Quasi-Experimental Design, Constituency, and Advancing Women’s Interests: Reexamining the Influence of Gender on Substantive Representation

• Author: Erin E. O'Brien,Jason A. MacDonald
• Date: 01 June 2011
• DOI: 10.1177/1065912909354703
• Published date: 01 June 2011
• Subject Matter: Articles

Political Research Quarterly
64(2) 472 –486
Quasi-Experimental Design,
© 2011 University of Utah
Reprints and permission:
sagepub.com/journalsPermissions.nav
Constituency, and Advancing
DOI: 10.1177/1065912909354703
http://prq.sagepub.com
Women’s Interests: Reexamining
the Influence of Gender on
Substantive Representation
Jason A. MacDonald1 and Erin E. O’Brien2
Abstract
Research investigating whether female legislators provide more effective substantive representation on women’s issues
than their male colleagues faces a significant methodological hurdle. Models used to estimate the effect of gender
on representation inevitably omit constituency variables that affect the character of legislators’ decisions and are
also correlated with gender, potentially biasing the estimates of the effect of gender. Employing a quasi-experimental
research design as an alternative strategy, the authors remove this hurdle and estimate the influence of gender on
representation free from this potential bias. The authors find that gender does affect representation and observe
critical mass effects.
Keywords
Congress, descriptive representation, substantive representation, women and politics, agenda setting
Do female members of Congress provide more effective
This conclusion comes from studies that employ an
representation of women’s interests than their male col-
array of methodological strategies. Nonetheless, these
leagues? Prior research generally indicates “yes” with
strategies regularly fail to differentiate between the effect
debate as to whether having a substantial number of
of a legislator’s gender on her or his representational
women in the legislative body, or a “critical mass,”
behavior and the demands of one’s constituency. The
encourages congresswomen to devote their scarce time
inferential dilemma is profound: “The omission of con-
and legislative resources to enhancing the quality of rep-
stituency preferences from models of gender differences
resentation that they provide women (e.g., Bratton 2005;
in legislators’ policy attitudes” may produce premature
Grey 2006; Reingold 1992, 2000; Saint-Germain 1989;
conclusions where “the impact of gender may have been
Thomas 1994). Scholars observe this benefit of descrip-
overestimated” (Poggione 2004, 306). The relatively few
tive female representation across a host of issues (Barrett
studies that work to account for constituency demands in
1995; Carey, Niemi, and Powell 1998) but especially
assessing women’s legislative behavior thus offer a sub-
with respect to women’s and feminist issues (e.g., Burrell
stantial leap forward for understanding women’s
1994; Carroll 2001; Dolan 1997; Swers 1998, 2002).
substantive representation (e.g., Bratton and Haynie
Female legislators also voice distinct policy priorities
1999; Poggione 2004; Swers 1998, 2000; Thomas and
from men (e.g., Thomas 1994; Thomas and Welch 2001),
Welch 1991; Tolleson-Rinehart 2001). Inferential dilem-
view women outside their district as constituents (Carroll
mas persist, though, as these inquiries rely almost solely
2002), and expand the agenda to include women’s per-
spectives (e.g., Dodson 2005, 2006; Kathlene 1994;
1West Virginia University, Morgantown, WV, USA
Reingold 1996; Rosenthal 2000; Wolbrecht 2002, 193).
2University of Massachusetts Boston, Boston, MA, USA
In summary, “women legislators tend, more often than
Corresponding Author:
men, to make priorities of issues important to women and
Jason A. MacDonald, Department of Political Science, West Virginia
to introduce and successfully usher those priorities
University, Morgantown, WV 26506-6317, USA
through the legislative process” (Dodson 2001, 226).
Email: Jason.MacDonald@mail.wvu.edu

---

MacDonald and O’Brien
473
on standard sampling techniques and multivariate
differences in men and women’s constituencies explain
models. Valid data that accurately capture the full politi-
the relationship between gender and legislators’ prefer-
cal landscape and constituency characteristics for all 435
ences, rather than gender itself.” Palmer and Simon
districts simply do not exist. This means inevitable prob-
(2008, 202) add more recent credence to this concern,
lems of measurement error and an inability to determine
finding that female candidates cluster in particular dis-
how omitted variable bias is at play (Clarke 2005, 2009).
tricts and that these districts share attributes that make
This article capitalizes on experimental logic to offer a
them particularly favorable toward electing women.
different empirical solution to the problem of controlling
Therefore, it is plausible that women elected from “favor-
for constituency. We utilize a quasi-experimental research
able” districts legislate differently because of the district’s
design composed of longitudinal data. The sample
unique policy demands rather than their gender.
includes all pairs of House members serving during the
This possibility demonstrates the importance of sepa-
1970s, 1980s, and 1990s during which a district transi-
rating the effect of gender and district characteristics and
tioned from having a woman serve it immediately before
suggests why those studies that began to take constitu-
and/or immediately after a man (omitting transitions in
ency into effect advanced the literature. Thomas and
which districts are changed due to redistricting). After
Welch (1991), for example, controlled for the degree of
identifying these pairs, we recorded the number of bills
urbanization in a district. Swers (1998, 2002) controlled
sponsored by the members on “social welfare” and “fem-
for district effects by including in her models the percent-
inist” issues during the relevant congressional sessions
age of district vote going to Clinton and Perot in 1992, an
employing Swers’s (2002) criteria for coding these issues.
urbanicity measure, a southern dummy variable, the per-
Our research design, then, allows us to isolate variance in
centage of African Americans in the district, and the
representatives’ behavior explained by gender from the
district’s median household income. Poggione (2004)
variance in behavior explained by representatives’ con-
took district into account by controlling for average
stituents, providing an estimate of the effect of gender on
household income, the percentage unemployed in the dis-
behavior that is less clouded by concerns about correla-
trict, the percentage of district residents who earned a
tion with omitted constituency variables than virtually
college degree, and urbanicity. Some who push the repre-
any previous research.
sentational question to simultaneously examine gender
In addition, we take advantage of our ability to control
and racial effects have included the percentage of a dis-
for the influence of lawmakers’ constituencies on their
trict that is African American and district income (Bratton,
behavior by extending our analysis to assess whether con-
Haynie, and Reingold 2006) as well as the size of the
gresswomen provide enhanced representation to women
largest city in the district, majority black district, and
when a higher percentage of their colleagues are women
various interaction variables (Bratton and Haynie 1999).
(see Kanter 1977). Although there is a considerable body of
The question of whether women provide more effec-
research on whether such a “critical mass” effect exists, this
tive substantive representation because of their gender,
research does not examine this question with the effects of
differences in constituencies, or both is not, however,
gender and constituency on representatives’ behavior dis-
settled. It is not credible to conclude from the indepen-
connected. Since our analysis does so, our findings provide
dent variables listed that the degree to which members’
novel insight on this “critical mass” question.
constituencies are favorable toward the representational
behaviors that serve as the dependent variables in these
Quasi-experimental Design and the
studies is measured exhaustively.2 For example, does the
level of support that President Clinton received (Swers
Empirical Foundations of Gender and
2002) indicate how predisposed constituents across
Critical Mass Effects
House districts are to addressing social welfare policy
challenges and/or endorsing feminist policies (the number
of such bills sponsored by members is a dependent vari-
The Difficulty of Isolating the Effects of Gender
able in some of the studies described previously)?
and Constituency on Representation
President Clinton won states, and congressional districts
within states, across the South, and received similar sup-
The difficulty for understanding how gender influences how
port in states, and congressional districts, in the Northwest
well legislators represent women does not stem from a lack
and Northeast. But individuals who voted for Clinton in
of outcome variables—almost all levels of legislative activ-
the South did not, on average, react similarly to Clinton
ity have been studied.1 Rather, as Poggione (2004, 306)
voters in the northeast when the new president endorsed
argues, the omission of constituency characteristics from
a “Don’t ask, don’t tell” policy regarding the ability of
these models leaves open the possibility that “systematic
homosexuals to serve in the military.

---

474
Political Research Quarterly 64(2)
Clearly, there is a lot more to a district...