explain why the polls got it wrong: (1) many respondents misled pollsters, perhaps worried that they would appear to be bigots if they expressed their real beliefs; (2) more effective efforts by the Yes on 8 campaign lowered support for same-sex marriage; (3) a principled opposition by some same-sex marriage opponents to writing discrimination into the constitution declined over the year; and (4) survey respondents misunderstood Proposition 8 and changed their positions as they became more aware of its meaning.

We find the most support for the fourth hypothesis and conclude that Proposition 8 passed because most Californians oppose same-sex marriage.


In examining the data, they found that:

·         “Though most respondents’ positions on Proposition 8 fit with their beliefs about same-sex marriage, the number of ‘errors’ is striking: across the three polls that asked about both, 8 to 21% of same-sex marriage supporters said they would vote for Proposition 8, and 11 to 25% of opponents [of same-sex marriage] planned to vote against it.” [from the second version of the paper]

·         “Proposition 8 opponents did a little better than they should have and that the 48-52 loss overstates current support for same-sex marriage” and “the explanation appears to be confusion [wrong-way voting] rather than a principled stand against ‘writing discrimination into the constitution.’” [from the first version of the paper]

Based on the data they examined, Lewis and Gossett explicitly rejected the alternative explanation that some same-sex marriage opponents had a principled opposition to writing discrimination into the constitution. See page 24 of their paper for their dissection of this possibility, available for download online at http://papers.ssrn.com/Sol3/papers.cfm?abstract_id=1451709. This finding is supported as well by the Lake data, which shows that wrong-way voting declined substantially over time and fell over time at a similar rate for both sides. This is more consistent with the steady but incomplete resolution of voter confusion than it is with any attempt (or success) by No on 8 to make a “protect the constitution” argument.

Additionally, Nate Silver, writing in his blog fivethirtyeight.com, relied upon election results from the previous votes cast in thirty-one states on marriage issue ballot measures. He noted that No on 8 received more votes than his model predicted. He did not explore at length the explanation for his finding, except to speculate in passing that perhaps the No on 8 campaign had not gotten all the credit it may have deserved for keeping the margin closer than he would have expected. But his finding would be equally consistent with wrong-way voting benefitting No on 8 having the effect of narrowing the margin.

Č
Ċ
David Caldwell,
Aug 3, 2010, 12:39 PM