Methodology Used to Calculate Voter Movement

For the purposes of this report, “movement” among voters -- that is, the change of aggregate opinion about same-sex marriage as a result of the Prop 8 campaign -- is calculated by comparing the results of the “Be Clear” question of Lake's September 22-24 poll, just before the two sides started airing television ads, and the final Lake tracking poll administered October 28-30. The Be Clear question asks: "Just to be clear, is your vote to eliminate marriage for gay and lesbian couples in the state of California or NOT to eliminate marriage for gay or lesbian couples in the state of California?" This question filters out the self-correction of wrong-way voting, ie the movement that occurs when a voter who is on one side or the other learns the meaning of Yes and No and changes his or her position on the horse-race question to comport with his or her pre-existing policy view. Movement on the Be Clear question isolates shifts in opinion on the public policy of allowing same-sex marriage.

 Using Lake’s data has two important strengths. First, Lake alone covers the period containing the vast majority of the paid TV advertising campaign, from September 24 until just a few days before Election Day. Second, comparing Lake data to other Lake data has one crucial strength. Any errors Lake might have made—eg determining the composition of the likely voter electorate, or in question wording—would tend to cancel one another out as we examine Lake’s estimated electorate over time. Using Lake data across the board allows us to isolate that movement in the face of other potential imperfections in the data.

The chart below presents both the raw polling data and change in margin for each demographic subgroup available from the Lake crosstabs. The columns on the right side of the chart labeled “Raw Polling Data” show the percentage of voters polled who want to eliminate or keep same-sex marriage for October 30 and September 22. Sample size of each group is in the right-most column. All groups with a sample size under 100 have a gray fill to warn that they provide less certain data.

The columns on the left side chart the net movement over this time period. The “margin favoring elim” for both dates is calculated by subtracting “%who want to keep” from “percent who want to elim”. The difference is the margin favoring eliminate. The key findings of the chart are in the center grayed column, “Change in Margin Favoring Elim.” This is calculated by finding how much “Eliminates” lead grew by subtracting the September 22 margin from the October 30 margin, ie column 1 minus column 2.

Below the chart is an explanation of how use change in margin to calculate approximate votes as well as net gains and losses for several demographic groups.

 


 

 

 

Just to be clear, is your vote to eliminate marriage for gay and lesbian
couples in the state of California or NOT to eliminate marriage for gay or lesbian couples in the state of California?

 

 

 

 

Changes in the Margin - Net Gains & Losses

Raw Polling Data

 

 

 

 

30-Oct

22-Sep

October minus September

30-Oct

30-Oct

30-Oct

22-Sep

22-Sep

22-Sep

 

** gray fill indicates a sample size less than 100

Margin Favoring Elim

Margin Favoring Elim

Change in Margin Favoring Elim

Points Gained by Elim

Points Gained by Keep

% Who Want to Elim

%Who Want to Keep

Sample Size

% Who Want to Elim

%Who Want to Keep

Sample Size

 

Total

13

3

10

6

-4

52

39

975

46

43

800

 

Men

14

9

5

4

-1

52

38

465

48

39

384

 

Women

12

-2

14

7

-7

51

39

510

44

46

416

Age Groups

Under 30

-14

-8

-6

0

6

40

54

148

40

48

112

30 - 39

14

-15

29

17

-12

53

39

138

36

51

120

40 - 49

14

-7

21

13

-8

54

40

192

41

48

144

50 - 64

13

8

5

2

-3

51

38

264

49

41

224

65 & over

28

22

6

2

-4

58

30

204

56

34

184

Under 35

-6

-9

3

3

0

43

49

199

40

49

168

35 - 49

13

-10

23

15

-8

54

41

279

39

49

208

Under 50

6

-10

16

11

-5

50

44

478

39

49

376

50 & over

20

14

6

2

-4

54

34

468

52

38

408

Age & Gender

Men <50

11

1

10

9

-1

53

42

243

44

43

193

Women <50

0

-22

22

12

-10

46

46

235

34

56

183

50 & over Men

15

17

-2

-2

0

51

36

212

53

36

187

 

 

 

 

 

 

 

 

 

 

 

 

 

 

50 & over Women

24

13

11

5

-6

57

33

257

52

39

221

Under 35 Men

-3

9

-12

-2

10

46

49

95

48

39

85

Under 35 Women

-8

-28

20

9

-11

41

49

104

32

60

84

35 - 49 Men

20

-5

25

16

-9

57

37

148

41

46

108

35 - 49 Women

6

-16

22

14

-8

50

44

131

36

52

99

Education Level

H.S. or Less

29

12

17

9

-8

60

31

226

51

39

152

Post H.S.

24

16

8

5

-3

57

33

297

52

36

248

Non-college grad

26

14

12

7

-5

58

32

523

51

37

400

College graduate +

-4

-9

5

4

-1

44

48

436

40

49

383

4-year College grad

2

-12

14

8

-6

47

45

316

39

51

240

Post-graduate

-16

-5

-11

-3

8

38

54

120

41

46

142

Non college men

25

26

-1

1

2

57

32

247

56

30

187

Non college women

28

5

23

11

-12

59

31

276

48

43

213

College men

1

-8

9

6

-3

47

46

210

41

49

188

College women

-7

-9

2

2

0

42

49

226

40

49

195

Party Affiliation

Strong Democrat

-13

-27

14

7

-7

40

53

341

33

60

282

Weak Democrat

-15

-37

22

13

-9

37

52

111

24

61

94

Weak Republican

39

35

4

6

2

66

27

88

60

25

82

Strong Republican

52

56

-4

-3

1

71

19

222

74

18

198

Democrat

-14

-29

15

8

-7

39

53

452

31

60

376

Indep.

20

-6

26

16

-10

55

35

180

39

45

120

Indep and Leans

12

-7

19

12

-7

51

39

270

39

46

208

Republican

49

50

-1

0

1

70

21

310

70

20

280

Party & Gender

Democratic men

-10

-24

14

7

-7

40

50

208

33

57

169

Democratic women

-17

-34

17

9

-8

38

55

244

29

63

207

Indep. men

16

4

12

11

-1

54

38

91

43

39

60

Indep. women

23

-17

40

21

-19

56

33

89

35

52

60

Indep. men w leans

10

3

7

8

1

51

41

132

43

40

116

Indep. women w leans

16

-19

35

18

-17

52

36

138

34

53

93

Republican men

47

50

-3

0

3

70

23

149

70

20

141

Republican women

50

49

1

0

-1

69

19

161

69

20

139

Ethnicity

White

15

-2

17

9

-8

53

38

634

44

46

519

African American **

15

11

4

3

-1

51

36

59

48

37

64

Latino

21

11

10

7

-3

56

35

165

49

38

136

Asian **

4

7

-3

-1

2

48

44

49

49

42

53

White men

18

1

17

10

-7

55

37

304

45

44

240

White women

12

-4

16

8

-8

51

39

331

43

47

279

Latino men

24

25

-1

0

1

55

31

73

55

30

79

Latino women

18

-8

26

15

-11

56

38

92

41

49

57

Ethnicity &…

White <50

5

-15

20

12

-8

49

44

278

37

52

211

White 50+

23

7

16

8

-8

57

34

343

49

42

298

Latino <50

18

-3

21

14

-7

56

38

106

42

45

91

Latino 50+

23

38

-15

-9

6

54

31

54

63

25

43

White Democrat

-24

-48

24

11

-13

34

58

241

23

71

211

White Indep.

14

-18

32

19

-13

52

38

116

33

51

68

White Republican

51

46

5

4

-1

71

20

261

67

21

226

African American Dems **

12

11

1

4

3

52

40

44

48

37

50

Latino Democrat

10

-5

15

9

-6

50

40

120

41

46

83

Latino Ind. and Repub

49

38

11

6

-5

71

22

40

65

27

49

Marital Status

Married

20

10

10

5

-5

55

35

586

50

40

480

Unmarried

-3

-8

5

4

-1

44

47

360

40

48

311

Single

-16

-20

4

3

-1

38

54

201

35

55

133

Separated/Divorced

41

-18

59

31

-28

65

24

70

34

52

66

Widowed

26

23

3

2

-1

58

32

58

56

33

79

Married men

22

15

7

4

-3

56

34

285

52

37

252

Married women

18

5

13

7

-6

55

37

300

48

43

228

Unmarried men

-1

-3

2

4

2

46

47

169

42

45

128

Unmarried women

-4

-13

9

5

-4

43

47

191

38

51

183

Parental Status

(Child under 18) Yes

24

2

22

13

-9

59

35

295

46

44

252

(Child under 18) No

6

2

4

3

-1

48

42

657

45

43

535

Father

34

14

20

13

-7

65

31

137

52

38

125

Mother

15

-11

26

13

-13

53

38

159

40

51

127

Dem/Ind Mothers < 45

0

-42

42

24

-18

47

47

~100

23

65

~ 100

Dem/Ind Fathers < 45

29

-4

33

16

-17

62

33

~100

46

50

~ 100

Childless man

5

5

0

1

1

47

42

319

46

41

251

Childless woman

9

0

9

5

-4

50

41

338

45

45

284

Location

LA County

10

5

5

1

-4

48

38

254

47

42

206

Orange

37

16

21

13

-8

64

27

89

51

35

75

Inland Empire

20

22

-2

-3

-1

56

36

89

59

37

73

San Diego

10

23

-13

-5

8

51

41

85

56

33

71

Sacramento

8

15

-7

0

7

49

41

108

49

34

87

Bay Area

-2

-33

31

16

-15

45

47

213

29

62

173

                                        Media Market (MM) &…  

MM Los Angeles

18

10

8

3

-5

53

35

457

50

40

373

MM San Francisco - Oakland - San Jose

-5

-31

26

13

-13

43

48

198

30

61

166

MM Sacramento - Stockton - Modesto

8

14

-6

1

7

49

41

108

48

34

88

MM San Diego

11

20

-9

-4

5

51

40

82

55

35

67

MM Other Central

36

27

9

9

0

66

30

82

57

30

62

MM Los Angeles Men

24

15

9

5

-4

57

33

220

52

37

181

MM Los Angeles Women

12

4

8

3

-5

50

38

237

47

43

193

MM San Francisco - Oakland - San Jose Men

-13

-32

19

14

-5

41

54

92

27

59

77

MM San Francisco - Oakland - San Jose Women

3

-30

33

13

-20

46

43

106

33

63

90

MM Los Angeles under 35

0

-9

9

7

-2

46

46

97

39

48

77

MM Los Angeles Under 50

17

-4

21

13

-8

55

38

227

42

46

163

MM Los Angeles 50 & over

20

21

-1

-3

-2

53

33

216

56

35

204

MM San Francisco - Oakland - San Jose Under 50

-22

-44

22

14

-8

37

59

99

23

67

77

MM San Francisco - Oakland - San Jose 50 & Over

13

-20

33

16

-17

52

39

92

36

56

88

MM Los Angeles White

17

5

12

4

-8

53

36

270

49

44

226

MM San Francisco - Oakland - San Jose White

-3

-36

33

17

-16

45

48

130

28

64

111

MM Sacramento - Stockton - Modesto White

16

16

0

4

4

53

37

76

49

33

68

Pres Vote Intention

(Plans to vote) McCain

66

55

11

8

-3

80

14

313

72

17

277

(Plans to vote) Obama

-15

-34

19

9

-10

38

53

538

29

63

387

(Plans to vote) Undecided

-4

7

-11

-7

4

37

41

71

44

37

91


A Detailed Look at theMethodology Used to Calculate Voter Movement

Basic Calculations

To calculate the net swing among each voter subgroup listed in the chart above, we compared the net change in margin between the September 22 and October 30 tracking polls. “Margin” means the difference between the percentage supporting a ban on same-sex marriage and those opposing a ban on same-sex marriage.

Example: Consider a voter subgroup that favored the marriage ban by 45-35 in the September poll and then by 60-30 in the October poll.

·         The subgroup experienced a 20-percentage-point swing. The swing is the difference between the October margin (60%-30% = 30%) and the September margin (45%-35% = 10%).

·         The estimated percentage of voters in the subgroup who changed their minds would be half of the swing, in this case 10%. To calculate: (60-30 = 30%) - (45-35 = 10%) = 20%, divided by two, yields 10%.

·         The estimated number of voters in the subgroup who changed their minds is calculated by applying the percentage above (10%) to the absolute number of voters in the subgroup. If this group contained 1,000,000 voters, multiplying that number by 10% = estimated net movement of 100,000 voters who moved toward favoring the ban on same-sex marriage.

·         The estimated number of votes represented by the 20-point swing is 200,000. Ie the movement in this voter subgroup resulted in a change of 200,000 in the net margin of victory for the Yes side.

The report sometimes refers to this change as a "200,000-vote swing," or sometimes refers to it as an estimated 100,000 voters who changed their minds.

The Advantages of Thinking about Voter Movement in terms of “Swing”

Advantage 1: It is easy to understand

The “200,000-vote swing” often lends the greatest insight when we are examining election results. If a ballot measure passes, as Prop 8 did, with 7,001,084 voters voting Yes and 6,401,482 voters voting No, most commonly people would say that it passed by 600,000 votes, even though 300,000 voters changing their minds from Yes to No would have been sufficient to change the outcome. Alternatively, it would have taken 600,000 additional non-voters voting No to change the outcome.

Some campaign activities -- eg turning out supporters -- yield a 1-vote swing per voter: turning out a voter who would not have voted, but will vote for your side, increases your margin (or decreases the opposition's margin) by one vote. Other campaign activities – eg those that persuade a voter to change his or her mind on an issue, or that correct a voter's wrong-way vote -- yield a 2-vote swing per voter: persuading a voter subtracts a vote from the opposition's total and adds a vote to your total.

Advantage 2: The swing is knowable with greater precision than the number of voters who changed their minds.

Saying that this net change in votes -- a 200,000-vote swing -- resulted from 100,000 people changing their minds is an oversimplification. Each voter (assuming the voter is certain to cast a ballot) changing his or her mind results in a two-vote swing -- it adds a vote to the side to which the voter changes, and subtracts a vote from the other side.

But there are other possibilities that create two-vote swings. One person might change his/her mind from undecided to Yes (creating a 1-vote swing), and another might change his/her mind from No to undecided (creating another 1-vote swing). Or one No voter might become disinterested in the election and decide not to vote instead of voting, while a Yes voter who was formerly disinterested becomes interested and decides to vote. Undoubtedly, a portion of the swing measure represents these occurrences. But in the Prop 8 campaign, probably the majority of the movement was a result of people moving within the electorate: most voters were voting, and there’s no reason to believe that the number of people entering and exiting the likely electorate was large, or that they would have tended to lean one way or the other.

The Methodological Assumption made in this report to allocate Undecided Voters

This report’s methodology assumes that an undecided voter is equally likely to cast a Yes vote or a No vote, and thus that moving from Yes to undecided is a 1-vote swing toward No, while moving from undecided to Yes is a 1-vote swing toward Yes. In some campaigns, this is not true; undecided voters tend to break disproportionately in one direction. The data from the Prop 8 campaign is far from conclusive: some data suggest that undecided voters broke toward Yes; some data suggest that they broke toward No. One could argue in favor of either position defensibly, and the truth may differ across demographic subgroups. This report makes the simplest assumption: that undecided voters divided equally.

Some might argue that undecided voters would abstain rather than divide equally. In the case of Prop 8, however, this view is not persuasive: the campaign’s polling showed approximately 10% of voters undecided throughout, yet fewer than 3% abstained from voting on Prop 8. But probably some undecided voters abstained (and also true that some decided voters probably abstained due to ballot spoilage or dropoff). The model used in this report does not account for this.

Others might argue that undecided voters would break in the same ratio as decided voters, eg in a group where Yes had a big lead, undecided voters might break toward Yes. There is no evidence for or against this in the case of Prop 8; it might or might not be true. The method used in this report has the advantage of simplicity of calculation and explanation.

Both of the above arguments would suggest the same calculation: ignoring undecided voters (or, to put it another way, reallocating them proportionally to decided voters). This formulation is often used in ballot measures, because many voters skip ballot questions on which they are undecided. On most ballot questions, however, undecided voters are largely voters with very low information about the particular ballot measure on which they are voting; they are not so much undecided (torn) as uninformed. Prop 8 was not that kind of measure.

In any case, the numbers would not change markedly if either of these methods were used. In the example above, this alternative methodology would estimate the above swing to be 20.8% (56.25%-43.75% =12.5% to 66.67%-33.33% = 33.3%), for a change in margin of 20.8%, rather than 20%, barely altering our calculation. So this alternative method of calculation would impact the findings of this report very little.

Methodological limitations in drawing conclusions from ballots cast by different voter subgroups

Below are this report’s approximations translating net movement among subgroups polled into ballots cast. For the most part, the quantifiable ballots cannot be added between demographic groups. For instance, if we know 500,000 parents moved toward the anti-gay side, and 700,000 Democrats did the same, the overlap between the categories prevents us from being able to define which voters are included in both categories. For mutually exclusive demographics, it is permissible to add between groups. Lake’s question to establish parental status is binary and separates people into either “childless” or “parent,” the latter defined as someone with a child under 18 living at home. Thus we can add the movement among the childless subgroup and the parent subgroup with confidence there aren’t overlapping voters.

All subgroup sample size is taken from the final Lake tracking poll ending on October 30. This is the most accurate estimation of electoral make-up available as it is the closest data point to election day. For consistency, this report applied the ratio of subgroup to total sample within Lake polling to total ballots cast to estimate the electoral size or each subgroup. Example: parents make up 30% of the total sample for the October 30 Lake Poll. We assume that parents also make up 30% of ballots cast election day. There were 13,743,177 total ballots cast on election day, enabling this report to estimate parents were responsible for 30% of the votes or 4,122,953 ballots.


Voter Movement by Change in Margin

Total Population

 

Total Ballots Cast on Election Day : 13,743,177
Change in Margin: 10% pt change in margin from (3) to (13)
“Net Movement”:1,374,318
“Movers” (1/2): a minimum of 687,159

 

Parental Status

 

Parents : Based on “child under 18 Yes”

 

Parents: 30% of those polled on Oct 30th

Total Parental Ballots: total ballots ( 13,743,177) x % of electorate (.30 ) = 4,122,953

Change in margin among parents: 22% change in margin from (2) to (24)

“Net Movement”: 907,050

“Movers” (1/2): a minimum of 503,525

 

Mothers :Based on “child under 18 Yes” and reported gender

 

Total Parental Ballots: total ballots ( 13,743,177) x % of electorate (.30 ) = 4,122,953

Mothers are 54% of parents polled on Oct 30th

Total Mother Ballots: total parent ballots (4,122,953) x % mothers ( .54) = 2,226,395

Change in margin among mothers: 26% pt change in margin from (-11) to (15)

“Net Movement”: 578,862

Movers” (1/2): a minimum of 289,431

 

Fathers : Based on “child under 18 Yes” and reported gender

 

Fathers are 46% of parents polled on Oct 30th

Total Father Ballots: total parent ballots (4,122,953) x % fathers ( .46) = 1,896,558

Change in margin among fathers : 20% pt change in margin from (14) to (34)

“Net Movement”: 379,311

“Movers” (1/2)– a minimum of 189,655

 

 

 


 

Parental Status Continued

Democratic & Independent Mothers < age 45

 

Total Parental Ballots : total ballots ( 13,743,177) x % of electorate (.30 ) = 4,122,953

Dem & Ind Parents < 45 are 44% of parents polled on Oct 30th

Mothers are 59% of Dem & Ind parents

Total Dem/Ind<45 Mother Ballots :total parental ballots (4,122,953) x % DemInd<45 (.44) x % mothers (.59) = 1,070,318

Change in margin among Dem/Ind<45 Mothers :42% pt change in margin from (-42) to (0)

“Net Movement”: 449,533

“Movers” (1/2): a minimum of 224,766

 

Democratic & Independent Fathers < age 45

 

Total Parental Ballots: total ballots ( 13,743,177) x % of electorate (.30 ) = 4,122,953

Dem & Ind Parents < 45 are 44% of parents polled on Oct 30th

Fathers are 41% of Dem & Ind parents

Total Dem/Ind<45 Father Ballots: total parental ballots (4,122,953) x % DemInd<45 (.44) x % Fathers (.41) = 743,780

Change in margin among Dem/Ind<45 Fathers: 33% pt change in margin from (-4) to (29)

“Net Movement” : 245,447

“Movers” (1/2):a minimum of 122,724

 

“Childless”: Based on “Child under 18 no”

 

“Childless”: 70% of those polled on Oct 30th

Total “Childless” Ballots: total ballots ( 13,743,177) x % of electorate (.70) = 9,620,224

Change in margin among “Childless”: 4% pt change in margin from (2) to (6)

“Movement” : 384,808

“Movers” (1/2) : a minimum 192,404

 


 

Race

 

White :

 

Whites account for 65% of those polled on Oct 30th

Total Ballots: total ballots ( 13,743,177) x % of electorate(.65) = 8,658,202

Change in margin among Whites : 17% pt change in margin from (-2) to (15)

“Movement”: 1,471,894

“Movers” (1/2) : a minimum of 735,947

 

African American:

 ** movement within margin of error

 

(movement based off of small sample size <100, ) ( time captured is Oct 2 – 0ct 30)

 

African Americans account for 6% of those polled on Oct 30th

Total Ballots: total ballots ( 13,743,177) x % of electorate(.06) = 824,590

Change in margin among African Americans: 4% pt change in margin from (11) to (15)

“Movement” : 32,984

“Movers” (1/2) : a minimum of 16,492

 

Asian:

** movement within margin of error

 

(movement based off of small sample size <100, ) ( time captured is Oct 2 – 0ct 30)

 

Asians account for 5% of those polled on Oct 30th

Total Ballots: total ballots ( 13,743,177) x % of electorate(.05) = 687,159

Change in margin among Asians: (-3%) pt change in margin from (7) to (4)

“Movement”: 20,615

“Movers” (1/2) : a minimum 10,307

 

Latino:

 

Latinos account for 17% of those polled on Oct 30th

Total Ballots: total ballots ( 13,743,177) x % of electorate(.17) = 2,336,340

Change in margin among Latinos :10% pt change in margin

“Movement” : 233,634

“Movers” (1/2) :116,817

 

 


 

Party

 

Democrat:

 

Democrats account for 46% of those polled on Oct 30th

Total Democrat Ballots:total ballots ( 13,743,177) x % of electorate(.46) : 6,321,861

Change in Margin among Democrats : 15% pt change in margin from (-29) to (-14)

“Movement” : 948,280

“Movers” (1/2) : a minimum of 474,140

 

Independent:

 

Independents account for 18% of those polled on Oct 30th

Total Independent Ballots: total ballots ( 13,743,177) x % of electorate(.18) = 2,473,772

Change in margin among Independents : 26% pt change in margin from (-6) to (20)

“Movement”: 643,181

“Movers” (1/2) : a minimum of 321,590

 

Republican:

** movement is within margin of error

 

Republicans account for 32% of those polled on Oct 30th

Total Republican Ballots: total ballots ( 13,743,177) x % of electorate(.32) = 4,397,817

Change in margin among Republicans: (-1) %pt change in margin from (50) to (49)

“Movement”:  43,987

“Movers” (1/2) : a minimum of 21,989