Approaches to Studying Policy Representation

• Date: 01 February 2016
• DOI: http://doi.org/10.1111/lsq.12110
• Published date: 01 February 2016
• Author: David E. Broockman

DAVID E. BROOCKMAN

Stanford Graduate School of Business

Approaches to Studying Policy

Representation

Some studies of policy representation test hypotheses about the relationship

between citizens’ views and elites’ positions on multiple issues by proceeding one issue

at a time. Others summarize citizens’ and elites’ preferences with “ideology scores” and

test hypotheses with these. I show that approach is flawed. It misinterprets citizens’ ide-

ology scores as summaries of policy preferences, but these scores actually measure

ideological consistency across areas: how often citizens’ ideal policies are liberal or con-

servative. Examples show how attending to this distinction overturns conventional

wisdom: legislators appear similarly moderate as citizens, not more extreme; however,

politically engaged citizens appear especially moderate.

Scholars typically employ one of two approaches when studying

political elites’ representation of public opinion. A f‌irst approach tests a

hypothesis about the relationship between citizens’ opinions and elites’

positions one issue at a time on each of many issues and then examines

how well the hypothesis held across issues. For example, Lax and Phillips

(2012) examine how likely policies are to become law at various levels

of public support. Gilens (2012) compares how likely policies are to

become law depending on levels of public support among those of vari-

ous incomes. And Lenz (2012) examines whether citizens tend to adopt

politicians’views on a variety of issues.

1

A second approach f‌irst computes measures of individual citizens’

and individual politicians’ overall “ideologies” based on their positions

on many policies and tests a hypothesis with these ideology scores. In

such analyses, citizens’ and elites’ policy views are both summarized by

a point on a liberal-conservative index, and their locations on this index

are then compared. For example, in their inf‌luential analysis, Bafumi and

Herron (2010) estimate individual-level “ideal points” for both survey

respondents and their Members of Congress. Bafumi and Herron (2010)

use these estimates to assess correspondence between representatives’

policy decisions and their constituents’ policy preferences, f‌inding that

over 90% of voters are more moderate than legislators but that donors

and primary voters are similarly extreme.

LEGISLATIVE STUDIES QUARTERLY, 41, 1, February 2016 181

DOI: 10.1111/lsq.12110

V

C2016 Washington University in St. Louis

Research summarizing voter preferences with ideological scales

has long been conducted (e.g., Enelow and Hinich, 1984) but has

burgeoned in recent years (Barber 2014; Barber

a 2015; Bond and

Messing 2015; Bonica 2013; Caughey 2014; Caughey and Warshaw

2014; Clinton 2006; Ellis and Stimson 2012; Erikson, MacKuen, and

Stimson 2002; Gerber and Lewis 2004; Griff‌in and Newman 2005, 2007;

Masket and Noel 2012; Pan and Xu 2015; Peress 2013; Rogowski and

Tucker 2014; Saiegh 2015; Shor, Berry, and McCarty 2010; Shor 2013;

Stone and Simas 2010; Tausanovitch 2014; Tausanovitch and Warshaw

2014a, 2014b). Expressing a growing sentiment regarding the superiority

of the ideological approach for studying citizens’ policy preferences, Lo,

Proksch, and Gschwend write that “research on elections and party

competition is unthinkable without measures of the ideological positions

of voters” (2014, 205).

This article argues that this ideological approach for studying

voter preferences has crucial and underappreciated f‌laws. The ke y issue

I explore is that ideological scales tend to capture citizens’ degree of

ideological consistency across policy domains (e.g., “this citizen has

liberal views on two-thirds of issues”) but say little about citizens’

views within domains, on issues themselves (e.g., “this citizen supports

state-sponsored healthcare”).

To appreciate this distinction between consistency across domains

and views within domains, consider a common use of ideological scales:

comparing how “extreme” legislators’ policy positions are relative to

citizens’ views. In studies that employ ideological scales for this task,

individual voters’ or politicians’ “extremity” is typically based on the

extremism of their score on an ideological index estimated from

responses to many binary survey items (for citizens) or votes across

many roll calls (for legislators) (e.g., Bafumi and Herron 2010; Barber

2014; Masket and Noel 2012; Peress 2013; Rogowski 2014; Shor 2013).

With this methodology in mind, examine the political preferences and

survey responses given by one voter and one legislator in two issue areas

shown in Table 1. The legislator in Table 1 has consistently conservative

but fairly moderate positions in both issue areas. However, because the

legislator comes down on the conservative side of both issues, the legisla-

tor would appear as conservative as possible on an ideological index

created from these two votes. On the other hand, the voter has extreme

views in both policy areas. Nevertheless, one liberal response and one

conservative response earns her a score at the middle of the index. Literature

on this topic would thus deem the voter moderate despite her thor-

oughly extreme views and the legislator an extremist despite his

moderate positions. But this voter is not really ideologically moderate,

182 David E. Broockman

she is ideologically mixed; and the legislator is not ideologically extreme,

he is ideologically consistent. Interpreting ideological scales as measuring

views on issues themselves rather than ideological consistency can thus

mislead even simple descriptions of individual’s policy preferences.

This distinction has more general consequences for the study of

congruence between voters and their representatives with ideological

scales, questions such as “how well [a politician] represents” his or her

constituency on policy matters (e.g., Bafumi and Herron 2010) or which

constituents a politician represents best (e.g., Griff‌in and Newman 2007;

Tausanovitch 2014).

A second hypothetical illustrates these implications. Suppose you

are a Member of Congress representing a district with f‌ive voters. In the

upcoming session of Congress, you will be asked to cast a roll-call vote

on f‌ive issues. Imagine that, wanting to maintain congruence with district

opinion, you conduct a poll of the f‌ive voters in your district on these

f‌ive issues. The results appear in Table 2. In each cell in Table 2, a 0 cor-

responds to a conservative view on a policy and a 1 corresponds to a

liberal view.

The results of the poll give clear guidance about how to vote

congruently. On each of the f‌ive issues, a majority of your constituents

TABLE 1

Example: Studying Extremism with Ideological Scales

Voter Legislator

Ideal Gay Rights

Policy

Do not allow gays to

teach in public schools

(Extremely conservative)

Civil unions only, no

same-sex marriage

(Moderately conservative)

Answer to survey

question / Roll call

vote: “Should gay

marriage be illegal?”

Yes Yes

Ideal Immigration

Policy

Open borders;

unlimited immigration

(Extremely liberal)

Limit low-skilled immigration

with border protections

(Moderately conservative)

Answer to survey

question / Roll call

vote: “Should

immigration be

restricted?”

No Yes

One-Dimensional

Ideology Estimated

From Survey

Responses / Votes

Moderate Extreme Conservative

183Studying Policy Representation