league of women voters of north carolina et al v. … · news, michael barone of the almanac of...

Multiple DocumentsPart Description1 30 pages2 Exhibit 1 - First Trende Rep.3 Exhibit 2 - Trende Supplement4 Exhibit 3 - Second Trende Rep.5 Exhibit 4 - Trende Dep. Excerpts6 Exhibit 5 - Burden Sur-Rebuttal7 Exhibit 6 - Burden Rep.8 Exhibit 7 - Gronke 2014 Sur-reply9 Exhibit 8 - Hood Dep. Excerpts10 Exhibit 9 - Stewart 2014 Surrebuttal11 Exhibit 10 - Dep. Ex. 11712 Exhibit 11 - Dep. Ex. 11613 Exhibit 12 - Dep. Ex. 11014 Exhibit 13 - Dep. Ex. 10915 Exhibit 14 - Bernstein et al.16 Exhibit 15 - McKee et al.17 Exhibit 16 - Dep. Ex. 378

LEAGUE OF WOMEN VOTERS OF NORTH CAROLINA et al v. STATE OF NORTH CAROLINA et al, Docket No. 1:13-cv-00660

© 2015 The Bureau of National Affairs, Inc. All Rights Reserved. Terms of Service // PAGE 1

http://www.bloomberglaw.com/ms/public/document/Terms_of_Service

EXHIBIT 3

Case 1:13-cv-00660-TDS-JEP Document 283-3 Filed 06/23/15 Page 1 of 79

IN THE UNITED STATES DISTRICT COURT FOR THE MIDDLE DISTRICT OF NORTH CAROLINA

NORTH CAROLINA STATE CONFERENCE OF THE NAACP, et al.,

Plaintiffs,

v. PATRICK LLOYD MCCRORY, in his official capacity as Governor of North Carolina, et al.,

Defendants.

) ) ) ) ) ) ) ) ) ) )

1:13CV658

LEAGUE OF WOMEN VOTERS OF NORTH CAROLINA, et al.,

Plaintiffs, and LOUIS M. DUKE, et al.,

Plaintiffs-Intervenors,

v. THE STATE OF NORTH CAROLINA, et al.,

Defendants.

) ) ) ) ) ) ) ) ) ) ) ) ) ) ) )

1:13CV660

UNITED STATES OF AMERICA,

Plaintiff,

v. THE STATE OF NORTH CAROLINA, et al.,

Defendants.

) ) ) ) ) ) ) ) )

1:13CV861

1


SECOND DECLARATION OF SEAN P. TRENDE

Sean Patrick Trende, first being sworn, deposes and says as follows:

1. I am over 18 years of age and am competent to testify regarding the matters

discussed in this declaration.

2. I am a recognized expert in the fields of psephology, voter behavior, voter

turnout, polling, and United States demographic trends and political history, with particular

emphasis on Southern politics.

3. I have been retained in this matter to provide expert testimony. I am compensated

at a rate of $300 per hour, excluding travel time.

4. My curriculum vitae is attached to this declaration as Exhibit 1.

5. A list of materials upon which I relied in support of this declaration are attached

as Exhibit 2.

EXPERT CREDENTIALS

6. I have studied and followed United States elections on both a full-time and part-

time basis for almost twenty years.

7. I received a B.A. from Yale University in 1995, with a double major in history

and political science.

8. I received a J.D. from Duke University in 2001.

9. I also received an M.A. from Duke University in 2001, in political science.

10. I joined RealClearPolitics in January of 2009 as their Senior Elections Analyst. I

assumed a full-time position with RealClearPolitics in March of 2010.

11. RealClearPolitics is one of the most heavily trafficked political websites in the

world. It serves as a one-stop shop for political analysis from all sides of the political spectrum

and is recognized as a pioneer in the field of poll aggregation. It is routinely cited by the most

2


influential voices in politics, including David Brooks of The New York Times, Brit Hume of Fox

News, Michael Barone of The Almanac of American Politics, Paul Gigot of The Wall Street

Journal, and Peter Beinart of The New Republic.

12. My main responsibilities with RealClearPolitics consist of tracking, analyzing,

and writing about elections. I also am in charge of rating the competitiveness of House of

Representatives races, and collaborate in rating the competitiveness of Presidential, Senate and

gubernatorial races. As a part of carrying out these responsibilities, I have studied and written

extensively about demographic trends in the country, exit poll data at the state and federal level,

public opinion polling, as well as voter turnout and voting behavior.

13. I am also a Senior Columnist for Dr. Larry Sabato’s “Crystal Ball.” I began

writing for the Crystal Ball in January of 2014.

14. The overarching purpose of my writings, both at RealClearPolitics and the Crystal

Ball, is to try to convey more rigorous statistical understandings of elections than is typically

found in journalistic coverage of elections to a lay audience.

15. I am the author of The Lost Majority: Why the Future of Government is up For

Grabs and Who Will Take It. The book offers a revisionist take on realignment theory. It argues

that realignments are a poor concept that should be abandoned. As part of this analysis, it

conducts a thorough analysis of demographic and political trends beginning around 1920 and

continuing through modern times.

16. I authored a chapter in Larry Sabato’s Barack Obama and the New America: The

2012 Election and the Changing Face of Politics, which discussed the demographic shifts

accompanying the 2012 elections. I also authored a chapter in Sabato’s forthcoming The Surge:

3


2014’s Big GOP Win and What It Means for the Next Presidential Election, which discusses

demographics and Electoral College shifts.

17. I co-authored the 2014 Almanac of American Politics. The Almanac is considered

the foundational text for understanding congressional districts and their representatives, as well

as the dynamics in play within those districts. PBS’s Judy Woodruff described the book as “the

oxygen of the political world,” while NBC’s Chuck Todd noted that “[r]eal political junkies get

two Almanacs: one for the home and one for the office.” My focus was researching the history of

and writing descriptions for many of the newly-drawn districts, including those in North

Carolina.

18. I have spoken on these subjects before audiences from across the political

spectrum, including at the American Enterprise Institute, the CATO Institute, the Bipartisan

Policy Center, and the Brookings Institution. In 2012, I was invited to Brussels to speak about

American elections to the European External Action Service, which is the European Union’s

diplomatic corps.

19. It is my policy to appear on any news outlet that invites me, barring scheduling

conflicts, and I have appeared on both Fox News and MSNBC to discuss electoral and

demographic trends. I have spoken on a diverse array of radio shows such as First Edition with

Sean Yoes, the Diane Rehm Show, the Brian Lehrer Show, the John Batchelor Show, and Fox

News Radio. I have been cited in major news publications, including The New York Times, The

Washington Post, The Los Angeles Times, The Wall Street Journal, and USA Today.

20. I sit on the advisory panel for the “States of Change: Demographics and

Democracy” project. This three-year project is sponsored by the Hewlett Foundation and

involves three premier think tanks: The Brookings Institution, the American Enterprise Institute,

4


and the Center for American Progress. The group takes a detailed look at trends among eligible

voters and the overall population, both nationally and in key states, in an attempt to explain the

impact of these changes on American politics, and to create population projections, which the

Census Bureau abandoned in 1995.

21. I previously authored an expert report in Dickson v. Rucho, No. 11-CVS-16896

(N.C. Super Ct., Wake County), which involved North Carolina’s 2012 General Assembly and

Senate maps. Although I was not called to testify, it is my understanding that my expert report

was accepted without objection.

22. I also previously authored an expert report in NAACP v. Husted, No. 2:14-CV-

404 (S.D. Ohio), which involved a challenge to multiple changes to Ohio’s voter laws, including

a reduction in early voting days and elimination of an overlapping period of registration and

early voting.

23. I am personally agnostic on the wisdom and utility of these particular laws and

regulations, which raise a host of normative issues that go beyond the realm of objective expert

testimony. With that said, the likelihood that these laws will have a significant impact on actual

voting, and specifically on minority voting, is certainly overstated by plaintiffs.

Introduction and Response to New Reports

24. Plaintiffs’ experts have utilized a variety of styles for their reports. For purposes

of consistency and simplicity, I will refer to the initial reports filed at the preliminary injunction

phase of this litigation as “PI Reports.” Sur-rebuttal reports will be referred to as “SR Reports.”

Those recently submitted for the trial phase will be referred to as “Trial Reports.”

25. Plaintiffs’ experts made strong claims at the preliminary injunction phase

regarding the extensive burdens that HB 589 would place on voters in the state. Dr. Charles

5


Stewart concluded that in the 2010 election: (1) the elimination of same-day voting would have

burdened 7,460 African American voters; (2) the elimination of out-of-precinct voting would

have burdened 2,479 voters; and (3) the reduction in early voting would have burdened 195,020

voters. Stewart PI Report ¶21. For the 2012 elections, he calculated that those numbers would be

31,127 voters, 2,079 voters, and 736,286 voters, respectively. Id. ¶19. This represented in excess

of 70% of the African-American electorate. Id.

26. Dr. Gronke was more explicit: “I know of no empirical argument by which one

could conclude that African-American voters – or any voters for that matter – will successfully

adjust to 40% fewer early voting days.” Gronke PI Report ¶52. Dr. Gronke continued: “I

conclude from the analysis in the Report that, because same-day vote registration has been

shown to be a strong and consistent predictor of higher turnout, the elimination of same-day

registration during the election process, whether during one-stop voting or on Election Day, will

lower turnout overall.” Id. ¶54.

27. Dr. Barry Burden embraced the “calculus of voting” approach, explaining that

“[s]mall changes in benefits or costs may alter the likelihood of voting directly. The decision to

vote is sensitive enough to costs that even Election Day weather has been shown to depress

turnout.” Burden PI Report at 3. “Scholarly research has demonstrated how increasing the costs

of voting depresses voter turnout.” id.

28. If we accept the argument along these lines – if costs are increased, turnout will

drop, all other things being equal – then if turnout increases after a reform, we must demonstrate

through some sort of rigorous analysis that all things were not equal, and that these differences

were sufficient to overcome the hypothesized burden. Otherwise, we should conclude that our

assumptions about the burden imposed are incorrect. Given that, under plaintiffs’ theory, almost

6


the entire African-American electorate was expected to see its costs of voting rise, there would

have to be a particularly compelling explanation regarding the magnitude of changes.

29. Last summer, this Court denied a motion to enjoin defendants from implementing

HB 589. The law, for the most part, went into effect prior to the election. Yet far from seeing

evidence of a substantial burden on African American voters in the form of decreased turnout,

African-American turnout was robust. Around 41% of the eligible population cast ballots for the

highest office on the ballot, the highest share in a midterm election since the Gantt-Helms race of

1990, and the second highest in midterm elections since at least 1980.

https://www.google.com/search?q=harvey+gantt&ie=utf-8&oe=utf-8.

30. We can see this effect among registered voters, broken down by race:

7


50.0%

45.0%

40.0%

35.0%

30.0%

25.0%

20.0%

15.0%

10.0%

5.0%

0.0%

Percent of Registered African-Americans, Non-Hispanic Whites and Other Voters Casting Ba llots, 2010-2014

42.2%

40.4°

19.5%

14.9%

25.0% 22.5%

46.8%

45.7°

18.0% 16.2%

.2010 . 2014

28.7% 29.3° 26.5%

26.01

17.4%

17.8°

8.8%

31. Note that, despite the changes from HB 589, African-American usage of early

voting increased.

32. The share of African-Americans voters that tumed out continued its long-tenn

trend of improving. It continued to converge on non-Hispanic white tumout, eclipsing the

differential that was displayed in the competitive 2002 senate race. What was a ten-point

difference in tumout in 2002 grew to 12 points in 2006 (when there was no top-of-the-ticket

race), before shrinking to five points in 2010, and again to three points post-ref01m, in 2014.

8


White and Africa n-American Voter Tu rnout In Recent North Ca rolina Elections

SO%

4S% 44% 43% 42%

40% 36%

37%

3S%

30%

2S%

20%

1S%

10%

S%

0%

2002 2006 2010 2014

• Share of Whites Voting • Share of African Americans Voting

33. Faced with this outcome, plaintiffs' expe1is claim that all things were not, in fact,

equal with 2010, and claim that the surge in 2014 is mostly the result of a highly competitive

senate campaign in the state. Campaign effects now dominate.

34. This is a plausible explanation, but it cannot be applied solely to 2014. During the

2000s, patties increasingly utilized eru·ly voting and similru· laws not to increase tumout, but to

lock voters in. Campaign stimuli helped prod groups to use eru·ly voting and other refonns;

campaign stimuli are now helping them adapt to the new legal regime. This is not some revealed

preference, but rather a response to exogenous forces. Faced with a new voting regime, we

would then expect these exogenous forces to change their strategy to alter voting behavior, and it

apperu·s that they did, successfully .

35. Plaintiffs' expelts gave a nod to the idea that campaign effects mattered, but

barely referenced it in their initial analyses. The word "Obama" appears four times in Dr.

Stewrut's rep01t, three times in Dr. Burden 's rep01i, and is absent from Dr. Gronke's. This makes

9


it more difficult for them to claim now that campaign effects overwhelmed the burden created by

HB 589, especially given the magnitude of the burden that they theorized.

36. Regardless, this outcome gives the Court some idea of the actual scale of these

campaign effects, as well as of the burden that HB 589 places on the electorate. That party

efforts can largely overwhelm – or at least obscure – the detrimental effect that these laws

supposedly had suggests that the burden HB 589 placed upon voter turnout is likely much

smaller than plaintiffs’ experts suggest, relative to the various factors that go into turnout.

37. Of course, there are better ways of fleshing this out: By comparing what occurred

in North Carolina to what occurred in other states. I have attempted such a comparison, and

found no statistically significant relationship between these laws and African-American turnout.

Unfortunately, the Current Population Survey data for 2014 have not yet been released, which

would best enable me to update my findings.

38. I will update my findings from my PI Report in the following sections to the best

of my ability. Before that, it is worth commenting upon some of the conclusions of plaintiffs’

experts in their Trial Reports.

39. In his latest report, Dr. Gronke claims that 2008 was more competitive than 2012.

Gronke Trial Report ¶10. The outcome was closer in 2008, but this analysis illustrates the

limitations of using outcomes as a measurement of competitiveness. If anything, for most of the

cycle, polling suggested a closer race in North Carolina in 2012 than 2008. Compare

http://www.realclearpolitics.com/epolls/2008/president/nc/north carolina mccain vs obama-

334.html with

http://www.realclearpolitics.com/epolls/2012/president/nc/north carolina romney vs obama-

1784.html. This showed up in spending as well. In 2008, the parties spent around $22 million in

10


the state. http://www.cnn.com/ELECTION/2008/map/ad.spending/. In 2012, they spent almost

$100 million in the state in advertising alone. http://www.washingtonpost.com/wp-

srv/special/politics/track-presidential-campaign-ads-2012/. In truth, the state was much more

competitive overall in 2012 than 2008.

40. Dr. Gronke writes that “[a]ll other things being equal, we should expect higher

overall turnout levels when comparing 2014 to 2010, even in the face of legal changes that may

have hindered the ability of individual citizens to cast an early vote.” Gronke Trial Report ¶ 21.

While I agree that campaign effects matter a great deal, the notion that campaign effects were

strong enough to overwhelm the effects of this law in 2014, but were secondary considerations

from 2000 to 2012, is a mere assertion here, and is difficult to square with the magnitude of the

burden that plaintiffs’ experts have suggested. We are offered no basis for evaluating at what

point we should expect campaign spending to overwhelm such a burden.

41. Dr. Gronke does offer a controlled experiment with his transition analysis, but it

ultimately only helps prove the point that African Americans were, in fact, able to navigate the

new voting rules well. He notes that 39.41% of African American voters who cast a one stop

ballot in 2012 failed to vote in 2014, while only 31.86% of non-hispanic Whites failed to vote.

He then asserts, with appropriate caveats, that this provides “some evidence that, contrary to the

claim that voters can easily adapt to a shorter period of time for early voters, African-American

voters may have been less able to adapt than were White voters.” Id. ¶ 28.

42. There are two problems with this. First, this disproportionate drop-off is exactly

what we would expect, even without a change in laws. African-American voters typically drop

off at disproportionate rates from white voters when one transitions to a midterm electorate from

a presidential one. Using the exit polls for guidance, the African-American share of the electorate

11


increased from 2012 to 2014 in only three states (of 19) where we have exit polling data for both

2012 and 2014: Florida, Illinois and Ohio. The latter datapoint is especially interesting, because

Ohio cut early voting such that it no longer overlapped with registration (thereby eliminating a

brief window of de facto same day registration), and u·ansitioned fi:om an exu·emely competitive

presidential campaign in 2012 to a gubem atorial campaign that was decided by over thirty points

in 2014. The African American share of the electorate decreased in eight states and was stable in

the remainder (in most of those states, the share was too small to drop off substantially).

40%

35%

30%

25%

20%

15%

10%

3% 5% 3

i1 0%

AK

African American Shares of Electorate, 2012 to 2014, Exit Polling

Data

8%8%

I 3% 2%

3%

;~ I CA co FL lA IL

16% 14%

5% 6%

13% 1%

:L

36%

23%

21%

2%

16%

13% 13%

KS ME M l MN MS NC NH NY OH OR PA VA WI

.2012 . 2014

43. Second, and cmcially, Dr. Gronke overlooks Election Day voters, who would

provide us with a conu·ol group that would make comparing 2012 to 2014 somewhat workable.

After all, both early voters and Election Day voters are subjected to similar "get out the vote"

efforts, adve1i ising, hectoring from neighbors, and whatever other campaign effects are at work.

Election Day voters were not, however, subjected to the same bm dens plaintiffs associate with a

reduction in early voting hom s and the elimination of same day regisu·ation. Indeed, under Dr.

12


Stewart’s calculations described above, any burden imposed by HB 589 falls disproportionately

on those who voted early in 2012, and by multiple orders of magnitude, at that.

44. But our control group actually fared worse than the voters who saw the biggest

changes to their previous method of voting. Almost 41% of non-Hispanic white Election Day

voters from 2012 failed to return in 2014, while almost 53% of African-American Election Day

voters from 2012 did not vote in 2014. In other words, 12.1% more African Americans than

whites failed to return to vote in the control group, versus 7.7% more African Americans than

whites in the early voting group.

29.7%

40.8%

37.3%

52.9%

7.7%

12.1%

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

Share of 2012White EarlyVoters whoDNV in 2014

Share ofWhite

Election DayVoters whoDNV in 2014

Share of BlackEarly Voterswho DNV in

2014

Share of BlackElection DayVoters whoDNV in 2014

DifferenceBetween

White andBlack DNV

Rates, EarlyVoters

DifferenceBetween

White andBlack DNV

Rates,Election Day

Voters

Percentage of 2012 Voters who Did Not Vote (DNV) in 2014, by Race and Voting Method

13


45. Plaintiffs’ experts might respond that early voters are more likely to be

enthusiastic voters, and are hence less likely to drop out than Election Day voters. But this is not

entirely helpful; if early voters tend to be more “sophisticated” voters, then they would be better

prepared to shoulder any burden that HB 589 places upon them than Plaintiffs credit. Moreover,

while this might explain why African-Americans and non-Hispanic whites who voted early

returned at higher rates than African-Americans and non-Hispanic whites, respectively, who

voted on Election Day, it doesn’t explain the differences in the gaps between African Americans

and non-Hispanic whites in the two groups. As Dr. Gronke correctly observes, it is hard to say

anything with certainty here, but to the extent the Court can draw conclusions from the transition

analysis, it actually suggests that any burden imposed by these laws is at best minimal.

46. Dr. Gronke also claims that “there is a racial disparity in transitions to other modes of

voting. While 36.84% of White one-stop voters in 2012 cast a one-stop ballot in 2014, only

33.64% of African American voters did so…30% of White one-stop voters in 2012 cast an

Election Day ballot in 2014, whereas only 26.5% of African American one-stop voters did

so…White voters in 2014 showed a greater ability to vote early in-person again and to switch to

alternative methods of voting.” Id. ¶ 26.

47. But this disparity is entirely a function of the higher African American rate of not

voting, which as we noted above, is endemic to midterm elections. African-Americans who

returned to vote proved to have just as much ability as non-Hispanic whites to navigate the new

legal regime. Of whites who voted early in 2012 and returned in 2014, 54.7% voted early again,

while 45.3% voted on Election Day. Of African Americans who voted in 2012 and returned in

2014, 54.9% voted early again, while 45.1% voted on Election Day. African-American Election

14


Day voters actually showed greater ability than whites to switch to alternative measures of

voting: 14.6% switched to early voting, while only 11.4% of non-Hispanic whites did so.

48. Dr. Gronke states that “[t]here were 73,975 domestic absentee (“by mail”) ballots cast

in North Carolina in 2014, just 1.62% of the 4,542,488 total ballots cast.” Id. n.24. Dr. Gronke

appears to have inadvertently used the 4.5 million ballots for 2012 as his denominator, rather

than the 2.9 million ballots cast in 2014. Around 2.6% of voters used absentee ballots in 2014.

This is a small amount, but it is up from 2010.

49. Finally, it is worth observing that plaintiffs’ experts continually emphasize the

importance of habit in voting. But this is never fully explained, and the claims that African-

American voting habits are “fragile,” Burden PI Report at 6, or that African-Americans are

habituated to early voting after two presidential elections, are simply offered as assertions. How

many elections do you have to have before you become habituated to a mode of voting? How

important is habit in voting? Is someone who shows up 17 days before Election Day, only to find

that he can’t vote for another week really disincentivized from voting ten days before Election

Day? Is this disincentive really comparable to the disincentive imposed upon someone who

shows up on Election Day, only to find that her polling place has moved? These are not trivial

questions, especially given that there has been an election held under the new laws, which would

now be creating habits in the opposite direction that a return to the previous status quo would

disrupt. But no analysis is ever given to try to answer these questions.

50. This result is ultimately unsurprising. It is consistent with the “conventional

wisdom that suggests that the vast majority of votes cast by either of these methods simply

represent voters switching their mode of casting a ballot, not additional votes.” Jan E. Leighley

& Jonathan Nagler, Who Votes Now? Demographics, Issues, Inequality, and Turnout in the

15


United States, 111 (2014). Given that “[e]arly voting depresses net voter turnout” by “robbing

Election Day of its stimulating effects,” and that “[t]his depressant effect is only partially offset

if [same day registration] is present or if [Election Day registration] offers a vehicle for the last-

minute mobilization of marginal voters,” we should expect something along these lines. Barry

Burden, et al., Election Laws, Mobilization, and Turnout: The Unanticipated Consequences of

Election Reform, 58 Am. Jrnl. of Poli. Sci. 95, 108 (Winter 2014). This conclusion has

“upend[ed] the conventional view that anything that makes voting easier will raise turnout.” Id.

The only “[c]onsistent way to increase turnout is to permit Election Day registration,” id. which

North Carolina did not have, and will not have if plaintiffs obtain the relief they seek.

OPINION 1: THE VOTING REFORMS CONTAINED IN HB 589 PLACE THE STATE

WITHIN THE MAINSTREAM REGARDING THESE PRACTICES.

51. Dr. Gronke once described voting systems in the United States as a “bewildering

array of different electoral systems.” Paul Gronke, Early Voting Reforms and American

Elections, 17 Wm. & Mary Bill Rts. J. 423, 425 (2008). This is certainly true. But while no two

states are exactly alike in their approach to election procedure—and in several states, procedures

are ultimately set at the county level, setting up an even wider variety of approaches—we can

nevertheless draw some broad, useful conclusions about state approaches to voter identification

laws, early voting, and registration practices. In this opinion, I attempt to summarize, in useable

form, these exceedingly diverse systems into a few basic sorts of laws that are employed by

different states. I then place the changes to North Carolina’s election laws into this national

context.

52. In support of this opinion, I reviewed the relevant sections of the legal codes for all

fifty states and the District of Columbia, reports published by the National Conference of State

16


Legislatures, the relevant elections websites of the fifty states and the District of Columbia, and

materials provided by counsel which were largely duplicative of the above. On certain occasions,

where the laws were ambiguous, I contacted the Secretary of State office directly. If voting was

administered at the county level in a state, I contacted the boards of elections for some of the most

populous counties.

53. Because I am evaluating legal systems here, for purposes of this opinion, states

are coded with respect to present law. So while, for example, Massachusetts will not actually

implement early voting until 2016, an early voting law is nevertheless now on the books in

Massachusetts, and it is coded accordingly. I sought to reflect the government’s intent to utilize

early voting. Similarly, California has yet to implement its same-day registration laws, though

they were passed in 2012. Nevertheless, it is coded as having same-day registration, because

such a law is on the books, reflecting the government’s intent is to utilize same-day registration

laws.

54. For the sake of convenience and readability, rather than repeatedly referencing

“states and the District of Columbia,” I will simply reference “voting jurisdictions” or

“jurisdictions” when I am referring to a group that includes both states and the District of

Columbia.

Photo ID

55. Presently, nineteen jurisdictions require citizens to present photographic

identification prior to voting. These states are Alabama, Florida, Georgia, Hawaii, Idaho,

Indiana, Kansas, Louisiana, Michigan, Mississippi, New Hampshire, North Carolina, Rhode

Island, South Carolina, South Dakota, Tennessee, Texas, Virginia and Wisconsin.

Out-of-Precinct Balloting

17


56. Presently, thirty-one states reject a ballot cast in the wrong precinct: Alabama,

Arizona, Arkansas, Delaware, Florida, Hawaii, Idaho, Indiana, Iowa, Kentucky, Maine,

Michigan, Minnesota, Mississippi, Montana, Nebraska, Nevada, New Hampshire, North

Carolina, North Dakota, Ohio, Oklahoma, South Carolina, South Dakota, Tennessee, Texas,

Vermont, Virginia, Utah, Wisconsin and West Virginia.

57. A few of these states allow for Election Day registration, which would soften any

negative impact arising from ballots being cast in the wrong precinct.

58. Two states, New York and Missouri, allow for a ballot to be counted if it is cast in

the wrong precinct but in the correct polling place – e.g., if a polling place houses multiple

precincts and a voter is sent to the wrong line, her vote will still be counted. Because this is a

narrow exception and is more restrictive than the previous law in North Carolina, these states are

classified as refusing to count votes cast in the incorrect precinct.

59. Two more states, Connecticut and Massachusetts, will count ballots cast in the

wrong precinct so long as they are cast in the correct town or city. Because this is a narrow

exception – many precincts in Connecticut and Massachusetts are entire towns – and is more

restrictive than the previous law in North Carolina, which granted a countywide exception these

states are classified as refusing to count votes cast in the incorrect precinct.

60. Sixteen states, plus the District of Columbia, count ballots cast in the wrong

precinct. Within this group, there exists a wide array of approaches. Some states, such as

Louisiana, will only count votes cast for federal races, and then only if the voter voted in the

correct parish. Others, such as Maryland, do not have a “correct-county” requirement and will

count votes cast in any race where the voter was eligible to cast a ballot. Regardless, because

18


these states count votes cast for at least some races if they are cast in the incorrect precinct, they

were all counted as states allowing out-of-precinct voting.

In-Person Early Voting

61. For purposes of this report, I included under the rubric of “early voting” both

states that have traditional early voting, as well as states that allow persons to apply for an

absentee ballot without an excuse and cast that ballot at the same time. Although there are

frequently some differences in administration, these distinctions are not enough to warrant

separating the two systems.

62. For the sake of convenience and readability, I use the term “early voting” as

shorthand for no-excuse in-person early voting, regardless of the format, even though traditional

absentee voting could technically be considered “early voting.”

63. Certain judgment calls had to be made in the coding of state early voting laws.

Some states do not have a defined time period for early voting. In these instances, a call to the

secretary of state or board of elections often helped to clarify things. For example, a call to the

Idaho Secretary of State’s office suggested that most counties begin their in-person early voting

between two and three weeks before the election. I use 18 days for early voting, representing an

average of two and three weeks (rounded up).

64. The following chart summarizes the length of early voting periods in United

States jurisdictions.

19


20


65. At the time of the preliminary injunction hearing, the median jurisdiction in

America allowed for an early voting period of between 12 days and 13 days; passage of the

HB589 moved North Carolina closer to the median number of early voting days offered in this

country. Because Massachusetts has passed an early voting law of greater than the median, that

median has since moved to 14 days. Regardless, granting plaintiffs the relief that they seek

would not move North Carolina any closer to the median early voting period.

66. Sixteen states do not allow for early voting.

67. Two of these states, Oregon and Washington, conduct their elections almost

entirely by mail. How to classify these states is a judgment call. On the one hand, these states do

not offer any in-person voting, which necessarily would preclude them from offering any in-

person early voting as well. On the other hand, the fact remains that a voter in Oregon and

Washington cannot go to a voting center and cast an early ballot. Because a voter cannot go to a

center to vote in-person, and because the standard sought by Plaintiffs here is to require that such

days be made available, I have concluded that these states ought not be counted as offering early

voting. See also Jan Leighly and Jonathan Nagler, Who Votes Now? Demographics, Issues,

Inequality and Turnout in the United States 119-20 (2014) (explaining the dilemma, but

ultimately coding Oregon as a no-early voting state. Washington State is not discussed).

68. One state, Colorado, has adopted “vote-by-mail” for upcoming elections. It will,

however, maintain traditional polling places as well, although they will function at the county

level, rather than the precinct level. It is therefore counted as allowing for early voting.

69. Only 15 states provide for early voting periods in excess of 16 days, equaling or

exceeding the early voting period plaintiffs ask this Court to require North Carolina to provide.

21


70. Of course, it is not merely the length of North Carolina’s early voting period that

is at issue in this litigation. Most states do not make early voting available on every day during

the time period. Many states exclude intervening days on the weekends, either by statute, state

regulation, or county regulation. States with longer early voting periods frequently will not be

open on Columbus Day.

71. Figure 2 summarizes the actual number of early voting days available in U.S.

voting jurisdictions in 2014:

22


23


72. These calculations do involve some judgment calls. In many states, early voting

availability changes from election-to-election, or is set at the county level, rather than the state

level.

73. For an example of how these issues present themselves, and how I approached

them, consider the following: In Massachusetts, the early voting period runs from eleven

business days before the election to the second business day before the election. This is a period

of 15 days. However, it only requires that the early voting take place during normal business

hours. Localities may, however, opt to allow for weekend voting. Because we don’t yet know

whether localities will, in fact, allow for weekend voting, the period in some places could be as

short as 11 days.

74. On the other hand, North Dakota allows for 15 days of early voting, while Texas

offers 12 days. Early voting decisions, however, are made at the county level. It was not

practical to contact all 53 counties in North Dakota, to say nothing of all 254 counties in Texas,

to determine how many days of early voting they would actually offer.

75. Instead, for North Dakota, I visited the County Auditors’ websites for some of the

most populous counties in the state: Cass (Fargo), Ward (Minot) and Grand Forks (Grand Forks).

I also spoke with persons in the County Auditor’s offices, and learned that the most days of

actual early voting offered was five, with two of the three offering only four days. North Dakota

is nevertheless coded as allowing for five days of early voting, since there is evidence that a

significant portion of the population can vote for at least five days.

76. For Texas, the Elections Division confirmed that while many smaller counties do

not allow for early weekend voting, many of the larger counties do. A call to Harris (Houston)

and Dallas (Dallas) County election officials confirmed that these large jurisdictions did, in fact,

24


offer early voting on weekends. I therefore coded the state as allowing for early voting for the

full twelve days.

77. In general, I tended to err on the side of inclusivity – that is, if there was evidence

that a substantial portion of the state’s population was allowed the full early voting time period, I

coded the state as allowing for the full early voting time period. Likewise, in the absence of

evidence that a substantial portion of the state’s population was not allowed the full early voting

time period, I coded the state as allowing for the full early voting time period.

78. The median jurisdiction allows for 11 actual days of early voting.

79. Only fourteen states provide 17 or more actual days of early voting.

80. In reviewing states’ early voter laws, I analyzed Figure 11 of the initial report of

Plaintiffs’ expert Charles Stewart, which suggests that that North Carolina moved from being in

the mainstream of early voting laws to being an outlier when it shortened the number of days

available for in-person early voting. Dr. Stewart did not update Fig. 11 for his updated report.

81. My review of state law brought to light several differences with Dr. Stewart’s

characterization of early voter laws.

82. First, although North Carolina may have moved from the median insofar as states

that allow early voting are concerned, Figure 11 excludes states that simply do not allow for in-

person early voting and therefore do not place North Carolina in a national context.

83. Dr. Stewart’s chart excludes one state that does allow for early voting, Montana. I

likewise overlooked Montana in my initial report, but it has been updated here.

http://sos.mt.gov/elections/FAQ/index.asp.

84. Dr. Stewart’s charts also seemingly include at least one jurisdiction that allows for

early voting, but only with an excuse. For example, Virginia law does allow for 43 days of

25


absentee voting; however, it does so only if a voter falls into certain narrow categories of excuses

(e.g., if the voter will be out of state, Code § 24.2-700(1); is disabled, sick or pregnant, id. §

24.2-700(4); is awaiting trial, id. § 24-2-700(5); or has a religious objection, id. § 24-2-700 (8)).

85. Virginia does not allow no-excuse early voting: in person, absentee, or otherwise.

Any comparison to the system in place in North Carolina is necessarily apples-to-oranges,

especially when one considers that North Carolina actually has one of the most permissive no-

excuse absentee ballot programs in place in the United States.

86. Dr. Gronke claims that “a conversation with any county official in Virginia would

have helped Mr. Trende understand that Virginia has substantial early voting de facto.” Gronke

SR ¶61. Although it is true that in recent presidential elections (though not in off-year elections),

rates of absentee usage in some Northern Virginia counties grew into the 30% range, the

statewide average remains around 12% in presidential elections and five percent in off-year

elections – much more comparable to the rates in absentee voting states than in-person early

voting states. https://voterinfo.sbe.virginia.gov/election/DATA/2012/68C30477-AAF2-46DD-

994E-5D3BE8A89C9B/Official/1 s.shtml; http://www.electproject.org/2012 early vote.

87. But this is irrelevant. First, the goal here is not to verify or critique Dr. Stewart’s

data, but rather to provide the Court with an accurate assessment of how common the laws at

issue in this case really are. Whether some voters use the law in ways in which it is not intended

is not at issue here. Second, this is not unique to Virginia, as seen in the early vote rates in other

states that do not have early voting as defined in this case.

http://www.electproject.org/2012_early_vote (e.g., Missouri, where eleven percent of ballots

were cast early in 2008). Third, Dr. Gronke’s website lists Virginia as requiring an excuse, and

26


color codes the state differently from states with no-excuse in person voting; we all seem to

agree that this is the proper coding. http://reed.edu/earlyvoting/calendar/.

88. My goal here is not to offer any sort of “sensitivity analysis,” by testing whether

his conclusion would change by altering assumptions about the number of early voting days

available in states where judgment calls were made. Stewart SR ¶ 26. I agree with Dr. Stewart

that, aside from his choice to exclude states that do not have any early voting, the alternate

coding choices do not change the median outcome.

89. The point of the exercise is much simpler than any of this: The chart included in

my report had differences with Dr. Stewart’s. Some differences were substantial in magnitude.

Some were relatively minor. Regardless, they deserved an explanation. See Trende Declr. ¶ 30

(“An explanation of why there are differences in our charts, some of which are truly judgment

calls, follows.”).

90. The real dispute here is whether we should include states that did not have early

voting in the chart. It is better, at least in this legal context, to include states with no early voting

in our analysis. My logic here is simple: When possible, an analyst should examine all data

available for his or her analysis, rather than drawing potentially arbitrary lines. Stewart SR ¶81.

91. In effect, my chart rebuts the claim that “HB 589 significantly reduces the number

of days available for early voting in North Carolina — so much so that it moves North Carolina

from the middle-of-the-pack of states that offer early voting (in terms of how many days are

available), to one of the most restrictive.” Stewart PI Report ¶ 129. It is important for the Court

to know that the caveat here – “of states that offer early voting” – is not a throwaway line.

Instead, it is crucial to Dr. Stewart’s ability to draw his conclusion. Including this caveat makes

27


the difference between North Carolina moving toward the midpoint of American early voting

laws or moving to the extremes.

92. The job of the expert witness is ultimately to “help the trier of fact to understand

the evidence or to determine a fact in issue.” Fed. R. Evid. 702. If the Court asked “with respect

to the early voting practices of other states, where did North Carolina stand before it passed HB

589, and where did it stand after it passed HB 589?” any candid answer would have to account

for the states that do not offer early voting at all. Put differently, had North Carolina eliminated

early voting altogether, an expert who answered the Court’s question without acknowledging any

of the states that have some form early voting would put her or his credibility seriously at risk.

93. The above approach does not “make all states with in-person early voting, even

those with severely restricted numbers of early voting days, appear to offer even more days of

early voting than they actually do.” Stewart SR ¶31. North Carolina is depicted as offering ten

days of early voting, regardless of how many states are included on the chart. While the relative

positioning of North Carolina changes with the inclusion of states that offer no early voting, the

absolute number of days of early voting remains unchanged. More importantly, if this assertion

were correct, the principle would also work in reverse: excluding the states that offer no early

voting would make the remaining states look as if they offered fewer days of early voting than

they actually do.

94. It is likely true that there are numerous qualitative differences between states that

offer no days of early voting and those that offer a limited number of early voting days. Stewart

SR ¶33 (claiming that states with early voting enjoy “improvements in security, registration list

maintenance, and access to the polls for people who need assistance.”) But if anything, the

argument that the relationship between the number of days of early voting offered and the

28


benefits a state enjoys from early voting might be non-linear – that is, that the benefits are “front-

loaded” toward the first day of early voting that is adopted – counsels in favor of including

observations regarding states that do not offer early voting at all. Since North Carolina retains

some early voting, it would be important for the Court to know to know that it was retaining the

disproportionate benefits enjoyed by offering even a few days of early voting. Regardless, even

if true, Dr. Stewart’s claim does not support his conclusion that the Court should not be aware of

states that offer no early voting when analyzing the median number of early voting days

available in the country.

Same Day Registration

95. For purposes of this opinion, it is useful to distinguish between same-day

registration and Election Day registration. See Burden, et al, at 96-97.

96. Election Day registration allows for voters to register to vote on the actual

Election Day. That is to say, in 2014, in a state with Election Day registration, a voter will be

able to register to vote on November 8, 2016.

97. Same-day registration, by contrast, allows for voters to register to vote during the

early voting period.

98. Even within categories, states employ differing approaches. For example, Alaska

allows for both same-day and Election Day registration, but only for President. Rhode Island

allows for Election Day registration for President. Because this is a stricter standard than the one

plaintiffs would have this Court adopt for North Carolina, they are not counted as offering

Election Day or same-day registration.

99. To provide another example of the variety of approaches utilized by states with

Election Day and/or same-day registration, Montana allows for what it calls late registration. But

29


voters cannot register at the polling place – in other words, it does not provide for the “one-stop

shopping” that plaintiffs hope to preserve here. It is not coded as a same-day registration state.

100. These variations can be difficult to resolve. Prior to 2014, Ohio had a so-called

“Golden Week,” where the early voting period and the normal registration period overlapped for

a brief period. As Dr. Gronke noted, “Ohio has [] been treated as a same day registration state in

academic literature on the subject, including work cited by Mr. Trende himself.” Gronke SR ¶37.

But in truth, there is a disagreement among political scientists regarding to how to code the state.

Burden, et al., do count it as an early voting state, but Larocca and Klemenski do not. Compare

id. at 99, with Roger Larocca and John S. Klemenski, Election Reform and Turnout in

Presidential Elections, 11 State Pols. & Policy Q., 76 (Mar. 2011) (“Although Ohio offered

simultaneous registration and early in-person voting in 2008, we have not included Ohio

residents in this interaction because Ohio permitted same-day registration only within a more

limited and remote window of thirty to thirty-five days before Election Day.”).

101. I concurred with Larocca and Klemenski’s coding because it was more

appropriate for purposes of this litigation. Plaintiffs seek 17 days of same day registration,

running up almost to Election Day, after the close of registration. This is substantively different

from the previous legal regime in Ohio: six days of same day registration a month before an

election, coinciding with the “normal” registration period.

102. North Dakota does not technically have Election Day or same-day registration,

but that is a function of the fact that it has no registration law at all. It is counted having same-

day registration and Election Day registration.

103. Vermont does not specifically have a same-day registration law, and is not coded

as having same-day registration by Larocca and Klemenski. Like Ohio, Vermont’s voter

30


registration period simply overlaps with its lengthy early voting period for a time. Vermont is

therefore not coded as having same day registration for purposes of this litigation, because the

remedy that plaintiffs seek here would bring about a substantively different result.

104. The following table summarizes the basic approaches to same-day registration in

various jurisdictions:

31


Election Day Registration

EDR + SDR Neither

CT CA AL NJ ID CO AK NM NH DC AZ NY

HI AR NC IA DE OH IL FL OK ME GA OR MN IN PA ND KS RI WI KY SC WY LA SD MD TN MA TX MI UT MS VT MO VA MT WA NE WV NV

105. Thirty-seven jurisdictions fail to offer Election Day registration or same-day

registration for all offices.

Pre-Registration

106. Ten states, plus the District of Columbia, allow pre-registration at the age of

sixteen.

107. North Dakota does not have a voter registration law. It is therefore treated as

allowing for pre-registration at age 16.

Summary

108. The following figure summarizes the above by displaying the number of states

that employ various iterations of the five laws detailed above.

109. Illinois does not require photographic identification to vote, allows out-of-precinct

voting, allows for same-day registration, and has an early voting period in excess of 16 days. It

32


does not, however, allow for pre-registration at the age of 16. It therefore employs four of the

approaches to voting that plaintiffs would require North Carolina to adopt. It is therefore coded

as a “four”.

110. Alabama, by contrast, requires photographic identification to vote, does not allow

out-of-precinct voting, does not have an early voting period in excess of 16 days, does not allow

for pre-registration at the age of 16, and does not allow for same-day registration. It is therefore

coded as a “zero”. North Carolina is likewise coded as a “zero.”

111. No United States jurisdiction currently has all five voter laws that the plaintiffs

would require the state of North Carolina to adopt.

33


112. Five jurisdictions have adopted any four voter laws that the plaintiffs would

require the state of North Carolina to adopt. They are California, Colorado, the District of

Columbia, Illinois, and Wyoming.

113. California, Wyoming and Illinois do not have pre-registration at age 16, while

Colorado and the District of Columbia do not offer in person early voting in excess of sixteen

days.

114. Five jurisdictions have adopted any three voter laws that the plaintiffs would

require the state of North Carolina to adopt. They are Iowa, Maine, Maryland, Minnesota and

North Dakota.

115. Nineteen jurisdictions have adopted any two forms of voter law that the plaintiffs

would require the state of North Carolina to adopt. They are Alaska, Arizona, Connecticut,

Delaware, Georgia, Hawaii, Idaho, Louisiana, Massachusetts, Montana, Nebraska, New Jersey,

New Mexico, Ohio, Oregon, Pennsylvania, Rhode Island, Vermont and Washington.

116. Fourteen jurisdictions have adopted any one voter law that the plaintiffs would

require the state of North Carolina to adopt. They are Arkansas, Florida, Indiana, Kansas,

Kentucky, Missouri, Nevada, New Hampshire, New York, Oklahoma, South Dakota, Utah, West

Virginia and Wisconsin.

117. Eight jurisdictions have not adopted any of the voter laws that the plaintiffs would

require the state of North Carolina to adopt. They are Alabama, Michigan, Mississippi, North

Carolina (as of today), South Carolina, Tennessee, Texas and Virginia.

118. The experiments of the North Carolina legislature during the 2000s effectively

created an outlier in United States election law, at least insofar as the laws at issue in this

litigation are concerned. Collectively, they rendered North Carolina the only state in the union

34


that did not require a photographic identification to vote, that counted ballots cast in the wrong

precinct, that allowed for a period of early voting in excess of 16 days, that allowed voters to

register during early voting, and that allowed voters to pre-register at the age of sixteen.

Responses to Plaintiffs’ Experts Additional Criticisms

119. Several plaintiffs’ experts fault this opinion for not considering a variety of other

factors, including the length of the registration period and whether they allow student or

government photo identification. E.g., Lichtmann SR at 9-10. Indeed, it is true that there are an

almost limitless number of ways voting laws could be examined, some of which might result in a

different result. But this would be a flawed approach in this context, where we are only

concerned with five practices challenged by plaintiffs.

120. Some of the distinctions are simply difficult to draw. For example, North Dakota

is classified as a state that does not require photographic identification, because you can utilize a

student certificate or certificate from a long-term care facility to vote. North Dakota is not a

photographic identification state, but the fact that a voter may not (as of this writing) use a bill or

bank statement arguably distinguishes it from other non-photo identification states.

121. Similarly, I could have drawn upon the endless list of options to improve North

Carolina’s standing. I could have included the length of no-excuse absentee voting time periods,

which would improve North Carolina’s positioning since it has one of the lengthiest absentee

balloting periods in the country. However, plaintiffs do not challenge the 60 day no-excuse

absentee voting period, so I did not consider it.

122. It would also improve North Carolina’s standing to explore the fact that North

Carolina is part of a minority of states that offers early voting on weekends, and a super-minority

of states that offer early voting on Sundays.

35


123. In addition to the sixteen jurisdictions that offer no early voting days, another

eleven jurisdictions – Montana, Oklahoma, Utah, Wisconsin, North Dakota, Idaho, Georgia,

Arizona, Maine, Wyoming, and Vermont – do not offer any early voting on the weekends, either

by statute, statewide regulation, or county regulation.

124. In addition to the 27 jurisdictions that either offer no early voting days, or that fail

to offer any early voting days on the weekends, Louisiana, West Virginia, Hawaii, the District of

Columbia, Kansas, Arkansas, Colorado, New Mexico, Tennessee, Nebraska, and South Dakota

do not offer early voting on Sundays.

125. In addition to North Carolina, only eleven states – Maryland, Minnesota,

Massachusetts, Florida, Texas, Nevada, Ohio, Alaska, Illinois, Indiana, and Iowa offer early

voting on Sundays. I did not find proof of Sunday voting in California, but did not find any

reason to doubt that Sunday voting was offered either.

126. Additionally, I could have taken up Dr. Gronke’s suggestion to look at the

number of early voting centers available. Gronke Surrebuttal ¶31. He notes that Texas has 860

offices for 254 counties, and that Nevada has 136 locations for 13 counties. My home state of

Ohio has one office per county, for a total of 88 centers, while Florida, whose early voting

practices have been a point of contention in this litigation, has 330 centers for 67 counties.

http://election.dos.state.fl.us/pdf/Early-Voting-2014-General-Final-English.pdf. Compared to

these states, North Carolina would look strong, with 368 locations for 100 counties, similar to the

“county-to-center ratio” for Texas, and much higher than Ohio’s. Or we could look at it in terms

of total ballots cast. North Carolina would offer one center for every 7,600 votes cast in 2014,

versus one center for every 18,000 votes cast in Texas or Florida, and one center for every

36


34,700 votes cast in Ohio (Nevada would lead, with one center for every 680 votes). But the

number of centers is not at issue in this case.

127. Rather than impose ad hoc cutoffs in my coding, I simply looked at the relief

sought by plaintiffs in their complaint. So, for example, the NAACP is not merely complaining

that the provisions of H.B. 589 that require photographic identification without allowing for the

use of student identification violate the Voting Rights Act. Their complaint is ultimately

broader: “the provisions of H.B. 589 that impose voter-identification requirements that eliminate

same-day voter registration, that reduce the number of days for early-voting, that prohibit out-of-

precinct voting, and that expand the number of poll observers and the numbers of people who

can challenge ballots will result in the dilution of African-American voting strength and will, in

dependently and collectively, deny African-Americans meaningful access to the political

process.” First Am. Compl. ¶ 113. The relief they seek is to do away with the photographic

identification requirement altogether. That is what I considered.

128. To be clear, my argument here is not that “if a state has a record of innovations in

voting laws that operate well and encourage turnout, that this state should abolish these laws in

order to return to the ‘mainstream.’” Gronke SR ¶58. This is an unwarranted claim. I expressly

disclaim any notions on what a state should do, as I have no strong opinions in this regard. See

supra ¶ 22.

129. While it may well be true that “[t]here is far too much variability in the mix of

laws and regulations regarding elections in our federalized election system in the United States to

expect a precise match to North Carolina’s mix of laws.” Gronke SR ¶ 58, there is no reason this

would be true given the limited number of laws at issue in this case.

37


130. While it might be of academic interest to offer “a systematic, comprehensive

review of all laws or practices that might be implemented by a state to encourage greater turnout

by its voters,” Stewart SR ¶ 48, no one has performed such an analysis, despite ample

opportunity to do so, and we are left without any proof that such an analysis would actually alter

the conclusion.

131. The laws were selected with a particular purpose – they are the laws the plaintiffs

would have this Court impose on North Carolina. Regardless of the outcome of this case, North

Carolina will not have, to use an additional example, Election Day registration in 2016.

Informing the Court about North Carolina’s lack of an Election Day registration law would

therefore not assist it in determining “a fact in issue.”

132. Dr. Stewart observes “[i]t is unsurprising that the state most like North Carolina

on this measure is North Carolina.” It is unclear why this should be unsurprising. Plaintiffs have

defined the five voting practices that are relevant to this litigation, and there is no reason why

North Carolina should be the only state that has these five practices. In fact, under current law,

the state most like North Carolina is not necessarily North Carolina – seven states share its

approach with respect to the laws at issue in this case. Moreover, had plaintiffs not challenged

North Carolina’s decision to reverse itself on pre-registration, North Carolina would have

resembled California and Wyoming under its previous legal regime. Trende Declr. ¶ 54.

133. Finally, Dr. Stewart claims that the methodology “would be unacceptable for

publication that required peer review.” It isn’t, however, entirely clear that this would be the case

if the purpose of the article were to examine HB 589; I would have no trouble approving this

approach to that particular problem were I a peer reviewer. Also, while one engaged in a broad

academic inquiry would certainly want to inquire as to permanent absentee voting laws

38


elsewhere, such an inquiry would likely not assist this Court, since these types of laws are not at

issue under the terms of plaintiffs’ Amended Complaint(s).

39


OPINION 2: THE DATA DO NOT SUPPORT PLAINITFFS’ CONCLUSIONS REGARDING THE EFFECTS OF HB589 ON MINORITY VOTING

134. Since the reforms that plaintiffs seek to re-establish in North Carolina were

implemented, minority turnout appears to have increased overall. It might be tempting, then, to

conclude that the early voting law itself played a role in the increase in turnout. Similarly, we

might be tempted to conclude that eliminating these laws would suppress African-American

turnout.

135. We might also be tempted to conclude that the increased use of early voting, same

day registration, and so forth, by African-Americans in presidential elections represents a

permanent shift in voting preferences.

136. Some experts, such as Dr. Leloudis, make such causal connections explicit. See,

e.g., Leloudis Report at 30 (“As illustrated in the graphic below, the net effect of these reforms

was steady improvement in voter turnout, particularly after the introduction of out-of-precinct

voting. The reforms also opened the way for African American turnout to soar to historic highs

in the 2008 and 2012 elections, when voters rallied behind the candidacy of Barack Obama, who

would become the first African American president of the United States.”); id. at 38 (“The law

eliminates the first week of early voting, same-day registration, and straight-ticket voting.

Numbers from the 2008 election in North Carolina suggest that these changes will have a

disproportionately negative effect on black voter participation.”).

137. But it would be a mistake to draw such definitive conclusions here, at least

without considering additional questions. The problem with the theory that, for example,

shortening early voting would present a burden upon African Americans and adversely impact

African-American turnout is that it assumes largely stagnant responses to changing incentives.

Even accepting that African-Americans disproportionately use early voting in North Carolina,

40


there is little evidence that these voters would fail to adjust their behavior in response to new

laws and regulations, and vote during the ten days of early voting remaining (or on Election

Day). Indeed, to the extent we have useable evidence from the 2014 elections, it suggests that

this is exactly what occurred in North Carolina. See supra ¶¶ 41-48.

138. It isn’t just North Carolina, though. Ohio shortened the number of days available

for early voting from 35 days to 23 days between 2010 and 2014, such that it no longer overlaps

with the regular registration period. Wisconsin also shortened its early voting period. Yet

African-Americans constituted 15% of the electorate in Ohio in 2010 and four percent in

Wisconsin; these numbers increased to 16% and six percent, respectively. Note that Ohio’s

elections were substantially less competitive in 2014 than 2010, while Wisconsin’s elections

were similarly competitive in both years, yet we saw improved African-American turnout,

relatively speaking. This provides a limited counterexample to plaintiffs’ suggestions that

increased burdens were imposed, but turnout effects enabled plaintiffs to overcome them.

139. Much, if not all, of the increase in African-American participation during the

1990s and 2000s, as well as the increased use of early voting by African-Americans vis-à-vis

non-Hispanic Whites, appears to be a function of exogenous forces converging upon the state. To

draw firm conclusions about African-American voting patterns in North Carolina, one must at

least attempt to take into account the vast sums of money poured into the state, the historic nature

of the Obama candidacy, and the changing strategies of the parties.

140. In my opinion, it is impossible to do a proper analysis without at least attempting

to take these factors into account. This is basically conceded by plaintiffs’ experts, who typically

agreed that campaign effects played a role, and who now point to those effects as an explanation

for why African-American turnout increased in the wake of HB 589’s implementation. Yet none

41


has undertaken the sort of rigorous attempt to separate the effects of campaigns from the effects

of legal regimes that would be necessary here.

141. A failure to take account of these factors hopelessly complicates any attempt to

draw the conclusion that increased utilization of these laws represents a revealed preference

among African-Americans that will continue to manifest in future elections. It may well be that

these effects are mostly derivative of the fact that African-American voters were uniquely

enthusiastic about the Obama candidacy in 2008 and 2012, and that Democratic campaigns in

2008, 2012 and 2014 funneled unprecedented resources into North Carolina and other states in

an attempt to convince African-American voters (and Democratic voters more generally) to turn

out to vote and to vote early in particular, and that they were successful in doing so.

142. As the laws change, however, we would expect the campaigns to alter their

strategies. That appears to be what happened here.

143. This opinion functions in two fashions, both of which are important to understand

and to distinguish between. It first operates in a defensive sense against attempts to draw

conclusions regarding how African-American voting patterns in North Carolina might be tied to

the laws affected by HB 589. It does so by offering up theories of alternate causation. In other

words, it argues that such conclusions cannot be drawn until one at least attempts to deal with the

countervailing theories presented. It functions in this sense as a critique of plaintiffs’ experts’

approach. In the absence of rigorous attempts by plaintiffs to rule out these countervailing

theories, their claims become substantially weaker.

144. Second, it goes a step further and presents evidence that these theories of alternate

causation might well be correct. It shows that the increase in African-American turnout

42


witnessed in North Carolina was replicated nationally, yet does not seem to correlate nationally

with the availability of the laws at issue in this case in a state.

145. It is worth noting, however, that the second function is in many ways gratuitous,

and is not necessary for rebutting plaintiff’s claims.

146. Almost every study of the interaction between legal regimes and voting habits,

from Raymond Wolfinger and Steven Rosenstone’s seminal Who Votes? onward, includes at

least some attempt to control for campaign effects and to examine the effects of laws in a

national context. Without this, plaintiffs effectively offer a case study of North Carolina voting

habits, which is an insufficient basis from which to draw firm conclusions about voting in North

Carolina.

The Current Population Survey Data, and a Comparison of Two States

147. The problem with this transition from theory to practice is neatly summed up by

the succeeding four charts, which rely on data obtained from the Current Population Survey, or

CPS. The CPS is a monthly survey, conducted by the United States Census Bureau, which is

usually used to calculate the unemployment rate.

148. But every even numbered year, a few weeks after the general election, the CPS

asks respondents about their participation in those elections. The results are published in a

document with the title “Voting and Registration in the Election of [year].” These results are

available online at

https://www.census.gov/hhes/www/socdemo/voting/publications/p20/2012/tables.html. To

access other years, a researcher can change the year in the URL. In other words, if “2012” in the

URL is changed to “2008,” the 2008 results will be brought up.

43


149. The CPS has an en01m ous sample size, which allows for significantly more

granular analysis than researchers can obtain from other surveys, such as the exit polls or the

American National Election Study.

150. The CPS is admittedly imperfect, both because some respondents retum the

smvey but fail to answer the questions about voting, and because respondents tend to overstate

the rate at which they participate in elections. Some of these changes are likely systemic, albeit at

a national level. See, e.g., http://www.huffingtonpost.com/michael-p-mcdonald/20 12-tumout-

race-ethnict b 3240179.html.

151. fu addition, there is a break in the data series beginning in 1996, when the census

began to make available data for citizenship. But for generating cross-state comparisons, it is the

best dataset we have.

152. Consider the following two charts. They represent the share of the African

American population that is registered to vote, as well as the share of the African American

population that rep01ted patticipating in the 2012 elections in two states.

Percent of Africa n Americans Registered, State A and State B

100%

80% ~ ...............

, , - -, - _...,......_ -, ,

60% , - , --40%

20%

0%

1980 1988 1992 1996 2000 2004 2008 2012

- s tate A - - StateB

44


90%

80%

70%

60%

SO%

40%

30%

20%

10%

0%

Percent of African Americans Voting, State A and State B

---1980 1988 1992 1996 2000 2004 2008

- state A - - State B

2012

153. As you can see, beginning in the middle of the 1990s, state "B" catches up with

state "A" in tenns of African American registration and voting patticipation rates. After that ,

African American regisu·ation and voting patt icipation increase at roughly the same rates in both

states.

154. The following two chatts represent the satne data as the preceding two chatt s,

only with the states labeled:

100%

80%

60%

40%

20%

0%

Percent of Af rican Americans Registered, Mississippi and North Carolina

---

1980 1988 1992 1996 2000 2004 2008

- Mississippi - North Carol ina

45

2012


90%

80%

70%

60%

SO%

40%

30%

20%

10%

0%

Percent of Af rican Americans Voting, M ississippi and North Ca rolina

---1980 1988 1992 1996 2000 2004 2008

- Mississippi - - North Carolina

2012

155. The black line represents African American pruticipation in voter registration and

actual voting in the state of Mississippi. The dashed line represents African American

pruticipation in voter registration and actual voting in the state ofN01th Carolina.

156. During this time period, N01th Cru·olina greatly relaxed its resu·ictions on voter

regisu·ation and voting, and experienced an increase in African American pruticipation

subsequent to the enactment of these laws. E.g. , Steven F. Lawson, PI Rep01i ~~ 10, 11.

157. But dm1ng the srune time period, African American pruticipation increased in

Mississippi at a similar rate. Yet Mississippi is a state that does not count ballots cast out-of-

precinct, does not allow for eru·ly voting in excess of sixteen days -indeed it doesn't allow for

early voting at all- does not allow for srune-day voter regisu·ation, and does not allow pre-

regisu·ation at the age of sixteen.

158. In fact, the three states where black tmn out was the highest - Wisconsin,

Mississippi, and N01th Carolina (according to an att icle Dr. Leloudis cites to as evidence that

there is a causal linkage between North Cru·olina's laws and African-American tumout) - had

what we might tetm middling, low, and highly petmissive voting regimes in place, respectively.

46


Likewise, the states with the lowest percentage of African American voter turnout – Arizona,

Arkansas and Washington – have legal regimes that might be characterized as middling, low, and

middling in terms of permissiveness. http://www.thewire.com/politics/2013/05/black-turnout-

2012-state-by-state-maps/65053/, cited in Declaration of James Leloudis at n.85.

159. If plaintiffs’ experts’ theory is correct, what we would expect to see in such

circumstances is that African American participation would increase dramatically in North

Carolina, while lagging badly in Mississippi. Instead, despite two entirely different voting

regimes, both states experienced similar increases in African American registration and

participation.

160. The foregoing is not meant as a dispositive analysis, but rather illustrates the

difficulty with making the jump from theory to practice that plaintiffs’ experts attempt to make.

Why should we expect that, in the absence of the laws passed by the North Carolina legislature

around the beginning of the twenty-first century, African Americans in North Carolina would

behave any differently than African Americans in Mississippi?

161. These simple observations – that there are states where these sorts of laws have

not been implemented yet African-American participation has flourished, and that this might

interest the Court – have proved remarkably controversial with Drs. Stewart and Gronke. As the

first sentence in the preceding paragraph makes clear, the Court ought not draw firm conclusions

from this. It merely illustrates the dangers in drawing causal connections based on the

experience of a single state, and the need for engaging in looking at what occurred in other states

at the same time. Plaintiffs’ experts seem to largely agree on the broad point.

162. This is not a “useless exercise,” as Dr. Gronke suggests. Gronke SR at 3. To

understand the usefulness of this, consider the beginning of the Erickson and Minnite article on

47


voter identification. They relate an anecdote about the difficulties that a group of nuns

encountered while trying to vote. Robert S. Erikson and Lorraine C. Minnite, “Modeling

Problems in the Voter Identification-Voter Turnout Debate,” 8 Election L.J. 85, 85 (2009). The

point of the anecdote is not to prove definitively that it is hard to detect any effect of voter

identification laws simply because a group of nuns had difficulty voting. Rather, it illustrates the

broader point of the article and introduces the reader to the subject.

163. Dr. Gronke also suggests that I erred by not re-weighting the CPS data, in

accordance with the suggestion of Aram Hur & Christopher H. Achen, Coding Voter Turnout

Responses to the Current Population Survey, 77 Pub. Op. Q. 985:991-92 (Winter 2012).

164. I disagree, for four reasons. First, this is a new technique, published in late 2013,

shortly before the initial expert reports were made available. I am unsure whether Dr. Burden’s

recent article on the subject uses reweighted data, but other work does not appear to have used

reweighted data. See, e.g., Leighley & Nagler at 18-23; Robert S. Erikson & Lorraine C.

Minnite, Modeling Problems in the Voter Identification-Voter Turnout Debate, 8 Elec. L.J. 85:

99 (2009), yet is considered reliable. Second, dropping non-respondents assumes, in effect, that

they voted proportionally to the respondent population, which may or may not be a good

assumption, especially among key demographic groups. Likewise, reweighting the topline to

actual turnout numbers reweights all demographic groups equally, which may or may not be a

good assumption. Third, the “topline” CPS numbers, whatever their flaws, nevertheless represent

official numbers published by a government agency, and may be more appropriate to use in a

federal court.

165. Finally, as I explained at my deposition, the differences between the weighted and

re-weighted series tend to be systemic (in fairness to Dr. Burden, who questions my conclusion

48


in his sur-rebuttal, I did not explain this thoroughly in my initial report). For example, African-

American turnout in the District of Columbia drops more than ten points in each every year

between 2004 and 2012 when the data are reweighted. But it tends to drop at a relatively

constant rate; the standard deviation of the drops is about 2.8 points for both whites and blacks in

DC. Much of this large drop in turnout that appears when we reweight would simply shift the

lines upward or downward on the chart, or would subtract out in the regression analysis. See also

Leighely & Nagler 20 (“We have no reason to believe that misreporting rates are not stable over

time.”)

166. Dr. Gronke notes that, using 2000 as a starting point, he sees different rates of

change between North Carolina and Mississippi. Gronke rebuttal p6. I chose 1996 because it

antedated the introduction of no-excuse early voting in North Carolina. But again, the point here

is not to create a dispositive analysis. Had Dr. Gronke gone with 2004 as his baseline, he would

have found the differences to be nearly identical (124.1% for Mississippi versus 124.7% for

North Carolina). But in any event, this is exactly why one needs to conduct a more detailed

analysis.

167. It is true, as Dr. Gronke suggests, that Alabama looks very different from North

Carolina or Mississippi, despite not having implemented any reforms. There are also many

reasons that the turnout rates between Mississippi and North Carolina could be similar. This is

true – in fact, it proves the broader point that we need to test many states with various controls.

168. One could also layer the District of Columbia over the map as an example of a

jurisdiction that implemented several reforms, yet did not have much of an increase in African-

American turnout. The point is, that you can, and should, continue layering states, until you

have them all on your chart. Then you should add controls, as Dr. Gronke suggests, and see what

49


the resulting line looks like. This is effectively what we do when we conduct a regression

analysis. What Dr. Leloudis and others have done, however, is simply place North Carolina on

the chart, and declare that the laws from the 2000s have helped improve turnout. This is

insufficient.

169. Again, the point of the Mississippi example is to illustrate for the Court why we

can’t simply look at the trend in North Carolina and assume that the increased African American

turnout that correlates with the passage of these laws implies a causal relationship. We have to

do more to draw these sorts of conclusions. We do not seem to disagree here.

170. It is simply untrue to claim, as Dr. Stewart does, that I “draw[] conclusions about

the relationship between the ‘permissiveness’ of voting regimes by conducting an informal

review of African American turnout rates compared to an informal classification of states as

“middling, low, and high” on a scale of voting regime permissiveness.” Stewart SR ¶62.

171. In fact, my declaration expressly disclaims any attempt to draw firm conclusions

in the paragraph immediately following the one Dr. Stewart cites. See Trende Declr. ¶ 80 (“The

foregoing is not meant as a dispositive analysis, but rather illustrates the difficulty with making

the jump from theory to practice that plaintiffs’ experts attempt to make.”).

172. I employ a descriptive approach at this point in my analysis to rebut a descriptive

approach used by a different plaintiff’s expert. The reference to the top three and bottom three

turnout states is followed by a citation to The Atlantic Wire. The citation includes a reference to

the Leloudis Declaration. If a careful reader then followed the citation to the Leloudis

declaration, he or she would then find Dr. Leloudis arguing: “These reforms opened the way for

black turnout to soar to historic highs in the 2008 and 2012 elections, when voters rallied behind

the candidacy of Barack Obama, who would become the first African American President of the

50


United States. In 2012, according to data compiled by The Atlantic Wire, just over 80 percent of

eligible blacks in the state cast a ballot, the second highest percentage in the nation, after

Mississippi at 82.4 percent.” Id. at 31.

173. Dr. Leloudis is drawing a causal connection between North Carolina’s vote

reforms and the fact that North Carolina had the second-highest African American turnout in the

country, which is a powerful rhetorical point. It was therefore important to point out that the state

with the highest African American turnout, Mississippi, employed none of the reforms that

plaintiffs would have this Court compel North Carolina to adopt, and also to point out that the

state with the third-highest African American turnout adopted only some of the laws that North

Carolina adopted. At the same time, the states at the “bottom” employed some of the laws that

North Carolina adopted. This is also a useful illustration because it paves the way for the more

rigorous investigation derived from the regression analysis.

174. It may be true that coding Wisconsin as moderately restrictive “would surprise

most experts on election administration.” Stewart SR ¶63. While this may be true, most experts

on election administration work in academic settings and are primed to think of these states in

terms of generalized election laws, and not solely in terms of the laws that plaintiffs have

narrowed this case down to. In terms of the laws relevant to this case, Wisconsin is middle-of-

the-pack.

175. Finally, Dr. Burden notes that the Senate factors would apply differently in

Mississippi and other states. Burden SR at 4. I do not believe that a report filed in Mississippi

would make substantially different claims about the status of African-Americans in Mississippi.

But the point here was not to dispose of any issue.

Historic African American Participation Nationally, and within North Carolina

51


176. The increased participation of African American voters in North Carolina has to

be viewed in the national context, as well as in the broader context of the state.

177. The story of African American voting participation in the United States is one of

gradual improvement and increasing parity with non-Hispanic white voting participation.

178. Figure 7 displays the rate at which African Americans participated in United

States presidential elections, from 1972 through 2012, according to CPS data. Prior to 2000, the

data represent voting as a share of the adult population, while afterwards, the data represent

voting as a share of the adult citizen population.

179. Participation is roughly flat from 1972 through 1996. Then, beginning in 1996, it

begins to climb upward gradually, improving from 51 percent in 1996, to 57% in 2000, to 60%

in 2004, to 65% in 2008, to 66% in 2012.

180. During this time period, African American participation has converged on the

non-Hispanic white participation nationally. But that convergence has likewise been gradual.

According to the CPS data, African American voting participation rates were 82% of those of the

52


non-Hispanic white population in 1984, 87% in 1988, 85% in 1992, 90% in 1996, 92% in 2000,

89% in 2004, 98% in 2008, and 104% in 2012.

181. The steady correlation between African Americans’ convergence upon non-

Hispanic white levels of participation is tight. If one conducts a simple regression analysis with

year as the independent variable and the African American participation rate as a share of the

non-Hispanic white participation rate, one explains 81.5% of the variance, and the time variable

is highly significant (p=.001); it is, if you will, outside the error margin.

182. The increase in African American participation in North Carolina is likewise a

gradual phenomenon, as illustrated by Figure 6 above.

183. Figure 9 illustrates African American voting rates in North Carolina and in the

United States of America.

53


90%

80%

70%

60%

SO%

40%

30%

20%

10%

0%

African America n Voter Participation Nationally and North Carol ina

,. ,. - --1

> , -- --- "' --

1980 1988 1992 1996 2000 2004 2008

- Nationally - - North Carolina

,_

2012

184. The two trendlines largely move in tandem. Eight-four percent of the movement

in the rate of African American voter participation in N01ih Carolina can be explained simply with

respect to national trends.

185. Dr. Gronke states " [h]ere we leam that African-American pmiicipation has been

increas ing over time nationally as well as in N01i h Carolina." Gronke sur-rebuttal at 4. He

continues "This is not a multivm·iate analysis that attempts to conu·ol for other vm·iables that

could potentially affect tumout, confmmding his analysis. The national u·end does not "explain"

the N01ih Cm·olina u·end, if by "explain" Mr. Trende means "causes," as per conventional

scientific usage. !d. at 5.

186. To be clear, when I say "explains the variance" as in pm·agraph 181, or when I say

"explains" a few paragraphs later, I am not asse1i ing causation, as plaintiffs' expe1is suggest, but

am rather refening to the r-squm·e, which is also "variance explained." See Thomas H.

Wonnacott & Ronald J. Wonnacott, Introduct01y Statistics for Business & Economics, 4th ed.

497 (1990) ("Since R2 gives the prop01i ion of total SS that is explained by all the regressors, it

measures how well the multiple regression fits the data."); William E. Griffiths, R. Calier Hill &

54


George G. Judge, Learning and Practicing Econometrics 253 (1993) (“Given these definitions, a

measure of the proportion of variation in y explained by x is [formula omitted] where R2 is called

the coefficient of determination. The closer R2 is to one, the better the job we have done in

tracking yt . . .”).

187. The inference to be drawn here is not “that because African American

participation rates have grown steadily over time, it is impossible for election laws to have had

an influence on rising African American participation.” Stewart Declr. ¶ 65. He amply

demonstrates over three pages, that this would be an inappropriate inference to draw. He simply

reads too much into this section. A conclusion that it was impossible for election laws to have

contributed to the convergence of African-American participation on non-Hispanic white

participation would be a critical one for this case. It would have been stated explicitly. This is

not my conclusion.

188. The inference one is actually supposed to draw from this regression analysis is

simply that African-American voter participation did, in fact, converge on non-Hispanic white

participation over time, and that this was a gradual development. It is easy to look at a line chart

and see trends that are not there. More importantly, it is easy to look at a line chart and see

stronger trends than are really there. The purpose of the time series regression and data supplied

was to demonstrate the strength and “steadiness” of the correlation. Had we found weaker

results, we might conclude that our mind was creating a pattern that the data didn’t support.

North Carolina’s Emergence As a Target State

189. While the changes to North Carolina’s voting laws were taking place, outside

forces were conspiring to alter the makeup of the North Carolina electorate. To the surprise of

many, North Carolina became a swing state.

55


190. This can be best analyzed by examining the state’s Partisan Index. By Partisan

Index, I mean how Democratic or Republican the state is, when compared to the country as a

whole.

191. A state’s Partisan Index is computed by subtracting the share of the state that

voted for the Republican presidential candidate from the share of the nation that voted for

Republican presidential candidate. For purposes of these calculations, third parties are excluded.

192. To illustrate the utility of the Partisan Index, consider the following example. In

1984, Ronald Reagan won 51.4% of the two-party vote in Massachusetts. In absolute terms, one

could consider Massachusetts a swing state. But no one would have considered Massachusetts a

swing state, because it had two Democratic senators, a Democratic governor, and an

overwhelmingly Democratic legislature. Ten of the state’s eleven congressional districts elected

Democrats, and the one Republican, Silvio Conte, was very liberal Republican.

193. Moreover, one would conclude that, using absolute terms, the state has swung

wildly toward Democrats in the interim, since Barack Obama won 61.8% of the two-party vote

in the state in 2012.

194. But Reagan’s 51.4% win in Massachusetts has to be viewed in the context of his

winning 59.2% of the two-party vote nationally. Compared to the country as a whole,

Massachusetts actually had a Democratic lean of 7.8 points in 1984.

195. Likewise, Obama’s 61.7% win in Massachusetts has to be viewed in the context

of his winning 52% of the two-party vote nationally. Compared to the country as a whole,

Massachusetts actually had a Democratic lean of 9.8 points in 2012. Viewed in this light,

Massachusetts has actually had relatively stable politics since 1984, with only a slight shift

toward Democrats.

56


196. Figure 10 shows the Partisan Index of North Carolina from 1972 through 2012.

197. In 1976, and again in 1980, the state was roughly four points more Democratic

than the country as a whole. Over the next four cycles, the state gradually trended toward

Republicans. By 1996, the state was 7.3 points more Republican than the country as a whole.

198. But in 2000, the state began moving back toward Democrats; it was only 6.7

points more Republican than the country as a whole that year. It did so despite the fact that

neither the Bush campaign nor the Gore campaign spent significant sums of money in the state.

See David B. Magleby, ed., Financing the 2000 Election, 98.

199. This movement was noticed by Democrats and political demographers, who

speculated that the growth of suburbs around Charlotte and especially in the Research Triangle

Park area would make the state increasingly fertile ground for the Party of Jackson. John Judis

and Ruy Teixeira, writing in 2002, observed that Democrats like Jim Hunt and John Edwards

were creating a new rapport with upscale voters in what they called “postindustrial areas” or

“ideopolises.” As they wrote “[s]ince 1988, all these areas have become more Democratic.

Dukakis lost Mecklenburg County 59-40 percent in 1988, but Gore lost it only 51-48 in 2000,

57


even though he did not campaign in the state. The Democrats’ edge in Durham County increased

from 54-45 percent to 63-35 percent over the time period.” John Judis & Ruy Teixeira, The

Emerging Democratic Majority 111 (2002). They predicted that over the course of the next

decade, the state would lean Republican, but would become very competitive. Id. 116.

200. In part because of the popularity of this book, Democratic interest in the state

increased. In 2004, John Kerry placed Sen. John Edwards of North Carolina on the Democratic

ticket as his running mate. There was some hope that Edwards would make the state competitive.

See http://mediamatters.org/research/2004/07/07/cnn-read-minds-not-polls-wrote-off-kerry-

edward/131397. Ultimately, neither the Bush campaign nor the Kerry campaign spent on

advertising in North Carolina.

http://edition.cnn.com/ELECTION/2004/special/president/campaign.ads/. Nevertheless, the state

continued to move toward Democrats, voting only five points more Republican than the country

as a whole.

201. In 2008, Barack Obama invested heavily in the state of North Carolina, both in

the crucial Democratic primary and in the general election. Obama spent over $15 million in

advertising, compared to just $3.8 million in advertising by the McCain campaign.

http://www.cnn.com/ELECTION/2008/map/ad.spending/

202. At the same time, the Obama campaign invested heavily in on-the-ground voter

mobilization efforts. Across the country, the Obama campaign established nearly twice as many

field offices as the McCain campaign. Seth E. Masket, “Did Obama’s Ground Game Matter? The

Influence of Local Field Offices During The 2008 Presidential Election,” 73 Public Opinion

Quarterly, 1026 (2009). These offices ultimately had a statistically significant effect on

Democratic turnout, sufficient to flip the state of North Carolina. Masket at 1032.

58


203. These offices focused on local registration drives and on early voting in

particular, in an attempt to “bank” votes for Election Day. Jim Messina, Obama’s then-chief of

state, in a discussion about the campaign’s concerns about holding big rallies in the wake of John

McCain’s “Britney Spears” commercial, commented “[f]inally, at the end of September we got

back to saying, ‘Look, we’re gonna do this again because we need to push early voting,’ and if

you’re gonna push early voting and voter registration you’ve gotta do big events.”

http://www.newyorker.com/reporting/2008/11/17/081117fa fact lizza?currentPage=all.

204. This was part of what national field director Jon Carson called the “Starbucks

strategy.” As he explained “[w]e wanted offices everywhere because, despite all the focus on the

online network, the truth of the matter was where we had offices and volunteers were working

together with staff, we got work done.” Kathleen Hall Jamieson, Electing the President 2008, 43

(2009).

205. Carson continued “Our Get out the Vote (GOTV) efforts were the culmination of

this empowerment strategy. We went after weak voting Democrats wherever they were. In

southeast Ohio, we were knocking on doors in precincts that had never seen it before. Our

organization allowed us to do that.” Id. 45.

206. Early voting in particular was a target of the Obama campaign. Carson stated

“[i]f you’re a fired up 64-year-old grandmother, and we told you how to vote early, you’re going

to make sure your sporadic voting kids and grand kids are doing it too. So that was the strategy.”

Id. at 46.

207. Notably, in states that didn’t have early voting, the Obama Administration

employed different strategies to get out the vote, such as traditional advertising, in an attempt to

stave off any late shift in voting patterns. Id. at 47.

59


208. The Obama campaign also worked diligently at registering new voters, African

Americans in particular. E.g., http://www.washingtonpost.com/wp-

dyn/content/article/2008/04/27/AR2008042702272.html,

http://usatoday30.usatoday.com/news/politics/election2008/2008-11-02-vote_N.htm?csp=34,

209. As a result of these efforts, North Carolina, compared to the country as a whole,

moved further toward the Democrats. Although it remained 3.5 points more Republican than the

country as a whole, the state of the economy and widespread discontent with the Bush

Administration created a national atmosphere that enabled Obama to win a state, in absolute

terms, that was 3.5 points more Republican than the country as a whole.

210. In 2012, both campaigns spent heavily on advertisements in North Carolina.

Republicans spent $57 million, while Democrats spent $40 million.

www.cnn.com/ELECTION/2008/map/ad.spending/. This time, Republicans notched a victory in

the presidential election. Nevertheless, once national effects are taken into account, the state

continued its march toward Democrats. The state leaned toward Republicans by three points.

211. The shift in North Carolina voting patterns cannot be evaluated in a vacuum, and

cannot readily be chalked up to the increasingly liberal voter access laws enacted during this

time. The shift in North Carolina was predicted by demographers without respect to these laws,

begins prior to the full enactment of these laws, and continues even after they are fully in place.

212. Instead, the movement of North Carolina toward Democrats – as well as the

increased African American participation in elections – is a complex phenomenon, which owes

much to demographic trends among whites in urban areas, the decisions of the various parties

and campaigns to target the state, and to target certain sorts of turnout at issue in this case in

particular. It is not simply that African Americans expressed a preference for early voting. The

60


Obama team knew that a vote that was cast could never be revoked, and that “banking” early

votes was the most surefire way to keep that from occurring. Because African Americans are

one of the most heavily Democratic constituencies in the country, this was one of the substantial

foci of Team Obama’s efforts.

213. Thus, these effects are present even in states without liberal voter laws. In

Georgia, for example, political scientists have concluded that “the 13% increase in black turnout

(see Table 2) essentially closed the racial gap in voting, and these high rates of mobilization

speak to the significance of the opportunity to elect the first African American president.” Seth

C. McKee, M.V. Hood, III, and David Hill, “Achieving Validation: Barack Obama and Black

Turnout in 2008,” State Politics & Policy Quarterly (12:1), 16.

214. This dovetails with what we observed above: The increase in African American

participation in North Carolina cannot be divorced from national effects. It was broad based, and

was largely commensurate with national and historic trends.

215. Dr. Gronke responds that “[m]oreover, contrary to Trende’s claims, trends in

early voting cannot be explained by factors such as “demographic trends” among “whites in

urban areas.”” Gronke SR ¶ 44. This is not my claim, and his responses that follow are therefore

irrelevant.

216. In effect, I am claiming that the theory of plaintiffs’ experts is essentially

“underdetermined.” To put it differently, if one explores African American turnout and

registration patterns in presidential elections, but neglects to explore the impact of the increased

competitiveness of presidential races in the state, the tens of millions of dollars that began to

pour into the state, or the fact that the first African American presidential candidate was on the

ticket, who specifically targeted minority voter turnout through early voting processes in the

61


state, then one has not adequately explored African American turnout and registration patterns. It

is plaintiffs, not defendants, who bear the burden of putting together an analysis that adequately

controls for such factors.

Cross-State Comparison

217. Perhaps the best way to summarize the difficulty in conducting an analysis that

pays little attention to outside effects and the potential change in behavior of voters in response

to different incentives is to look at African American turnout in every state with respect to the

various laws that plaintiffs would require the State of North Carolina to adopt.

218. Figure 12 lists the states for which the CPS has an unbroken data series of African

American voter participation between 2000 and 2012. The second column – by which the data

are sorted – shows the increase in African American voter participation between 2000 and 2012.

The third column shows the number of laws that plaintiffs would require the State of North

Carolina to adopt that were in place in each state. Because so many voter identification laws

have been enjoined or were only passed in the wake of the 2012 elections, such laws are

excluded from this analysis.

State Change in African American

Participation, 2000-2012 Number of Laws Adopted Alabama 5.4% 0 Arizona 12.0% 1

Arkansas -3.2% 0 California 6.6% 2 Colorado 6.5% 1

Connecticut 9.6% 1 DC 5.0% 3

Delaware 8.5% 1 Florida 4.7% 1 Georgia 11.4% 2 Illinois 3.3% 2 Indiana 16.7% 1 Kansas -8.8% 1

62


Kentucky 4.3% 0 Louisiana 6.7% 1 Maryland 2.4% 2

Massachusetts 6.1% 0 Michigan 1.7% 0

Minnesota 17.0% 1 Mississippi 24.0% 0 Missouri -2.4% 0 Nevada 21.2% 0

New Jersey 16.1% 1 New York 12.8% 0

North Carolina 32.3% 4 Ohio 16.4% 1

Oklahoma 10.3% 0 Pennsylvania 2.0% 1

South Carolina 8.1% 0 Tennessee 8.5% 0

Texas 4.5% 0 Virginia 13.3% 0

Washington 9.7% 2 Wisconsin 13.0% 1

219. North Carolina witnessed the largest increase in African American participation

since 2000. But the states that experienced the second- and third- largest increase in African

American participation had adopted none of the laws that plaintiffs would require North Carolina

to adopt.

220. Note too that Indiana had the fifth-largest increase in African American

participation in the nation during this time, despite the implementation of a voter identification

law that many feared would substantially inhibit African American participation. E.g., Brief of

Amicus Curiae, United States Congressman Keith Ellison, in Support of Petitioners, Crawford v.

Marion County Election Bd. (Nov. 12, 2007). In fact, from 2004 to 2008, Indiana’s African

American participation rates increased 7.1%, outstripping both North Carolina’s increase from

that cycle (3.7%) and the national average (4.7%). They increased another 7% from 2008 to

63


2012. As Dr. Hood will doubtless explain, Georgia saw increased African-American turnout

despite the imposition of a “strict” identification law.

221. But rather than eyeball the data, the best way to test for any effects here is to

conduct a regression analysis. A simple regression does reveal a positive correlation between the

number of laws a state passes and the increase in African American participation between 2000

and 2012. The effect is small, however, and is not statistically significant (p=.17). If you will, it

is within the “margin of error.” Nor is much of the variance explained (r2=.028). In plain English,

there isn’t a statistically significant relationship between the number of laws and the change in

African-American turnout over time.

222. But adding in a few basic controls weakens, rather than strengthens, the

relationship between the liberality of voting laws and the increase in African American turnout.

For example, drawing on my experience as a psephologist and knowledge of United States

elections, I classified states as either contested or uncontested in both 2000 and 2012. States that

were contested in 2000 but uncontested in 2012 (Arkansas, California, Missouri, New Mexico,

Oregon, Tennessee and Washington) were coded as -1, while states that were contested in 2012

but not in 2000 (Colorado, Nevada, North Carolina and Virginia) were coded as 1. The

remainder was coded as zero.

223. Re-running the regression with this control in place reduces the significance of the

“laws” variable further (p=.232), while target state status becomes highly significant (p=.0087).

The r2 jumps to .199.

224. Moreover, a state that had lower African American turnout in 2000 necessarily

had more room to grow, and might be expected to enjoy more of a surge in turnout. If we add a

control for African American turnout in 2000, both it and the target state variable return as

64


statistically significant (p=.014 and p=.012, respectively). The “laws” variable still is not

(p=.237). The r2 increases to .331.

225. Dr. Gronke suggests that perhaps we should adopt a lower test for significance.

Gronke Surrebuttal ¶ 38. But one of his fellow experts has rejected conclusions drawn at this

level, using a similar technique with a similar number of observations. Erickson & Minnite at

94-97 (finding that effects that include some accompanied by a p of .34, .3, .25, .31, and .29 are

“statistically inconclusive” and “not close to statistical significance”). I do not believe this

opinion would be accepted for peer review claiming a p of .237, as such results would be

considered unreliable.

226. Likewise, Dr. Gronke protests that the opinion is not suitable for peer review

because it layers multiple laws in a single ranking system. Gronke SR ¶30. The idea is that a

ranking variable assumes that going from zero laws to one law has the same effect as going from

three laws to four laws. This is a fair objection, and if anything, it adds additional uncertainty to

already-uncertain findings about the effects of voting laws. But it does not render the opinion

unreliable. In fact, the Erickson and Minnite piece, which was published in the journal Dr.

Gronke now edits, includes a 8-tier ranking of voter ID laws. It assumes that the difference

between a state that simply requires a voter to state their name and a state that requires the voter

to sign a book is the same as the difference between a state that requires non-photographic

identification and one that requires a photo identification. Erickson & Minnite at 90.

227. In paragraphs 8- 22, Dr. Stewart begins his critique of my approach by explaining

the general approach used by many social scientists to voting questions, following Wolfinger and

Rosenstone’s classic Who Votes? I found the inclusion of this discussion odd for three reasons.

65


228. First, none of plaintiffs’ experts conduct the sort of analysis that Wolfinger and

Rosenstone conducted. For example, Wolfinger and Rosenstone do not focus on a single state.

They rely upon CPS data, albeit at the individual level, to conduct the sort of cross-state survey

that I fault plaintiffs’ experts for failing to conduct. Examine Table 1 in Dr. Stewart’s

Surrebuttal. Note that Wolfinger and Rosenstone include variables regarding the presence of a

gubernatorial race on the ballot. They understood the necessity of at least attempting to control

for campaign effects, something for which plaintiffs’ experts fail to control. In other words,

under Dr. Stewart’s suggestion, no expert has conducted a proper inquiry into how voting laws

affect turnout here. The Court is left without any guidance as to how the laws in question will

ultimately affect actual turnout.

229. Second, and perhaps most importantly, Dr. Stewart never explains why it would

be improper to use aggregate data to explore the effects of these laws on voting in this

circumstance. In other words, he simply says “this is how it has always been done,” without

explaining why it must be done this way. Our concern here is narrower: what are the effects of

the laws at issue in this litigation on minority voting turnout. If plaintiffs are correct that these

laws impose a substantial, insuperable burden on minority voting, then we should see lower

minority turnout, all other things being equal, in states that have the legal regime that North

Carolina has in place. See also Erikson and Minnite at 89 (“If voter ID laws suppress turnout, the

relationship should be negative: increased voter ID requirements should be associated with lower

voting rates.”). That is precisely what my regression analysis sought to test, and Dr. Stewart

offers no compelling reason why it is inappropriate in this particular context.

230. Third, Dr. Stewart’s ultimate conclusion – that my approach fails to include the

hallmarks of the central research on the literature – is surprising, and even jarring. Stewart

66


describes those hallmarks as (1) a theoretical grounding in the idea of the “representative

citizen”; (2) disaggregation, that is, exploring the effects of individual election laws on turnout,

rather than lumping them together; and (3) a careful attention to appropriate statistical controls.

Stewart SR ¶¶12-14. This is jarring for two reasons. First, my core critique of plaintiffs’ experts’

approach is that they themselves have not paid careful attention to appropriate statistical controls.

There is little attention dedicated to examining what is occurring in other states that have both

similar and dissimilar laws. Nor is there much attention paid to campaign effects. As but one

illustration of how gross this deficiency is, Dr. Stewart’s Declaration and the Appendices total

213 pages of text largely dedicated to exploring African-American participation in North

Carolina elections, focusing mostly (though hardly exclusively) on the 2008 and 2012 elections.

Yet the word “Obama” appears only four times: Three times in an explanation of a regression

equation where he uses the president’s vote share as a measurement tool for the county’s

partisanship (Stewart Declr. ¶ 190, Exhibits at 163), and once in his bibliography (Exhibits at

17).

231. Stewart’s comments are also jarring because the approach used in my Report is

substantially similar to one used by a different plaintiff’s expert in a published work. In Voter ID

Laws and Voter Turnout, Robert Erikson and Lorraine Minnite argue that the use of aggregate

data should be preferred when measuring the impact of statewide laws, and that it would be a

“wrong path” to utilize the approach suggested by Dr. Stewart. They write: “We might even have

been tempted into using our 64,000-plus respondents as our units rather than our 51 states. It is

worthwhile considering how we would have been led astray.” Erikson and Minnite at 92.

232. They argue that if we are testing the effects of state policy choices on turnout,

using individual data artificially inflates the number of observations, thus artificially increasing

67


the statistical significance of our conclusions. Id. at 92. (“The whole approach [using individual

data], even with robust standard errors, is the wrong solution for dealing with our state-level

policy variable, as the standard errors are still seriously deflated compared to what they should

be.”).

233. Not only do they use aggregate data, but their approach is, at its core, more

“crude” than mine (to use Dr. Stewart’s word). Erikson and Minnite use a simple bivariate OLS

regression equation. Id. at 90 (“The setup then is a bivariate analysis for 50 state observations.

We perform OLS regression equations where the dependent variable is change in turnout. The

independent variable is the change in voter identification legislation, either as the change score

on the 0–7 scale or the presence or absence of change.”).

234. In plain English, they simply look at the changes in state voting laws between

2002 and 2006, place them on a scale of zero to seven, and see how these changes correlate with

changes in turnout levels. Erikson and Minnite do make use of individual CPS data, but only to

inform an alternative measurement of turnout, where the aggregated state turnout rates are

adjusted to conform to the expected values based upon their demographic profiles. But they note

that these demographic adjustments have little impact on their regression output. Id. at 93.

(“Still, the gains from the lesser variance turn out to be slight.”). See also Jan Leighly and

Jonathan Nagler, Who Votes Now? Demographics, Issues, Inequality and Turnout in the United

States (2014) (explaining the limitations of a logit or probit regression in this context).

235. Dr. Stewart asserts that there is a problem with my methodology because my

analysis only “pertain[s] to a limited number of states.” Stewart SR ¶62. The reason that certain

states are excluded is that the census only prints aggregate data for African Americans for a

limited number of states. This makes sense; some of these states include fewer than 20 African

68


Americans in their samples; the “error margins” there run on the order of +/-20%. Montana had

no African-Americans in its sample in 2012. Introducing datapoints with such a significant

amount of uncertainty behind them would unwise and could skew the regression analysis. I opted

to use the Census Bureau’s cutoff. More importantly, no expert has adduced any evidence

suggesting that the inclusion of these states would alter the results of the analysis.

236. Dr. Stewart asserts in paragraph 77 that “’[f]irst, if we are to understand whether

there are disparate effects of election laws based on race, we need a comparison of the relative

differences in black and white participation rates across states.” Dr. Stewart (Dr. Gronke also

makes this argument) would prefer that my analysis demonstrate that there is no differential

impact on African American turnout. My analysis instead goes a step further and demonstrates

that there is no statistically significant impact on African American turnout whatsoever.

237. In paragraph 78, Dr. Stewart states that I am not utilizing a multivariate approach,

and that I was wrong to apply the controls one-at-a-time. The controls are applied one-at-a-time

because of our unique context here: to illustrate for the Court the effect of applying these

controls. The final reported results include all of the controls.

238. In paragraph 79, Dr. Stewart complains that I did not code my independent

variable as a change in laws, instead referencing whether a law was on the books as of Election

Day. He offers many reasons for doing so, all of which I considered. But I opted to ignore these

theoretical reasons for a practical one: the Obama campaign utilized existing laws, such as early

voting, where available, in new ways. In other words, in this particular instance, it didn’t matter

when a state adopted the law, only that it was, in fact, on the books.

239. In a footnote, Dr. Stewart claims that political scientists define a competitive state

with reference to “how close a state’s presidential election returns are to a 50/50 split” (what I

69


call the “partisan index). A state may be close, however, but if a campaign does not spend money

on the state, there will be no campaign effects. There are numerous examples of races that appear

competitive by the outcome but were not treated as such by the parties or campaigns: for

example, the Virginia Senate race in 2014, the Maryland governor’s race in 2014 and the

Kentucky Senate race in 2004. Likewise, some races look substantially less competitive than

they really are: the North Carolina Senate race in 2008, the New Hampshire Senate race in 2010,

the Kentucky Senate race in 2010, and the Iowa Senate race in 2014 are all prime examples here.

240. Dr. Gronke believes that each law should be tested separately. First, with only 34

observations, that is probably unwise to have six independent variables. It is my understanding

that plaintiffs’ theory of the case is that the law has cumulative impact; testing the laws together

may well make the most sense in this context.

241. Dr. Gronke also argues that it would be better to use the change in number of

laws, rather than the number of laws. But that depends on how you think the laws work. My

theory is that Democrats, particularly the Obama campaign, began using already-existing early

voting laws to get their voters to the polls early, in an attempt to ensure that they did not change

their votes. Under that theory it would have been inappropriate to treat a state that introduced

early voting in the 1990s (like Colorado or Iowa) differently from a state that introduced it in

2008. Looking simply at the change in days, in other words, would assume that the increased

availability of early voting in North Carolina in the early 2000s had no impact on turnout in

2012. That may be true, but it is not an assumption that would be favorable to plaintiffs.

242. In the end, these are mostly judgment calls, of the type that anyone trying to

explain something like voting behavior with numbers has to make inevitably. But rather than

simply leaving it at that, we can take things a step further. We can take Dr. Stewart’s suggestion

70


and take the change in PVI from year-to-year as our measure of competitiveness. We can then

run the regression multiple ways: Using weighted and re-weighted data, using black turnout

alone and black turnout in terms of white turnout, using number of early voting days and the

length of the early voting period, and using laws as they exist on the final election day and using

the change from an early voting period. In the interest of time and space, we will run 2004 to

2008, 2008 to 2012, and 2004 to 2012. We will also test early voting alone, as it has by far the

highest alleged burden. Dr. Gronke is also correct that there is not, as of this time, enough

variance in the number of same day registration laws adopted in this time period to draw

conclusions when it is tested on an individual level. This gives us 48 regression analyses.

Adding 2000 to the dataset would increase it to 150, and running it for all four variables would

increase that to almost 500.

243. The point of the exercise is simply to illustrate that plaintiffs’ differing

conclusions do not alter ultimate outcomes. We can use any combination of their suggested

measurements, and there is still no statistically significant relationship present.

71


Version Re-

Weighted Bl or Bl/W Years

New only?

Days or Period?

Pos or Neg? coeff. t-stat

Regression (a) No Bl '04-'08 Yes Days Neg -2.90E-05 -0.016

Regression (b) No Bl '04-'08 Yes Period Neg -4.76E-06 -0.003

Regression (c) No Bl '04-'08 No Days Pos 0.0001173 0.90351

Regression (d) No Bl '04-'08 No Period Pos 1.84E-05 0.019

Regression (e) No Bl '08-'12 Yes Days Pos 0.0004192 0.217

Regression (f) No Bl '08-'12 Yes Period Pos 0.0007775 0.357

Regression (g) No Bl '08-'12 No Days Pos 0.0003263 0.249

Regression (h) No Bl '08-'12 No Period Pos 9.99E-05 0.085

Regression (i) No Bl 04-'12 Yes Days Pos 0.001048 0.709

Regression (j) No Bl 04-'12 Yes Period Pos 0.001021 0.743 Regression (k) No Bl 04-'12 No Days Pos 0.0006231 0.484 Regression (l) No Bl 04-'12 No Period Pos 0.0003911 0.339

Regression (m) No Bl/W '04-'08 Yes Days Neg -0.0002566 -0.088

Regression (n) No Bl/W '04-'08 Yes Period Neg -0.0003133 -0.108

Regression (o) No Bl/W '04-'08 No Days Pos 0.0004762 0.309

Regression (p) No Bl/W '04-'08 No Period Pos 0.0003498 0.232

Regression (q) No Bl/W '08-'12 Yes Days Neg -0.0007807 -0.255

Regression (r) No Bl/W '08-'12 Yes Period Neg -5.59E-05 -0.016

Regression (s) No Bl/W '08-'12 No Days Pos 0.001536 0.755

Regression (t) No Bl/W '08-'12 No Period Pos 0.001152 0.634

Regression (u) No Bl/W 04-'12 Yes Days Pos 0.002072 0.961

Regression (v) No Bl/W 04-'12 Yes Period Pos 0.001995 0.998

Regression (w) No Bl/W 04-'12 No Days Pos 0.001973 1.044

Regression (x) No Bl/W 04-'12 No Period Pos 0.001533 0.903

Regression (y) Yes Bl '04-'08 Yes Days Neg -0.00184 -0.845

Regression (z) Yes Bl '04-'08 Yes Period Neg -0.001915 -0.884

Regression (aa) Yes Bl '04-'08 No Days Pos

0.0002326 0.212

Regression (bb) Yes Bl '04-'08 No Period Pos

0.0001705 0.158

Regression (cc) Yes Bl '08-'12 Yes Days Pos 0.000887 0.433

Regression (dd) Yes Bl '08-'12 Yes Period Pos

0.001869 0.818

Regression (ee) Yes Bl '08-'12 No Days Neg

-0.0009847 -0.714

Regression (ff) Yes Bl '08-'12 No Period Neg -0.0008798 -0.717

Regression (gg) Yes Bl 04-'12 Yes Days Pos 0.002008 1.44

Regression (hh) Yes Bl 04-'12 Yes Period Pos

0.00176 1.347

Regression (ii) Yes Bl 04-'12 No Days Neg -0.0003864 -0.311

Regression (jj) Yes Bl 04-'12 No Period Neg -0.0005031 -0.453

72


244. As the table shows, in none of these situations do we achieve a statistically

significant effect. Note that the regressions that utilize the two criteria favored by plaintiffs’

experts (regressions (ll) and (pp)), the coefficient is positive in the first instance and negative in

the second; in neither instance does it approach significance.

245. Again, engaging in this regression analysis is somewhat superfluous for defense

purposes. Whatever flaws plaintiffs’ experts may believe it may have, it serves to illustrate the

sort of analysis in which plaintiffs ought to have engaged to prove a relationship between these

laws and turnout.

The Florida Example

246. Plaintiffs’ experts do offer one cross-state example, drawing parallels between the

consequences observed in Florida’s decision to shorten early voting and the likely outcome in

North Carolina.

247. While it is always risky to draw broad conclusions based upon a single

observation, there are additional reasons to believe that the Florida example is not apposite.

Regression (kk) Yes Bl/W '04-'08 Yes Days Neg -0.003436 -0.91

Regression (ll) Yes Bl/W '04-'08 Yes Period Neg -0.0036 -0.958

Regression (mm) Yes Bl/W '04-'08 No Days Pos

0.0005675 0.287

Regression (nn) Yes Bl/W '04-'08 No Period Pos

0.00045 0.233

Regression (oo) Yes Bl/W '08-'12 Yes Days Pos

0.001659 0.409

Regression (pp) Yes Bl/W '08-'12 Yes Period Pos

0.003316 0.732

Regression (qq) Yes Bl/W '08-'12 No Days Neg

-0.001517 -0.555

Regression (rr) Yes Bl/W '08-'12 No Period Neg -0.001258 -0.518

Regression (ss) Yes Bl/W 04-'12 Yes Days Pos 0.00363 1.286

Regression (tt) Yes Bl/W 04-'12 Yes Period Pos 0.003049 1.155

Regression (uu) Yes Bl/W 04-'12 No Days Neg

-0.0002987 -0.118

Regression (vv) Yes Bl/W 04-'12 No Period Neg -0.0004967 -0.219

73


248. The Florida example is most thoroughly explained in the reports of Drs. Gronke

and Stewart. E.g., Gronke Declr. ¶¶30-38. To summarize briefly, Florida shortened its early

voting time period before the 2012 elections, which may have contributed to a lengthening of

lines in early voting on Election Day. A working paper from Drs. Gronke and Stewart concludes

that the result of this was a decline in the number of votes cast during the early voting period,

from 2,663,995 to 2,380,165 in 2012. This decline disproportionately affected African

Americans.

249. Dr. Stewart notes that the decline did not actually come from the voters who

voted on the days where early voting was cut. These voters were the most committed voters.

Stewart Declr. ¶203. This makes intuitive sense – the most committed voters were the most eager

to vote, and turned out the earliest. When the early voting period was shortened, they were not

perturbed by this. It was the less-motivated voters who had waited until late in the cycle to vote

in 2008 who did not return.

250. Obviously sorting out causation is difficult to accomplish in a situation like this,

but we are left wondering: Why should we necessarily conclude that these less motivated voters

from 2008 stayed home because of the lines? See Stewart Declr. ¶ 206. Perhaps they stayed

home simply because they were less motivated. After all, 2012 saw lower turnout nationally that

in 2008.

251. Significantly, although early voting was down in Florida, overall voting was not,

suggesting an overall shift in voting strategies from early voting to Election Day voting. See

http://uselectionatlas.org/. This contrasts with a generalized decline in turnout nationwide:

Early versus Total Voting in Florida, 2008-2012

Early Voting

Florida Total Voting in

Florida Total Voting in

USA 2008 2,663,995 8,412,248 131,473,705

74


2012 2,380,165 8,492,175 129,232,106 Absolute Change -283,830 79,927 -2,241,599

Percent Shift -10.7% 1.0% -1.7% 252. Of course, Florida’s population grew during the interim (as did the United States’

population as a whole). Overall, voter participation was down in Florida. But the CPS data

suggest that the decline was concentrated among non-Hispanic whites, rather than African

Americans. From 2008 to 2012, participation was down 3.3% among non-Hispanic whites, but

only .9% among African Americans.

253. This suggests an alternate narrative. Rather than long lines incentivizing marginal

early voters to stay home, marginal early voters failed to vote because they were marginal voters,

and the dreary 2012 campaign failed to match fully the excitement of the historic 2008

campaign. In other words, the overall decline in African American participation in Florida early

voting was not a manifestation of longer lines, but was simply a manifestation of declining

interest overall. To the extent that it was a manifestation of longer lines, the voters who were

interested in voting shifted their voting strategies accordingly. Again, if plaintiffs’ overall theory

were correct, we should expect to see more significant declines overall among African

Americans than non-Hispanic whites, given the former’s supposed preference for early voting as

a means of exercising their franchise.

254. This is not to suggest that Drs. Gronke and Stewart are conclusively wrong. These

may be two distinct phenomena – one set of inputs may have led to the decline in participation

among African Americans while another set of inputs may have encouraged the decline in

participation among whites. This does, however, illustrate the difficulty in drawing broad

conclusions on the basis of a single observation.

75


255. Dr. Stewart claims that I have misrepresented his and Dr. Gronke’s research in

Florida. Dr. Stewart first claims that by referring to those who voted the earliest in 2008 as “the

most committed voters,” I have “disparage[d] the civic commitment of those who voted later in

the early voting period, as well as others who voted using other modes.” Stewart Surrebuttal ¶85.

It was not my intention to disparage voters’ sense of civic commitment, or anything else about

voters for that matter. With that said, Dr. Stewart clearly establishes that the voters who voted

the earliest in 2008 were the most likely to return to vote early in 2012, regardless of lines. Those

who voted later in 2008 did not return to vote early in 2012, and “many of the latest early voters

in 2008 failed to vote in 2012” at all. Stewart Declr. ¶ 204. While Dr. Stewart might prefer a

different word choice, given that both groups faced similar conditions, yet voted at different

rates, we can draw reasonable inferences about underlying motivation.

256. Dr. Stewart fails to read my declaration closely when he claims I claim that he

and Dr. Gronke “characterize voters who voted in 2008 but not in 2012 as ‘less motivated.’”

Stewart SR at ¶86. That is my suggestion, contrasted with the explanation he offers. Trende

Declr. ¶ 140 (“Obviously sorting out causation here is difficult, but the above does raise the

question: Why should we necessarily conclude that these less motivated voters from 2008 stayed

home because of the lines? See Stewart Declr. ¶ 206. Perhaps they stayed home simply because

they were less motivated to vote.”).

257. Notably, Dr. Stewart does not engage my core critique here: That despite all of

this, African American turnout fell less in Florida from 2008 to 2012 than did non-Hispanic

white turnout.

Conclusions

76


258. Three conclusions follow from this investigation. First, plaintiffs ask this Court to

force North Carolina to become the only state in the nation with the combination of vote reforms

they urge here.

259. Second, plaintiffs’ experts fail to control adequately for national effects in their

analysis, and fail to place North Carolina’s surge in minority participation during the 2000s in

any sort of context. We simply do not have sufficient data to conclude that North Carolina’s vote

reforms will significantly impact minority groups’ voting habits. Indeed, the similar experience

of other states with less permissive regimes suggests otherwise.

260. Third, the 2014 elections provide further reason to believe that these laws do not

negatively impact African-American turnout.

261. The point of this report is to place the conclusions of plaintiffs’ experts in a wider

context – both in terms of the elections that they analyze and the scholarly literature as a whole.

The real-world evidence in support of their theories is, at best, thin. Their conclusions about

North Carolina are effectively built around the effects of the Obama campaign (and now,

Democratic campaigns). When the overall data for North Carolina are placed in the context of

the conclusions of other scholars’ work, it is simply impossible to say with any reasonable

degree of certainty that the changes to North Carolina’s voting laws will decrease minority

participation.

77


Pursuant to 28 U.S. C. § 1746, I declare under penalty of perjury under the laws of the United

States of America that the foregoing statements are true and correct.

A This the J£. day of March, 2015.

78


General Information

Court United States District Court for the Middle District of NorthCarolina; United States District Court for the Middle District ofNorth Carolina

Federal Nature of Suit Civil Rights - Voting[441]

Docket Number 1:13-cv-00660

LEAGUE OF WOMEN VOTERS OF NORTH CAROLINA et al v. STATE OF NORTH CAROLINA et al, Docket No. 1:13-cv-00660

© 2015 The Bureau of National Affairs, Inc. All Rights Reserved. Terms of Service // PAGE 81

http://www.bloomberglaw.com/ms/public/document/Terms_of_Service