Varied Effects of Policy Cues on Partisan Opinions

58
The Variable Effect of Policy Cues on Partisan Opinions Matt Grossmann Assistant Professor of Political Science, Michigan State University 303 S. Kedzie Hall East Lansing, MI 48823 (517) 355-7655 [email protected] Abstract: Although citizens often arrive at the same views as their political party’s leaders, they also respond to information about policy targets and effects. Accounting for political context encourages a variable view of how partisanship shapes opinions in policy debates. In three survey experiments associated with policies supported by both Democrats and Republicans, I find that

Transcript of Varied Effects of Policy Cues on Partisan Opinions

The Variable Effect of Policy Cues on Partisan Opinions

Matt GrossmannAssistant Professor of Political Science,

Michigan State University303 S. Kedzie Hall

East Lansing, MI 48823(517) [email protected]

Abstract:

Although citizens often arrive at the same views as their

political party’s leaders, they also respond to information about

policy targets and effects. Accounting for political context

encourages a variable view of how partisanship shapes opinions in

policy debates. In three survey experiments associated with

policies supported by both Democrats and Republicans, I find that

The Variable Effect of Policy Cues on Partisan Opinions 1

both aspects of policy argumentation and the actors making the

arguments can enable partisanship to affect public opinion. This

process is highly conditional, however: sometimes polarization

occurs only with the presence a single politician; in other

areas, polarization is likely following presentation of evidence

by either partisan side. The effect of policy information on

partisan polarization is variable across political and policy

contexts.

The Variable Effect of Policy Cues on Partisan Opinions 2

American public opinion is often polarized based on partisan

attachments. The mere appearance of President Barack Obama behind

a policy initiative seems to direct partisans to each side of the

debate. Are policy opinions thus mere artifacts of partisan

attitudes? There are certainly cases where they seem to be driven

by little else, but not all issue areas follow this pattern.

Sometimes the American public does unite behind particular policy

options and respond to information about policy effectiveness; at

other times, partisans remain divided regardless of the evidence

and even without a clear divide among partisan proponents. Policy

choices are discerned through partisan lenses, but that is hardly

the same as saying that information is uniformly irrelevant. I

ask how different policy cues affect partisan polarization.

Democrats and Republicans perceive policy issues differently

and listen to different sets of political elites. As they

encounter political debate, they often divide into two

constituencies sharing the views of their favored leaders. Not

all issues polarize, but division in contested issue areas

usually follows partisan divisions (Zaller 1992). As party

leaders advocate on behalf of policy proposals, frame the terms

The Variable Effect of Policy Cues on Partisan Opinions 3

of political debate, and take credit for achievements, citizens

use pre-existing partisan preferences to arrive at the same views

as their party’s leaders. This is the view, anyway, of an

enduring framework for understanding public opinion: the public

interprets politics through their “partisan hearts and minds”

(Green, Palmquist, and Schickler 2002). Policy details, in this

story, matter less than competing partisan messages.

Yet an expanding area of scholarship argues that, regardless

of partisan affiliation, public opinion on policy issues is

responsive to the specific target populations of policy proposals

and information about their potential effects (Schneider and

Ingram 1993; Lau and Schlesinger 2005). Public opinion is

susceptible to change based on the arguments used by proponents

and opponents, especially when they tap underlying attitudes

toward social groups and institutions (Nelson and Kinder 1996;

Jerit 2009). Although opinions are often biased based on partisan

and ideological predispositions, there is no immutable

relationship between one’s general side in the political debate

and their opinions about any specific proposal. Differences in

opinions may develop based on who pays attention and what values

The Variable Effect of Policy Cues on Partisan Opinions 4

are tapped by policy arguments and may thus change with the

political territory and the evolution of policy debate (Bullock

2011).

The availability of randomized survey experiments has

enabled this debate to proceed beyond generic claims and toward

specific examples, exploring how policy argumentation affects

public opinion. Yet these experiments have been subject to

criticism on the grounds that they inappropriately attribute

causal influence (Gaines, Kuklinkski, and Quirk 2007) and lack

external validity (Barabas and Jerit 2010). Their results are

often dependent on the specific political context in which they

are embedded. The popular response to this criticism has been to

modify experiments to make them as generalizable as possible and

verify their effects with other data sources, but the attempted

extensions have not shown much consistency.

I pursue a different approach: accepting the conditionality

of partisan response to policy frames, I more fully embed

experiments in particular contexts. I use ambiguities in popular

discussion to assess the effects of different aspects of policy

and their associated political dynamics. If public response is

The Variable Effect of Policy Cues on Partisan Opinions 5

variable based on political debate, the framing and credit

claiming choices of the partisan sides that present relevant

information to citizens should be important for shaping public

opinion. Sometimes policy cues can direct attention to particular

target populations that may be more appealing to partisans on

each side. Sometimes well-known proponents of a policy (like the

President) can be polarizing, whether or not equivalent partisan

proponents have the same effect. The presence of evidence for

policy effectiveness may also trigger partisanship, regardless of

who presents it. Because partisanship already structures many

opinions, we should expect significant variation in how much

policy cues affect the structure of variation in opinion. Some

issues are easily polarizing, others only in limited

circumstances. Information presented by political elites may

polarize or depolarize opinion.

To assess the influence of policy cues on partisan opinions,

I take advantage of three policy issue debates in the U.S. state

of Michigan concerning Medicaid, transportation funding, and the

auto industry bailout. Each case featured a critical ambiguity in

the partisan composition of proponents: both Democratic and

The Variable Effect of Policy Cues on Partisan Opinions 6

Republican officials were voicing support or claiming credit for

the same policies, especially Democratic President Barack Obama

and Republican Governor Rick Snyder. Via randomized experiments

in a statewide telephone survey, I manipulate how respondents

think about the beneficiaries and sponsors of policy proposals

and the responsible parties for policy outcomes. This enables an

investigation of three different paths for opinion change in

policy debates, where I find multiple types of systematic

partisan responses.

The analysis proceeds in several parts. First, I review

claims and evidence about the relationship between partisanship

and issue attitudes. Second, I analyze research on the effects of

policy cues regarding the beneficiaries, proponents, and credited

actors for policy proposals. Third, I argue that taking account

of specific political context helps reconcile previous findings

and encourages a variable view of how partisanship shapes

opinions in policy debates. Next, I present the data and evidence

from my original survey experiments.1 I then analyze the results

of three survey experiments covering partisan opinion on (a) the

target populations of Medicaid expansion, (b) proponents of

The Variable Effect of Policy Cues on Partisan Opinions 7

gasoline tax increases for road infrastructure, and (c) claimants

of credit for the auto industry bailout. The results show that

partisanship regularly structures how citizens respond to policy

cues but not in consistent form. There are at least three

variations in partisan responses to policy debates, including how

cues alter partisan predispositions.

Partisanship and Issue Attitudes in Public Opinion

The American public often knows little about the stakes or

sides of policy debates (Delli Carpini and Keeter 1997). Few

Americans pay enough attention to be aware of the political

leaders’ stands on each issue area or the reasons for their

disagreements (Hibbing and Theiss-Morse 2002) As a result, many

American have unstable, ambivalent, or unclear attitudes about

policy issues (Zaller 1992). Opposing arguments sent by each

party’s leaders help polarize the electorate by convincing

partisans to support their leaders’ views (Zaller 1992).

To the extent that citizens form policy opinions,

partisanship plays a prominent role. Americans, have “partisan

hearts;” their main political predispositions are a consequence

The Variable Effect of Policy Cues on Partisan Opinions 8

of their partisan and social group attachments (Green, Palmquist,

and Schickler 2002). They also have “partisan minds:” they

interpret policy cues and messages through partisan lenses, often

reaching different conclusions in the face of similar evidence

(Green, Palmquist, and Schickler 2002).

Partisans tend to credit their own officials with policy

successes and the opposition party with failure (Haider-Markel

and Joslyn 2003), counter arguing against any evidence that the

other party’s performance was associated with positive news. They

respond differently to the same proposals made by members of

different parties and the same types of candidates when they are

associated with different partisan labels (Goren, Fredrico, and

Kittilson 2009; Rahn 1993). The main differences in individual

opinion on policy issues (Green, Palmquist, and Schickler 2002),

individual votes on statewide initiatives (Branton 2003), and

electorates across the states correspond to a partisan and

ideological dimension (Erickson, Wright, and McIver 1994).

Political beliefs and identities become much better indicators of

policy preferences as policy environments become more polarized

(Garner and Palmer 2010). Elite polarization along partisan lines

The Variable Effect of Policy Cues on Partisan Opinions 9

influences individual attitudes and policy preferences

(Hetherington 2009).

If partisanship is a central determinant of public opinion,

it should manifest in several ways on survey experiments. First,

the public should divide into two opposing groups on policy

issues tied to each party, despite differences in information

presented. Second, respondents should become polarized in the

face of partisan proponents of policy or partisan attempts to

take credit for policy. Third, non-partisan information and

references to group beneficiaries should have less effect on

public opinion and produce less polarization. Although there have

been some findings of induced polarization through survey

experiments and partisan cues (Mondak 1993; Goren, Fredrico, and

Kittilson 2009), there is also some evidence that information and

thematic frames provided to voters may be more important (Bullock

2011; Jerit 2009).

Cues about Policy Beneficiaries, Proponents, and Credit Claimers

Investigations of public opinion on policy issues show

numerous effects beyond partisanship. Most Americans are unaware

The Variable Effect of Policy Cues on Partisan Opinions 10

of current government policy and even less aware of potential

proposed solutions to address social problems (Delli Carpini and

Keeter 1997). Even if they do not have firm opinions on policy

issues, citizens have access to a variety of considerations in

addressing an issue that comes to their attention via a survey

question. Survey experiments prime specific considerations,

making respondents more likely to think of policy in different

terms (Nelson and Kinder 1996). Respondents provide different

opinions when they learn how specific policy proposals relate to

their values (Lau and Schlesinger 2005). By providing policy

cues, experiments make different factors more important

considerations in opinion formation, affecting respondent beliefs

about policy results (Jerit 2009).

What types of cues are likely to affect public opinion?

Nelson and Kinder (1996) argue that public opinion is commonly

“group-centric:” policy opinions follow attitudes toward “the

social groups perceived as the beneficiaries of policy.” Their

survey experiments demonstrate that issues can be framed to

enhance or detract from respondents‘ tendencies to evaluate

policy on the basis of group benefits. One way to stimulate

The Variable Effect of Policy Cues on Partisan Opinions 11

“group-centric” policy opinions is to directly manipulate a

policy’s target population. Frames stimulating symbolic political

beliefs have been found far more influential than indicators of

direct self-interest (Lau and Schlesigner 2005). Although many

experiments show that policy support responds to target

populations, most are designed to assess the role of racism and

social desirability, rather than the multiple constituencies of

most government social programs (Gaines, Kuklinski, and Quirk

2007).

Slothuus and De Vreese (2010) argue that the partisanship of

respondents and policy proponents should affect respondents’

susceptibility to policy frames. They find that partisanship

conditions how citizens respond to policy argumentation,

especially on issues that are widely debated. According to

Slothuus and De Vreese (2010), citizens are more accepting of

frames forwarded by co-partisan sponsors and may counter-argue

against an opposition proponent making a similar argument,

decreasing their support for the same policy. Their experiments

have the same frame forwarded by different parties on two issues:

support for contracting in-home care and a trade agreement,

The Variable Effect of Policy Cues on Partisan Opinions 12

finding greater partisan framing effects in the case of in-home

care (and supposing that the difference would be apparent on all

conflict-ridden issues).

Beyond the proponents, the information itself seems to

influence survey respondents. In direct comparisons, attitudes

change at least as much in response to learning relevant policy

information as from learning which parties support each policy

(Bullock 2011). In the experiments that demonstrate public

response to partisan cues, some respondents may be using the cues

to infer information about the beneficiaries of policy rather

than simply adopting the positions of their party elites. When

citizens learn directly about the beneficiaries through policy

details, they respond less to party cues (Bullock 2011).

Citizens seek credible information, especially when

knowledge is limited (Druckman 2001). The effects of cues on

opinions may be non-uniform across the population, differing

based on whether respondents lack political sophistication or pay

less attention to government (Lau and Schlesigner 2005). Opinion

change that can be replicated in the real world may only be

The Variable Effect of Policy Cues on Partisan Opinions 13

observable among the most engaged subset of respondents (Barabas

and Jerit 2010).

There are significant limitations to survey experiments. The

effects of hearing a short snippet of information may be

transitory (Gaines, Kuklinski, and Quirk 2007). For respondents

to learn these policy details in the course of political debate,

they would have to pay attention, remember, and reassess their

views; this may be asking a lot. Natural experiments that attempt

to replicate the findings of contemporaneous survey experiments

do not show the same effects on policy opinions (Barabas and

Jerit 2010). Only a subset of respondents may change their

beliefs about policy in response to learning the information

provided in a survey experiment through media coverage in the

real world; among that subset, changes in beliefs may not lead to

changes in policy opinions (Barabas and Jerit 2010). Policies

designed to create supportive constituencies can develop and

sustain public support over time, even if the messages used in

the initial political debate over those policies are less

persuasive (Jacobs and Mettler 2011).

The Variable Effect of Policy Cues on Partisan Opinions 14

Taking Account of Political Context

Survey experiments likely cue messages that respondents have

already encountered and serve to highlight which considerations

are already apparent in respondents’ thinking about an issue

(Gaines, Kuklinski, and Quirk 2007). When results indicate that

respondents in one treatment do not differ significantly from

those in the control condition, it may indicate that the

consideration primed by that treatment is already included in

most respondents’ views on the issue (Jerit 2009). Studying

changes in opinion requires attention to both the short-term

effects of arguments used in political debate and the long-term

effects of learning about policy results (Jacobs and Mettler

2011).

Previous studies of the effects of frames on policy opinions

recommend searching for the popular arguments used in elite

discourse via a search of public records and advocate statements

to see if citizens change opinions after they learn the specific

claims made in policy debates (Nelson and Kinder 1996). This

approach suffers from possible pre-treatment, where some

respondents have already heard an argument in popular debate

The Variable Effect of Policy Cues on Partisan Opinions 15

(Jerit 2009). As a result, other scholars recommend approaches

that enable better generalization from the arguments presented to

citizens in survey experiments to what would happen if public

officials presented those same arguments (Gaines, Kuklinski, and

Quirk 2007).

I pursue an alternative approach, taking advantage of

ambiguities in political discourse and policy argumentation to

isolate the potential effects of hearing about different aspects

of policies from different proponents. Rather than attempt to

divorce away from political context, I use specific political

contexts to clarify the circumstances under which opinion

differences manifest in response to policy cues. By abandoning

the assumption that citizens respond equivalently to the same

type of source cue or policy frame regardless of circumstances,

scholars can tell a more nuanced story of the many possible

responses in different political contexts.

A context sensitive approach requires taking advantage of

real-world political circumstances where citizens might be

treated to multiple types of political arguments. Rather than

have a Democratic and Republican candidate both make a statement

The Variable Effect of Policy Cues on Partisan Opinions 16

that only a Democrat would normally make, researchers can find

examples of where voters might hear equivalent arguments from

different politicians. Cues regarding target populations and

credit claiming should likewise attend to contexts where multiple

constituencies might benefit from a policy or where multiple

elites might try to take credit for policy achievements. This

requires taking advantage of situations that mimic natural

experiments where citizens hear the cues used in survey questions

(Barabas and Jerit 2010). Cues can help guide survey respondents

to think in terms of one aspect of a policy that they might

encounter in political debate.

I take advantage of the political context surrounding three

policy issue debates in Michigan: Medicaid expansion, gasoline

taxes for road funding, and the auto industry bailout. All three

cases feature a partisan ambiguity regarding responsibility.

Republican Governor Rick Snyder, Democratic President Barack

Obama, conservative interest groups like the Chamber of Commerce,

and liberal interest groups like the AFL-CIO were all

simultaneously promoting Medicaid expansion and gasoline taxes;

they were also taking credit for the success of the automotive

The Variable Effect of Policy Cues on Partisan Opinions 17

industry bailout. Proponents of Medicaid advocate it as both a

policy for poor families and for elderly nursing home care,

enabling a comparison of target populations. In the case of

gasoline taxes, I can honestly manipulate the proponents to

include Democratic and Republican elected officials and liberal

and conservative interest groups. In the auto bailout case, I can

realistically manipulate who takes credit for the policy and

whether respondents hear any evidence of its success.

I do not expect a uniform response to policy cues across

these issue areas. Medicaid expansion is generally popular, but

some constituencies (like the elderly) may be more popular than

others (like poor families). Gasoline taxes are generally

unpopular, but each proponent may stimulate different partisan

considerations. One elected official (like President Obama) may

be more polarizing than others. Because the auto bailout is

already widely considered a success in Michigan, it may matter

less who takes credit for it afterwards.

I also expect heterogeneity in the degree of polarization

surrounding each issue area. Gasoline taxes seem the most subject

to potential polarization, but the variety of their proponents in

The Variable Effect of Policy Cues on Partisan Opinions 18

Michigan may help undermine the role of traditional political

divisions over taxation. The auto bailout and Medicaid may not

instinctively produce opposing Democratic and Republican

opinions, unless the specific cues highlight partisan

considerations. The contexts of each policy debate enable an

investigation into whether opinion is inherently polarized by

partisanship as well as how cues about the beneficiaries and

proponents of policy shape potential polarization.

Data and Methods

The experiments were part of the 62nd State of the State

Survey conducted by the Institute for Public Policy and Social

Research at Michigan State University. The quarterly survey uses

a stratified random sample with both landline telephones and cell

phones. The survey included interviews with 1,015 Michigan

residents from June 12 – August 13, 2012. Survey data,

instruments, and documentation are available at

http://ippsr.msu.edu/SOSS.

These data draw on public opinion in a single state, but

Michigan is as an excellent case for assessing the interaction

The Variable Effect of Policy Cues on Partisan Opinions 19

between partisanship and policy opinions. Michigan has a

Republican governor that is pursuing policies sometimes

associated with Democrats and shares some priorities with

Democratic President Obama. It is also the state most affected by

the automobile bailout, a policy shaped by a Republican and

Democratic president and supported by the state’s entire

congressional delegation (but most associated with President

Obama). This makes Michigan a particularly useful case for

observing the relationship between policy cues, partisanship, and

public opinion in specific policy proposals and outcomes. At the

time of the survey, Republicans were in firm control of both

houses of the state legislature and the governor’s office, even

though Democrats substantially outnumber Republicans in the

public. The state was still facing an uneven economic recovery,

but no issue dominated state politics.

I conducted three survey experiments within a broader survey

that asked residents about their opinions of state government and

their quality of life. The question assessing support for

Medicaid expansion began “Medicaid is a joint state and federal

program that pays providers to deliver health care services.”

The Variable Effect of Policy Cues on Partisan Opinions 20

Respondents were asked “How much do you favor or oppose

increasing government funding for Medicaid?” The response choices

were “strongly favor, somewhat favor, neither favor nor oppose,

somewhat oppose, and strongly oppose.” Respondents could also

volunteer that they “don’t know” or refuse to answer the

question, although only 12 respondents out of 1,015 selected

these options. Before answering the question, a random number

assigned uniquely to each respondent determined which additional

piece of information respondents received. In the control

condition, they received no additional information. In the for

poor condition, they were told “It covers basic health care for

the poor” or “It covers basic health care for poor families with

children.”2 In the for elderly condition, respondents were told “It

covers nursing home care for the elderly.” Both treatments

provide true information; most Medicaid recipients are poor

families but most Medicaid dollars are spent caring for the

elderly. Treatments that provide information that forces

respondents to consider different dimensions of a policy issue is

more likely to reflect underlying value conflicts and lead to

opinion divergence (Jerit 2009). This is a standard experimental

The Variable Effect of Policy Cues on Partisan Opinions 21

manipulation of target population made possible by an ambiguity

in the policy’s actual beneficiaries. Both Republican Governor

Snyder and Democratic President Obama are actively promoting a

dramatic expansion of Medicaid.

The question assessing support for gasoline taxes for road

construction began “Gasoline tax revenues, the primary source of

road funding, are declining because newer cars get better gas

mileage.” Respondents were asked “How much do you favor or oppose

funding repairs and maintenance of roads and bridges by

increasing gasoline taxes?” The response choices were the same

five-category scale from “strongly favor” to strongly oppose”

used in the Medicaid question. A different random number again

assigned each respondent to an experimental treatment. In the

control condition, they were told that “some people argue that we

must maintain funding to repair our roads.” In the Obama supports

condition, respondents were told that “President Barack Obama

argues that we must maintain funding to repair our roads.” In the

Snyder supports condition, “Governor Rick Snyder” replaced Obama as

the source of this argument. In the business supports condition,

“Business leaders” made the same argument; “Union leaders” made

The Variable Effect of Policy Cues on Partisan Opinions 22

the argument in the union supports condition. The only manipulation

was thus to the source of the argument, not the information

itself. All information again reflects political discourse in the

state, including the strange bedfellow coalition of unions,

corporations, and Democratic and Republican officials supporting

a policy with low public support.

The question assessing auto industry bailout approval

advised all respondents that “George W. Bush and Barack Obama

provided federal financing to General Motors and Chrysler. Both

companies restructured in managed bankruptcies.” They were asked:

“How much do you approve or disapprove of the decision to provide

government financing for the auto industry restructuring?” The

response choices were “strongly approve, somewhat approve,

neither approve nor disapprove, somewhat disapprove, or strongly

disapprove.” Before the question, respondents were assigned to a

treatment group. In the control condition, respondents were given

no additional information. In the evidence condition, they were

told: “Detroit area unemployment has since fallen from 16% to

9.2%. Some say the local economy is improving but others say it

is still in tough shape.” In the Republican evidence condition, the

The Variable Effect of Policy Cues on Partisan Opinions 23

same fact was attributed to “A Republican Congressmen from

Michigan” while “others say” was replaced with “some Democrats

say.” For the Democratic evidence condition, I used the same

structure but opposite party affiliations. This experiment thus

enabled a test of the effect of hearing the additional

information about policy success as well as the effect of the

party taking credit for that success or disparaging it. There are

examples of all four pairs of credit attribution and blame in the

state’s news coverage of the auto bailout.

Due to random assignment, the respondents in each

experimental condition are representative of the Michigan

population and do not differ substantially in their demographics

or political attitudes. I rely on data regarding the political

orientations of respondents to assess the factors leading to

support for each policy. My measures of partisan identification

and political ideology use two-part question formats using the

standard categories from the National Election Studies.

Respondents are placed on seven-category equidistant scales from

strong Republican to strong Democrat and from very conservative

to very liberal. Respondents are also asked whether they like or

The Variable Effect of Policy Cues on Partisan Opinions 24

dislike labor unions and whether they like or dislike for-profit

companies; I created five-category equidistant scales from “like

a lot” to “dislike a lot” based on the answers to these

questions. I am thus able to assess the role of political

ideology and partisanship along with attitudes concerning the

specific groups in the policy information cues.

I also assess whether the extent to which frames change

political opinions is dependent on the political sophistication

of respondents. Some people pay a lot of attention to politics

and the news media and may already have been exposed to some of

the informational cues that we use. I measured political

information directly through a two-question quiz asking

respondents to correctly identify the Vice President of the

United States (72% could) and the party in control of the state

legislature (62% could). I construct a 3-point scale between 0

and 1 based on the number of correct responses. This helps assess

whether each treatment produced a stronger or weaker effect among

those who are more exposed to government and policy debate.

Although some previous studies find interactions between

partisanship and political information, additional models showed

The Variable Effect of Policy Cues on Partisan Opinions 25

no effects for this interaction in the Medicaid or gasoline tax

experiments.3 Although the survey included a standard battery of

demographic questions, none of the other attributes had large and

consistent relationships with the answers to the policy

questions. Across all of these demographic factors, there was no

evidence that randomization failed to produce equivalent groups

across treatments and controls.

Partisan Opinion on the Target Populations of Medicaid

To assess the determinants of support for Medicaid

expansion, I converted the five-category question to a zero to

one scale with equidistant positions from “strongly favor” (one)

to “strongly oppose” (zero). Figure 1 illustrates the average

scores on this measure for different groups of respondents. I

divide all respondents based on the experimental treatment that

they received; I also provide scores for Republicans and

Democrats who received each treatment. Democrats were

substantially more supportive than Republicans in all three

experimental conditions. Respondents who were told that the

elderly were the target population were much more supportive,

The Variable Effect of Policy Cues on Partisan Opinions 26

especially among Republicans. Support for Medicaid expansion was

generally high, except among Republicans who did not hear about

any target population.

[Insert Figure 1]

The differences between partisan groups and experimental

conditions are both statistically significant. Table 1 uses

multivariate models to assess the strength of the relationships.4

The first model includes the entire sample of respondents, with

dummy variables comparing each of the experimental treatments to

the control treatment where respondents heard no target

population. The final three columns cover separate models for

each of the three treatment groups. Both the for poor and the for

elderly treatments significantly increased support for Medicaid

expansion. More democratic and liberal respondents were more

supportive of Medicaid in all three experimental treatments,

though the effect of partisanship was diminished in the for elderly

group. The model associated with the elderly target population

also explains less variation in opinion overall.

[Insert Table 1]

The Variable Effect of Policy Cues on Partisan Opinions 27

Figure 2 plots the predicted values of Medicaid expansion

support associated with each partisan category in each

experimental treatment, holding all other variables at their

mean. All three treatments show a substantial partisan effect.

Yet the slope is largest for the control condition and the

intercept is lowest; the opposite is true of the for elderly

treatment. As a result, the effects of the treatment are only

apparent on the Republican side of the scale and only large in

the case of strong Republicans. Democrats are not much affected

by learning about the different target populations.

[Insert Figure 2]

This suggests that cues bringing different target

populations to mind can affect citizen opinions, but perhaps

differentially based on whether partisan tendencies already

support program expansion. Republicans may be given a reason to

support the policy, whereas Democrats may not need to be

reassured about who benefits. These affects were apparent even

absent any manipulation of the partisan identity of proponents of

Medicaid expansion.

The Variable Effect of Policy Cues on Partisan Opinions 28

Partisan Opinions on Gasoline Taxes for Road Funding

I observed a different pattern of partisan opinions in the

experiment on gasoline taxation. Figure 3 graphs the basic

results, again using a zero-to-one scale from “strongly oppose”

to “strongly favor.” I divide respondents based on their self-

reported partisanship and experimental treatment. Although

Democrats were more supportive overall, the control condition

showed no significant partisan difference. Whereas support from

businesses, unions, and Governor Rick Snyder did not seem to

affect support, the condition highlighting Obama’s support for

gas taxes did polarize Democrats and Republicans. Overall,

support for gas tax increases was low, even given the question’s

emphasis on road funding and repair.

[Insert Figure 3]

Table 2 uses similarly constructed models to assess support

for gas tax increases among respondents. I include a model of the

complete sample as well as one for three of the five experimental

treatments (the business supports and union supports conditions did not

show significant differences from the control group and are not

shown). The complete model again includes indicator variables

The Variable Effect of Policy Cues on Partisan Opinions 29

comparing each experimental treatment to the control group; no

treatment significantly increased support. The effects of

traditional political variables in predicting support for gas tax

increases were quite muted. In fact, the models for the control

group and the Snyder supports condition explained essentially no

variation in support for gas taxes. In contrast, partisanship is

significantly associated with support for gas taxes in the Obama

supports condition; that model also explains significant variation

in opinion.

[Insert Table 2]

The polarizing effect of hearing that Obama supports gas

taxation for road funding is apparent in Figure 4, which graphs

the predicted values of support for the policy among partisan

categories for the three experimental treatments (holding all

other variables at their mean). The control and Snyder supports

conditions show flat lines. The effect of partisanship in the

Obama supports condition is pronounced, with strong Republicans

converging on opposition and strong Democrats becoming quite

supportive.

[Insert Figure 4]

The Variable Effect of Policy Cues on Partisan Opinions 30

This shows a second type of effect of policy cues on

partisan opinions: support from a particular highly visible actor

may be polarizing, even when other actors do not show such

effects. I took advantage of a real political context where both

the Republican governor and the Democratic president, along with

traditionally Democratic and Republican interest groups,

supported a policy that the public often opposes (a highly

visible tax). Support from a Republican governor or conservative

or liberal interest groups did not change opinions or stimulate

polarization. Even though taxes are often polarizing, only

support from President Obama caused Republicans to gravitate

toward their traditional anti-tax stance.

Partisan Approval of the Auto Industry Bailout

The third experiment tapped citizen support for a policy

that had already been enacted and was reportedly associated with

some economic success: federal support for the auto industry

bailout. This policy was quite popular in Michigan, the home of

the American auto industry including the two corporations who

were bailed out by the federal government and restructured in

The Variable Effect of Policy Cues on Partisan Opinions 31

bankruptcy. Figure 5 illustrates levels of approval for the

policy on a five-category equidistant scale ranging from

“strongly disapprove” (zero) to “strongly approve” (one). I again

separate respondents based on their self-described partisanship

as well as the treatment group to which they were randomly

assigned. The results show that Democrats are significantly more

supportive than Republicans but the difference in support is not

as large in the control condition.

[Insert Figure 5]

Table 3 presents multivariate models of support for the auto

industry bailout with the same variables used in Table 2. The

complete model compares each experimental treatment to the

control group (where I provided no information about the success

of the policy and no arguments in favor or opposed). There are no

significant direct effects of any treatment. Polarization along

liberal-conservative ideological lines was apparent in all

conditions, including the control, as was diverging opinion based

on favorability toward labor unions (a key beneficiary of federal

support for the auto restructuring). Political knowledge

significantly raises bailout approval, but only among those who

The Variable Effect of Policy Cues on Partisan Opinions 32

hear evidence of its success. The control condition showed no

divergence based on partisanship.5 Partisanship became much more

polarizing in the Democratic evidence and Republican evidence

conditions. This means that, even when respondents were informed

that a Republican congressmen was touting the policy’s effects

and Democrats were still unsatisfied, Democrats became much more

likely to support the policy. In this case, the polarization

seemed to come from the presentation of the evidence itself and

the cueing of the Detroit comeback that the policy helped enable.

[Insert Table 3]

Figure 6 visualizes these differing effects of partisanship

in the experimental conditions. In the control group, Democratic

and Republican respondents were nearly equal in their support. In

all of the other treatments, Democrats were much more supportive.

Whether the evidence of the Detroit comeback was presented by a

Democratic or Republican congressman, it stimulated Democrats to

approve and Republicans to disapprove of the policy. This was

also true in the neutral evidence condition, where respondents

heard the same information attributed to no party. Although it is

not pictured, the slope of the relationship in the neutral

The Variable Effect of Policy Cues on Partisan Opinions 33

evidence condition was even larger than that in the conditions in

which partisan officials presented the same evidence.

[Insert Figure 6]

This third outcome of partisan opinion formation based on

policy cues demonstrates that partisan polarization may not only

arise from specific cues based on party support for policy

options, but also based on cues regarding aspects of a policy

even when no party is mentioned. Republicans may be less likely

to respond to evidence of economic success during the Obama

administration, especially for a policy that was publicly

associated with Obama. Alternatively, Republicans in Michigan may

have responded negatively to information about the policy’s

effect on Detroit, a largely Democratic area of the state

(although the evidence presented referenced the “Detroit area”

rather than the city itself). Partisan polarization in opinions

is not always an outcome of partisan cues; it can even result

from evidence of policy success.

The Varied Effects of Policy Cues on Partisan Opinions

The Variable Effect of Policy Cues on Partisan Opinions 34

The three different survey experiments, each contextualized

in contemporary Michigan policy debates, produced three different

types of effects on partisan opinions. One treatment in the

Medicaid experiment (cueing the elderly as the target population)

reduced polarization. One treatment in the gas tax experiment

(cueing Obama’s support for the policy) increased polarization,

even when equivalent cues for Republican politicians and liberal

and conservative interest groups did not. All treatments that

provided information on the auto industry bailout experiment

increased party polarization, whether or not partisans presented

the information.

This combination of results suggests that both partisanship

and policy cues play a role in public opinion on policy issues.

Issues differ based on their initial level of polarization and

the extent to which they are associated with a particular

partisan politician. The effectiveness of cues differs based on

the political context, even when Democratic and Republican

officials are supporting the same policies or arguing over who

should get credit for the same policy success. Some policies are

highly associated with a particular politician regardless of

The Variable Effect of Policy Cues on Partisan Opinions 35

frame. Others can be framed as an idea of a particular leader,

stimulating associated polarization even if proposals made by

other partisan politicians and interest groups are not

polarizing.

Together, these results suggest that partisan polarization

is neither an inevitable consequence of partisan cues nor an

intrinsic characteristic of policy debates. Public opinion on

policy issues can be shaped by information about the proponents

or beneficiaries of policy, evidence about policy effectiveness,

and the identities of those making the arguments. Yet no response

is inevitable; the public may not initially divide into two

opposing partisan groups, polarize in response to partisan

proponents of policy, or depolarize when presented with policy

information. No one mechanism is central to how policy cues

affect levels of citizen support for policy proposals or to

polarization in attitudes between Democrats and Republicans.

One previous survey experiment that directly compared the

effects of policy frames and partisan sponsors on policy opinion

and polarization found that partisanship consistently conditions

the effects of policy frames (Slothuus and de Vreese 2010). Based

The Variable Effect of Policy Cues on Partisan Opinions 36

on only two issue areas, the authors posited a general difference

in the degree of partisan response by categorizing them as

conflict or consensus issues. The research conducted here

challenges the notion that any one or two types of issue

categories should govern partisan response to policy cues. I

found three different patterns of partisan response in three

issue areas that all featured some conflict but also bipartisan

consensus among a set of prominent proponents. The results were

also inconsistent across partisan sponsors within the same

experiment and issue area, with President Obama stimulating more

polarization than other partisan actors.

Survey experiments that take account of political context

can thus add nuance to debates about public response to source

cues and policy frames. This investigation manipulated the

sources of arguments, the target populations of policy, and the

identity of those claiming credit for policy success, finding

three different patterns of response. All three were realistic

portrayals of the arguments made in state policy debates, taking

advantage of fortuitous circumstances in Michigan that enabled

investigation of policies supported by both Democrats and

The Variable Effect of Policy Cues on Partisan Opinions 37

Republicans. Partisanship played an important role in public

opinion in all three issue areas, but did not have the same

consistent effects, either in the baseline condition or in

response to policy cues. Rather than expecting policy frames or

source cues to have uniform effects, scholars can take advantage

of real variation in the proponents, credit claimers, and

potential beneficiaries of policy proposals to investigate how

public opinion responds to information and partisan cues. The

potential responses may be more variable than scholars initially

supposed.

If responses by U.S. citizens to policy and partisan cues

are this variable, there is unlikely to be a generalizable

pattern of public response to policy debates across countries. In

addition to variation in partisan identification and political

information in industrialized nations, there is also important

variation across policy issue areas; details of policy debates

and specific partisan proponents affect how citizens respond to

information that is presented to them by political elites. Since

three policy debates in one U.S. state produced three distinct

patterns of responsiveness, scholars should expect to continually

The Variable Effect of Policy Cues on Partisan Opinions 38

modify theories of policy frames and partisan polarization when

applying them to specific policy contexts rather than developing

one theory to explain how citizens react to policy information or

partisan division across nations, issues, or time periods.

All this means that research on policy cues and polarized

responses to policy issues may not be treated as evidence for

generalized patterns. Researchers cannot yet say whether these

three experiments will prove exceptions to a rule or illustrative

of regular differences across policy areas. Most prior research

assesses responses to information or policy proponents in only

one issue area and context; we should not assume that any of

these particular responses are universal. The research presented

here tested partisan responses to three different issue areas: in

all cases, partisanship made an important difference in how

citizens responded to policy information and choices but no case

looked alike. Researchers will need to assess the importance of

specific polarizing figures, target populations, and the

presentation of evidence in other issue areas and political

contexts to see if the responses here are common. These results

only show that all three patterns are possible and none are

The Variable Effect of Policy Cues on Partisan Opinions 39

universal. Nonetheless, the lesson is a useful one: the effects

of policy cues on partisan polarization are quite variable and

unlikely to be reduced to one or two variations on a theme.

Citizens often view policy through partisan lenses. Both

aspects of policy argumentation and the actors making the

arguments can enable partisanship to become the primary

determinant of public opinion. Yet this process is highly

conditional, sometimes specific to a single polarizing politician

and other times likely in the face of any policy argumentation or

evidence. Whether scholars manipulate arguments directly or wait

for variation in political debate, public opinion on policy

issues responds to both political and policy context.

Bibliography

Barabas, Jason and Jennifer Jerit. 2010. “Are Survey Experiments

Externally Valid?” American Political Science Review 104 (2): 226-

242.

Branton, Regina P. 2003. “Examining Individual-Level Voting

Behavior on State Ballot Propositions.” Political Research

Quarterly 56 (3): 367-377.

Bullock, John. 2011. “Elite Influence on Public Opinion in an

Informed Electorate.” American Political Science Review 105 (3):

496-515.

Delli Carpini, Michael X. and Scott Keeter. 1997. What Americans

Know About Politics and Why It Matters. New Haven: Yale University

Press.

Druckman, James N. 2001. “On the Limits of Framing Effects: Who

Can Frame?” Journal of Politics 63 (4): 1041-1066.

Erickson, Robert S., Gerald C. Wright, and John P. McIver. 1994.

Statehouse Democracy: Public Opinion and Policy in the American States.

Cambridge: Cambridge University Press.

The Variable Effect of Policy Cues on Partisan Opinions 1

Gaines, Brian J., James H. Kuklinski, and Paul J. Quirk. 2007.

“The Logic of the Survey Experiment Reexamined.” Political

Analysis 15(1): 1-20.

Garner, Andrew and Harvey Palmer. 2011. “Polarization and Issue

Consistency Over Time.” Political Behavior 33 (2): 225-246.

Goren, Paul, Christopher M. Federico, Miki Caul Kittilson. 2009.

“Source Cues, Partisan Identities, and Political Value

Expression.” American Journal of Political Science 53 (4): 805–820.

Green, Donald, Bradley Palmquist, and Eric Schickler. 2002.

Partisan Hearts and Minds: Political Parties and the Social Identities of Voters.

New Haven: Yale University Press.

Haider-Markel, Donald P. and Mark R. Joslyn. 2003. “Gun Policy,

Opinion, Tragedy, and Blame Attribution: The Conditional

Influence of Issue Frames.” Journal of Politics 63 (2): 520-543.

Hetherington, Marc J. “Review Article: Putting Polarization in

Perspective.” British Journal of Political Science 39 (2): 413-448.

Hibbing, John R. and Elizabeth Theiss-Morse. 2002. Stealth

Democracy: Americans’ Beliefs About How Government Should Work.

Cambridge: Cambridge University Press.

The Variable Effect of Policy Cues on Partisan Opinions 2

Jacobs, Lawrence and Suzanne Mettler. 2011. “Why Public Opinion

Changes: The Implications for Health and Health Policy.”

Journal of Health Politics, Policy and Law 36 (6): 917-933.

Jerit, Jennifer. 2009. “How Predictive Appeals Affect Policy

Opinions.” American Journal of Political Science 53 (2): 411-426.

Lau, Richard R. and Mark Schlesinger. 2005. “Policy Frames,

Metaphorical Reasoning, and Support for Public Policies.”

Political Psychology 26 (1): 77-114.

Mondak, Jeffery J. 1993. “Public Opinion and Heuristic Processing

of Source Cues.” Political Behavior 15 (2): 167-192.

Nelson, Thomas E. 2004. “Policy Goals, Public Rhetoric, and

Political Attitudes.” Journal of Politics 66 (2): 581-605.

Nelson, Thomas E. and Donald R. Kinder. 1996. “Issue Frames and

Group-Centrism in American Public Opinion.” Journal of Politics 58

(4): 1055-1078.

Rahn, Wendy M. 1993. “The Role of Partisan Stereotypes in

Information Processing about Political Candidates.” American

Journal of Political Science 37 (2): 472-496.

The Variable Effect of Policy Cues on Partisan Opinions 3

Schneider, Anne and Helen Ingram. 1993. “Social Construction of

Target Populations: Implications for Politics and Policy.”

American Political Science Review 87 (2): 334-347.

Slothuus, Rune and Claes H. de Vreese. 2010. “Political Parties,

Motivated Reasoning, and Issue Framing Effects.” Journal of

Politics 72 (3): 630-645.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge:

Cambridge University Press.

The Variable Effect of Policy Cues on Partisan Opinions 4

Figure 1: Support for Medicaid Expansion by Treatment and

Partisanship

All

Democrats

Republicans

0

0.2

0.4

0.6

0.8

ControlFor PoorFor Elderly

Support for Medicaid expansion is coded as a 5-category strongly favor to strongly oppose scale with equidistant positions between 0 and 1.

The Variable Effect of Policy Cues on Partisan Opinions 5

Figure 2: Predicted Support for Medicaid by Treatment and

Partisanship

Stron...

Neither

Stro...

00.10.20.30.40.50.60.70.80.9

Control

For Poor

For Elderly

Support for gas tax increases for road construction is coded as a 5-category stronglyfavor to strongly oppose scale with equidistant positions between 0 and 1. Party identification is coded on a 7-point scale with equidistant positions. All other variables from Table 2 are held at their means.

The Variable Effect of Policy Cues on Partisan Opinions 6

Figure 3: Support for Gasoline Tax Increases by Treatment and

Partisanship

All

Democrats

Republicans

0

0.2

0.4

0.6

ControlObama SupportsSnyder SupportsBusiness SupportsUnions Support

Support for gas tax increases for road construction is coded as a 5-category strongly favor to strongly oppose scale with equidistant positions between 0 and 1.

The Variable Effect of Policy Cues on Partisan Opinions 7

Figure 4: Predicted Support for Gasoline Tax Increases by

Treatment and Partisanship

Strong

Republican

Neither

Strong

Democrat

00.10.20.30.40.50.60.7

ControlObama SupportSnyder Support

Support for gas tax increases for road construction is coded as a 5-category strongly favor to strongly oppose scale with equidistant positions between 0 and 1. Party identification is coded on a seven-point scale with equidistant positions. All other variables from Table 2 are held at their means.

The Variable Effect of Policy Cues on Partisan Opinions 8

Figure 5: Approval of Auto Industry Bailout by Treatment and

Partisanship

All Democrats Republicans0

0.10.20.30.40.50.60.70.80.9

ControlEvidenceDemocratic EvidenceRepublican Evidence

Approval of the auto industry bailout is coded as a 5-category strongly approve to strongly disapprove scale with equidistant positions between 0 and 1.

The Variable Effect of Policy Cues on Partisan Opinions 9

Figure 6: Predicted Approval of Auto Industry Bailout by

Treatment and Partisanship

Strong...

Neither

Stron...

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

ControlDemocratic Evidence

Approval of the auto industry bailout is coded as a 5-category strongly approve to strongly disapprove scale with equidistant positions between 0 and 1. Party identification is coded on a seven-point scale with equidistant positions. All other variables from Table 3 are held at their means.

The Variable Effect of Policy Cues on Partisan Opinions 10

Table 1: Models of Support for Medicaid Expansion

The table reports ordinary least squares regression coefficients with standard errors in parentheses. Support for Medicaid expansion is coded as a 5-category strongly favor to strongly oppose scale with equidistant positions between 0 and 1. The first model is the complete sample, with each treatment compared against the control groups; each of the other models refers to a different experimental treatment. *<.05 **<.01 ***<.001

CompleteSample

ControlGroup

HealthCare forPoor

Care forElderly

Treatment:Basic Health Care for Poor

.052*(.025)

Treatment:Nursing Home Care for the Elderly

.087**(.029)

Party Identification (Democratic +)

.052***(.006)

.058***(.012)

.053***(.008)

.038**(.012)

Ideology(Liberalism +) .026***

(.007).028*(.013)

.024**(.009)

.030*(.013)

Political Knowledge - .019

(.014).021

(.027)- .041*(.019)

- .029(.028)

Constant .318 .267 .403 .464Adjusted R2 .17 .18 .19 .12N 907 237 442 211

The Variable Effect of Policy Cues on Partisan Opinions 11

Table 2: Models of Support for Gasoline Tax Increases for RoadMaintenance

The Variable Effect of Policy Cues on Partisan Opinions 12

The table reports ordinary least squares regression coefficients with standard errors

CompleteModel

ControlGroup

ObamaSupports

SnyderSupports

Treatment:Obama Supports .054

(.037)

Treatment:Snyder Supports - .003

(.037)

Treatment:Businesses Support

.004(.035)

Treatment:Unions Support - .030

(.037)

Party Identification (Democratic +)

.011(.007)

- .003(.017)

.069***(.016)

.004(.018)

Ideology(Liberalism +) .018*

(.007).022

(.016).013(.015)

.025(.019)

Political Knowledge .054***

(.015).037

(.035).006(.032)

.032(.039)

Union Favorability .063

(.038).022

(.086).081(.086)

- .015(.094)

Business Favorability - .004

(.040).078

(.091).153(.084)

- .080(.099)

Constant .176 .244 - .031 .296Adjusted R2 .05 .00 .22 .00N 887 162 166 173

The Variable Effect of Policy Cues on Partisan Opinions 13

in parentheses. Support for gas tax increases for road construction is coded as a 5-category strongly favor to strongly oppose scale with equidistant positions between 0and 1. The first model is the complete sample, with each treatment compared againstthe control groups; each of the other models refers to a different experimental treatment. *<.05 **<.01 ***<.001

The Variable Effect of Policy Cues on Partisan Opinions 14

Table 3: Models of Approval of Auto Industry Bailout

The table reports ordinary least squares regression coefficients with standard errors in parentheses. Approval of the auto industry bailout is coded as a 5-category

CompleteModel

ControlGroup

Democratic

Evidence

Republican

EvidenceTreatment:Neutral Evidence

.004(.012)

Treatment:Democratic Evidence

.046(.012)

Treatment:Republican Evidence

- .005(.012)

Party Identification (Democratic +)

.051***(.012)

- .002(.012)

.068***(.012)

.062***(.012)

Ideology(Liberalism +) .033***

(.012).049***(.012)

.034**(.012)

.033*(.012)

Political Knowledge .044**

(.012)- .010(.012)

.060*(.012)

.033(.012)

Union Favorability .220***

(.012).310***(.012)

.175**(.012)

.128*(.012)

Business Favorability .117***

(.012).127

(.012).188**(.012)

.113(.012)

Constant .019 .225 - .096 .085Adjusted R2 .30 .20 .35 .30N 887 209 239 219

The Variable Effect of Policy Cues on Partisan Opinions 15

strongly approve to strongly disapprove scale with equidistant positions between 0 and 1. The first model is the complete sample, with each treatment compared against the control groups; each of the other models refers to a different experimental treatment. *<.05 **<.01 ***<.001

1 All replication data is available online via the Institute for

Public Policy and Social Research at:

<http://ippsr.msu.edu/soss/sossdata.htm>. Accessed 9 July 2013.

2 These were initially two separate conditions but I observed no

significant difference across the two experimental treatments.

Separate analysis of respondents in each experimental treatment

revealed no significant difference in the determinants of their

opinions.

3 There was an interaction effect (discussed below) for political

information and partisanship in the control condition of the auto

bailout experiment.

4 I use ordinary least squares regression for ease of interpretation.

The response categories for all three experiments could also be

considered an ordinal scale. I ran the same models using ordered

logit and observed no major differences in the size of effects or the

patterns of statistical significance. There was little evidence of

departures from linearity across the response categories.

5 An additional model showed an interactive effect of political

information and partisanship in the control condition. Among those

who already possessed high levels of political information, Democrats

were substantially more supportive than Republicans. This finding is

consistent with the effects of the experiment: partisans who obtain

more information are more polarized.