Reassessing Public Support for a Female President

Abstract

We re-deploy a list experiment conducted a decade ago to reassess the degree to which the American public opposes electing a woman as president. We find that opposition has been cut in half from approximately 26% to 13%. In addition, opposition is now concentrated in specific sociodemographic categories rather than being evenly distributed. Newly developed statistical methods that permit multivariate analysis of list experiment data reveal that resistance has all but disappeared among Democratic-leaning groups in the electorate. These patterns appear to reflect the reduction of uncertainty among groups most favorable toward the recent success of Democratic women.

Full text

SHORT ARTICLE
Reassessing Public Support for a Female President
Barry C. Burden, University of Wisconsin–Madison
Yoshikuni Ono, Tohoku University
Masahiro Yamada, Kwansei Gakuin University
We re-deploy a list experiment conducted a decade ago to reassess the degree to which the American public opposes
electing a woman as president. We find that opposition has been cut in half from approximately 26% to 13%. In ad-
dition, opposition is now concentrated in specific sociodemographic categories rather than being evenly distributed.
Newly developed statistical methods that permit multivariate analysis of list experiment data reveal that resistance has
all but disappeared among Democratic-leaning groups in the electorate. These patterns appear to reflect the reduction
of uncertainty among groups most favorable toward the recent success of Democratic women.
More than two centuries after the country’s found-
ing, a woman has yet to be elected president of the
United States. With Hillary Clinton becoming the
nation’sfirst female major party nominee for that position,
it is worth reconsidering how willing the American public is
to vote for a female presidential candidate. We re-deploy an
experiment conducted a decade ago and show that opposi-
tion to a female president has been cut in half. Using newly
developed multivariate statistical methods, we find that the
opposition in the electorate now varies tremendously across
subpopulations in ways that reflect experiences within the
political parties in recent years.
Because of the potential for social desirability effects in
surveys, it is difficult to assess public acceptance of a female
president by asking people directly. Respondents opposed
to seeing a woman in the White House are likely to bow to
prevailing social norms and falsely report that they are will-
ing to endorse a female president. Alternative methods are
needed to elicit true attitudes on sensitive questions such as
these. One of the options to do so is a “list experiment.”
The list experiment was introduced in political science by
Kuklinski, Cobb, and Gilens (1997) in their study of racial
attitudes. The idea behind it is to avoid social desirability by
allowing respondents to endorse unpopular opinions indi-
rectly. Each respondent is given a list of items and asked how
many they find objectionable. A random half of the sample
is given a list with common irritants such as “Requiring seat
belts to be used when driving”and “Large corporations pol-
luting the environment.”The other half of the sample is given
the same list but with an additional item that is a sensitive
topic. Respondents in both conditions are asked how many
of the items in the lists they saw bothered them. The differ-
ence between the mean number of items selected by the con-
trol and treatment groups is an estimate of what share of the
population was bothered by the item of interest. Because
respondents were not asked which items bothered them but
merely how many, the experiment allows respondents to keep
their unpopular opinions private while also allowing research-
ers to estimate bias.
Streb et al. (2008) used a list experiment for the first time
to measure bias against a female president. In March 2006,
they presented respondents in a national telephone survey
with a list of statements and asked how many of the state-
ments made them “upset.”Survey respondents were ran-
Barry C. Burden (bcburden@wisc.edu) is professor of political science and director of the Elections Research Center at the University of Wisconsin–
Madison, 53706. Yoshikuni Ono (onoy@law.tohoku.ac.jp) is professor of political science at the School of Law at Tohoku University, Japan. Masahiro
Yamada (myamada@kwansei.ac.jp) is professor of political science at Kwansei Gakuin University, Japan.
This research was financially supported by the Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Scientific Research (26285036;
26780078) and the Kwansei Gakuin University Research Grant. Yoshikuni Ono also received the JSPS postdoctoral fellowship for research abroad. The study
was approved by the research ethics board of Kwansei Gakuin University. Data and supporting materials necessary to reproduce the numerical results inthe
paper are available in the JOP Dataverse (https://dataverse.harvard.edu/dataverse/jop). An online appendix with supplementary material is available at
http://dx.doi.org/10.1086/691799.
The Journal of Politics, volume 79, number 3. Published online May 4, 2017. http://dx.doi.org/10.1086/691799
q2017 by the Southern Political Science Association. All rights reserved. 0022-3816/2017/7903-0022$10.00 1073
This content downloaded from 192.218.160.009 on December 10, 2017 22:01:13 PM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

domly assigned to a control group or a treatment group.
The control group received four statements. The treatment
group’s list added a fifth statement: “A woman serving as
president.”(The full protocol is provided in the appendix,
available online.) The Streb et al. study yielded two impor-
tant findings.First,themeannumberofitemsselectedwas2.16
in the control condition compared to 2.42 in the treatment
condition, for an overall difference of .26, or 26% of respon-
dents being upset about a woman president. That percentage
was much higher than the approximately 10% share of the
public that opposed a qualified female presidential candidate
when asked about it directly. Second, the prevalence of being
upset about a female president was quite stable across a range
of demographic groups. Views were no different between men
and women, those with more or less education, or people of
different age groups.
A variant on this list experiment was fielded by Benson,
Merolla, and Geer (2011), in which respondents were pre-
sented with statements such as “I could not support a woman
for President.”Posing the statements in this negative fashion
yielded an estimated bias of 11% in 2007 and of 17% in 2008,
although neither was statistically significant. However, it is
difficult to compare these results directly to those of the Streb
et al. study because the sample sizes are much smaller, the
study did not examine differences across groups (aside from
born-again southerners), and the question wordings were
substantially different.
STUDY DESIGN
Our study was fielded in March 2016 using a protocol nearly
identical to that of Streb et al. (2008).
1
Because our survey
was conducted via the Internet rather than by phone, one
might be concerned that differences in mode would con-
found a comparison of the two sets of results. However, when
research has found differences due to mode, self-administered
Internet surveys generally result in higher levels of socially
undesirable characteristics (Lind et al. 2013). Thus, any de-
cline we observe in opposition to a female president probably
understates rather than overstates changes over time.
At the time that Streb et al. conducted their analysis,
studies using list experiments did little more than compare
means between control and treatment groups. This was done
across various subpopulations one at a time rather than si-
multaneously. For example, Streb et al. examined the ex-
perimental effects by income and by age independently, even
though the two variables are probably correlated. In recent
years, however, more sophisticated techniques have been
developed to permit multivariate modeling so that the ef-
fects of being in a demographic or social group can be more
accurately and efficiently estimated after controlling for con-
founding variables. This is especially useful in our reassess-
ment because our survey includes a set of highly relevant at-
titudinal variables such as party identification that were not
considered by Streb et al. (2008). We introduce these multi-
variate models in the following section.
Our theoretical expectations consider how the publiclearns
from experiences with people of different backgrounds in
public life. Attitudes toward social groups and acceptability
of various demographic characteristics in the public sphere
often change rapidly in response to real world experiences.
Research has shown that the election of a black mayor re-
duces public opposition to black candidates in future elec-
tions (Hajnal 2007). This appears to happen because observ-
ing a person in public life reduces uncertainty about how a
member of that group would act in office. Since the Streb
et al. experiment was conducted in 2006, several women
have made their way into high-profile political positions.
Two women were appointed to the Supreme Court. Sarah
Palin was chosen as a vice presidential running mate on the
Republican ticket. Nancy Pelosi served as speaker of the
house, putting her second in line to the presidency. Most
notably, in 2008, Hillary Clinton nearly became the first
major party presidential nominee. Ten years ago, respon-
dents would have been required to imagine a hypothetical
woman in the White House; today that imaginary leap is
much easier to make. Indeed, Clinton was an active candi-
date for president at the time our study was fielded. We ex-
pect that these experiences have changed public attitudes, es-
pecially among Democratic constituencies, given that Clinton
and Pelosi are both Democrats. To examine this more com-
pletely, we included a wider range of covariates in our survey,
namely, party identification and race/ethnicity.
UNIVARIATE RESULTS
Table 1 presents the main univariate results from our ex-
periment alongside the results from Streb et al. (2008). The
broad finding is that the overall level of being upset about a
woman president has been cut in half, from 26% to 13%
(technically, a proportion of .126). We are confident about
the comparability of the two studies because the mean num-
ber of selected items in the control condition is nearly iden-
tical (2.17 in our study vs. 2.16 in theirs). The main difference
is in the treatment condition, where the number of selected
items has shrunk from 2.42 to 2.30. It seems that events over
1. The introductory script differs slightly. The full wording is provided
in the appendix.
1074 / Reassessing Public Support for a Female President Barry C. Burden, Yoshikuni Ono, and Masahiro Yamada
This content downloaded from 192.218.160.009 on December 10, 2017 22:01:13 PM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table 1. Opposition to a Female President, by Various Subgroups
Demographic Control Condition Treatment Condition Difference Streb et al. (2008) Estimate
All respondents 2.17 2.30 12.6* 26.0***
(.04) (.04) (5.5) (4.9)
Male 2.03 2.29 25.7*** 26.0***
(.05) (.06) (8.3) (7.2)
Female 2.30 2.31 .1 25.6***
(.05) (.05) (7.3) (6.6)
No BA degree 2.23 2.41 18.3** 23.2***
(.05) (.06) (7.5) (6.3)
BA or above 2.09 2.16 7.5 26.4***
(.06) (.06) (8.2) (7.8)
18–29 years old 2.14 2.42 28.4* 24.9*
(.08) (.09) (12.7) (12.1)
30–50 years old 2.22 2.43 20.2** 35.9***
(.06) (.06) (8.5) (8.1)
51–65 years old 2.16 2.10 25.9 22.2*
(.07) (.07) (10.0) (10.7)
66 years old or above 2.07 2.06 -1.3 12.3
(.11) (.15) (18.1) (9.5)
Lower and lower-middle class 2.29 2.39 9.9 26.8*
(.07) (.07) (9.7) (12.2)
Middle class 2.15 2.23 8.4 28.8**
(.05) (.06) (7.5) (9.1)
Upper-middle and upper class 2.03 2.32 29.1* 29.3*
(.10) (.11) (14.7) (14.3)
South 2.26 2.26 -.1 31.8***
(.07) (.07) (9.3) (9.3)
Non-South 2.13 2.32 19.4** 23.6***
(.04) (.05) (6.9) (5.8)
White 2.18 2.28 10.2
(.04) (.05) (6.6)
Black 2.04 2.17 13.0
(.11) (.13) (17.2)
Hispanic 2.21 2.63 41.1**
(.10) (.12) (15.5)
Other race/ethnicity 2.27 2.13 -13.8
(.14) (.15) (20.6)
Democrat 2.24 2.27 3.2
(.06) (.06) (8.1)
Republican 2.11 2.39 27.4**
(.07) (.09) (10.8)
Independent 2.16 2.27 11.2
(.07) (.08) (10.9)
Note. Entries in the first two columns are mean number of items, with standard errors in parentheses. Social class is measured as explicit categories in our
study but is measured by annual income groupings in the Streb et al. study.
*p!.05.
** p!.01.
*** p!.001.
This content downloaded from 192.218.160.009 on December 10, 2017 22:01:13 PM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

the past decade have lessened bias against a female presi-
dential candidate substantially.
2
As noted above, this runs
contrary to what one would expect from survey mode effects
alone. Instead, the change is consistent with quickly moving
attitudes due to observing high-profile political women such
as Clinton and Pelosi.
Our experiment turned up another intriguing finding.
Whereas the demographic groups analyzed in the Streb
et al. study displayed nearly equal levels of bias, those groups
(and others) now display tremendous variation. For exam-
ple, Streb et al. found, surprisingly, that men and women
showed equal negativity toward a female president. We find
the same level of bias among men as they did (26%), but bias
among women has disappeared.
3
Streb et al. also found no
differences by educational attainment, whereas our experi-
ment shows less bias among the college-educated.
The differences we uncover are consistent with our ex-
pectation that bias has decreased mainly among Democratic-
leaning groups. Self-identified Democrats do not show sig-
nificant levels of hostility toward a female president, whereas
Republicans do. Independents are placed somewhere in be-
tween Democrats and Republicans. The growing acceptance
of a woman in the White House has taken place almost ex-
clusively among subpopulations that are most favorable to
the Democratic women who have had the most success in
high-level political offices.
MULTIVARIATE RESULTS
As compelling as these univariate results are, they capture
descriptive differences across groups in isolation without con-
sidering overlapping group memberships. They also make
inefficient use of the data. Fortunately, a new class of esti-
mators now permits multivariate analysis in a form that is
analogous to familiar regression models applied to traditional
data sets. These models essentially generalize the difference
of means approach by efficiently modeling the joint distri-
bution to allow for control for multiple explanatory vari-
ables simultaneously. We implement the maximum likeli-
hood models developed in Blair and Imai (2012) and Imai
(2011) and refer readers to those sources for derivation of the
estimator.
4
The full regression results, whose coefficient estimates
appear in the appendix, are provided in graphical form in
figure 1. The figure indicates the estimated proportions of
respondents opposing to a female president (with lines rep-
resenting 95% confidence intervals). Several of the descrip-
tive univariate results continue to hold with multivariate
controls, but some conclusions must be revised. We con-
tinue to find that men are more opposed to a female presi-
dent than are women, albeit with an estimated difference of
13.1 percentage points, which is about half of the simple dif-
ference of means. A similar consistency holds with regard to
age, where older respondents are more acceptant of a female
president. Republicans also remain more biased against a fe-
male president than do Democrats and Independents; esti-
mated differences in the multivariate results are 17.4 and
12.4 percentage points, respectively. In addition, social class
still has a significant effect on the estimated proportions of
respondents opposing to a female president. Perhaps sur-
prisingly, people who claim to be “upper class”are more
hostile to a female president than those in “lower class.”It is
possible that class is a proxy for ideology. In further analy-
sis, we confirmed that perceived social class and conservative
ideology are in fact positively correlated. Based on our theo-
retical orientation, we conjecture that conservatives who
view themselves as upper class are more uncertain about
what election of a female president would mean for their
ideological interests.
Some results that appeared counterintuitive in table 1 are
now more sensible once multivariate controls are in place.
Whereas the univariate results suggested that Hispanics are
more opposed to a female president than are whites and blacks
and that southerners are less opposed than nonsoutherners,
both of these apparent effects disappear under further scru-
tiny. Figure 1 shows no differences across race/ethnicity or
region.
5
The opposition to a female president observed among
lower-educated respondents in the univariate results also be-
comes less evident once we control for other confounding
factors.
It is possible that the attitudes we are measuring are little
more than a “Hillary effect.”Rather than capturing general
views about a woman as president, we might instead be tap-
ping into views of Hillary Clinton as the most likely female
president. Streb et al. suggest that a test of this idea would
include a control for attitudes toward Clinton in the model.
We have done that, re-estimating the multivariate model af-
ter including a measure of favorability toward Hillary Clin-
ton. Results reported in the appendix show that attitudes
2. The 13.4 percentage point difference-in-difference between the Streb
et al. estimate and ours is significant at pp.07 by two-way ANOVA.
3. The difference between the two estimates for women is itself sta-
tistically significant at p!.05.
4. We implement the constrained model in version 8.0 of the package
“list”written for R.
5. The results remained the same substantively when using “blacks
and others”(instead of “Hispanic and others”) as the reference category in
the regression model.
1076 / Reassessing Public Support for a Female President Barry C. Burden, Yoshikuni Ono, and Masahiro Yamada
This content downloaded from 192.218.160.009 on December 10, 2017 22:01:13 PM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

toward Clinton play little role in whether respondents are
upset by a female president. Other substantive results remain
unchanged.
CONCLUSION
Having re-deployed the Streb et al. (2008) study, we now
have two data points that bracket important political changes
for women in public life. The public is now less hostile to-
ward a female president overall, but levels of resistance in the
electorate have become uneven, with Democratic-leaning
groups showing the lowest levels of opposition with other
groups little changed.
Researchers should fully explore other techniques for
eliciting opinions on sensitive issues. These procedures in-
clude the randomized response technique and endorsement
experiments (e.g., Blair 2015), alternative versions of the list
experiment (Benson et al. 2011), the implicit association test
(IAT) that measures automatic attitudes (toward female
leaders) at the unconscious level (e.g., Mo 2015), and “face
saving”methods that allow people to justify their prefer-
ences with explanations (Krupnikov, Piston, and Bauer 201 6)
or attribute their views to others. On this latter idea, we
highlight recent surveys fielded by CNN that have asked
respondents whether they believe the “country”is “ready”to
Figure 1. Multivariate estimates of opposition to a female president. Dots represent estimated proportions of respondents upset by a female president, and
lines are 95% confidence intervals from the regression model in table A2 in the appendix.
Volume 79 Number 3 July 2017 / 1077
This content downloaded from 192.218.160.009 on December 10, 2017 22:01:13 PM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

have a woman in the White House. We suspect that focus-
ing on a generalizable “other”rather than the respondent
provides a face-saving way to report one’s views more accu-
rately. As we show in the appendix, the list experiment out-
come (13% opposition) is indeed closer to the most tempo-
rally proximate CNN poll (19%) than the more traditional
question responses that ask about biases directly.
ACKNOWLEDGMENTS
We thank Rikhil Bhavnani for helpful comments, Yusaku
Horiuchi for sharing his R scripts, and Masahiro Zenkyo for
his assistance with the data collection process.
REFERENCES
Benson, Brett V., Jennifer L. Merolla, and John G. Geer. 2011. “Two Steps
Forward, One Step Back? Bias in the 2008 Presidential Election.”
Electoral Studies 30:607–20.
Blair, Graeme. 2015. “Survey Methods for Sensitive Topics.”APSA Com-
parative Politics Newsletter 25:12–15.
Blair, Graeme, and Kosuke Imai. 2012. “Statistical Analysis of List Ex-
periments.”Political Analysis 20:47–77.
Hajnal, Zoltan. 2007. Changing White Attitudes toward Black Political
Leadership. New York: Cambridge University Press.
Imai, Kosuke. 2011. “Multivariate Regression Analysis for the Item Count
Technique.”Journal of the American Statistical Association 106:407–16.
Krupnikov, Yanna, Spencer Piston, and Nichole M. Bauer. 2016. “Saving
Face: Identifying Vote Responses to Black Candidates and Female Can-
didates.”Political Psychology 37:253–73.
Kuklinski, James H., Michael D. Cobb, and Martin Gilens. 1997. “Racial
Attitudes and the ‘New South.’” Journal of Politics 59:323–49.
Lind, Laura H., Michael F. Schober, Frederick G. Conrad, and Heidi
Reichert. 2013. “Why Do Respondents Disclose More W hen Computers
Ask the Questions?”Public Opinion Quarterly 77:888–935.
Mo, Cecilia Hyungjung. 2015. “The Consequences of Explicit and Implicit
Gender Attitudes and Candidates Quality in the Calculations of Voters.”
Political Behavior 37:357–95.
Streb, Matthew J., Barbara Burrell, Brian Frederick, and Michael A.
Genovese. 2008. “Social Desirability Effects and Support for a Female
American President.”Public Opinion Quarterly 72:76–89.
1078 / Reassessing Public Support for a Female President Barry C. Burden, Yoshikuni Ono, and Masahiro Yamada
This content downloaded from 192.218.160.009 on December 10, 2017 22:01:13 PM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).