Present at the Destruction what I learne
Political scientists envision elections as sets of classic puzzles: Can party
elites decide primary elections? Why do people vote at all given the high
costs and low benefits? Why don’t more women run for office? Debates over
answering these questions sustain many researchers’ careers, although as a
specialist in foreign policy I mostly keep up with them by skimming articles
and reading the Monkey Cage. But these puzzles suddenly became of more
than academic interest to me when my mother (hereafter “M”) decided to
run for a seat on the three-member county commission that runs
Middletown County, a community of just under 200,000 people in Indiana.
As the family’s most accomplished (and only) political scientist, I felt an
obligation to contribute to her campaign. I felt confident that I could be
useful. As a grad student teaching research methodology, I had used the
experiments of Gerber and Green (2000) and Nickerson (Nickerson 2008)
and the stories collected in Issenberg (2012) as examples of the real-world
impact of the best political science research. And so I set about
summarizing the newest research and applying its lessons to the myriad
tasks of a campaign: developing and maintaining a voter database;
designing mailers and Facebook ads; and coordinating calling campaigns—
all while also performing my duties as a first-year assistant professor.
I want to share what I learned from this unusual experience. This article is
not “research” in a standard hypothesis-testing or even participantobservation mode. But it is also not a memoir, since I skip most of the
details and almost all of the best stories. It is probably most accurately
thought of as an applied literature review helping to show the strengths and
weaknesses of a field through personal application. I found that political
science has made great strides in rigorously identifying variations in voter
behavior, particularly the marginal effects of turnout strategies. Yet I also
found that Americanist political science is less prepared to grapple with
broader contextual factors, including an electorate that is distrustful of
politics and often hostile to campaigning. Put another way, my experience
validated the insights both of works like Eitan Hersh’s Hacking the
Electorate (2015), a sophisticated quantitative book that argues that
modern campaigns perceive voters as data, and of those like Kathy
Cramer’s The Politics of Resentment (2016), an ethnographic work that
argues that how citizens see themselves shapes how they regard politics
and governance.
This article speaks to three distinct audiences. To specialists in the
campaigns-and-elections literature, I want to provide a reminder of how
ordinary campaigns work, an example of how research findings can (and
can’t) be applied, and to point out research questions that need addressing.
For consumers of the campaigns-and-election literature, I want to add an
informed perspective showing the importance—and limitations—of
specialists’ contributions. And for the discipline as a whole, I want to argue
for the importance of integrating theory with broader observations about
the health of American democracy. The lessons of 2016 include not just
what I learned from my mother’s campaign but those we need to process
from Donald Trump’s victory—including, perhaps, an appreciation of why
research programs and campaign strategies that assume the long-term
stability of institutions might prove to be self-negating prophecies.
Running Scared in the Dark: Uncertainty During the Primary
On the surface, my mother’s campaign was straightforward. The campaign
began with her announcement on November 5, 2015, that she would run for
county commissioner. She was a “quality candidate”—she had not only won
county-wide office before, but she had previously held this seat from 2005
to 2007, when she joined Governor Mitch Daniels’s administration as a state
agency head. On the other hand, her last outing—a 2010 race for state
representative—had ended in narrow defeat (she ran against a senior
Democrat in the most heavily Democratic district in Southern Indiana and
lost by fewer than 200 votes). Her general-election opponent seemed
formidable. A Democrat and the incumbent, he would be seeking his third
four-year term on the commission. Before that, he had served several terms
as a district City Council member for Middletown City, the largest city in
Middletown County (and the third-largest city in Indiana). It would be a
tough campaign: the county went for Romney over Obama 54-44 but Obama
over McCain by 50.6-48.2. Even with favorable baselines, winning the
Republican nomination did not mean a cakewalk in the general election.
The paragraph above lays out most of the salient data points that political
scientists would normally ask about. Given the conventional wisdom about
turnout and partisan identification, it’s likely that the predicted probabilities
favored M’s victory. But, of course, the map is not the territory and the
predicted distribution is only a suggestion of what the real one looks like. In
fact, the most important lesson I took from advising my mother’s campaign
was that we never quite knew what was going on. Even as the campaign
hummed along seemingly from strength to strength, persistent doubts
nagged: how much more money could the campaign raise? Would M get the
newspaper’s endorsement? Did a phone call from a television reporter
presage a good story for M or for her opponent? How could the campaign
respond to opponents’ strategies—and how could it disrupt their plans?
After all, regardless of the longstanding debate of whether “campaigns
matter”, any campaign more or less has to act on the presumption that it
does. These ordinary uncertainties were a lived contradiction to the wellidentified strategies we relied on from the turnout literature.
The biggest uncertainties affected the primary election. The mayor of
Middletown City is the most visible local officeholder in the county; the
incumbent mayor, as a Republican, is also therefore the most influential
local Republican. During the year preceding her announcement, M had
publicly criticized his plans for redevelopment as wasteful and ineffective.
In response, the county party chairman had symbolically expelled her from
the party. (The local Republican women’s club followed suit via a
handwritten note on expensive stationery.) Not coincidentally, several days
after M’s announcement, A, a young White male attorney and political
neophyte, suddenly announced that he would run against her in the primary.
We learned that the mayor’s team—who had raised over a million dollars to
support his 2015 re-election—was fundraising for A in earnest. Another
complication arose when a third candidate, B, emerged; B was a retired
White woman and Tea Party activist who had appeared on Glenn Beck’s
show.
All of these complicated the campaign’s decisions. As a devoted student of
Bawn et al’s theory of parties (2012), I was nervous about how M would fare
if partisans coalesced around the mayor’s signals.1 Would it be good or bad
for her if the election turned into a referendum on the mayor instead of a
focus on M’s record? (Indeed, the local media bought into the “referendum”
narrative early on, to my irritation.) But even beyond that, uncertainties
loomed: what would matter more in the primary: the fact that M had vastly
higher name recognition or the fact that A would enjoy a massive
fundraising advantage? Similarly, given the dynamics of the Republican
presidential primary, would M’s record of public service hurt her relative to
A’s youth and B’s conservative activism? Moderate Indiana Senator Richard
Lugar’s 2012 primary defeat by conservative insurgent Richard Mourdock
loomed over the primary (Hershey 2012). Finally, a quirk of the alphabet
added one last worry: ballot-order effect. Meredith and Salant’s (2013) find
that the first-positioned candidate gains (in this case, B) at the expense of
the median-ranked candidate (in this case, M). In a close election, every
vote could matter (Ho, & Imai 2008).
1 My impressions here contradict Enos and Hersh (2015), who find that campaign
operatives typically overestimate their chances of victory. Perhaps being literally remote
from the campaign helped; perhaps routinely giving lessons on overconfidence and
groupthink and disaster properly re-set my priors; perhaps I am just neurotic.
We couldn’t afford polling to answer these questions. (Our total budget for
the primary was only a few thousand dollars; fundraising dried up suddenly
after the mayor’s candidate entered the race.) Nor is it obvious that polling
would have helped. What we needed to know was how voters would change
their mind as they got to know the new candidates. Perhaps A and B would
flame out; perhaps they would display previously unknown political talents.
It is likely that polls at the beginning of that process would have been as
informative as the mid-2015 polls that showed Hillary easily defeating
Bernie or Jeb handily winning the Republican nomination.
Macropolitical factors compounded these uncertainties. Indiana’s primary
comes late (in 2016, it was held on May 3), and so it had not hosted a
contested Republican presidential race since Ronald Reagan defeated
Gerald Ford in 1976, 51%-49%. We had expected that 2016 would be similar
—that Jeb Bush or Marco Rubio would wrap up the nomination by early
March. The campaign prepared for a turnout in line with historical
experiences of about 7,000 to 10,000 primary voters, an assumption that
guided our targeting for months. By mid-April, however, it became clear
that Indiana would feature Ted Cruz’s last stand against Donald Trump. As
one of the largest cities in the state, candidates and surrogates (including
former Indiana University basketball coach Bob Knight) descended on
Middletown County.2 That suddenly threw into question all of our
fundamental assumptions.
I drew on the 2008 Indiana Democratic primary turnout surge3 and surges
in other 2016 Republican primaries to estimate the turnout surge that a
competitive Republican primary could bring. With only a handful of data
2 I suspect that the “surprisingly competitive” primaries like Indiana’s 2008 and 2016
primaries could prove a boon to students seeking as-if random assignments for salience
and turnout’s effect on down-ballot races.
3 On Indiana’s 2008 presidential primary, see Hershey (2008).
points, I guesstimated that Republican turnout would range from between
14,000 and 40,000 voters, most likely nearer 20,000 to 25,000 voters.
Despite all of my graduate training’s emphasis on getting the most precise
estimate possible, adjusting ourselves for a big range proved Tukey’s
dictum about the value of an approximate answer to the right question. This
was a broad range, but it spurred changes in outreach strategies to reach a
less well-identified electorate. Our data vendor relied on primary voting
history to identify partisans—exactly as Hersh (2015) describes. 4 One
consequence of low Republican primary turnout over four decades was that
we had poor data to guide us about who was a Republican. We “knew” only
20,000 identified Republican voters going into an election in which as much
as twice as many might turn out. 5 (In the event, Republican turnout totaled
26,772.) We therefore shifted tactics from reliance on pure voter contact,
including buying ads on conservative talk radio (apparently grabbing
timeslots the Cruz campaign wanted a few days later), and sending mailers
to a much larger population.
In the end, M received 49.35 percent of the vote to A’s 38.32 percent and
B’s 12.33 percent. Figure 1 displays M’s precinct-level share of the top-two
candidates’ votes plotted against Trump’s share of the Republican vote
(Trump carried the county with 54.71 percent, trailed by Cruz’s 34.97
percent and 8.16 percent for Kasich). Bivariate OLS regression suggests
that every additional percentage point of Trump turnout predicted a 0.34
percentage point increase in M’s vote share ( .096; p
elites decide primary elections? Why do people vote at all given the high
costs and low benefits? Why don’t more women run for office? Debates over
answering these questions sustain many researchers’ careers, although as a
specialist in foreign policy I mostly keep up with them by skimming articles
and reading the Monkey Cage. But these puzzles suddenly became of more
than academic interest to me when my mother (hereafter “M”) decided to
run for a seat on the three-member county commission that runs
Middletown County, a community of just under 200,000 people in Indiana.
As the family’s most accomplished (and only) political scientist, I felt an
obligation to contribute to her campaign. I felt confident that I could be
useful. As a grad student teaching research methodology, I had used the
experiments of Gerber and Green (2000) and Nickerson (Nickerson 2008)
and the stories collected in Issenberg (2012) as examples of the real-world
impact of the best political science research. And so I set about
summarizing the newest research and applying its lessons to the myriad
tasks of a campaign: developing and maintaining a voter database;
designing mailers and Facebook ads; and coordinating calling campaigns—
all while also performing my duties as a first-year assistant professor.
I want to share what I learned from this unusual experience. This article is
not “research” in a standard hypothesis-testing or even participantobservation mode. But it is also not a memoir, since I skip most of the
details and almost all of the best stories. It is probably most accurately
thought of as an applied literature review helping to show the strengths and
weaknesses of a field through personal application. I found that political
science has made great strides in rigorously identifying variations in voter
behavior, particularly the marginal effects of turnout strategies. Yet I also
found that Americanist political science is less prepared to grapple with
broader contextual factors, including an electorate that is distrustful of
politics and often hostile to campaigning. Put another way, my experience
validated the insights both of works like Eitan Hersh’s Hacking the
Electorate (2015), a sophisticated quantitative book that argues that
modern campaigns perceive voters as data, and of those like Kathy
Cramer’s The Politics of Resentment (2016), an ethnographic work that
argues that how citizens see themselves shapes how they regard politics
and governance.
This article speaks to three distinct audiences. To specialists in the
campaigns-and-elections literature, I want to provide a reminder of how
ordinary campaigns work, an example of how research findings can (and
can’t) be applied, and to point out research questions that need addressing.
For consumers of the campaigns-and-election literature, I want to add an
informed perspective showing the importance—and limitations—of
specialists’ contributions. And for the discipline as a whole, I want to argue
for the importance of integrating theory with broader observations about
the health of American democracy. The lessons of 2016 include not just
what I learned from my mother’s campaign but those we need to process
from Donald Trump’s victory—including, perhaps, an appreciation of why
research programs and campaign strategies that assume the long-term
stability of institutions might prove to be self-negating prophecies.
Running Scared in the Dark: Uncertainty During the Primary
On the surface, my mother’s campaign was straightforward. The campaign
began with her announcement on November 5, 2015, that she would run for
county commissioner. She was a “quality candidate”—she had not only won
county-wide office before, but she had previously held this seat from 2005
to 2007, when she joined Governor Mitch Daniels’s administration as a state
agency head. On the other hand, her last outing—a 2010 race for state
representative—had ended in narrow defeat (she ran against a senior
Democrat in the most heavily Democratic district in Southern Indiana and
lost by fewer than 200 votes). Her general-election opponent seemed
formidable. A Democrat and the incumbent, he would be seeking his third
four-year term on the commission. Before that, he had served several terms
as a district City Council member for Middletown City, the largest city in
Middletown County (and the third-largest city in Indiana). It would be a
tough campaign: the county went for Romney over Obama 54-44 but Obama
over McCain by 50.6-48.2. Even with favorable baselines, winning the
Republican nomination did not mean a cakewalk in the general election.
The paragraph above lays out most of the salient data points that political
scientists would normally ask about. Given the conventional wisdom about
turnout and partisan identification, it’s likely that the predicted probabilities
favored M’s victory. But, of course, the map is not the territory and the
predicted distribution is only a suggestion of what the real one looks like. In
fact, the most important lesson I took from advising my mother’s campaign
was that we never quite knew what was going on. Even as the campaign
hummed along seemingly from strength to strength, persistent doubts
nagged: how much more money could the campaign raise? Would M get the
newspaper’s endorsement? Did a phone call from a television reporter
presage a good story for M or for her opponent? How could the campaign
respond to opponents’ strategies—and how could it disrupt their plans?
After all, regardless of the longstanding debate of whether “campaigns
matter”, any campaign more or less has to act on the presumption that it
does. These ordinary uncertainties were a lived contradiction to the wellidentified strategies we relied on from the turnout literature.
The biggest uncertainties affected the primary election. The mayor of
Middletown City is the most visible local officeholder in the county; the
incumbent mayor, as a Republican, is also therefore the most influential
local Republican. During the year preceding her announcement, M had
publicly criticized his plans for redevelopment as wasteful and ineffective.
In response, the county party chairman had symbolically expelled her from
the party. (The local Republican women’s club followed suit via a
handwritten note on expensive stationery.) Not coincidentally, several days
after M’s announcement, A, a young White male attorney and political
neophyte, suddenly announced that he would run against her in the primary.
We learned that the mayor’s team—who had raised over a million dollars to
support his 2015 re-election—was fundraising for A in earnest. Another
complication arose when a third candidate, B, emerged; B was a retired
White woman and Tea Party activist who had appeared on Glenn Beck’s
show.
All of these complicated the campaign’s decisions. As a devoted student of
Bawn et al’s theory of parties (2012), I was nervous about how M would fare
if partisans coalesced around the mayor’s signals.1 Would it be good or bad
for her if the election turned into a referendum on the mayor instead of a
focus on M’s record? (Indeed, the local media bought into the “referendum”
narrative early on, to my irritation.) But even beyond that, uncertainties
loomed: what would matter more in the primary: the fact that M had vastly
higher name recognition or the fact that A would enjoy a massive
fundraising advantage? Similarly, given the dynamics of the Republican
presidential primary, would M’s record of public service hurt her relative to
A’s youth and B’s conservative activism? Moderate Indiana Senator Richard
Lugar’s 2012 primary defeat by conservative insurgent Richard Mourdock
loomed over the primary (Hershey 2012). Finally, a quirk of the alphabet
added one last worry: ballot-order effect. Meredith and Salant’s (2013) find
that the first-positioned candidate gains (in this case, B) at the expense of
the median-ranked candidate (in this case, M). In a close election, every
vote could matter (Ho, & Imai 2008).
1 My impressions here contradict Enos and Hersh (2015), who find that campaign
operatives typically overestimate their chances of victory. Perhaps being literally remote
from the campaign helped; perhaps routinely giving lessons on overconfidence and
groupthink and disaster properly re-set my priors; perhaps I am just neurotic.
We couldn’t afford polling to answer these questions. (Our total budget for
the primary was only a few thousand dollars; fundraising dried up suddenly
after the mayor’s candidate entered the race.) Nor is it obvious that polling
would have helped. What we needed to know was how voters would change
their mind as they got to know the new candidates. Perhaps A and B would
flame out; perhaps they would display previously unknown political talents.
It is likely that polls at the beginning of that process would have been as
informative as the mid-2015 polls that showed Hillary easily defeating
Bernie or Jeb handily winning the Republican nomination.
Macropolitical factors compounded these uncertainties. Indiana’s primary
comes late (in 2016, it was held on May 3), and so it had not hosted a
contested Republican presidential race since Ronald Reagan defeated
Gerald Ford in 1976, 51%-49%. We had expected that 2016 would be similar
—that Jeb Bush or Marco Rubio would wrap up the nomination by early
March. The campaign prepared for a turnout in line with historical
experiences of about 7,000 to 10,000 primary voters, an assumption that
guided our targeting for months. By mid-April, however, it became clear
that Indiana would feature Ted Cruz’s last stand against Donald Trump. As
one of the largest cities in the state, candidates and surrogates (including
former Indiana University basketball coach Bob Knight) descended on
Middletown County.2 That suddenly threw into question all of our
fundamental assumptions.
I drew on the 2008 Indiana Democratic primary turnout surge3 and surges
in other 2016 Republican primaries to estimate the turnout surge that a
competitive Republican primary could bring. With only a handful of data
2 I suspect that the “surprisingly competitive” primaries like Indiana’s 2008 and 2016
primaries could prove a boon to students seeking as-if random assignments for salience
and turnout’s effect on down-ballot races.
3 On Indiana’s 2008 presidential primary, see Hershey (2008).
points, I guesstimated that Republican turnout would range from between
14,000 and 40,000 voters, most likely nearer 20,000 to 25,000 voters.
Despite all of my graduate training’s emphasis on getting the most precise
estimate possible, adjusting ourselves for a big range proved Tukey’s
dictum about the value of an approximate answer to the right question. This
was a broad range, but it spurred changes in outreach strategies to reach a
less well-identified electorate. Our data vendor relied on primary voting
history to identify partisans—exactly as Hersh (2015) describes. 4 One
consequence of low Republican primary turnout over four decades was that
we had poor data to guide us about who was a Republican. We “knew” only
20,000 identified Republican voters going into an election in which as much
as twice as many might turn out. 5 (In the event, Republican turnout totaled
26,772.) We therefore shifted tactics from reliance on pure voter contact,
including buying ads on conservative talk radio (apparently grabbing
timeslots the Cruz campaign wanted a few days later), and sending mailers
to a much larger population.
In the end, M received 49.35 percent of the vote to A’s 38.32 percent and
B’s 12.33 percent. Figure 1 displays M’s precinct-level share of the top-two
candidates’ votes plotted against Trump’s share of the Republican vote
(Trump carried the county with 54.71 percent, trailed by Cruz’s 34.97
percent and 8.16 percent for Kasich). Bivariate OLS regression suggests
that every additional percentage point of Trump turnout predicted a 0.34
percentage point increase in M’s vote share ( .096; p