You are on page 1of 38

2010 TURNOUT: QUANTIFYING THE QUANDARY

How Voting in 2010 will be Unlike any Previous Midterm in Recent History and How it will be Exactly the Same

AUTHORS:
NIRMAL MANKANI, NEW ORGANIZING INSTITUTE
ETHAN ROEDER, NEW ORGANIZING INSTITUTE

SEPTEMBER 2010

CONTACT

ETHAN ROEDER
DIRECTOR OF DATA, TECHNOLOGY, AND ELECTION ADMINISTRATION
NEW ORGANIZING INSTITUTE
ETHAN@NEWORGANIZING.COM
CONTENTS
Introduction ............................................................................................................................................................................................ 3

Key Findings for 2010 .................................................................................................................................................................... 3

Key Findings For All MidTerm Elections .................................................................................................................................... 4

Drop-off Trends in All Midterm Elections ......................................................................................................................................... 5

Length of Residence .................................................................................................................................................................... 6

Age ................................................................................................................................................................................................. 10

Age and Length of Residence................................................................................................................................................ 13

Educational Attainment ........................................................................................................................................................... 13

Gender .......................................................................................................................................................................................... 15

Citizenship Type ........................................................................................................................................................................... 16

Hispanic Origin............................................................................................................................................................................. 17

African-Americans...................................................................................................................................................................... 19

Unmarried Women ..................................................................................................................................................................... 20

Union Members ........................................................................................................................................................................... 21

Partisanship ................................................................................................................................................................................... 23

Using Off-Year Elections to Predict Voting in 2010..................................................................................................................... 24

Enthusiasm Gap and Voting............................................................................................................................................................ 30

Methodology ....................................................................................................................................................................................... 35

Historical Rates of Participation – Data Sources and Assumptions ................................................................................... 35

Enthusiasm Gap – Data Sources ................................................................................................................................................ 36

Off-Year Elections – Data Sources and Assumptions ............................................................................................................ 37

Acknowledgements .......................................................................................................................................................................... 38

2
INTRODUCTION

Politics is an industry defined as much by blunders as successes. Turnout predictions are an effort to bring
structure to the madness of elections and, all too often, they only contribute to the confusion. It is said,
famously, that John Kerry exceeded his vote goal in Ohio in 2004 while Barack Obama fell short of his in 2008 –
a narrative that runs perfectly contrary to the actual results. The paradox of turnout is this: our predictions are
just as likely to be wrong as they are essential to running an effective campaign.

2010 is an election year no political analyst with any shred of humility would propose to fully understand. The
political winds are blowing with such force, already dislodging some of America’s most established elected
officials, that any prediction of the result of November 2nd is bound to be little more than a squinting-eyed
transcription of a newspaper from across a football field.

Will 2010 really be unlike any previous American election? Will all of the rules be broken; what are the rules,
anyway? Will African-American turnout drop precipitously to pre-2006 levels? Will every young person with a
social life avoid the polls like the plague? Or can we learn from past elections that, in fact, although many
things about 2010 will be utterly unique, there are some things we can still reliably predict?

This original research investigates the boundaries of what we can know about turnout in 2010. By analyzing
historical evidence and current indicators, we quantify the drop-off in voter participation among demographic
segments from presidential election years to midterm election years and evaluate how these trends will play
out in 2010. We explore how, if at all, the historic turnout figures from the presidential election in 2008 will alter
the landscape of voter participation in 2010. Using a wide range of data sources, we gain a better
understanding of historical turnout and drop-off among underrepresented groups of voters, the utility of the
enthusiasm question in surveys of public opinion to predict partisan turnout, and how turnout in the 2009 off-
year elections can inform us about 2010.

KEY FINDINGS

KEY FINDINGS FOR 2010

 We expect African-Americans to drop off at a higher rate in 2010 than in previous midterm elections. A
precinct-level analysis of the off-year elections in Virginia and New Jersey indicates that heavily African-
American precincts tended to drop off more in the 2009 off-year elections. We did not find similar
patterns in the 2005 off-year elections.

 We expect Democratic voters to drop off at a higher rate in 2010 than in previous midterm elections. The
same precinct-level analysis of the off-year elections in Virginia and New Jersey indicates that, on
average, precincts that support Democratic candidates at higher rates tended to drop off more in the
2009 off-year elections. We did not find similar patterns in the 2005 off-year elections.

 We cannot find a link between the “enthusiasm gap” and base turnout. The unprecedented
advantage that Republicans currently enjoy in self-reported enthusiasm potentially indicates a large
increase in support among all voters for Republican candidates this November, however we cannot
find a link between this enthusiasm gap and turnout. In other words, although we find a relationship
between the enthusiasm gap and partisan performance, the enthusiasm gap does not seem to predict
share of self-identified partisans among all voters. The context in which many commentators use the
enthusiasm gap – as a way to speculate about base turnout – is not supported by the evidence
available to us.1

It should be noted that, based on historical patterns, we do expect a decrease in turnout among self-identified
1

Democrats. However the enthusiasm gap does not offer us insight into the magnitude of this difference in 2010.

3
KEY FINDINGS FOR ALL MIDTERM ELECTIONS

 Young people tend to drop off at very high rates in midterm elections. Age is not only highly correlated
with turnout, but there has been a strong relationship between age and drop-off in midterm election
years. The average turnout rate for 18-24 year olds in midterm elections is about half of their turnout rate
in Presidential elections, whereas the average turnout rate in midterm elections for individuals over 65 is
approaching 90% of their turnout rate in Presidential elections. This suggests that young people may be
a high value turnout target for organizations dedicated to increasing civic participation among
underrepresented groups of people in midterm elections.

 Recent movers have also dropped off at very high rates in midterm elections. Individuals who have
lived in the same place for a longer period of time tend to drop off much less frequently in midterm
elections. We observe some of this difference in voter registration, where recent movers tend to register
to vote more during Presidential election cycles than they do during midterm election cycles. However,
even among just registered voters, length of residence is still highly correlated with midterm drop-off. For
organizations dedicated to increasing civic participation among underrepresented groups of people,
recent movers may be a high value target for both voter registration and turnout efforts in midterm
elections.

 African-Americans, Hispanics, unmarried women, and people with less than a bachelor’s degree tend
to drop off at higher rates in midterm elections than the general public. Each of these groups is also
underrepresented among all registered voters. These demographics may also be worthwhile targets for
organizations dedicated to increasing civic participation among underrepresented groups of people.
However, the magnitude of drop-off among these groups is not nearly as pronounced as it is among
young people and recent movers.

 On average, self-identified Republicans make up a larger share of all voters in midterm elections than
they do in Presidential elections. With the exception of 2006, since 1994 the Republican share of voters
has always increased in midterm elections and decreased in Presidential elections.

 Off-year elections such as those from 2005 and 2009 are indicators of drop-off in midterms. Our analysis
shows that voting trends observed from a presidential to an off-year (such as drop-off among voters in
Virginia from 2008 to 2009) are correlated with voting trends in the subsequent midterm election (in this
example, 2010).

4
DROP-OFF TRENDS IN ALL MIDTERM ELECTIONS

In order to understand how, if at all, 2010 may be unique from other midterm elections we must first quantify
what is consistent across all midterm elections and what is not. Have some demographic groups historically
voted at higher rates than others? Over time, are turnout rates changing at a faster rate for some groups than
others? Do certain demographic groups drop off at higher rates in midterm elections? For programs that
target underrepresented groups (such as young people or minorities), identifying which groups are very likely to
drop off in midterm elections is particularly relevant.

To assess these patterns, we use the U.S. Census’ Current Population Survey (CPS) and the National Election
Pool (NEP) exit polls from 1994 to 2008. Our selection of data sources and assumptions are explained in the
Methodology appendix.

TURNOUT AND DROP-OFF BY DEMOGRAPHIC GROUPS

One of the most reliable patterns in American politics is the decrease in voter participation from presidential to
midterm elections.

Figure 1.1: Turnout rates among the entire U.S. Citizen 18+ population in Presidential and midterm elections between
1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Over the eight elections between 1994 and 2008, average turnout during midterm elections is fourteen
percentage points lower than average turnout during Presidential elections2. Voter participation in Presidential
elections has been steadily, albeit modestly, on the rise during this period. Participation has gone up an
average of two percentage points per presidential election year since 1996. We do not see a similar pattern in
midterm elections in which turnout rates have remained relatively constant.

2 New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

5
Entire Citizen 18+ Population

Figure 1.2: Retention and average turnout by demographic group between 1994 and 2008. Source: New Organizing Institute
analysis of Census CPS Voting and Registration Supplement data.

Midterm retention (average midterm turnout rate ÷ average presidential turnout rate) for the entire citizen 18+
population is 76%.3 If retention among a demographic cohort is lower than 76% it means that cohort drops off
at a higher rate in midterm elections than the general voting eligible population.

Figure 1.2 displays retention and average turnout among every demographic group we studied. The lower on
the chart a group is, the more they tend to drop off in midterm elections. Individuals over the age of 45,
individuals who have a bachelor’s degree or higher, union members, people who have lived at their residence
for over five years, and men all tend to drop off less in midterm elections than the general population.

LENGTH OF RESIDENCE

People who have lived at their residence for a shorter period of time tend to both be registered and vote at
lower rates. For example, people who have lived at their residence between one and six months have been

3 The Methodology appendix discusses how “retention” is calculated and should be interpreted.

6
registered at an average rate of 55% and voted at a rate of 39% in the eight elections between 1994 and 2008,
whereas people who have lived at their residence over five years have been registered at an average rate of
82% and voted at a rate of 68% in the same period.

VOTING RATES AND DROP-OFF

Figure 1.3: Turnout rates by length of residence for the entire U.S. Citizen 18+ population in elections between 1996 and
2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Figure 1.4: Midterm voting retention rate (avg. midterm voting rate / avg. presidential voting rate) by length of residence
between 1998 and 2008 for U.S. Citizen 18+ population. Source: New Organizing Institute analysis of Census CPS Voting
and Registration Supplement data.

There is still a very strong relationship between length of residence and drop-off in midterm elections.
Individuals who have lived at their residence for a shorter period of time are far more likely to vote in a

7
Presidential election than a midterm election. The midterm turnout rate among people who have lived at their
residence for less than six months is about half of the turnout rate of the same group in Presidential elections,
whereas the turnout rate among people who have lived at their residence for five years or longer in midterm
elections is closer to 80% of that group’s Presidential year turnout rate.

REGISTRATION RATES

Figure 1.5: Average registration rate by length of residence for the entire U.S. Citizen 18+ population during
elections between 1996 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and
Registration Supplement data.

People who have recently moved are much less likely to be registered to vote regardless of the election year
(see Figure 1.5). This stands to reason as recent movers would almost always be required to take the active
step of re-registering at their new address.

Figure 1.6: Midterm registration rates by length of residence between 1998 and 2008 for U.S. Citizen 18+
population. Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Depressed registration rates among recent movers is even more pronounced in midterm elections (see Figure

8
1.6). This suggests that there is more at play than the mere inconvenience of re-registering at a new address;
perhaps recent movers are also less likely to be motivated by an upcoming election in a new state or county. It
is also possible, however, that the behavior of other demographic segments such as young people and single
people (who are more likely to move more frequently) are also being reflected in this measurement. Whatever
the underlying cause, these findings suggest that recent movers may be worthwhile registration targets during
midterm elections in terms of increasing voter participation among underrepresented groups.

DROP-OFF AMONG REGISTERED VOTERS

Figure 1.7: Average midterm voting retention rates (avg. midterm voting rate / avg. presidential voting
rate) from 1998 - 2008 by length of residence for registered voters only. Source: New Organizing
Institute analysis of Census CPS Voting and Registration Supplement data.

If we consider turnout rates among just registered voters, we still see large differences in midterm drop-off by
length of residence. In other words, even after accounting for the fact that recent movers are less likely to be
registered to vote, recent movers are still more likely to drop off in midterm elections.

9
AGE

Older people tend to vote at higher rates. For example, people aged 55 and up have voted at an average
rate of 66% in the eight elections between 1994 and 2008, whereas people aged between 18 and 24 have
voted at an average rate of 31% in the same period.

VOTING RATES AND DROP-OFF

Figure 1.8: Turnout rates by age among the entire U.S. Citizen 18+ population in elections between 1994 and
2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Figure 1.9: Midterm voting retention rate (avg. midterm voting rate / avg. presidential voting rate) by age
between 1994 and 2008 for U.S. Citizen 18+ population. Source: New Organizing Institute analysis of Census
CPS Voting and Registration Supplement data.

10
There is a very strong relationship between age and drop-off in midterm elections. Younger people are far
more likely to vote in a Presidential election than a midterm election. The midterm turnout rate among people
aged 18-24 is about half of the turnout rate of the same group in Presidential elections, whereas the turnout
rate among people aged 65+ in midterm elections is approaching 90% of that group’s Presidential year turnout
rate.

Also, interestingly, turnout has increased over time at a faster rate among the young. On average during this
period, turnout has increased 2.2 percentage points per election among individuals between the ages of 18
and 24, 1.52 percentage points per election among individuals between the ages of 25 and 34, and 0.8
percentage points or less per election among individuals 35 years old or greater.

REGISTRATION RATES

Figure 1.10: Average registration rate by age for the entire U.S. Citizen 18+ population during elections
between 1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration
Supplement data.

Younger people are less likely to be registered to vote, regardless of election year (see Figure 1.10).

11
Figure 1.11: Midterm registration retention rates by age between 1998 and 2008 for U.S. Citizen 18+
population. Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Even accounting for this difference, young people are even less likely to register to vote in midterm years than
they ordinarily would be in Presidential years (see Figure 1.11).

DROP-OFF AMONG REGISTERED VOTERS

Like length of residence, even when we consider turnout rates among registered voters, we still observe large
differences in midterm drop-off by age.

Figure 1.12: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting rate) by
age between 1994 and 2008 for registered voters only. Source: New Organizing Institute analysis of
Census CPS Voting and Registration Supplement data.

12
AGE AND LENGTH OF RESIDENCE

Because age and length of residence are the strongest predictors of drop-off, and highly correlated with
turnout, age and length of residence together may help identify individuals even more likely to drop off in a
midterm election.

Turnout among 18 – 24 year olds with a length of residence between 1 and 6 months averaged at a rate of
30%, but with a residency of 5 years or longer turnout among 18 – 24 year olds increases to an average rate of
41% (see Figure 1.13). Based on historical patterns, 18-24 year olds with 1-6 month residency are 73% as likely to
turn out as their counterparts with 5 years or longer residency.

Similarly, turnout among people aged 65+ with a length of residence between 1 and 6 months averaged at a
rate of 53%, but with a residency of 5 years or longer turnout among people aged 65+ increases to an average
rate of 74%. Based on historical patterns, people aged 65+ with 1-6 month residency are 72% as likely to turn
out as their counterparts with 5 years or longer residency.

Average Turnout by Age and Length of Residence, 1996 - 20084

1-6 months 7-11 months 1-2 years 3-4 years 5 years or longer
18 - 24 30% 30% 35% 37% 41%
25 - 34 39% 41% 49% 54% 54%
35 - 44 43% 47% 53% 61% 66%
45 - 54 46% 52% 56% 64% 72%
55 - 64 52% 56% 64% 70% 76%
65+ 53% 53% 61% 70% 74%
Figure 1.13: Average turnout by age and length of residence for U.S. Citizen 18+ population. Source: New
Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Length of residence and age combined are extremely predictive of midterm drop-off. Interestingly, within age,
length of residence as asked by the Census becomes less useful in identifying differences in midterm retention
(see Figure 1.14), but still has predictive power. In other words, as available to us, age is a stronger predictor of
drop-off than length of residence.

Midterm Voting Retention by Age and Length of Residence, 1998 - 2008

Figure 1.14: Midterm voting retention by age and length of residence for U.S. Citizen 18+ population.
Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

EDUCATIONAL ATTAINMENT

More educated people tend to vote at higher rates. People with a bachelor’s degree or higher have voted at
an average rate of 71% in the eight elections between 1994 and 2008, whereas people without a high school
diploma have voted at an average rate of 34% in the same period.

4 Length of residence less than one month excluded due to sample size constraints. Length of residence question not asked
in 1994.

13
VOTING RATES AND DROP-OFF

Figure 1.15: Turnout rates by educational attainment among the entire U.S. Citizen 18+ population in elections
between 1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration
Supplement data.

Figure 1.16: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting rate) by
educational attainment between 1994 and 2008 for U.S. Citizen 18+ population. Source: New Organizing
Institute analysis of Census CPS Voting and Registration Supplement data.

People with a bachelor’s degree tend to drop off at a lower rate than people with less education. Retention
for people with a bachelor’s degree or higher is 82%, whereas retention for people with less than a bachelor’s
degree is around 75%.

14
GENDER

Turnout across gender is very similar, but women vote at a slightly higher rate than men. Women have voted at
an average rate of 55% in the eight elections between 1994 and 2008, whereas men have voted at an
average rate of 53% over the same period.

VOTING RATES AND DROP-OFF

Figure 1.17: Turnout rates by gender among the entire U.S. Citizen 18+ population in elections between
1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration
Supplement data.

Although women vote at a higher rate than men (Figure 1.17), they also tend to drop off at a slightly higher rate
in midterm elections (Figure 1.18). Men have a midterm retention of 78%, but women have a midterm retention
of 75%.

Figure 1.18: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting rate)
by gender between 1994 and 2008 for U.S. Citizen 18+ population. Source: New Organizing Institute
analysis of Census CPS Voting and Registration Supplement data.

15
CITIZENSHIP TYPE

People who were born in the United States vote at a higher rate than naturalized citizens. U.S. born citizens
turned out to vote at a rate of 55% between 1994 and 2008, whereas naturalized citizens voted at a rate of 46%
over the same period.

VOTING RATE AND DROP-OFF

Figure 1.19: Turnout rates by citizenship type among the entire U.S. Citizen 18+ population in elections
between 1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and
Registration Supplement data.

Figure 1.20: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting rate)
by citizenship type between 1994 and 2008 for U.S. Citizen 18+ population. Source: New Organizing
Institute analysis of Census CPS Voting and Registration Supplement data.

Naturalized citizens also drop off slightly at a slightly higher rate than U.S. born citizens. Midterm retention
among U.S. born citizens is 77%, and retention among naturalized citizens is 74%.

16
HISPANIC ORIGIN

Hispanics have historically voted at lower rates than non-Hispanics. In the eight elections between 1994 and
2008, Hispanics have voted at a rate of 40%, whereas non-Hispanics have voted at a rate of 55%.

VOTING RATES AND DROP-OFF

Figure 1.21: Turnout rates by Hispanic origin among the entire U.S. Citizen 18+ population in
elections between 1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting
and Registration Supplement data.

Hispanics are also more likely to drop off in midterm elections than non-Hispanics. Hispanics have a midterm
retention of 70%, but non-Hispanics have a midterm retention of 77%.

Figure 1.22: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting
rate) by Hispanic origin between 1994 and 2008 for U.S. Citizen 18+ population. Source: New
Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

17
HISPANIC SHARE OF ELECTORATE

Despite voting at lower rates than the general public, Hispanics are growing particularly fast as a share of the
electorate (defined here as the universe of people who are eligible to vote). In 1994, Hispanics were 5.8% of all
people eligible to vote. By 2008, their share of the electorate increased to 9.5% (see Figure 1.23).

Figure 1.23: Hispanic share of eligible voters and share of voters between 1994 and 2008 for U.S. Citizen 18+
population. Source: New Organizing Institute analysis of Census CPS Voting and Registration Supplement
data.

We see a similar pattern when we examine Hispanic share of voters over time (see Figure 1.24). In 1994,
Hispanics were 4.2% of all voters. By 2008, their share of voters increased to 7.4%.

Figure 1.24: Hispanic share of voters between 1994 and 2008 for U.S. Citizen 18+ population. Source: New
Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

18
AFRICAN-AMERICANS 5

African-Americans have historically voted at lower rates than the general population. In the eight elections
between 1994 and 2008, African-Americans have voted at a rate of 50%, whereas the Citizen 18+ universe has
voted at a rate of 55%. However, in 2008, the African-American turnout rate exceeded that of the general
population.

VOTING RATES AND DROP-OFF

Figure 1.25: Turnout rates for African-Americans among the entire U.S. Citizen 18+ population in elections
between 1994 and 2008. Source: New Organizing Institute analysis of Census CPS Voting and Registration
Supplement data.

5The Current Population Survey changed possible responses to the race question in 2003, so results from 2004-2008 are not
directly comparable with results from 1994-2002.

19
Figure 1.26: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting
rate) for African-Americans between 1994 and 2008 for U.S. Citizen 18+ population. Source: New
Organizing Institute analysis of US Census CPS Voting and Registration Supplement data.

African-Americans tend to drop off in midterm elections at a higher rate than the general population. African-
Americans have a midterm retention of 70%, while retention among the Citizen 18+ universe is 76%.

UNMARRIED WOMEN

Unmarried women have historically voted at a lower rate than the rest of the population. In the eight elections
between 1994 and 2008, single women have voted at a rate of 47%, whereas everyone else has voted at a
rate of 56%.

VOTING RATES AND DROP-OFF

Figure 1.27: Retention for single women between 1994 and 2008 for U.S. Citizen 18+ population. Source:
New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Although the difference in turnout rates between unmarried women and everyone else is typically near eleven
percentage points, in the past two Presidential elections the gap has been closer to six percentage points.

20
Figure 1.28: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting
rate) for single women between 1994 and 2008 for U.S. Citizen 18+ population. Source: New
Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Unmarried women are also more likely to drop off in midterm elections than everyone else. Unmarried women
have a midterm retention of 70%, but everyone else has a midterm retention of 78%.

UNION MEMBERS

Union members have historically voted at a higher rate than the rest of the population. In the eight elections
between 1994 and 2008, union members have voted at a rate of 63%, whereas everyone else has voted at a
rate of 52%.

VOTING RATES AND DROP-OFF

Figure 1.29: Turnout rates by union membership among the entire U.S. Citizen 18+ population in
elections between 1994 and 2008. Source: New Organizing Institute analysis of Census CPS
Voting and Registration Supplement data.

21
Figure 1.30: Midterm voting retention rates (avg. midterm voting rate / avg. presidential voting
rate) by union membership between 1994 and 2008 for U.S. Citizen 18+ population. Source:
New Organizing Institute analysis of Census CPS Voting and Registration Supplement data.

Union members are less likely to drop off in midterm elections than everyone else. Union members have a
midterm retention of 80%, while everyone else has a midterm retention of 73%.

Other than in 2006, the gap in turnout rates between union members and everyone else has been steadily
decreasing. Turnout among union members has increased at a rate of .8 percentage points per election,
whereas turnout among everyone else has increased at a rate of 1.6 percentage points per election during the
same period.

22
PARTISANSHIP

The Census CPS does not ask respondents their partisanship. However, we can use the NEP exit polls to
estimate the composition of voters by self-identified partisanship for each election between 1994 and 2008.

Figure 1.31: Share of voters by partisanship between 1994 and 2008. Source: New Organizing Institute analysis of NEP exit
poll data.

Although the distribution of self-identified partisanship among the entire population, not just voters, changes
over time, we still see some evidence of partisan bias in midterm elections. With the exception of 2006, the
Republican share of voters has always increased in midterm elections and decreased in Presidential elections.
On average, self-identified Democrats are 38.4% of all voters in Presidential elections, but 37.1% of all voters in
midterm elections. On the other hand, on average self-identified Republicans are 34.6% of all voters in
Presidential elections, but 36.4% of all voters in midterm elections.

Overall, we find that key constituencies that typically support Democrats such as young people, single women,
African-Americans, and Hispanics all tend to drop off at higher rates in midterm elections. Finally, we find
evidence of a partisan turnout bias that favors Republicans in midterm elections more than it does in
Presidential elections. Democrats are more likely to be disproportionately hurt by midterm drop-off than
Republicans are.

23
USING OFF-YEAR ELECTIONS TO PREDICT VOTING IN 2010

Although general trends in drop-off are informative, we want to determine whether there is something unique
about turnout in the 2010 midterm elections beyond general historical patterns. Using polling data, such as the
enthusiasm gap, is one way to assess how 2010 turnout may be different from previous midterm elections.
Another approach to making inferences about turnout in 2010 is by examining drop-off in the 2009 off year
elections.

However, simply looking at drop-off from 2008 to 2009 is not useful, because we don’t necessarily know that
drop-off between 2008 and 2009 will help predict 2010 drop-off. Instead, we want to consider the elections in
2004, 2005, 2006, 2008, and 2009 together. If drop-off from 2004 to 2005 is in some way related to drop-off from
2004 to 2006, we will have more confidence that drop-off from 2008 to 2009 can be used to make inferences
about drop-off from 2008 to 2010.

We’re particularly interested in what drop-off among African-Americans will look like in 2010, especially in the
context of their unprecedented turnout rate in 2008. If you remember from Figure 1.25, 2008 was the first year
that African American turnout exceeded that of the general public. We would like to determine whether this
increase in participation will persist in future elections, or whether African-Americans will drop off at a higher
rate than they have in the past.

24
PRECINCT-LEVEL ANALYSIS

For the purposes of our analysis, we are considering precinct level turnout and election results from the 2004,
2005, 2006, 2008, and 2009 elections in New Jersey and Virginia. Data sources, assumptions, selection of states,
and how we calculate drop-off are discussed in the Methodology appendix.

First, we want to know whether drop-off from 2004 to 2005 can predict drop-off from 2004 to 2006 at the
precinct level. We plot precinct level drop-off from 2004 to 2006 against drop-off from 2004 to 2005 in Virginia
and New Jersey.
Correlation Between 04 to 05 Drop-off and 04 to 06 Drop-off – Virginia

Figure 3.1: Virginia – Drop-off from 2004 to 2006 vs. Drop-off from 2004 to 2005. When
we fit a linear model, P > |t| is less than 0.000, R^2 = .284. Source: New Organizing
Institute analysis of NCEC precinct level election data.

Correlation Between 04 to 05 Drop-off and 04 to 06 Drop-off – New Jersey

Figure 3.2: New Jersey – Drop-off from 2004 to 2006 vs. Drop-off from 2004 to 2005. When
we fit a linear model, P > |t| is less than 0.000, R^2 = .196. Source: New Organizing Institute
analysis of NCEC precinct level election data.

25
We find that there is a relationship, albeit noisy, between drop-off from 2004 to 2005 and drop-off from 2004 to
2006 at the precinct level. As expected, precincts with high drop-off from 2004 to 2005 tended to also have
higher drop-off from 2004 to 2006. This gives us more confidence that we can use 2009 drop-off at the precinct
level to speculate about 2010.

Next, we want to compare partisan drop-off from 2004 to 2005 to partisan drop-off from 2008 to 2009. To do
this, we create a generic measure of partisanship for each precinct by averaging their 2004 and 2008 support
for the Democratic nominee for President. Within each state, we rank order precincts by this generic
partisanship into quintiles. For each partisanship quintile, we calculate the average drop-off from 2004 to 2005
and compare it to the average drop-off from 2008 to 2009.

04 to 05 Drop-off by Precinct-level Partisanship Quintile – Virginia

Figure 3.3: Virginia – Average drop-off from 2004 to 2005 by partisanship quintile
(precinct level). Source: New Organizing Institute analysis of NCEC precinct level
election data.

04 to 05 Drop-off by Precinct-level Partisanship Quintile – New Jersey

Figure 3.4: New Jersey – Average drop-off from 2004 to 2005 by partisanship quintile
(precinct level). Source: New Organizing Institute analysis of NCEC precinct level
election data.

26
In both Virginia and New Jersey, drop-off from 2004 to 2005 appeared to be unrelated to whether precincts
tend to support Democratic candidates. We then turn our attention to drop-off between 2008 and 2009. If
drop-off in 2009 is different than drop-off in 2005 in both states, this is a hint that 2010 partisan drop-off may be
uniquely different than previous election years.

08 to 09 Drop-off by Precinct-level Partisanship Quintile – Virginia

Figure 3.5: Virginia – Average drop-off from 2008 to 2009 by partisanship quintile
(precinct level). Source: New Organizing Institute analysis of NCEC precinct level
election data.

08 to 09 Drop-off by Precinct-level Partisanship Quintile – New Jersey

Figure 3.6: New Jersey – Average drop-off from 2008 to 2009 by partisanship
quintile (precinct level). Source: New Organizing Institute analysis of NCEC precinct
level election data.

There does appear to be a relationship between 2009 drop-off and how Democratic precincts are. In general,
the more a precinct leans toward supporting Democrats, the more likely it was to drop-off in voting from 2008
to 2009.

27
We can also rank order precincts by the percentage of African-Americans who reside within them. Like
partisanship, we then compare this rank ordering to average drop-off from 2004 to 2005 and average drop-off
from 2008 to 2009.

04 to 05 Drop-off by Precinct-level Quintile of Percent African American – Virginia

Figure 3.7: Virginia – Average drop-off from 2004 to 2005 by quintile of percent
African American at the precinct level. The first quintile averages 1% African
American, the fifth quintile averages 61% African American. Source: New Organizing
Institute analysis of NCEC precinct level election data.

04 to 05 Drop-off by Precinct-level Quintile of Percent African American – New Jersey

Figure 3.8: New Jersey – Average drop-off from 2004 to 2005 by quintile of percent
African American at the precinct level. The first quintile averages 1% African
American, the fifth quintile averages 33% African American. Source: New Organizing
Institute analysis of NCEC precinct level election data.

Although we may see a slight pattern in Virginia, for the most part there does not appear to be a clear
relationship between percentage of African-Americans living in a precinct and drop-off from 2004 to 2005.

28
08 to 09 Drop-off by Precinct-level Quintile of Percent African American – Virginia

Figure 3.9: Virginia – Average drop-off from 2008 to 2009 by quintile of percent
African American at the precinct level. Source: New Organizing Institute analysis of
NCEC precinct level election data.

08 to 09 Drop-off by Precinct-level Quintile of Percent African American – New Jersey

Figure 3.10: New Jersey – Average drop-off from 2008 to 2009 by quintile of percent
African American at the precinct level. Source: New Organizing Institute analysis of
NCEC precinct level election data.

We observe a much stronger relationship between percentage of African-Americans living in a precinct and
drop-off from 2008 to 2009. Particularly in Virginia, the magnitude of difference in drop-off between the first
quintile (least African American precincts) and the fifth quintile (most African American precincts) is greater
than any other comparison we have considered.

We have established that drop-off from 2004 to 2005 is predictive of drop-off from 2004 to 2006. In other words,
drop-off from off year elections may help predict drop-off in midterm elections at the precinct level. We also
know that we observe drop-off among groups in 2009 that we did not in 2005 at the precinct level. In
particular, drop-off in 2009 was uniquely high among very Democratic and heavily African American precincts.
Given that drop-off from off-year elections appears to be predictive of drop-off in midterm elections, we
expect to see a similar decrease in rates of participation among very Democratic and heavily African-
American precincts in the 2010 midterm elections.

29
ENTHUSIASM GAP AND VOTING

Commentary on surveys of public opinion often refer to a partisan “enthusiasm gap;” this is a way to measure
the relative excitement of the base of each political party. Surveys ask both partisan self-identification as well
as a question such as “Compared to previous elections, are you more enthusiastic about voting than usual, or
less enthusiastic?” When examining the crosstabs for these two questions, if there is a difference in self-reported
enthusiasm by political party; an “enthusiasm gap” exists.

Simply running a Google search for the term “enthusiasm gap” will yield numerous examples of commentators
using measures of enthusiasm to make predictions about the magnitude of turnout differences between the
base of each political party. Chris Cilizza, a pundit who writes a Washington Post column and frequently
appears on cable news stations as a political commentator, wrote earlier this year, “When one party's base is,
to borrow a phrase, fired up and ready to go and the others isn't, major change can happen… an enthusiasm
gap favoring Republicans exists – and that's good news for the GOP heading into the fall6.”

There are several reasons to be skeptical about the utility of the enthusiasm gap in predicting differences in
partisan base turnout. First, self-identified Democrats and Republicans are subsets of the universe of survey
respondents. Making comparisons between these two groups on a particular measure introduces a much
higher margin of error than the survey has as a whole. Secondly, enthusiasm gaps within election cycles have
historically been highly volatile. By the time a survey is conducted close enough to an election for the
enthusiasm gap to be relevant, we have much better predictors of election outcomes. Despite these caveats,
however, we do find a correlation between enthusiasm and partisan performance.

ANALYSIS

To determine whether there is a link between the enthusiasm gap and partisan turnout, we use historical
enthusiasm data among-self identified partisans from Gallup surveys, the NEP exit polls, historical vote totals
compiled by the U.S. House of Representatives, and historical estimates of the voting eligible population (VEP).
Our selection of data sources is explained further in the Methodology appendix.

As enthusiasm for self-identified partisans increases, we should expect more voters to turn out and vote for
candidates of that political party. We want to answer the question “of everyone who could possibly vote for a
party’s U.S. House candidates, how many people ended up voting for those candidates?” To assess this, for
each election cycle we compare partisan enthusiasm to number of votes for that party’s House candidates as
a percentage of the voting eligible population. So for each election, we define:

Partisan Congressional Vote % = (number of House votes for party’s candidates ÷ size of voting eligible
population).

We chose House votes because we expect them to be, on balance, a more generic measure of national
partisanship than Senate, Gubernatorial, or Presidential races. Since 435 races nationwide gives us more
granularity than only evaluating statewide races, local peculiarities are better smoothed out and more salient
trends can be observed.

6Cilizza, Chris. "Morning Fix - June 21." Web. <http://voices.washingtonpost.com/thefix/morning-fix/1-2-weld-county-


district.html>.

30
Figure 2.1: Democratic Congressional Vote % vs. enthusiasm between 1994 and 2008. Source: New Organizing
Institute analysis of Gallup enthusiasm data and national vote totals for U.S. House candidates.

Figure 2.2: Republican Congressional Vote % vs. enthusiasm between 1994 and 2008. Source: New Organizing
Institute analysis of Gallup enthusiasm data and national vote totals for U.S. House candidates.

Surprisingly, among both Democrats and Republicans, there appears to be a linear relationship between self-
reported enthusiasm and votes for partisan candidates as a percentage of VEP. Both trendlines are significant
at alpha = 0.05, and for both parties self-reported enthusiasm explains nearly 70% of the variation in vote as a
percentage of VEP. Based on historical results, self-reported enthusiasm may be a useful tool in projecting
partisan votes as a percentage of VEP. However, among Republicans, the inclusion or exclusion of the 2004
election cycle (right most scatter point) would drastically change the coefficient of the model, so we feel less
confident in it than our linear model for Democrats.

31
We’re not just interested in whether enthusiasm correlates with historical Partisan Congressional Vote %, but
whether differences in enthusiasm (the “enthusiasm gap”) can predict differences in Partisan Congressional
Vote %. Although Democratic and Republican enthusiasm may individually help predict Partisan Congressional
Vote %, considering them as differences introduces a higher amount of potential variation. In addition, if
Democratic and Republican enthusiasm have been both highly correlated with each other as well as highly
correlated with national turnout overall, the enthusiasm gap would be much less useful in predicting differences
in Partisan Congressional Vote %.

Figure 2.3: Difference in Partisan Congressional Vote % vs. enthusiasm gap between 1994 and 2008. Source: New
Organizing Institute analysis of Gallup enthusiasm data and national vote totals for U.S. House candidates.

Despite these potential impediments to the utility of the enthusiasm gap, it appears to be highly predictive of
the differences in votes as a percentage of the entire VEP between Democrats and Republicans. Our model is
significant at alpha = 0.01, and explains 84% of the variation in Partisan Congressional Vote %.

Considering the number of votes for each party’s candidates, however, does not necessarily tell us whether
some self-identified partisans (or alternatively, a party’s “base”) turn out at a higher rate than the other party.
In an attempt to separate candidate support from partisan turnout, we compare the enthusiasm gap to
partisan share of voters for each election cycle. The NEP exit polls are our best way at determining this historical
partisan composition of voters.

32
Figure 2.4: Share of voters who identify as Democrats in NEP exit poll vs. enthusiasm gap between 1994 and
2008. Source: New Organizing Institute analysis of Gallup enthusiasm data and NEP exit polls.

Figure 2.5: Share of voters who identify as Republicans in NEP exit poll vs. enthusiasm gap between 1994 and 2008.
Source: New Organizing Institute analysis of Gallup enthusiasm data and NEP exit polls.

For both Democrats and Republicans, we cannot fit a linear model between the enthusiasm gap and self-
identified partisan share of voters (P > |t| is higher than 0.2 in both cases). In other words, we cannot use the
“enthusiasm gap” to predict base vote share as reported by the NEP exit polls for each election year.

This does not mean that a link between the “enthusiasm gap” and “base” partisan turnout does not exist, but
that we cannot find one with data available in the public domain. What we truly want to measure is what
percentage of a party’s members turned out to vote among everyone who self-identifies as a member of that
party for a given election. It’s possible that there could be more accurate ways to measure turnout among a
party’s base than share of self-identified partisans interviewed in the exit polls (for example if historical voterfiles
with stable indicators of partisanship existed).

33
Despite our initial skepticism, the enthusiasm gap appears to have utility in speculating about future elections
based on historical patterns. Historically the party with an advantage in self-reported enthusiasm also wins
more congressional votes. In addition, the size of the enthusiasm gap can be used to predict the magnitude of
difference in votes for Democratic and Republican U.S. House candidates.

Although the enthusiasm gap gives us insight into national election results (Figure 2.3), it does not appear to tell
us anything about which party’s base is particularly energized or when a party’s base votes at a higher rate
than the base of the other party (Figures 2.4 and 2.5). The context in which many pundits use it, that
enthusiasm among self-identified partisans is largely about base turnout and not candidate support, is not
supported by the evidence.

U.S. HOUSE VOTE MARGIN IN 2010

We can attempt to predict the difference in votes for Democratic and Republican U.S. House candidates in
2010 using recent enthusiasm gap data and our model from Figure 2.3. Applying our methodology to the most
recent enthusiasm data from Gallup7, Republicans currently enjoy a 21 percentage point advantage in self-
reported enthusiasm for the 2010 elections (incidentally, a higher Republican advantage than any other
election we evaluated going back to 1994). If historical patterns persist, this is a 5.3 percentage point advange
for Republicans in terms of number of votes as a percentage of VEP for U.S. House candidates. Figure 2.6 shows
how this projected advantage compares to previous elections.

Enthusiasm Gap, Difference in Partisan Congressional Vote %, and Seat Change by Election

Enthusiasm Votes as % of
Year Gap (D-R) VEP (D-R) Seat Change8
1994 -10% -3% R+54
1996 0% D+8
1998 -6% 0% D+5
2000 -12% 0% D+1
2002 -4% -2% R+8
2004 -4% -1% R+3
2006 9% 3% D+31
2008 20% 6% D+21
Figure 2.6: Enthusiasm gap, Dem – Rep House votes as percentage of VEP, and seat
change in U.S. House after election, 1994 - 2008. Source: New Organizing Institute
analysis of Gallup enthusiasm data and national vote totals for U.S. House
candidates.

However, the conventional wisdom regarding the enthusiasm gap focuses on its purported utility in speculating
about base turnout, not predict election outcomes. If our goal is simply to predict the results of the 2010
midterm elections, other measures such as the generic Congressional ballot question from opinion surveys, or
the condition of the economy, may have higher predictive power than the enthusiasm gap.9 Furthermore the
enthusiasm gap has historically been volatile prior to each election, which is another reason to turn to other
measures when attempting to make this prediction.

7 Gallup, Inc. "GOP Takes Unprecedented 10-Point Lead on Generic Ballot." Web. 1 Sep 2010.
<http://www.gallup.com/poll/142718/gop-unprecedented-lead-generic-ballot.aspx>. Averaged results from two most
recent surveys to calculate “enthusiasm gap.”
8 Redistricting sometimes makes comparisons at the Congressional district level across election years difficult.

9 Recent models using polling data indicate a 49 to 52 seat gain for Republicans in the U.S. House. A model using less

volatile data (economic indicators rather than polling) projects a 27 to 30 seat gain for Republicans
<http://www.huffingtonpost.com/2010/09/06/political-scientists-fore_n_706643.html>

34
METHODOLOGY

HISTORICAL RATES OF PARTICIPATION – DATA SOURCES AND ASSUMPTIONS

Voter files are the gold standard for turnout data as they provide individual level data on voter participation in
specific elections. In an ideal world, we would use voter files to estimate turnout rates and midterm drop-off for
various demographic groups over time. Unfortunately, we did not have access to snapshots of historical voter
files at the national level for this research. Degradation of the voter file occurs in a nonrandom way (i.e. voters
who move, die, or for any other reason are removed from a voter file), so the universe of people who exist on
the voter file today that were registered to vote in 1994 can’t be used to estimate turnout rates in 1994.
Because attrition from the voter file is biased, we also cannot make comparisons between turnout for previous
years and more recent elections based on current data.

US CENSUS CURRENT POPULATION SURVEY (CPS)

Instead, to determine historical rates of participation in Presidential and midterm elections, we use self-reported
turnout from the US Census’ Current Population Survey (CPS). Self-reported turnout is problematic because
people have a tendency to over-report their voting behavior. However, a cross-validation of voter registration
files and Current Population Survey data finds that the CPS and voter files are in general agreement about
turnout rates at the aggregate level10. We use the CPS when possible because we feel we can obtain
accurate aggregate estimates, even if individual level self-reported turnout is not accurate.

The US Census’ CPS is a monthly survey of around 57,000 nationally representative households, and has been
conducted for over 50 years. On a monthly basis, the CPS collects information on a variety of demographic
characteristics, including employment situation, gender, age, race, marital status, educational attainment.
Every two years, after the national elections in November, the CPS asks additional questions on voting behavior
and registration as part of their “Voting and Registration Supplement.”

To determine historical turnout and drop-off for various demographic groups, we built a dataset that combined
every CPS “Voting and Registration Supplement” dating back to 1994. Like the Census Bureau, we estimated
turnout as a percentage of the Citizen 18+ universe in the survey. Variables such as length of residence, age,
and education level are determined at the time of the survey, and do not carry across years.

NEP NATIONAL ELECTION EXIT POLLS

In some cases, variables that we wanted to study were not available in the CPS datasets. In particular, the CPS
has not reliably asked about income between 1994 and 2008, and does not ask about partisanship at all. In
these cases, we rely on election day exit polls conducted by the National Election Pool (NEP).

For each state, the NEP exit pollsters send interviewers to a randomly selected set of precincts, weighted by
prior turnout. Interviewers stand at these precincts and survey a random subset of voters as they are leaving
the polling place. The survey includes a range of questions on both demographics and political attitudes.
After voting is over, the exit poll is weighted to actual turnout counts and eventually the result of the election11.

Because the NEP surveys the universe of people who actually voted, it is often used to determine the
composition of the electorate. The aforementioned cross-validation of voter files and election survey

10 McDonald, Michael. "The True Electorate: A Cross-Validation of Voter Registration Files and Election Survey
Demographics." Public Opinion Quarterly. 71.4 (2007): 588-602. Print.
11 Blumenthal, Mark. "Exit Polls: What You Should Know." Web. 1 Sep 2010.

<http://www.pollster.com/blogs/exit_polls_what_you_should_kno_1.php>.

35
demographics indicates that there is some representational bias in the exit polls, even after weighting.
However, we’re interested in changes in who votes over time, not absolute levels of representation, and the
NEP exit polls serve this purpose.

TERMS AND ASSUMPTIONS

Ultimately, we’re not just interested in historical rates of participation across different demographic groups, but
also whether particular demographic groups tend to drop off at higher rates than others. To do this, we
compare average Presidential turnout rate to average midterm turnout rate across groups (as a proportion, not
percentage point difference). For example, if demographic group A historically votes at a rate of 60% in
Presidential elections and at a rate of 30% in midterm elections, and demographic group B votes at a rate of
60% in Presidential elections but at a rate of 45% in midterm elections, demographic group A has higher drop-
off than demographic group B. We define “retention” for a particular group as:

(Average midterm Turnout Rate) ÷ (Average Presidential Turnout Rate)

For the set of elections we are considering, this is:

(1994 Turnout + 1998 Turnout + 2002 Turnout + 2006 Turnout) ÷ 4


------------------------------------------------------------------------
(1996 Turnout + 2000 Turnout + 2004 Turnout + 2008 Turnout) ÷ 4

We assume that the lower the historical “retention” of a particular group, the more likely their participation rate
will drop off in future midterm elections in the absence of efforts specifically targeting that group.

If there is representational bias in the NEP exit polls even after weighting, we assume that this bias is similar
across election years. So if young people are over represented in the 2006 exit poll post weighting, we assume
that they are similarly over represented in the 2008 exit poll. For the groups we are examining, we care more
about observing changes in their share of voters over time than we do about precisely estimating their
representation in a particular election.

Finally, we assume that the US Census’ CPS provides reliable aggregate level estimates of turnout for various
demographic groups, even if self-reported turnout at the individual level is less reliable. If in some cases
aggregate level estimates of turnout may not be accurate, we assume that self-reported turnout is strongly
correlated enough with actual turnout to draw conclusions about how turnout among particular demographic
groups change over time, and whether particular subgroups have higher drop-off than others.

To compare the rate at which turnout for particular groups increase or decrease over time, we run a linear
regression on the group’s turnout over time and report the regression coefficient ß.

ENTHUSIASM GAP – DATA SOURCES

Among survey organizations who release their results in the public domain, Gallup has asked the enthusiasm
question in a consistent manner for the longest period of time. Since 1994, Gallup has asked their survey
respondents “Compared to previous elections, are you more enthusiastic about voting than usual, or less
enthusiastic?” Gallup data is also frequently cited in news articles about differences in partisan enthusiasm. For
election cycle we average the two most recent Gallup political surveys asking the enthusiasm question prior to
election day12, then subtract Republican % very enthusiastic from Democratic % very enthusiastic13.

12Enthusiasm data is not available for the 1996 election cycle. During the 2000 election cycle, the most recent enthusiasm
questions prior to election day were asked well before the November election, in January and March.

36
To assess the relationship between partisan enthusiasm and turnout, we compare the Gallup enthusiasm figures
we compiled to the share of voters by self-identified partisanship from the NEP exit polls. We also compare the
Gallup enthusiasm figures to national vote totals for Congressional candidates, from the “Statistics of the
Congressional Election” tables compiled by the U.S. House of Representatives14. Finally, we use historical Voting
Eligible Population (VEP) estimates from Dr. Michael McDonald’s “United States Election Project15.”

OFF-YEAR ELECTIONS – DATA SOURCES AND ASSUMPTIONS

We chose to conduct our analysis at the precinct level rather than individual level because we need
comparable data from the 2004, 2005, 2006, 2008, and 2009 elections.

Although we would have preferred to perform an individual level analysis on the voter file, we encounter a
similar challenge as we did with our analysis of historical drop-off: the group of people who exist on the voterfile
today and were registered to vote in 2004 is not comparable to the group of people who would have existed
on the voterfile in 2004. In other words, there would be no way to make an effective comparison between
drop-off from prior years and more recent drop-off. Similarly, we lack access to the exit poll datasets from New
Jersey and Virginia in 200516, so we have no way of knowing whether the drop-off we observe in the 2009 exit
polls is unusual or not.

We acknowledge that there are substantial challenges in using precinct level results to make predictions about
2010 turnout. First, patterns we observe using aggregate measures do not necessarily reflect upon individuals.
For example, just because we observe a particular turnout pattern among precincts with a high percentage of
African-Americans, it does not mean that we would observe the same patterns among African-Americans at
the individual level. Secondly, the composition of precints may change over time, which introduces additional
“noise” into our results.

Precinct level data comes from the National Committee for an Effective Congress (NCEC), who undertakes the
very difficult task of collecting and standardizing precinct level election results. We then matched precinct
level counts of the number of African-Americans from the Catalist voter file. “Drop-off” was calculated as the
percentage point difference between the turnout rates in two elections at the precinct level.

We chose to perform our analysis in New Jersey and Virginia. If we see similar patterns in drop-off in both states,
we can be more confident about our expectations in 2010. We did not choose special elections in house
races or Massachusetts because those races did not also occur in 2005.

13 Gallup enthusiasm data retrieved from


<http://www.gallup.com/poll/File/140816/2010_Voting_Enthusiasm_June_18_2010.pdf>
<http://www.gallup.com/poll/108322/election-enthusiasm-dips-after-primaries.aspx>
<http://www.gallup.com/poll/111115/democrats-election-enthusiasm-far-outweighs-republicans.aspx>
14 U.S. House of Representatives, . "Statistics of the Congressional Election." Web. 1 Sep 2010.

<http://clerk.house.gov/member_info/electionInfo/index.html>.
15 McDonald, Michael. "Voter Turnout Data 1980-2008." Web. 1 Sep 2010. <http://elections.gmu.edu/Turnout%201980-

2008.xls>.
16 2005 exit poll datasets from New Jersey and Virginia are not available in the Roper Center archive.

37
ACKNOWLEDGEMENTS

We would like to thank the many folks who gave generously of their time to contribute to this research and our
interpretation of it:

John Boney, Matt Lackey and Laura Reznick, AFL-CIO


Tom Bonier, National Committee for an Effective Congress (NCEC)
Talia Borodin, Catalist
Chris Kennedy, Alliance for Climate Protection
Matthew Loveless
Melissa Roy, Ruthie Posekany and Matthew Berg, The Atlas Project

38

You might also like