Content uploaded by Yoshi Kobayashi
Author content
All content in this area was uploaded by Yoshi Kobayashi on Dec 18, 2021
Content may be subject to copyright.
The Missing Fingerprints: U.S. Women Legislators and
International Development Aid
Katelyn E. Stauffer1, Yoshiharu Kobayashi∗2, Kelsey M. Martin-Morales1, Riley
Lankes1, Tobias Heinrich1, and Catherine R. Goodwin1
1University of South Carolina, Columbia, USA
2University of Leeds, Leeds, UK
December 12, 2021†
Abstract
There is a optimism that a growing number of women in political office will reorient
the focus of international politics towards more social and humanitarian issues. One
basis for this optimism are arguments that women legislators hold distinct foreign
policy preferences and act on them to affect changes in policy. However, we know
little about gender differences in behavior of individual legislators on these issues.
This study investigates the behavior of individual legislators of the United States,
the most important actor in international politics, in the context of development
aid. Analyzing a diverse set of legislative behaviors in the U.S. Congress, we find
no evidence that women legislators behave any differently than men with regards
to these issues. Beyond its contribution to our understanding of the making and
future of American foreign policy, this study contributes to broader debates about
women’s representation and foreign policy.
Keywords: gender; legislative behavior; foreign policy; foreign aid; the United States
∗Corresponding Author: Y.Kobayashi@leeds.ac.uk
†We thank Ben Fordham, Greg Vonnahme, Matt Winters, Amanda Clayton, participants at the 2019
Midwest Political Science Association Meeting, and participants at the 2019 European Conference on
Politics and Gender for useful suggestions and comments and Kenneth Lowande for sharing his data.
1
More women are gaining access to political power around the world, affording them in-
creased opportunities to shape domestic and foreign policies. In public and academic
discourses, there is optimism that this trend will shift international politics away from
military interventions to social and humanitarian issues (Slaughter, 2012). Findings in
the academic literature on gender and foreign policy may lend credence to this optimism.
Evidence across countries shows that greater women’s representation in a parliament is
associated with the country spending less on defense (Koch and Fulton, 2011; Clayton and
Zetterberg, 2018), contribute more against climate change (Mavisakalyan and Tarverdi,
2019), involving itself less in violent international conflict (Caprioli, 2000; Regan and
Paskeviciute, 2003), having lower tariffs (Imamverdiyeva et al., 2021), and contributing
more to global humanitarian actions (Breuning, 2001; Shea and Christian, 2017). While
these findings show a consistent pattern between gender, legislative representation, and
foreign policy, the underpinnings of this relationship are not well-understood.
One explanation for the observed, country-level associations is the women’s values
thesis, which contends that women politicians hold distinct foreign policy preferences on
which they act to influence policy (Breuning (2001) but see also Togeby 1994; Koch and
Fulton 2011; Hicks et al. 2016).1Such actions in parliament manifest themselves in pro-
motion of more humanitarian and less militarist foreign policy, which in turn effect policy
change. Researchers buttress the mechanisms behind the women’s value thesis by invok-
ing findings from public opinion surveys and lab experiments that find women to be more
altruistic and more supportive of peaceful and humanitarian foreign policy (Eichenberg,
2016; Andreoni and Vesterlund, 2001; Lizotte et al., 2020). However, we know remarkably
little about whether these results extrapolate to elite politicians’ preferences2and, impor-
tantly, whether and through which specific legislative venues women politicians actually
1An important counter-explanation to the women’s values thesis is the social equity thesis (Breuning,
2001; Lu and Breuning, 2014), which contends that countries valuing equality see both increases
in women representation in the legislature and higher levels of aid expenditure. While these two
possible explanations are often examined jointly, in this article we focus on the women’s values thesis
exclusively so that we can test it more rigorously.
2Exceptions are McGlen and Sarkees (1993), Holsti and Rosenau (1981), Imamverdiyeva et al. (2021),
Bashevkin (2014), and Bashevkin (2018), which provide mixed evidence.
1
affect such humanitarian foreign policy.3
In this study, we focus on politicians’ legislative behavior to directly test the key mech-
anism underlying the women’s value thesis, which underlies much of the recent research
linking parliamentary gender compositions and foreign policy. In the study most similar
to our own, Lu and Breuning (2014) also explore this question by comparing the relation-
ship between women’s inclusion in legislatures and key cabinet posts and aid expenditure.
They conclude that the relationship between gender and aid expenditures is likely rooted
in social equity, after observing that though women’s presence in legislative politics is
positively associated with foreign aid, women heads of foreign-policy related ministries
(presumably the women with the greatest policy influence in this area) are not. While
this research certainly provides insights into the present study, it does not allow us to
understand how gender may influence the specific behaviors of elites. Rather than mak-
ing cross-institutional comparisons, we directly examine individual legislators’ behaviors
across multiple fora.
We argue that it is crucial to analyze multiple channels through which legislators can
influence foreign policy-making. While roll-call voting is one obvious legislative activity
to examine (Bendix and Jeong, 2020), it is also one of the most visible legislative activities
and one that is highly structured so that legislators might be prevented from actuating
preferences that deviate from their political party’s (Snyder Jr and Groseclose, 2000). By
examining multiple forms of legislative behavior, including bill co-sponsorship, conduct
in hearings, and monitoring of the bureaucracy, this study provides a fuller analysis of
whether—and in what ways—women legislators may affect foreign policy.
We conduct this analysis in the context of U.S. development aid. We focus on the
United States because it is the country whose foreign policy has the most influence in
the world and which still has a relatively small, yet increasing, share of women in the
legislature. If the women’s values thesis holds, we would predict that U.S. foreign policy
will become more and more humanitarian at the margin. We think these stakes about
the future makes it an important case to study the mechanisms undergirding the women’s
3An exception is Bendix and Jeong (2020) who study roll-call voting on military matters in the U.S.
Congress and find limited evidence of gender-differences.
2
values thesis in this particular context.
We analyze data from the U.S. House of Representatives on roll-call voting (1981–
2008), bill co-sponsorship (1985–2008), participation in hearings (2007–2018), and moni-
toring of the aid bureaucracy (USAID, 2007–2010). Since these fora have varying levels
of visibility and discretion afforded to individual legislators, if the women’s value the-
sis holds, we expect to see gendered differences in behavior in at least one of our four
outcomes.
Our estimation strategy directly targets the differences between women and men legis-
lators beyond their gender, i.e. the selection effect. Indeed, unsurprisingly, our data reveal
pronounced gender differences in covariates, including revealed ideology, constituents’ in-
come, educational and ethnic profiles, and state-level ideological orientations. First, we
use entropy balancing to reweight observations (by year and party) so that covariate av-
erages are very similar across gender subsets (Hainmueller, 2012). Second, we also control
for these covariates when estimating the effect of gender on the various outcomes. Com-
bining entropy balancing and linear regression with covariates leads to a doubly robust
estimator (Zhao and Percival, 2017), improving the confidence in the inferences from these
data.
Our results provide no evidence that women act any differently than men in using
legislative levers to influence U.S. aid policy. Across a series of models, all differences in
behavior are tiny (mean estimates), at times opposite to the expected direction, and noisy.
Our analyses by party and time periods do not reveal any noteworthy heterogeneity in the
association between a legislator’s gender and behavior in the context of foreign aid. Our
additional analyses using using random forests, which account flexibly for interactions
and context-dependent gender differences, confirm our core findings (Breiman, 2001).
The results from the diverse set of legislative behavior provide little support for the
women’s values thesis, at least in the United States. Women legislators do not advocate for
development issues any differently than men. With the key mechanism of the women’s
values thesis not at play in the United States, the optimism for a more humanitarian
foreign policy as women enter parliament in the United States, the most powerful country
3
in the world and the largest overall donor of development aid, should be scaled back.
Though the nature of our analysis varies considerably, our substantive conclusions are in
line with Lu and Breuning (2014).
Even though our analyses examine legislative behavior and do not address preferences
per se, the results do lead us to question the validity of the assumptions underlying the
women’s value thesis. First, if women legislators have more humanitarian preferences,
then we would expect that they use at least one of legislative levers to influence aid pol-
icy, especially those that are less visible to party leadership and constituents. However,
women and men do not differ in promoting international development through less visible
activities, such as monitoring USAID. Second, experimental evidence from a companion
paper shows that U.S. voters neither punish nor reward any politician, women or men,
for advocating for international development at the expense of promoting other issues
(AUTHORS, 2021). Thus, there is no need for women politicians to fear electoral back-
lash from voters if they act on their humanitarian preferences and promote international
development.4Yet, our behavioral results show no gender differences in advocating for
international development. Our study challenges arguments that women legislators have
distinct, more humanitarian preferences than men in the context of foreign aid. We discuss
this point at length in the discussion section.
While our focus is on international development, our results also speak to women’s
representation and US foreign policy more broadly. Because aid is of relatively low salience
and is perhaps the foreign policy area least-associated with masculinity, the electoral and
institutional constraints faced by legislators should be weakest in this context. The fact
that we find no evidence of women legislators promoting aid suggests that we should not
expect gender differences to manifest in other foreign policy domains, where electoral,
partisan, and other institutional constraints will be stronger. Indeed, a study on voting
on military matters in the U.S. Congress finds that gender-differences largely disappear
after accounting for legislators’ party affiliations (Bendix and Jeong, 2020).
Beyond this contribution to our understanding of the present and future of U.S. foreign
4Similarly, if we had found differences in legislative behavior, electoral considerations would not
incentivize women to take pro-aid stances.
4
policy, the paper engages with broader debates over explanations for the observed relation-
ships between women representations and foreign policy (see also Lu and Breuning 2014).
Our results clearly demonstrate the usefulness in analyzing individual politicians to di-
rectly test key mechanisms in these explanations (Bendix and Jeong, 2020; Imamverdiyeva
et al., 2021). In addition, while our analyses are limited to the United States, our fail-
ure to find evidence for the woman’s value thesis call for more attention to alternative
explanations and to potential scope conditions.
Gender and Development Aid
The relationship between women representation and foreign policy has been documented
in a range of policy domains, including defense spending (Koch, 1997; Clayton and Zetter-
berg, 2018), tariffs (Betz et al., 2021; Imamverdiyeva et al., 2021), climate change cooper-
ation (Mavisakalyan and Tarverdi, 2019), use of force (Caprioli, 2000; Caprioli and Boyer,
2001; Regan and Paskeviciute, 2003), and humanitarian intervention (Shea and Chris-
tian, 2017). The common thread is that greater parliamentary representation for women
is associated with a more humanitarian and peaceful foreign policy. Prominently, many
studies report fairly consistent, positive associations between women’s seats shares in na-
tional legislatures and aid expenditures (Breuning 2001; Lu and Breuning 2014; Hicks
et al. 2016; Fuchs and Richert 2018; Yoon and Moon 2019; Okundaye and Breuning 2021,
but see Lundsgaarde et al. 2007; Fuchs et al. 2014).5A similar association is found with
higher aid quality, an important dimension of foreign aid which assesses how well a given
amount of aid is targeted to serve those most in need (Hicks et al. 2016; Heinrich and
Kobayashi 2021).6Unfortunately, our understanding of the mechanisms that produce
these associations is scant.
In her foundational work, Breuning (2001) provides two explanations for the observed
relationship between women’s representation and development aid, and almost all schol-
5These findings have led scholars to employ the share of women as an instrument for development
and democracy aid (Dietrich and Wright, 2014; Ziaja, 2020).
6Some evidence also suggests that the gender of international development ministers is related to aid
quality (Kleemann et al. 2016; Fuchs and Richert 2018; Dreher et al. 2015).
5
arship on the topic—whether implicitly or explicitly—draws on one of them. The first,
which is the focus of the present paper, is the women’s values thesis. It contends that
women politicians hold distinct foreign policy preferences, which manifest in observable
differences in legislative output. The second, the social equity thesis, argues that over-
arching societal preferences and attitudes towards equality lead to higher expenditures
on foreign aid and to more women in legislatures. In this paper, we focus on testing the
mechanisms of the women’s values thesis.
The women’s values thesis holds that women parliamentarians have more development-
minded preferences and act to further development goals. Research commonly draws on
evidence that women are more altruistic towards others, particularly in contexts where
social distance is high—as would be the case with foreign aid (Engel 2011; Eckel and
Grossman 1998)—and are more inequality-averse than men (Dufwenberg and Muren 2006;
Andreoni and Vesterlund 2001). Moreover, women hold more favorable attitudes towards
domestic policies that promote equity and equality (Inglehart and Norris 2000; Luttmer
and Singhal 2011) and development aid than men (Bauhr et al. 2013; Paxton and Knack
2012).7Most of these studies focus on ordinary citizens rather than elites but the few elite
studies that exist report mixed evidence (Holsti and Rosenau 1981; McGlen and Sarkees
1993; Bashevkin 2014).
On the other hand, the social equity thesis argues that overarching societal preferences
and attitudes towards social equity lead countries to invest more in foreign aid, while also
creating a political environment that is more conducive to women’s emergence in politics
(Breuning, 2001; Lu and Breuning, 2014). This explanation calls into question the claim
that women’s legislative actions would be different from men’s, implying that gender-
based compositional differences cannot cause differences in foreign aid outcomes, and
indeed foreign policy outcomes more broadly. In their study of the relationship between
gender and foreign aid expenditure, Lu and Breuning (2014) find that while the percentage
of women serving in parliament is associated with increased aid, female foreign ministers
7While women are more likely to agree that helping poor countries is normatively important, answers
become statistically indistinguishable from (sometimes even more negative than) men’s when asked
about economic aid specifically (Chong and Gradstein 2008; Heinrich et al. 2016).
6
are associated with decreases in aid. Because foreign ministers are more powerful than
rank and file parliamentarians, they conclude that this indicates the social equity thesis
is the more compelling explanation. Making comparisons across positions however may
miss potentially important nuances. Women are not randomly selected to hold ministerial
positions (see Goddard 2019), and it could be the case that the type of women who are
selected for these positions deviate from rank and file members in important ways.
Building on past research we contribute to this debate by focusing on the legislative
behaviors of individual legislators. Assuming that gendered mass attitudes extend to
elites, research has focused on women’s overall presence in the parliament rather than
focusing on the behavior of individual legislators that ought to give rise to the observed
country-level outcomes. Yet, the women’s value thesis posits a specific causal mechanism
based on individual politicians’ behavior in parliament. We argue that a shift towards
individual legislators is a productive way to evaluate the women’s value thesis and to
ultimately contribute to the broader debate about women’s representation and foreign
policy.
We are not the first to study the behaviors of individual legislators to shed light on the
relationship between gender and foreign policy. Notably, two prior studies systematically
examine individual behavior of legislators, but the evidence for the role of gender in for-
eign policy is mixed and limited in scope. Bendix and Jeong (2020) analyze roll-call votes
on security-related bills in the U.S. Congress and report limited evidence for gendered
differences in voting behavior after accounting for legislators’ routine backgrounds like
party. Angevine (2017) and Imamverdiyeva et al. (2021) analyze voting and sponsorship
of bills that deal with foreign women as policy targets and find evidence that women leg-
islators are more likely than men to vote on and introduce such legislation, respectively.
While these insights are useful, there are other fora available where legislators can influ-
ence policy-making, and it is unclear whether the results extend to other, broader policy
domains like foreign aid. Most notably, our study departs from these previous studies by
examining multiple forms of legislative behavior that vary in important ways, which we
outline below.
7
Legislative Behavior on Foreign Aid
One implication of the women’s values thesis is that if women’s presence causes increases
in aid expenditures, we should see men and women legislators engaging in observably dif-
ferent behaviors at the individual level. To test for such differences, in this paper we focus
our attention on the behaviors of legislators in the U.S. House of Representatives.Though
women remain underrepresented in the U.S., especially in comparison to other nations,
their inclusion has been increasing over the last several decades. Moreover, if the link
between women’s inclusion and foreign aid is rooted in individual-level differences, we
should still be able to observe differences in behaviors even in contexts such as Congress,
though as we note below it is important to account for the institutional and contextual
constraints (women) legislators face.
We focus our attention on the U.S. case for two reasons. First, there is great availability
of data across many forms of legislative behavior (e.g. roll call voting, hearings, bill co-
sponsorship, bureaucratic interventions) that are not readily available in other countries.
By examining multiple forms of legislative behavior, we acknowledge that there are many
paths through which legislators can effectuate their preferences, either through policy
discussion, creation, or implementation. Thus, in order to determine whether and how
women influence foreign aid expenditures, accounting for these multiple paths is essential.
Second, the U.S. House of Representatives has been researched more extensively com-
pared to other national parliaments, enabling us to draw on the extensive knowledge to
guide our inquiry. In particular, despite the strong partisanship and extent of women’s
underrepresentation in the U.S. Congress, existing work documents that gender differ-
ences do manifest in women’s behavior in the U.S. Congress in some instances: the types
of bills women (co)sponsor (Swers, 2002), the types of topics women discuss (Pearson and
Dancey, 2011), and the tone used to talk about women’s issues (Dietrich et al., 2019).
Collectively, this work suggests that while institutional constraints might shape the ways
in which women’s behavior in office can diverge from men’s, these factors are not so strong
as to create uniformity. In other words, if women hold distinct preferences on foreign aid,
as assumed by the women’s values theses, there should be some opportunity for them to
8
act on these preferences.
In our analysis, we examine four legislative behaviors, all of which could be used to
influence policy creation or implementation related to foreign aid: roll-call voting, bill
(co)sponsorship, participation in legislative hearings, and contacting bureaucratic agen-
cies. We select these behaviors because they vary considerably on two key dimensions:
visibility and partisan control. Partisanship is the dominant organizing feature in Ameri-
can politics, making it important to take visibility and institutional control seriously. For
example, behaviors that are highly visible, such as bill co-sponsorship, may disincentivize
women from acting on pro-aid preferences if those preferences diverge from her party.
Conversely, men might be incentivized to act in a pro-aid fashion (whether or not they
hold this preference) if it matches their party. In either case, this would reduce differences
in the behaviors observed from men and women legislators. In contrast, low-visibility be-
haviors, such as monitoring bureaucratic agencies, may allow differences in preferences to
manifest as these types of behaviors are less likely to be monitored by party leaders and
to draw ire in cases where representatives act in a manner discordant with the party.8
Likewise, the ability of political parties to structure behavior could influence the
whether we see women engaging in different behaviors than men, even if they hold differ-
ent preferences. For example, in the case of “agenda-response” behaviors such as roll-call
voting, parties have the ability to structure not only how bills are voted on, but which
bills come up for a vote (see Osborn 2012; Snyder Jr and Groseclose 2000). Because of
the high degree of control afforded to parties and the structured nature of the choice-set
offered to legislators in these contexts, party cohesion should be high which may limit the
observance of gendered differences in behavior. In contrast, “agenda-setting” behaviors,
such as bill (co-)sponsorship, are (relatively) less structured by parties in government and
offer legislators more discretion in pursuing their policy interests (see Osborn 2012).
In the sections that follow, we test for differences between men and women for each
of our four legislative behaviors. Roll-call voting—a highly visible, highly structured
act—represents the hardest case for the women’s values thesis. Bureaucratic oversight in
8Our data on such monitoring only became public via a Freedom of Information Act (FOIA) request
by Lowande (2018).
9
contrast represents the easiest test in that it is a low visibility act that is not structured
by party elites. By examining an array of legislative behaviors that vary across several
dimensions, we argue that if women legislators truly hold distinct preferences on foreign
aid, we should see differences emerge in at least one of the behaviors we look at.
Roll-call Voting
The first channel through which legislators can affect aid outlays that we study is roll-call
voting on foreign aid bills. Roll-call voting offers that hardest test for the women’s values
thesis as it is highly visible and structured by political parties (Snyder Jr and Groseclose,
2000; Osborn, 2012). These features should both act as constraints on legislators to act on
their preferences. Indeed, to the extent that scholars have observed gendered differences in
roll-call voting in Congress, the results have been mixed. While early scholarship found
that women were more liberal than their male counterparts (Welch, 1985; Frankovic,
1977), other research finds that, at this stage of the legislative process, gendered differences
are minimal, and factors such as partisanship and ideology are more important predictors
of women’s (and men’s) roll-call behaviors. Indeed, Frederick (2009) reports that to the
extent men and women do engage in different roll-call behaviors, these differences have
been diminishing over time as Congress has become increasingly ideologically polarized.
At the same time, scholars have observed differences in roll-call voting patterns in some
circumstances. For example, Swers (1998a) finds that after accounting for partisanship,
district characteristics, and a host of other controls, Congresswomen are more likely to
vote in support of women’s issues bills. Likewise, Frederick (2015) finds that while men
and women who represent similar districts have “virtually indistinguishable voting records
on the liberal-conservative policy dimension,” women are more supportive of legislation
dealing with women’s interests than men (pg. 103). While not necessarily categorized as
a women’s issue in much of the literature, we do note that foreign aid is often related to
women’s issues through its impact on women, children, and issues such as education and
healthcare. For that reason, we expect women to advocate for aid to a greater extent
than men through their roll-call votes, if women legislator hold more pro-aid preferences
10
compared to their male counterparts.
We start with roll-call data collected by Milner and Tingley (2015), the most com-
prehensive dataset for bills related to foreign aid in the House of Representatives. The
temporal domain spans the 97th to 110th Congresses. We restrict our attention to “eco-
nomic aid” bills which deal mostly with amendments to proposals on aid appropriations
that to seek a change in aid appropriations.9Some bills seek increases to aid generally,
whereas others decreases. For each legislator in a Congress and for each bill, we record
whether the vote is a ‘yay’ for an outcome that would lead to an “increase” in aid com-
pared to the opposite vote; whether one votes ‘nay’ on such outcome. Abstentions are
also possible. We aggregate the votes so that we have the shares of ‘yay’ and ‘nay’ votes
for each person in each Congress, which we multiply by 100 for each of reading.
We augment this dataset by adding a legislator’s gender as well as legislator-specific
and constituency-level data, which prior work has theorized and examined to have effects
on legislative action on foreign aid and on the probability of electing a woman represen-
tative. For legislator-specific variables, we include the legislator’s age and ideology (first
dimension DW-NOMINATE score). For district-level variables, we obtain the percentage
of people born abroad in the district; percent with B.A. and above degrees; percent his-
panic; percentage white; percent unemployed; and, percent with incomes of more than
$10k, $50k, and $100k per year.10 Finally, state-level ideological leanings on social and
economic issues influence who gets elected and what types of legislative behavior is sought.
Thus, we add the estimated social and economic liberalism scores provided by Caughey
and Warshaw (2016).
The resulting dataset has 4,232 legislator-Congress observations, spanning the 97th −
110th Congresses (1981–2008). 424 observations come from 136 unique women.11 A first
9We exclude bills on food and geopolitical aid. For one, food aid is known to be notoriously capture
of the agricultural industry; for the other, geopolitical aid does not clearly relate to the development
objectives implicit in the body of the work on the association between legislative compositions and
aid allocations.
10 These data come from Ella Foster-Molina who assembled them from U.S. census data (https:
//github.com/profEllaFM/congressData)
11 This dataset has minor missingness, which we remedy by using the multiple imputation implemen-
tation by Honaker and King (2010). Throughout, we conduct the estimations and calculations of
effects for each of the ten imputed datasets.
11
glance at the data suggests a strong imbalance by gender across all covariates (Table
A.1). If the imbalance is related to preferences on foreign aid as well, which we would
expect for several covariates, then this selection effect causes problems for the inference.
We take two approaches to remedy the issues, leading to a doubly robust estimation of
the effects of a legislator’s gender on roll-call voting behavior. First, we reweigh the male
legislators’ observations such that the averages of covariates match those in the women
legislators’ data.12 We use entropy balancing to balance the pre-specified moments of the
“untreated” sample (i.e. male legislators) to those of the “treated” sample (i.e. female
legislators) (Hainmueller, 2012). We conduct this reweighting within strata of party and
Congress, ensuring that women legislators are only compared against co-partisans in the
same Congress. The advantage of entropy balancing is that, unlike matching, it targets
covariate imbalance directly, a source of confoundedness and bias. Figure A.1 shows how
the sizable differences in the raw data largely disappear after reweighing.
Second, we control for indicators for party affiliation and different Congresses in a
simple specification and for the legislator-specific and district-specific covariates in a larger
specification.13 We estimate the models via OLS, bootstrap-clustering residuals by party-
Congress. We obtain estimates from each imputed data set, pooling the estimates for a
final result.
Table 1 gives the results for our main estimates. The first column shows the coeffi-
cient when regressing the percentage of ‘yay’ votes cast for aid-increase bills on a female-
indicator as well as controls for the Congress and the legislator’s party, i.e. using the
simple model specification. The coefficient is 0.8, which suggests that a woman legislator
casts less than one percentage-point more votes for a bill that would increase foreign aid.
To get a sense of the magnitude of this, consider that in 2021, there are 101 women in the
House of Representatives. Crudely, our estimates suggest that there would be one more
12 The specific covariates are: the home state’s social and economic liberalism, the first dimension
DW-NOMINATE score, age, and the district’s percentages of people that are foreign born, are
unemployed, hold a four-year college degree, and identify as white, and the percentages of households
that make at least 15k/ 25k/ 35k/ 50k/ 75k/ 100k U.S. dollars per year, respectively.
13 This larger control set uses the same variables that we used for reweighing except that the household
income and the unemployment percentages are replaced by a simple index of district prosperity. It
is created by first turning all the variables in z-scores and then taking the mean across them.
12
Voting yay on aid increase Voting nay on aid increase
Simple Detailed Simple Detailed
Gender
Coefficient 0.8 0.6 -1.6 -1.4
95% CI [-3.6; 5.1] [-3.3; 4.5] [-5.1; 2.0] [-4.6; 1.8]
S.E. (2.3) (2.0) (1.8) (1.6)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 3899 3899 3899 3899
# Women 424 424 424 424
# unique Women 136 136 136 136
Congresses 97–110 97–110 97–110 97–110
Table 1: Estimates for gender-effects in roll call voting on foreign aid using all observa-
tions.The table reports the estimated coefficients on legislator gender (1=female). The Simple model
specifications in Columns 2 and 4 includes indicators for party affiliation and Congresses. The Detailed
specifications in Columns 3 and 5 further include legislator-specific (age, DW-NOMINATE) and district-
specific variables (percentages foreign born, with college degrees, whites, unemployed, income of at least
$10k/ $50k/ $100k).
vote for an aid-increase bill compared to when the whole chamber consistented of men or
if there were no gender differences. The effect is tiny. Even though the point estimate is
in the direction suggested by the women’s values thesis, the uncertainty estimates show
that the coefficient is statistically insignificant as the confidence interval contains zero and
the standard error is almost four times the absolute value of the coefficient.14
The second outcome specification under the simple specifications shows a similar pat-
tern. The third column gives the estimates using the percentages of ‘nay’ votes on the aid
increases. While the change magnitude is larger than for the ‘nay’ percentages, the esti-
mate is still in the predicted direction but also small in magnitude and also statistically
insignificant. Adding the additional legislator-specific and district-specific covariates to
either model (columns 2 and 4) reduces the respective point estimate (absolute value) as
well as the standard error. Nonetheless, the estimates remain statistically insignificant.
14 We refrain from discussing any of the control variables because they were chosen to help with
inference on the gender indicator and not to have a substantive interpretation.
13
In the appendix, we show results from two subset analyses. First, in Table A.1, we
restrict our sample to legislators who are members of the Democratic Party. The four
results are qualitatively the same, with magnitudes shrunk. We refrain from analyzing
the subset of Republicans due the small number of women in it. Second, in Table A.5, we
give the results for more recent Congresses (106th and later). The results are once more
similar, with smaller magnitudes for ‘yay’ vote percentages and generally larger standard
errors.
Bill Co-sponsorship
Having uncovered no evidence of gendered differences in roll-call voting, we next ana-
lyze co-sponsorship of bills. Like roll-call voting, co-sponsorship is a highly visible act.
However, unlike roll-call voting, co-sponsorship offers legislators more flexibility. While
majority parties exert strong control over which bills are voted on, the decision to create
or sign-on to co-sponsor a piece of legislation and determine the content of the bill offers
legislators relatively more agency. Therefore, we might expect that differential prefer-
ences on aid to manifest in the types of policies legislators (co-)sponsor, even if we do not
observe differences in roll-call behaviors when these bills eventually come up for a vote.
Indeed in her study, Swers (1998b) finds the largest differences between Congressmen and
Congresswomen at this stage of the legislative process.
Existing evidence suggests that women use (co-)sponsorship as a means to promote
women’s issues in a way that is distinctive from Congressmen (Volden et al., 2016; Celis,
2006; Franceschet and Piscopo, 2008; Swers and Larson, 2005). Congresswomen are more
likely to (co-)sponsor feminist legislation (Wolbrecht, 2002; Swers, 2002), and more likely
to sponsor bills related to “women’s issues” such as education, healthcare, and childcare
(Swers, 2002, 2016; Swers and Larson, 2005). Angevine (2017) shows parallel patterns
in foreign policy, where women are more likely to sponsor bills where foreign women are
the policy target. Thus, if women legislators have different preferences on foreign aid,
(co-)sponsorship may be a viable avenue to pursue these preferences.
Unlike for roll-call votes, we could not draw on existing research to determine which
14
bills address foreign aid. We first identified bills of potential interest for us out of the
universe of bills. We started with the Congressional Bills Project (Adler and Wilkerson,
2013), which categorizes bills based on topic using the coding system of the Policy Agendas
Project/Comparative Agendas Project. For each Congress between the 97th and 110th , we
sampled 1/3 of all bills that are labeled either “International Affairs” or “Foreign Aid.”
We then hired workers via Amazon’s MTurk platform to crowd-code whether each bill
was about development aid or not, and if it was, we asked them to code whether the
bill increased, decreased, or left the level of aid the same. This gives us 135 aid increase
and 39 aid decrease bills, which we analyze separately. See Section B for details of this
crowd-sourced coding process. Then, we used the Cosponsorships Network Data compiled
by Fowler (2006a,b) to identified all sponsors and cosponsors for all bills to come before
Congress.
Consponsoring aid increases Consponsoring aid decreases
Simple Detailed Simple Detailed
Gender
Coefficient 0.2 0.2 0.0 0.0
95% CI [-0.4; 0.7] [-0.2; 0.5] [-0.6; 0.7] [-0.8; 0.8]
S.E. (0.3) (0.2) (0.3) (0.4)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 4656 4656 4656 4656
# Women 558 558 558 558
# unique Women 131 131 131 131
Congresses 99–110 99–110 99–110 99–110
Table 2: Estimates for gender-effects in cosponsoring legislation on foreign aid using all
observations.The table is constructed like Table 1.
As before, we aggregate the item-specific choices to the legislator-Congress level, cal-
culating the percentage of aid-increase and aid-decrease bills that a person (co)sponsored.
The covariates and model specifications are like before. In Figure A.1, we show that
entropy balancing again reduces gender imbalances in this data set.
15
The results echo those from the roll-call votes as Table 2 shows. Across the different
outcomes and the different model specifications for each, the point estimates are small in
magnitudes and statistically insignificant. The results for the subset of Democrats and
later time periods (Tables A.2 and A.6) look very similar. Even though, (co)sponsorship
affords legislators more freedom to act perhaps in opposition to party leadership, the
results are comparable to those for the roll-call analysis.
Congressional Hearings
So far we have found no evidence that women legislators are more likely to promote for-
eign aid compared to men. We now shift our attention to behaviors related to policy
implementation that are relatively less visible. Examining the implementation stage of
the policy process is essential because bureaucratic agencies handling aid have consid-
erable discretion over the implementation of aid policy (Van Belle, 2004; Arel-Bundock
et al., 2015; Fuchs and Richert, 2018). Legislative activities, such as making inquiries
about policy implementation in hearings and directly contacting bureaucratic agencies,
can essentially work as oversight over the aid bureaucracy (Milner and Tingley, 2015, Ch.
3). Thus, to the extent that women’s presence in legislators is linked to aid expenditures,
we must understand not only what bills are passed, but the practical reality of how these
bills are implemented.
We turn our attention to participation in hearings. Even though it is visible behavior,
most hearings are rarely attended by members of the committee and do not typically
gain attention from the public. In this sense, legislators may be more free to act on their
preferences in this forum as they are unlikely to attract attention.
We examine two questions in the context of hearings. First, we are interested in
whether, among all members of the committees holding a given hearing, women legislators
are more likely than men to participate in the hearing by making comments and inquiries
about policy implementation. Second, we examine whether women legislators express
greater support for aid than men given that they participate in those hearings.
We started by collecting transcripts of hearings in which senior personnel from US-
16
AID or the Millennium Challenge Corporation (MCC) appear accessing the ProQuest
Congressional Database. First, we obtained 120 hearing transcripts dating back to 1970
in which “USAID” and “Millennium Challenge” appeared as keywords. Next, we exam-
ined titles and synopses to find those that covered topics relevant to the study.15 This
lead to 25 hearings,16 a number that drops to twelve when restricting attention to those
held by the House of Representatives. While we can use these twelve for the attendance of
meetings, the number declines to ten when examining expressed attitudes by this analysis
necessitates that a hearing is attended by at least one woman and one man legislator.
Our first analysis concerns whether, among all members of the committees holding
a given hearing, women legislators were more likely to participate in it at all. For this,
we assemble a dataset with the potential attendee–hearing as the unit of analysis. This
yields a data set of 424 men and 63 women legislator observations, spanning the 110th to
115th Congresses (2007–2019). The outcome variable is a dummy variable indicating if
the legislator participated in the hearing or not, which we analyze with linear probability
models (multiplying the “1” by 100) on reweighed data.
The second analysis concerns the positivity toward aid expressed by the attending
legislators. Since attendance at these meetings is rare, the sample size drops dramatically
(56 and 16 men and women legislator observations). We measure the expressed sentiment
for each legislator in each hearing by coding random samples of speech fragments. For
every speaker in every hearing, the authors coded speech fragments (three sentences by an
attendee in a row) as negative/ neutral/ positive about development aid, or not about aid.
Three authors each coded about 15% of the 6,251 possible fragments, and two others about
10%. We then estimate the latent aid sentiment at the speaker-level in a hearing via a
measurement model to remove coder-specific idiosyncrasies (Caughey and Warshaw, 2015)
(see Section C for full details). Specifically, we use the positively-coded expressions among
15 For example, the hearing entitled “The FY2014 Budget Request—U.S. Foreign Assistance Priorities
and Strategy” was downloaded for use as the hearing focuses on USAID’s current priorities. On
the other hand, one entitled “Meeting the Challenges of the Millennium” was not as it contained
the relevant keywords but was not actually about foreign aid in any way. Further, if a transcript
contained testimony from a USAID administrator, but the testimony was about the current events
in a certain country and not primarily about aid, the transcript was omitted.
16 The estimation described below is run on the set of 25 hearings.
17
all coded utterances (by coder) to estimate positive aid attitudes. These latent attitudes
are scaled to a standard normal distribution. These latent attitudes are examined as
before, relying on reweighed data and linear regression models.17
Attend hearings on aid Support aid at hearings
Simple Detailed Simple Detailed
Gender
Coefficient 2.2 -0.6 -0.5 -0.5
95% CI [-9.4; 14.0] [-13.4; 11.1] [-1.5; 0.4] [-2.3; 1.6]
S.E. (6.2) (6.2) (0.5) (1.0)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 424 424 56 56
# Women 63 63 16 16
# unique Women 20 20 7 7
Congresses 110–115 110–115 110–115 110–115
Table 3: Estimates for gender-effects in attendance and attitudes in aid-related hearings
using all observations.The table is constructed like Table 1.
Table 3 gives the results. The gender effects on attending an aid-related hearing
are tiny once again as well as statistically insignificant from zero with standard errors
about three and ten times the sizes of the coefficients under the two specifications. The
expressed sentiments are lower for women legislators by about half a standard deviation
of the scale–a direction opposite to what the women’s values thesis predicts–but with
again large standard errors. The results for only Democrats (Table A.3) and only the
most recent years (Table A.7) are very similar.
Before moving on to our last analyses, a few remarks are in order. First, our data
collection revealed that only 25 hearings were relevant to development aid in the 23 year
17 As there is measurement uncertainty over our outcome variable, we use a non-parametric bootstrap,
taking one draw from the latent estimates. Results are averaged over these bootstraps and imputa-
tions. Further, due to the sample size, the entropy balancing proceeded without the Congress/party
subsetting and by using a smaller covariate set (state economic and social liberalism, percent unem-
ployed, age, ideology, dummy for whether one is a Democrat, percent of people with at least at a
4-year college degree, and percent who identify as white.
18
period we examined (Senate and House), and only 12 are usable for our research here.
This corroborates our claim that development aid is low-salience in the eyes of members
of the Congress, suggesting that the electoral and institutional constraints should be
relatively weak. That is, if women legislators have preferences for promoting aid, they
should feel relatively free to act on such preferences in the hearings that do take place.
Yet, the share of women attending those hearings is low. Second, it merits repeating how
few observations there are in the analysis of expressed sentiments. Presumably just a few
hearings taking place now or in the future could change the results we present.
Monitoring USAID
Our final analysis concerns lobbying bureaucratic agencies. Contacting and lobbying
bureaucracies on behalf of (groups of) constituents represents another opportunity for
legislators to shape policy implementation. While we found no gendered difference in
participation and expressed sentiments in Congressional hearings, we noted that these
are still context in which institutional structures are likely to shape legislative behavior.
Informal, direct contacts with the bureaucracy, however, is not a structured activity which
means legislators have far more autonomy in exercising their preferences. Moreover, such
activities are not visible meaning that the constraints legislators face when engaging in
these behaviors are minimal. However, such legislative requests can have influence on
policy decisions and implementations by U.S. bureaucracies (Mills and Kalaf-Hughes,
2015; Ritchie and You, 2019).
We anticipate that if women hold distinct foreign policy preferences and want to act
on such preferences, contacting agencies handling foreign aid should be one of the easiest
places to effect policy change. Evidence from Lowande et al. (2019) indicates that women
do indeed contact bureaucracies in a manner that differs from men, often in ways that
can be seen as acting for women’s interests. In the analysis that follows, we test whether
women similarly use this avenue to influence the implementation of pro-aid policy.
In the analysis of direct intervention with the bureaucracy, we probe whether women
legislators are more likely to monitor the primary bureaucracy for aid, USAID, than men
19
Contacting USAID (total) Contacting USAID (policy)
Simple Detailed Simple Detailed
Gender
Coefficient -0.9 -0.9 -0.3 -0.3
95% CI [-7.5; 5.7] [-8.0; 6.3] [-4.6; 3.9] [-4.7; 4.1]
S.E. (3.4) (3.7) (2.2) (2.2)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 723 723 723 723
# Women 144 144 144 144
# unique Women 85 85 85 85
Congresses 110–111 110–111 110–111 110–111
Table 4: Estimates for gender-effects in contacting USAID using all observations.The table
is constructed like Table 1.
legislators. Our analysis on bureaucratic oversight activities draws on data collected by
Lowande (2018) who filed Freedom of Information Act requests for records of contacts
made by individual members of Congress to USAID between 2007 and 2014.18 These
contacts are requests by legislators to elicit some type of response from USAID, including
congressional casework and general inquiries related to policy. A “casework” contact is
defined as a request made by a legislator on behalf of a particular constituent or a group
of constituents while a “policy” contact as an inquiry about policies but not serving
particular constituents. We analyze any request and only the policy-related subsets.
We examine these indicators of legislative activity the same as before. Once again,
entropy balancing reduced imbalances in this data set as well (Figure A.1). We use two
binary outcomes, namely whether one contacted USAID at all or specifically about policy.
For the ease of interpretation, we scaled the outcome by 100 so that contacting USAID
is a 100 and not doing so zero.
Table 4 gives the results for all House members. The first two columns examine any
18 Our own FOIA request to the Millennium Challenge Corporation receive no response beyond the
acknowledgment of the receipt of the request.
20
request made to USAID, the latter two only policy-related ones. Once again, the gender-
based differences are tiny, and confidence intervals include zero. The results based only
on Democrats (Table A.4) are very similar.
Discussion
Under the women’s value thesis, an increase in the number of women in a parliament
should lead to more and higher quality development aid because women politicians more
actively promote international development than their male counterparts. However, across
legislative modes of influence, regardless if using all available data or subsets by party or
more recent periods (106th Congress or later), women legislators are never observed to
be significantly more supportive of foreign aid. The results are from the U.S. House of
Representatives using a doubly robust approaching that first reweighs the gender-specific
subsets to be comparable (Zhao and Percival, 2017). In Section F in the appendix, we
present a robustness check analysis using random forest models to explore Congress-
specific gender differences; no noticeable deviations from the main results are found.
Our results reveal no gender-based differences in legislative behavior about foreign aid
and therefore provide no support for the key mechanism in the women’s value thesis, at
least in the important case of the United States. These behavioral results also have an
important implication for one of the assumptions underlying the women’s value thesis. If
U.S. women legislators had distinct preferences over aid policy, as assumed in the women’s
value thesis, but feared punishment by their party leadership or constituents, there are
channels, such as participating in hearings and directly contacting USAID, where their
activities are considerably less visible but still can have influence on policy. Yet, our
analyses of these fora show that men and women legislators do not behave any differently
even when there is no need to fear of punishment. Therefore, the room for maintaining
the assumption that women and men politicians have different preferences shrinks in light
of the evidence from the diverse forms of legislative behavior.
Of course, there are other potential reasons why U.S. women legislators might choose
not to act on their distinct preferences if they hold them. First, women politicians may
21
mask their true preferences for fear of electoral backlash. After all, the primary goal of
any politicians is to survive in office. Advocating for international development at the
expense of promoting other, salient issues might harm politicians electorally, women in
particular, by reinforcing feminine stereotypes that voters generally do not deem favor-
ably in selecting leaders (Eagly and Karau, 2002; Bauer, 2019; Vinkenburg et al., 2011).
Promoting international development may fall into the category of feminine stereotypes.
In a companion paper, we have examined this electoral backlash possibility by study-
ing whether it pays or harms a legislator’s election prospects to advocate for international
development (AUTHORS, 2021). Using a conjoint experiment, we analyze how voters
select legislators to represent them in Congress based on bundles of political messages by
legislators of differing genders (and parties). We find little evidence that any legislator,
man or woman, is systematically punished, or rewarded, for promoting international de-
velopment in their communication with constituents. Therefore, any legislator in the U.S.
should be able to advocate for international development without fear of electoral back-
lash. Yet, women legislators do not advocate for development issues any differently than
their male counterparts as the results in this paper shows. Thus, U.S. women legislators
are likely not masking their preferences for fear of sanction at the ballot box.
Second, women politicians may be masking their true preferences due to the lack of
critical mass of women necessary to translate women’s preferences into policy change.
However, critical mass theories have historically been contested (Sarah and Mona, 2008)
with some arguing that gender differences are actually more likely to manifest when women
are fewer in number (Crowley, 2004). We also note, that despite women’s underrepre-
sentation in the U.S. Congress, many studies do find evidence of gendered differences in
behavior (Swers, 2002; Pearson and Dancey, 2011; Dietrich et al., 2019; Frederick, 2015).
Our results using later years, when the share of woman legislators is higher, are the same
as before, suggesting that changes at the observed margins of the mass are not consequen-
tial. While we can only advance our indirect evidence contra the critical mass argument
here, the room for maintaining the foundational assumption of the women’s value thesis
is considerably small.
22
Conclusion
A growing number of women serve in politics in the United States and across the world,
a trend that is expected to continue. This raises an important question: will they re-
orient international politics more toward humanitarian issues? In this study, we directly
examine this question by studying individual legislators’ behavior in the United States,
a country for which the answer to this question is particularly relevant: it is among the
most influential countries and the share of women is parliament is still growing. However,
we find no gendered differences in how U.S. legislators approach development aid even
when they are able to exercise discretion and should be not be afraid of punishment by
party leaders (and constituents). These findings strongly suggest that the recent and fu-
ture increases in women’s representation in U.S. Congress are unlikely to lead to a greater
emphasis by the elected women on international development and more development aid
by the U.S. government.
This study also contributes to broader debates in the literature on gender and politics.
Researchers have grappled with, debated, and interrogated competing explanations to
explain observed relationships between gender, representation, and foreign policy. Our
findings provide no support for one prominent explanation, the women’s value thesis. An
alternative account is the social equity thesis (Breuning, 2001; Lu and Breuning, 2014;
Caprioli, 2000; Koch and Fulton, 2011; Brysk and Mehta, 2014), which holds that societal
attitudes towards social equity lead countries to pursue more humanitarian and peaceful
foreign policies, while also creating a political environment that is more conducive to
women’s emergence in politics. In line with past research (i.e. Lu and Breuning 2014),
our findings suggest that the social equity thesis is a more fruitful framework for future
research, which should also do more to examine societal preferences as determinants of
foreign policies.19
While the U.S. is the country with the largest aid budget, it is important to consider
to which extent the our results are transportable to other aid donors, such as Germany,
19 See however Imamverdiyeva et al. (2021) for legislative behavioral differences in the realm of tariffs
on internationally traded goods.
23
United Kingdom, Japan, and the European Union. One obvious feature of the U.S.
case might limit the transportability even though we are doubtful about this limit: U.S.
citizens are historically among the least enthusiastic supporters of development assistance
compared to people in other donors. Thus, the lack of electoral rewards or punishment for
promoting international development in the U.S. context may not be surprising (which we
find in the companion paper, see AUTHORS (2021)). By contrast, voters in other donors
are generally more supportive and may be more willing to reward politicians, especially
women, for emphasizing international development. While this scenario is possible, many
other donors have proportional electoral systems (and their variants) in which political
parties have strong influence on legislators’ behavior. This means that the role voters
play in shaping individual legislators’ behavior is relatively smaller compared to those in
the United States. That said, this alternative causal mechanism could be at work despite
the powerful role played by political parties. Research on a wider variety of donors would
generate useful insights into the relationship between gender and foreign aid.
More generally, our study demonstrates the benefits of shifting the level of analysis
from country-level to individual politicians. We call for future research to unpack the
“black box” and study women in foreign policy-making to better adjudicate between
competing explanations (Smith, 2020; Williams, 2017). At the same time, our findings
call for greater attention to the social equity thesis, which studies have so far tested at
the macro-level (Caprioli, 2000; Koch and Fulton, 2011; Brysk and Mehta, 2014). Here, a
shift towards the individual level would be productive as well. For example, in the context
of development aid, the thesis would imply several hypotheses, including that citizens
who support gender equality should be more likely to support development aid. A better
understanding of the micro-foundation of these prominent explanations is strongly needed
in the literature on gender and foreign policy.
24
References
Adler, E. S. and J. D. Wilkerson (2013). Congress and the Politics of Problem Solving.
Cambridge University Press.
Andreoni, J. and L. Vesterlund (2001). Which is the fair sex? gender differences in
altruism. The Quarterly Journal of Economics 116 (1), 293–312.
Angevine, S. (2017). Representing all women: An analysis of congress, foreign policy,
and the boundaries of women?s surrogate representation. Political Research Quar-
terly 70 (1), 98–110.
Arel-Bundock, V., J. Atkinson, and R. A. Potter (2015). The limits of foreign aid diplo-
macy: How bureaucratic design shapes aid distribution. International Studies Quar-
terly 59 (3), 544–556.
AUTHORS (2021). Does it pay or hurt electorally to promote international development?
evidence from a conjoint experiment. Working Paper.
Bashevkin, S. (2014). Numerical and policy representation on the international stage:
Women foreign policy leaders in western industrialised systems. International Political
Science Review 35 (4), 409–429.
Bashevkin, S. (2018). Women as foreign policy leaders: national security and gender
politics in superpower America. Oxford University Press.
Bauer, N. M. (2019). A feminine advantage? delineating the effects of feminine trait and
feminine issue messages on evaluations of female candidates. Politics & Gender , 1–21.
Bauhr, M., N. Charron, and N. Nasiritousi (2013). Does corruption cause aid fatigue?
public opinion and the aid-corruption paradox. International Studies Quarterly 57 (3),
568–579.
Bendix, W. and G.-H. Jeong (2020). Gender and foreign policy: Are female members of
congress more dovish than their male colleagues? Political Research Quarterly 73 (1),
126–140.
Betz, T., D. Fortunato, and D. Z. O’Brien (2021). Women’s descriptive representation
and gendered import tax discrimination. American Political Science Review 115(1),
307–315.
Breiman, L. (2001). Random forests. Machine Learning 45 (1), 5–32.
Breuning, M. (2001). Women’s representation and development assistance: a cross-
national study. Women & Politics 23 (3), 35–55.
Brysk, A. and A. Mehta (2014). Do rights at home boost rights abroad? sexual equality
and humanitarian foreign policy. Journal of Peace Research 51 (1), 97–110.
Caprioli, M. (2000). Gendered conflict. Journal of Peace Research 37 (1), 51–68.
Caprioli, M. and M. A. Boyer (2001). Gender, violence, and international crisis. Journal
of Conflict Resolution 45 (4), 503–518.
25
Caughey, D. and C. Warshaw (2015). Dynamic estimation of latent opinion using a
hierarchical group-level irt model. Political Analysis 23 (2), 197–211.
Caughey, D. and C. Warshaw (2016). The dynamics of state policy liberalism, 1936–2014.
American Journal of Political Science 60 (4), 899–913.
Celis, K. (2006). Substantive representation of women and the impact of descriptive
representation. case: The belgian lower house 1900-1979. Journal of Women, Politics
and Policy 28 (2), 85–114.
Chong, A. and M. Gradstein (2008). What determines foreign aid? the donors’ perspec-
tive. Journal of Development Economics 87 (1), 1–13.
Clayton, A. and P. Zetterberg (2018). Quota shocks: Electoral gender quotas and gov-
ernment spending priorities worldwide. The Journal of Politics 80 (3), 916–932.
Crowley, J. E. (2004). When tokens matter. Legislative Studies Quarterly 29 (1), 109–136.
Dietrich, B. J., M. Hayes, and D. Z. O’Brien (2019). Pitch perfect: Vocal pitch and
the emotional intensity of congressional speech on women. American Political Science
Review 113 (4), 941–962.
Dietrich, S. and J. Wright (2014). Foreign aid allocation tactics and democratic change
in africa. The Journal of Politics 77 (1), 216–234.
Dreher, A., K. Gehring, and S. Klasen (2015). Gesture politics or real commitment?
gender inequality and the allocation of aid. World Development 70, 464–480.
Dufwenberg, M. and A. Muren (2006). Generosity, anonymity, gender. Journal of Eco-
nomic Behavior & Organization 61 (1), 42–49.
Eagly, A. H. and S. J. Karau (2002). Role congruity theory of prejudice toward female
leaders. Psychological Review 109 (3), 573.
Eckel, C. C. and P. J. Grossman (1998). Are women less selfish than men?: Evidence
from dictator experiments. The Economic Journal 108 (448), 726–735.
Eichenberg, R. C. (2016). Gender difference in american public opinion on the use of
military force, 1982–2013. International Studies Quarterly 60 (1), 138–148.
Engel, C. (2011). Dictator games: A meta study. Experimental Economics 14 (4), 583–
610.
Fowler, J. H. (2006a). Connecting the congress: A study of cosponsorship networks.
Political Analysis 14 (4), 456–487.
Fowler, J. H. (2006b). Legislative cosponsorship networks in the us house and senate.
Social Networks 28 (4), 454–465.
Franceschet, S. and J. M. Piscopo (2008). Gender quotas and women’s substantive rep-
resentation: Lessons from argentina. Politics & Gender 4 (03), 393–425.
Frankovic, K. A. (1977). Sex and voting in the us house of representatives 1961-1975.
American Politics Quarterly 5 (3), 315–330.
26
Frederick, B. (2009). Are female house members still more liberal in a polarized era? the
conditional nature of the relationship between descriptive and substantive representa-
tion. In Congress & the Presidency, Volume 36, pp. 181–202. Taylor & Francis.
Frederick, B. (2015). A longitudinal test of the gender turnover model among us house
and senate members. The Social Science Journal 52 (2), 102–111.
Fuchs, A., A. Dreher, and P. Nunnenkamp (2014). Determinants of donor generosity: A
survey of the aid budget literature. World Development 56, 172–199.
Fuchs, A. and K. Richert (2018). Development minister characteristics and aid giving.
European Journal of Political Economy 53, 186–204.
Goddard, D. (2019). Entering the men’s domain? gender and portfolio allocation in
european governments. European Journal of Political Research 58 (2), 631–655.
Hainmueller, J. (2012). Entropy balancing for causal effects: A multivariate reweighting
method to produce balanced samples in observational studies. Political Analysis 20 (1),
25–46.
Heinrich, T. and Y. Kobayashi (2021). Evaluating explanations for selectivity in foreign
aid. Kyklos.
Heinrich, T., Y. Kobayashi, and K. A. Bryant (2016). Public opinion and foreign aid cuts
in economic crises. World Development 77, 66–79.
Hicks, D. L., J. H. Hicks, and B. Maldonado (2016). Women as policy makers and donors:
Female legislators and foreign aid. European Journal of Political Economy 41, 46–60.
Holsti, O. R. and J. N. Rosenau (1981). The foreign policy beliefs of women in leadership
positions. The Journal of Politics 43 (2), 326–347.
Honaker, J. and G. King (2010). What to do about missing values in time-series cross-
section data. American Journal of Political Science 54 (2), 561–581.
Imamverdiyeva, U., P. M. Pinto, and P. E. Shea (2021). A re-examination of gender,
descriptive representation, and trade policy. Unpublished.
Inglehart, R. and P. Norris (2000). The developmental theory of the gender gap: Women’s
and men’s voting behavior in global perspective. International Political Science Re-
view 21 (4), 441–463.
Kleemann, L., P. Nunnenkamp, and R. Thiele (2016). Gender inequality, female leader-
ship, and aid allocation: A panel analysis of aid for education. Journal of International
Development 28 (3), 376–395.
Koch, J. (1997). Candidate gender and women’s psychological engagement in politics.
American Politics Research 25 (1), 118–133.
Koch, M. T. and S. A. Fulton (2011). In the defense of women: Gender, office holding,
and national security policy in established democracies. The Journal of Politics 73 (1),
1–16.
27
Lizotte, M.-K., R. C. Eichenberg, and R. J. Stoll (2020). Citizen support for the pursuit
of gender equality in us foreign policy. Journal of Human Rights 19 (3), 291–306.
Lowande, K. (2018). Who polices the administrative state? American Political Science
Review 112 (4), 874–890.
Lowande, K., M. Ritchie, and E. Lauterbach (2019). Descriptive and substantive repre-
sentation in congress: Evidence from 80,000 congressional inquiries. American Journal
of Political Science 63 (3), 644–659.
Lu, K. and M. Breuning (2014). Gender and generosity: does women’s representation
affect development cooperation? Politics, Groups, and Identities 2 (3), 313–330.
Lundsgaarde, E., C. Breunig, and A. Prakash (2007). Trade versus aid: donor generosity
in an era of globalization. Policy Sciences 40 (2), 157–179.
Luttmer, E. F. and M. Singhal (2011). Culture, context, and the taste for redistribution.
American Economic Journal: Economic Policy 3 (1), 157–79.
Mavisakalyan, A. and Y. Tarverdi (2019). Gender and climate change: Do female parlia-
mentarians make difference? European Journal of Political Economy 56, 151–164.
McGlen, N. E. and M. R. Sarkees (1993). Women in foreign policy: The insiders. Rout-
ledge.
Mills, R. W. and N. Kalaf-Hughes (2015). The evolution of distributive benefits: The
rise of letter-marking in the united states congress. The Journal of Economics and
Politics 22 (1), 4.
Milner, H. V. and D. Tingley (2015). Sailing the water’s edge: the domestic politics of
American foreign policy. Princeton University Press.
Okundaye, G. and M. Breuning (2021). Is the expansion of women’s access to political
leadership rewarded? evidence from the allocation of us foreign aid. Journal of Women,
Politics & Policy 42 (3), 225–242.
Osborn, T. L. (2012). How Women Represent Women: Political Parties, Gender and
Representation in the State Legislatures. Oxford University Press.
Paxton, P. and S. Knack (2012). Individual and country-level factors affecting support
for foreign aid. International Political Science Review 33 (2), 171–192.
Pearson, K. and L. Dancey (2011). Elevating women?s voices in congress: Speech partic-
ipation in the house of representatives. Political Research Quarterly 64 (4), 910–923.
Regan, P. M. and A. Paskeviciute (2003). Women’s access to politics and peaceful states.
Journal of Peace Research 40 (3), 287–302.
Ritchie, M. N. and H. Y. You (2019). Legislators as lobbyists. Legislative Studies Quar-
terly 44 (1), 65–95.
Sarah, C. and L. K. Mona (2008). Critical mass theory and women’s political represen-
tation. Political Studies 56 (3), 725–736.
28
Shea, P. and C. Christian (2017). The impact of women legislators on humanitarian
military interventions. Journal of Conflict Resolution 61 (10), 2043–2073.
Slaughter, A.-M. (2012). Why family is a foreign policy issue, too. Foreign Policy (197),
73–73.
Smith, K. E. (2020). Missing in analysis: Women in foreign policy–making. Foreign Policy
Analysis 16 (1), 130–141.
Snyder Jr, J. M. and T. Groseclose (2000). Estimating party influence in congressional
roll-call voting. American Journal of Political Science 44 (2), 193–211.
Swers, M. L. (1998a). Are women more likely to vote for women’s issue bills than their
male colleagues? Legislative Studies Quarterly, 435–448.
Swers, M. L. (1998b). Are women more likely to vote for women’s issue bills than their
male colleagues? Legislative Studies Quarterly 23 (3), 435–448.
Swers, M. L. (2002). The difference women make: The policy impact of women in
Congress. University of Chicago Press.
Swers, M. L. (2016). Pursuing women’s interests in partisan times: Explaining gender dif-
ferences in legislative activity on health, education, and women’s health issues. Journal
of Women, Politics & Policy 37 (3), 249–273.
Swers, M. L. and C. Larson (2005). Women in congress: Do they act as advocates for
women’s issues? Women and Elective Office: Past, Present, and Future 2.
Togeby, L. (1994). The gender gap in foreign policy attitudes. Journal of Peace Re-
search 31 (4), 375–392.
Van Belle, D. A. (2004). Media, bureaucracies, and foreign aid: A comparative analysis
of the United States, the United Kingdom, Canada, France and Japan. Springer.
Vinkenburg, C., M. Van Engen, A. Eagly, and M. Johannesen-Schmidt (2011). An explo-
ration of stereotypical beliefs about leadership styles: Is transformational leadership a
route to women’s promotion? The Leadership Quarterly 22 (1), 10–21.
Volden, C., A. E. Wiseman, and D. E. Wittmer (2016). Women’s issues and their fates
in the united states congress. Political Science Research and Methods 6 (4), 679–696.
Welch, S. (1985). Are women more liberal than men in the us congress? Legislative
Studies Quarterly, 125–134.
Williams, K. P. (2017). Feminism in foreign policy. In C. G. Thies (Ed.), Oxford Research
Encyclopedia of Politics. Oxford University Press.
Wolbrecht, C. (2002). Explaining women’s rights realignment: Convention delegates,
1972–1992. Political Behavior 24 (3), 237–282.
Yoon, M. Y. and C. Moon (2019). Gender equality in politics at home and promotion of
gender equality in politics abroad: The role of bilateral official development assistance.
International Political Science Review 40 (2), 161–180.
29
Zhao, Q. and D. Percival (2017). Entropy balancing is doubly robust. Journal of Causal
Inference 5 (1).
Ziaja, S. (2020). More donors, more democracy. The Journal of Politics 82 (2), 433–447.
30
The Missing Fingerprints:
U.S. Women Legislators and Development Aid
Web Appendix
Not for Print Publication
Contents
A Balancing A.1
B Additional Details on Co-sponsorship Analysis A.2
C Additional Details on Hearing Analysis A.3
D Subset analysis using only Democrats A.6
D.1 Roll-call .....................................A.6
D.2 Co-sponsorship .................................A.7
D.3 Hearings.....................................A.8
D.4 USAIDContact.................................A.9
E Subset analysis using recent data (106th Congress and later) A.10
E.1 Roll-call .....................................A.10
E.2 Co-sponsorship .................................A.11
E.3 Hearings.....................................A.12
F Random Forest Estimates A.13
A Balancing
Roll Call
Cosponsor
Contact USAID
−0.50 −0.25 0.00 0.25 0.50 −0.50 −0.25 0.00 0.25 0.50 −0.50 −0.25 0.00 0.25 0.50
Foreign born
Unemployment
BA or more
Share white
Income, >15k
Income, >25k
Income, >35k
Income, >50k
Income, >75k
Income, >100k
Legislator Democrat
Social liberalism
Economic liberalism
Legislator Ideology
Age
Differences scaled by SD
Data Raw Weighted
Hearings, Attend
Hearings, Attitude
−0.50 −0.25 0.00 0.25 0.50 0.75 −0.50 −0.25 0.00 0.25 0.50 0.75
Unemployment
BA or more
Share white
Legislator Democrat
Social liberalism
Economic liberalism
Legislator Ideology
Age
Differences scaled by SD
Data Raw Weighted
Figure A.1: Covariate balance between men and women legislators across data sets. Each
panels a different data set used for analysis. The x-axis in each gives the standardized difference between
men and women legislators; dots show the differences in the raw data, triangles for the reweighed data
based on stratified entropy balancing.
A.1
B Additional Details on Co-sponsorship Analysis
To examine whether men and women co-sponsor aid legislation at different rates, we use
the Cosponsorships Network Data by Fowler (2006a, b). Using this data, we are able to
obtain the list of cosponsors on every piece of legislation to come before the U.S. House
of Representatives and Senate from the 93rd Congress to the 110th Congress. Bills of
potential interest were identified using information from the Policy Agendas Project. The
Policy Agendas Project classifies each piece of legislation as falling under one of twenty-
three potential topics, followed by a sub-topic classification. For example, a bill related
to foreign aid may be classified as falling under the topic of “International Affairs” and
the subtopic of “Foreign Aid.”
To identify which of these bills pertain to foreign aid (and the content of these bills), we
conducted crowd-coding using MTurk workers. For each Congress, we randomly sampled
one-third of the bills of potential interest for coding. Workers on MTurk were given brief
descriptions of a given bill from the Congressional Archive. After reading this summary,
workers were asked to identify: 1) whether the bill was related to foreign aid, and 2)
if so, did the bill increase aid, decrease aid, or did not influence the amount of aid.
Our procedure was designed to assign multiple workers to each bill. In cases where our
workers reached a consensus, bills were coded based on this consensus. If consensus was
not achieved, the researchers read the bill summary in question a made a determination
about the appropriate coding. All in all, we ended up with 39 decrease and 135 increases
bills.
A.2
C Additional Details on Hearing Analysis
The hearings used in this analysis are a collection of full–text transcripts released by U.S.
Congress and cataloged in the ProQuest Congressional Database. The database contains a
record of all Congressional hearing transcripts dating back to 1824, with the title, synopsis
of the topics covered, date, members, and full text of the hearing. We first filtered this
database to search for hearings dating back to 1970, then keyword-searched these hearings
for “USAID” and “Millennium Challenge.” This cast a wide net, yielding approximately
120 hearings in the date range containing those keywords. These were further filtered by
title and synopsis using a keyword search, downloading only the ones that cover topics
relevant to the study. For example, the hearing entitled “The FY2014 Budget Request—
U.S. Foreign Assistance Priorities and Strategy” was downloaded for use as the hearing
focuses on USAID’s current priorities. On the other hand, one entitled “Meeting the
Challenges of the Millennium” was not as it contained the relevant keywords but was not
actually about foreign aid in any way. Some transcripts were removed out due to a lack of
relevance for foreign aid. For example, if a transcript contained testimony from a USAID
administrator, but the testimony was about the current events in a certain country and
not primarily about aid, the transcript was omitted. In the end, 25 transcripts of hearings
were retained. While we estimate the measurement model below on these 25 hearings, the
usable number for inferential purposes falls to twelve for the study of whether committee
members show up as we only examine the U.S. House. For the expressed attitudes, the
number declines to ten because we require at least one woman and one man to be present
at the hearing.
Our interest lies in measuring the positivity toward aid that hearing attendees ex-
pressed. We split each speaker’s totality of remarks at a hearing into text segments
three sentences in length. Five of the authors coded these fragments without know-
ing the hearing, speaker, and date. Specifically, we coded using these instructions: “if
you can reasonably infer that the paragraph is about funds for promoting international
development—e.g. poverty alleviation, education, better access to clean water, etc.”; if
the answer is “yes”, the coder should judge whether the speaker is “defending develop-
ment aid or advocating for an increase or better use of aid” (positive), if the legislator “is
advocating for a reduction or withdrawal of aid” (negative), or if it is neutral.20
All in all, there are 6,251 such speech fragments. Three authors coded randomly about
15%, two about 10% of them. Many fragments were coded multiple times.
For each speaker–hearing, we estimate the latent expressed sentiment toward aid using
our coded fragments (three sentences). Let Yij denote the number of positive segments
out of Nij coded segments for speaker-hearing icoded by coder j. The probability that
20 Neutral should not be used if the statement contains positive and negative expresses. In such a case,
an overall assessment should be made.
A.3
a given segment of iis coded as positive by jis modeled as a function of the speaker’s
latent sentiment toward aid (θi) adjusted by a coder specific offset (κj) and scaled by
the variability of legislators’ expressions and coders’ judgements of iand j, respectively
(σi,τj). We use the normal cumulative density function as the link function to relate the
latent sentiment to the probability parameter in a Binomial distribution. Taken together,
we have
Yij ∼Bin(πij , Nij ),
with the key probability parameter modeled21 as
πij = Φ θi+κj
√σi+τj.
The scale of the parameter main interest, θi, the latent support for aid by speaker-hearing
(i), is set by assigning a standard normal prior.22 The model is estimated using JAGS. A
second model uses the sum of positive and neutral (ie. non-negative) codings as Yij .
Figure A.2 shows the summary of the results for the ten hearings that we are using
in the analysis. Each panels gives the estimates for one hearing; the speakers are on the
y-axis, the x-axis indicates θi.
21 The items in the denominator are restricted to be positive, and κjis given a N(0,1) prior.
22 The model is a close adaptation of Caughey and Warshaw (2015).
A.4
2017−11−11
2011−09−08
2013−04−25
2015−03−24
2009−04−28
2011−03−16
2011−05−10
2007−03−08
2007−06−28
2009−04−01
−2 −1 0 1 2 3
−2 −1 0 1 2 3 −2 −1 0 1 2 3
Victor Snyder
Robert J. Wittman
Susan A. Davis
Christopher H. Smith
Karen Bass
Ann Marie Buerkle
Joaquin Castro
Ron DeSantis
Albio Sires
Robin L. Kelly
Ileana Ros−Lehtinen
Thomas Tancredo
Edward R. Royce
Christopher H. Smith
Sheila Jackson Lee
Lynn C. Woolsey
Gregory W. Meeks
Steve Chabot
Donald A. Manzullo
David N. Cicilline
Mike Kelly
David Rivera
Christopher Murphy
Ted Poe
Connie Mack
Gary L. Ackerman
Gerald E. Connolly
Edward R. Royce
Jeff Fortenberry
Dana Rohrabacher
Donald Payne
Howard L. Berman
Christopher H. Smith
Ann Marie Buerkle
Brad Sherman
Ted S. Yoho
Paul Cook
Ami Bera
Tom Marino
Theodore E. Deutch
Eliot L. Engel
David N. Cicilline
Ted Poe
Gerald E. Connolly
Edward R. Royce
Dana Rohrabacher
Christopher H. Smith
Karen Bass
Ileana Ros−Lehtinen
Albio Sires
Jeff Fortenberry
Mike Pence
Jeff Flake
Dana Rohrabacher
Donald Payne
Howard L. Berman
Christopher H. Smith
Tom Lantos
Diane Watson
Ileana Ros−Lehtinen
Brian P. Bilbray
Gerald E. Connolly
Paul Hodes
Diane Watson
Russ Carnahan
Christopher H. Smith
Karen Bass
Daniel M. Donovan, Jr.
Joaquin Castro
Ami Bera
Christopher H. Smith
Karen Bass
Expressed aid support score
Attitude type Positive Positive/ Neutral
Figure A.2: Estimates of θfor every speaker by hearing. The dot denotes the median estimate,
the line segments the 95% central credible intervals. Black dots/ lines show results using only positive
utterances, whereas gray ones use non-negative instances.
A.5
D Subset analysis using only Democrats
D.1 Roll-call
Voting yay on aid increase Voting nay on aid increase
Simple Detailed Simple Detailed
Gender
Coefficient -0.6 -0.5 -0.4 -0.4
95% CI [-4.5; 3.4] [-5.0; 4.0] [-3.6; 2.8] [-3.9; 3.1]
S.E. (2.0) (2.3) (1.7) (1.8)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 2055 2055 2055 2055
# Women 283 283 283 283
# unique Women 88 88 88 88
Congresses 97–110 97–110 97–110 97–110
Table A.1: Estimates for gender-effects in roll call voting on foreign aid using only
Democrats.The table reports the estimated coefficients on legislator gender (1=female). The Sim-
ple model specifications in Columns 2 and 4 includes indicators for party affiliation and Congresses. The
Detailed specifications in Columns 3 and 5 further include legislator-specific (age, DW-NOMINATE) and
district-specific variables (percentages foreign born, with college degrees, whites, unemployed, income of
at least $10k/ $50k/ $100k).
A.6
D.2 Co-sponsorship
Consponsoring aid increases Consponsoring aid decreases
Simple Detailed Simple Detailed
Gender
Coefficient 0.3 0.3 -0.1 0.0
95% CI [-0.4; 1.0] [-0.1; 0.7] [-0.6; 0.5] [-0.9; 0.8]
S.E. (0.3) (0.2) (0.3) (0.4)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 2394 2394 2394 2394
# Women 371 371 371 371
# unique Women 86 86 86 86
Congresses 99–110 99–110 99–110 99–110
Table A.2: Estimates for gender-effects in cosponsoring legislation on foreign aid using only
Democrats.The table is constructed like Table 1.
A.7
D.3 Hearings
Attend hearings on aid Support aid at hearings
Simple Detailed Simple Detailed
Gender
Coefficient 4.1 -3.8 -0.6 0.0
95% CI [-10.7; 18.3] [-18.0; 10.1] [-2.1; 0.8] [-16.3; 16.6]
S.E. (7.5) (7.3) (0.7) (14.6)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 186 186 26 26
# Women 56 56 9 9
# unique Women 16 16 5 5
Congresses 110–115 110–115 110–115 110–115
Table A.3: Estimates for gender-effects in attendance and attitudes in aid-related hearings
using only Democrats.The table is constructed like Table 1.
A.8
D.4 USAID Contact
Contacting USAID (total) Contacting USAID (policy)
Simple Detailed Simple Detailed
Gender
Coefficient -3.3 -3.3 -1.1 -1.2
95% CI [-12.0; 5.4] [-13.7; 6.6] [-7.1; 5.0] [-6.7; 4.2]
S.E. (4.4) (5.1) (3.1) (2.8)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 383 383 383 383
# Women 107 107 107 107
# unique Women 63 63 63 63
Congresses 110–111 110–111 110–111 110–111
Table A.4: Estimates for gender-effects in contacting USAID using only Democrats.The
table is constructed like Table 1.
A.9
E Subset analysis using recent data (106th Congress
and later)
E.1 Roll-call
Voting yay on aid increase Voting nay on aid increase
Simple Detailed Simple Detailed
Gender
Coefficient -0.1 -0.2 -1.6 -1.5
95% CI [-8.4; 8.1] [-6.5; 6.1] [-9.1; 5.9] [-6.7; 3.5]
S.E. (4.2) (3.3) (3.8) (2.6)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 1108 1108 1108 1108
# Women 186 186 186 186
# unique Women 93 93 93 93
Congresses 106–110 106–110 106–110 106–110
Table A.5: Estimates for gender-effects in roll call voting on foreign aid using observations
since 106th Congress.The table reports the estimated coefficients on legislator gender (1=female). The
Simple model specifications in Columns 2 and 4 includes indicators for party affiliation and Congresses.
The Detailed specifications in Columns 3 and 5 further include legislator-specific (age, DW-NOMINATE)
and district-specific variables (percentages foreign born, with college degrees, whites, unemployed, income
of at least $10k/ $50k/ $100k).
A.10
E.2 Co-sponsorship
Consponsoring aid increases Consponsoring aid decreases
Simple Detailed Simple Detailed
Gender
Coefficient 0.2 0.2 -0.1 -0.1
95% CI [-0.6; 1.0] [-0.2; 0.5] [-0.5; 0.3] [-0.4; 0.2]
S.E. (0.4) (0.2) (0.2) (0.1)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 1851 1851 1851 1851
# Women 313 313 313 313
# unique Women 93 93 93 93
Congresses 106–110 106–110 106–110 106–110
Table A.6: Estimates for gender-effects in cosponsoring legislation on foreign aid using
observations since 106th Congress.The table is constructed like Table 1.
A.11
E.3 Hearings
Attend hearings on aid Support aid at hearings
Simple Detailed Simple Detailed
Gender
Coefficient 2.7 -0.2 -0.5 -0.5
95% CI [-8.4; 14.3] [-11.6; 11.2] [-1.4; 0.5] [-2.5; 1.4]
S.E. (5.9) (5.9) (0.5) (1.0)
Specification
Party control? X X X X
Congress control? X X X X
District controls? X X
Legislator controls? X X
Data
# Men 415 415 53 53
# Women 72 72 19 19
# unique Women 21 21 7 7
Congresses 110–115 110–115 110–115 110–115
Table A.7: Estimates for gender-effects in attendance and attitudes in aid-related hearings
using observations since 106th Congress.The table is constructed like Table 1.
A.12
F Random Forest Estimates
This section explores the robustness of our results by using a random forest model. This
analysis gives us two things. First, we can account for potential interactions between the
control variables and our “treatment” (whether a legislator is a woman or man). Second,
the first point lets us examine heterogeneity of effects over time, something that not only
gives us a richer descriptive patterns but also acts as a robustness check in itself.
For each outcome, we split the available data by gender, draw a non-parametric boot-
strap draw for each, and then estimate two random forest models (by gender).23 For each
Congress in the data, we take the woman legislators’ observed covariates and get a pre-
diction under each model. Last, we save the difference in means between the predictions.
This procedure is repeated 1,000 times for each outcome.
Figure A.3 shows the results for all outcomes, except for the sentiments expressed at
the hearings as the data set is too small for this type of analysis. Each panel gives the
results for one outcome. In each, the y-axis indicates the mean difference between the
estimate for the woman and man legislators. The x-axis denotes the Congress number.
Each dot gives the conditional result from one bootstrap iterations. Semi-transparency
and horizontal jitter were applied to reduce over-plotting issues.
23 We use the tuning parameter setting tuned on the full data.
A.13
Voting 'yay' on aid increase
Cosponsoring aid increase
Voting 'nay' on aid increase
Contact USAID (total)
Cosponsoring aid decrease
Attending aid hearings
Contact USAID (policy)
100 105 110 115
100 105 110 115
−20
−10
0
10
20
30
−20
−10
0
10
20
30
−20
−10
0
10
20
30
−20
−10
0
10
20
30
Congress
Female − Male estimate
Figure A.3: Gender effects in legislative behavior across Congresses; random forest es-
timates. Each panel plots the gender effect estimated for each Congress in the available data. The
y-axis shows either a difference in shares (cosponsoring, roll calls) or probabilities (attending hearings,
contacting USAID), multiplied by 100. Semi-transparency and horizontal jitter were applied to reduce
overplotting issues.
A.14
