1 The Effects of Unemployment on Voter Turnout in U.S. National Elections Matthew B. Incantalupo 1 Joint Degree Program in Politics and Social Policy Princeton University [email protected]Work in Progress: Please do not cite or share without permission. Abstract: This paper examines the effects of economic hardship on political participation. Economic hardship can impede political participation, but can also galvanize citizens to seek redress from government. In this paper, I study the effects of personal experience with unemployment on voter turnout in the United States. I argue that individuals perceive job loss as a personal problem in low-unemployment contexts and a social problem in high-unemployment contexts. Job loss is therefore a mobilizing experience when and where the unemployment rate is high. I test my theory using the Current Population Study Voting and Registration Supplement from 1978 through 2010. In a cross-sectional analysis, I find that the effect of unemployment on reported turnout is increasing in unemployment context. By exploiting the timing of job loss around Election Day as a quasi-experiment, I estimate that the causal effect of involuntary job loss on reported turnout ranges from about -5 percentage points in low-unemployment contexts to over 10 percentage points in high-unemployment contexts. I do not find similar effects for individuals who quit their jobs around Election Day or who lose their jobs well before Election Day. 1 I would like to thank Amy Lerman, Christopher Achen, Martin Gilens, Larry Bartels, Kosuke Imai, John Londregan, Kay Lehman Schlozman, Scott Abramson, Rikhil Bhavnani, Will Bullock, Stephen Chaudoin, Thomas Hayes, Sarah Hummel, and the participants in the Graduate Research Seminar in American Politics and the Political Methodology Colloquium at Princeton University for helpful feedback and support. All remaining errors and omissions are my own.
48
Embed
The Effects of Unemployment on Voter Turnout in U.S ... The Effects of Unemployment on Voter Turnout in U.S. National Elections Matthew B. Incantalupo1 Joint Degree Program in Politics
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
The Effects of Unemployment on Voter
Turnout in U.S. National Elections
Matthew B. Incantalupo
1
Joint Degree Program in Politics and Social Policy
The United States is recovering from its longest period of high unemployment since the
Great Depression in the aftermath of the economic downturn that began in late 2007.2 While the
link between high or rising unemployment and anti-incumbent voting is well-established in
political science (Mueller 1970; Kramer 1971; Lepper 1974; Hibbs 1979), most of the research
on the connections between economic conditions and political behavior focuses on
unemployment as a macroeconomic indicator and not as a form of personal economic hardship
experienced by some Americans and not others (e.g. Radcliff 1992). Additionally, the literature
on economic voting primarily considers how economic outcomes affect individual vote choice,
but rarely examines whether these outcomes affect political participation in the form of turning
out to vote. The majority of attempts to understand this relationship (Schlozman and Verba 1979;
Rosenstone 1982) predate the last time the unemployment rate reached recent heights and tend to
focus on unemployment solely during economic recessions.
This paper examines the effects of personal experience with job loss and unemployment
on Americans’ propensities of turning out to vote in national elections spanning 1978 through
2010, a period that includes both good and bad economic conditions. I develop a new theory of
political behavior following personal economic hardship that incorporates the broader economic
context in which Americans experience unemployment. Then, I utilize the instrument that the
federal government uses to calculate the unemployment rate on a monthly basis, the Current
Population Survey (CPS), jointly administered by the Bureau of the Census and the Bureau of
Labor Statistics to test this theory across multiple levels of analysis. By exploiting the quasi-
panel structure of the CPS and the timing of job loss, I estimate the causal effects of job loss on
2 At the worst point of this jobs crisis, October 2010, over 15.4 million Americans were unemployed, with over 40
percent unemployed for 27 weeks or longer periods of time.
3
voter turnout for a balanced sample of Americans who become unemployed in the weeks
surrounding Election Day.
I find that the economic and social contexts in which Americans experience
unemployment structure the ways in which their economic situations influence their political
behavior. Broadly speaking, personal experience with involuntary unemployment is a mobilizing
phenomenon that increases voter turnout in high-unemployment contexts and a demobilizing
phenomenon that decreases voter turnout in low-unemployment contexts. These findings have
important implications for how political scientists think about the phenomenon of
unemployment, particularly the significance of unemployment as a personal experience that has
systematic effects on how we engage in politics. From a methodological standpoint, this paper is
an example of how design-based inference can help us to establish the causal relationships
between personal economic conditions and political behavior.
The remainder of this paper proceeds as follows: In the next section, I summarize the
existing scholarly research on the effects of job loss on political behavior, namely political
participation. Following this, I articulate my main argument, that the context in which Americans
experience job loss influences the extent to which unemployment is a “socially-centered
problem,” that mobilizes individuals who experience it to engage in political participation
(Brody and Sniderman 1977). Then, I describe the data used in this analysis. I present a large,
cross-sectional analysis of the effects of unemployment on voter turnout. I find a robust
association between job loss and voter turnout that is consistent with the theory of
Unemployment-in-Context. Next, I introduce a design-based approach which carefully constructs
two groups into which citizens who experience job loss are quasi-randomly assigned within a
one-month window of time around Election Day. Here, I demonstrate that the design-based
4
approach reduces the problem of selection bias and produces a more balanced sample. I then
estimate the causal effect of job loss on individual propensity to turn out to vote and attempt to
address the causal mechanisms that contribute to my findings, particularly whether resources or
psychological factors underpin them. In this section, I also discuss the limitations and
implications of my findings. Finally, I conclude with a discussion of how my findings contribute
to our understanding of how personal economic hardship affects political behavior and with
suggestions for future research.
Previous Research
At present, gaps exist in our understanding of how unemployment affects political
attitudes and behavior in part because there has been so little attention paid to unemployment as
a personal hardship and the disparate contexts in which someone can experience unemployment
(when it is a widespread problem and when it is relatively uncommon). With the exception of
very recent scholarship emerging from the 2007 economic recession (Margalit 2013; Burden and
Wichowsky 2014; Owens and Pedulla 2013), existing theories of unemployment were created
and evaluated the last time the unemployment rate was as high as it is today, nearly three decades
ago
Mobilization
Unemployed Americans may engage in political behavior to lobby government to address
economic hardships such as high unemployment or inflation or to express their discontent with
their unfortunate situations and punish the members of government who they believe are
responsible for job loss and high unemployment (Lipset 1960; Schlozman and Verba 1979;
Healy 2009; Margalit 2011). This mobilization would be consistent with the notion of
retrospective “pocketbook voting” (Fiorina 1981; Bartels 2008; Bartels 2010) and findings that
5
voters are wont to punish elected officials for poor economic outcomes (Lau 1982; Owen 2011).
Similarly, voting as a means of expressing grievances with current conditions can be viewed as
an extension of political participation as a civic duty (Riker and Ordeshook 1968). Another
mechanism through which unemployment may directly influence political participation is by
way of policy feedback, through the receipt of unemployment benefits from the government.
Interacting with government through participation in a government program such as Social
Security has been shown to increase an individual’s likelihood of turning out to vote (Campbell
2002; Mettler and Soss 2004; Chen 2012).
Job loss can also provide non-psychological sources of voter mobilization; the
unemployed may become more politically active because they have more free time to follow
politics and participate (Charles and Stephens 2011). However, the link between free or leisure
time and political participation is rather weak (Verba, Schlozman, and Brady 1995; Putnam
2000). Furthermore, survey evidence from both the 1970s and the 2010s indicates that while
unemployed Americans have more leisure time and watch more television than gainfully
employed Americans, they dedicate less time to watching or reading the news (Scholzman and
Verba 1979; Krueger and Mueller 2011).
Withdrawal
The financial burden of unemployment and the demands and stresses of looking for new
work could lower one’s likelihood of participation (Schlozman and Verba 1979; Rosenstone
1982). Unemployment contributes to significant psychological distress and associates with
negative physical health outcomes (Warr 1987). The unemployed express lower levels of life
satisfaction, including satisfaction with income, satisfaction with their families, and satisfaction
with their personal accomplishments (Feather 1989; Schlozman and Verba 1979). Most
6
importantly, the unemployed report lower levels of self-esteem and personal efficacy, which can
later contribute to a general sense of apathy (Bandura 1982).
These findings point to unemployment as a demobilizing phenomenon. Americans who
experience job loss are unlikely to get sufficiently aroused to make it a point to head to the polls
on Election Day. To the extent that citizens perceive job loss as a personal problem with which
they must cope and that they must remedy, job loss will not increase their likelihood of voting
(Sniderman and Brody 1977). As Rosenstone concludes, “When a person experiences economic
adversity his scarce resources are spend on holding body and soul together – surviving – not on
remote concerns like politics” (1982, p. 26).
Unemployment in Context
In this section, I outline a new theory of how personal hardships interact with broader
social and economic trends to influence political behavior, which I call Unemployment-in-
Context. While I focus my discussion on the economic hardship of unemployment and the
political behavior of turning out to vote, the theory is similarly applicable to other forms of
economic hardship and other political outcomes. Unemployment represents a good test case
because it is an unambiguous personal hardship, as opposed to more subjective worries over not
making ends meet.3
Unemployment occurs under a variety of circumstances that should affect how it is
experienced, understood, and potentially incorporated into political attitudes and behavior.
Building off of Sniderman and Brody (1977), I argue that personal problems and hardships, such
as unemployment, can become politicized to the extent that individuals perceive them as
externally caused, pervasive in their communities, and resistant to individual influence or
3 While citizens may not be able to agree on what constitutes a struggling economy, perhaps due to partisan biases in
evaluating objective economic conditions (Bartels 2008), or agree upon a threshold for “low-income,” the definition
of unemployment is comparatively easier to understand.
7
remedy. Individuals form these perceptions in response to their local unemployment contexts,
which reflect the social and economic realities in which individuals must cope with their
economic hardships and take steps to remedy them.4 Local context and geographically narrow
interests have been shown to meaningfully affect political behavior (Cutler 2007). In short, job
loss in low-unemployment contexts is a demobilizing phenomenon. However, unemployment
becomes a socially-centered and politicized issue in high-unemployment contexts and should
contribute to political mobilization (Burden and Wichowsky 2014).
When an individual loses his or her job, “the person will undertake a causal search to
determine why the outcome occurred, especially if the job was an important part of that person's
life or if loss of the job was unexpected" (Feather 1990, p.68). Unemployed citizens learn about
the broader state of the economy through their family and friends, who might also be
unemployed (Paolino 2011), their job-seeking behavior (Mortensen 1984; Krueger and Mueller
2011), and other contextual factors such as nearby mass layoffs (Healy 2009) and plant closures
due to outsourcing decisions (Margalit 2011). The act of applying for unemployment benefits
can also aid in this process (Mutz 1994).
Importantly, the politicization of economic hardship is affected by real economic context,
not simply national economic conditions as reported in the news media or elsewhere.5 Reported
concern about unemployment is significantly affected by the unemployment rate but not
significantly affected by national news stories about unemployment (Behr and Iyengar 1985;
Iyengar and Kinder 1987). Jobless Americans should exhibit a distinct sensitivity to
4 This is a major distinction between my theory and the existing literature. I allow individuals to form perceptions
about their hardships, rather than classify some hardships as personally-centered (such as going through a divorce)
and some as socially-centered (such as monetary inflation) based on their content alone. 5 Of course, local and national economic conditions are correlated, but at any given point in time, we observe
substantial variation in unemployment across U.S. States. For example, during the worst months of the most recent
jobs crisis, the unemployment rate in Nevada approached 15%, while at the same time remaining around 4% in
North Dakota. Followers of European politics will note that the unemployment rate in Nevada at this time was
roughly equivalent to that of Greece.
8
unemployment context that alters the manner in which their own experiences with
unemployment affect their political behavior. This sensitivity should be consistent with the
mobilization that occurs when an individual is faced with a socially-centered problem (Brody
and Sniderman 1977; Sniderman and Brody 1977).
In low-unemployment contexts, job loss should have a demobilizing effect on voter
turnout on those who experience it. Unemployed Americans will not politicize their personal
hardships and perceive them as problems that should be addressed by government. They will
perceive being unemployed as their own problem and devote their efforts to the job search over
participating in politics, making voting a substitute for the job search and raising the costs of
voting (Downs 1957, Riker and Ordeshook 1968). In high-unemployment contexts, job loss
should make individuals more likely to vote than we would expect had they experienced no
unemployment. I expect individuals who experience job loss when and where the unemployment
rate is high to perceive their hardship as part of a broader social problem and be spurred by this
to vote. This is not because of increased leisure time or any other resource-based explanation but
is psychological and based on mobilization. In the remainder of this paper, I evaluate the
implications of Unemployment-in-Context and present evidence in favor of a mobilization-based
explanation for how job loss affects voter turnout.
Data
Focusing on the unemployed as a population of interest can be challenging for a variety of
reasons. First, since 1960 no more than 10.8 percent of the U.S. labor force has been considered
unemployed. Second, while the total proportion of jobless Americans has always been greater
than the unemployment rate6, it is difficult to differentiate between the long-term unemployed
6 The reported unemployment rate uses as its denominator the proportion of non-institutionalized Americans who
are in the labor force. Individuals who have left the labor force for any reason are not included, even if they have left
9
who may still be seeking employment, and those who are no longer part of the labor force
because they are discouraged. Third, as is the case with many embarrassing or socially
undesirable outcomes, we can expect self-reports of unemployment to be biased downward.
Finally, surveys that rely on simple random sampling or similar techniques tend to undersample
individuals with low socioeconomic status, a group that is more likely to be unemployed. Thus,
without a very large sample or a purposeful sampling design, it is difficult to attain precise data
about unemployed Americans.
The ideal data set for a study of the effects of job loss on Americans’ political attitudes
and behavior would be a very large longitudinal survey that frequently inquires about labor force
participation and includes various measures of public opinion and inquiries about political
participation. To the best of my knowledge, no such data set exists.7 In order to locate an
adequately-sized sample of unemployed Americans, I turn to the Bureau of Labor Statistics’
official instrument for calculating the monthly unemployment rate, the Current Population
Survey (CPS). The CPS is administered at the household level to about 50,000-60,000
households and asks about the employment status of each member of the household at least 15
years of age in the week containing the 12th
day of the month.
Dependent Variable
Most important for my purposes, the November CPS contains a Voting and Registration
Supplement in even-numbered years, which asks eligible respondents whether they (and other
members of their household) turned out to vote on Election Day. I exclude any respondents who
the labor force out of frustration due to long-term unemployment. For instance, Western and Pettit (2005)
demonstrates that the racial disparities in employment and wages would be even greater but for the exclusion of
incarcerated individuals from the labor force in reports published by the Bureau of Labor Statistics. 7 For example, while the American National Election Studies 2008-2009 Panel Study fielded 6 waves in 2008 alone,
its sample size was under 3,000 and it included just 2 inquiries to employment status nearly a year apart.
10
could not recall if they voted.8 This leaves a sample of 1,535,337 respondents, with 56,234
unemployed respondents, in 17 elections spanning 1978 through 2010. This time period includes
two elections that took place when the national unemployment rate was very high (1982 and
2010, with rates of 10.8 and 9.8 percent, respectively) and three elections that took place when
the unemployment rate was rather low, approaching what is considered full employment9 (1998,
2000, and 2006, with rates of 4.4, 3.9, and 4.5 percent respectively). In the next section, I offer
support for the theory of Unemployment-in-Context by presenting evidence that indicates that the
effect of unemployment on an individual’s likelihood of turnout to vote significantly varies with
the unemployment rate in his or her state at the time of the election.
Cross-sectional analysis
Unemployment-in-Context posits that unemployed Americans will be mobilized to turn
out to vote in high-unemployment contexts. In order to broadly test this claim, I rely on two
useful characteristics of the available data, First, the CPS household survey is both large in size
and a representative sample of U.S. states. This allows me to examine the effect of
unemployment on voter turnout in each state for each election separately (867 state-year cases in
total, counting the District of Columbia), without making any assumptions about how
unemployment in one state may be similar to unemployment in another state, which would be the
case if I pooled my data within election years or pooled across all states and elections. Second, I
am able to exploit the substantial variation in the unemployment rate across states at a given
period of time. For the remainder of this analysis, I operationalize “local unemployment context”
8 This differs from the approach used by the Census Bureau, which does not exclude these unsure responses when it
publishes turnout statistics using CPS data (Hur and Achen 2013). 9 The Full Employment and Balanced Growth Act of 1978 instructs the United States to strive for full employment,
which it defined as unemployment rates not more than 3% for persons aged 20 or over and not more than 4% for
persons aged 16 or over.
11
as the seasonally-adjusted unemployment rate in a respondent’s state of residence.10
Figure 1
displays the distribution of the seasonally-adjusted unemployment rate by state over the time
period of interest for this analysis.
[Fig. 1 here]
This substantial variation in the unemployment rate by state means that I will be able to
incorporate additional variation in local unemployment context, ranging from very low-
unemployment contexts (the lowest being New Hampshire in 1986, with an unemployment rate
of 2.1 percent) to very high-unemployment contexts (the highest being West Virginia in 1982, at
17 percent). If I am correct, unemployed Americans will be more likely to turn out to vote in
high-unemployment contexts, all else equal. As a descriptive first step, I plot the average turnout
rates for each state in each election as computed from the CPS household survey for employed
and unemployed respondents as a function of the unemployment context in Fig. 2.11
Because
turnout rates are uniformly higher during presidential elections and considerably more stratified
by age, income, and education (Wolfinger and Rosenstone 1980), I plot presidential and off-year
elections separately.
[Fig. 2 here]
The scatterplots in Fig. 2 demonstrate that turnout generally increases with
unemployment context, measured here as the seasonally-adjusted unemployment rate in each
state in November of each election year. In both presidential elections and off-year national
elections, reported turnout among the employed and the unemployed is higher on average in
10
State of residence is the best available geographical unit for this analysis because the CPS utilizes a representative
sample of each state. While there can be intrastate variation in unemployment, I am unable to consider smaller units
such as metropolitan areas and counties, because confidentiality laws prevent the BLS from sharing this information
for respondents in all but the largest metropolitan areas and counties. 11
For simplicity’s sake, I plot the raw means for each group without any weighting or adjusting.
12
high-unemployment contexts than in low-unemployment contexts. This suggests that difficult
economic times “raise the stakes” of elections and mobilize voters to participate.12
In both presidential and off-year elections, the turnout rate for unemployed respondents
(designated on the plot using x’s and summarized with dashed lines) increases more rapidly than
the turnout rate for the employed as a function of the local unemployment context. This is
especially true in presidential elections, in which a 1-point increase in the state-level
unemployment rate is associated with a roughly 1.3-point increase in turnout among unemployed
voters (t=3.85, p < .01). This is approximately 3 times larger than the corresponding effect of a
1-point increase in the unemployment rate on employed voters’ reported turnout rates. The
information presented in Fig. 2 indicates that unemployed citizens are more sensitive to local
economic context than employed citizens, particularly during presidential election years.
Multilevel Analysis
Turnout among unemployed Americans is positively associated with local unemployment
context. However, the data presented in Fig. 2 do not account for various underlying differences
between states and across election years that might confound the relationship between economic
conditions and aggregate voter turnout. To account for these differences, and to assess the effect
of unemployment across a range of economic contexts, I utilize the two-step, multilevel
approach that has been previously applied to questions of economic voting (Duch and Stevenson
2005) and voter turnout (Jusko and Shively 2005). Specifically, I estimate the following model
for each of the 867 elections (51 states over 17 election years) in my merged CPS data set: