Week 6 - Assignment: Examine the Impacts of Policies Implemented During the Great Recession and Week 7 - Assignment: Measure the Effects of Social Policies

Positive Returns and Equilibrium: Simultaneous

Feedback Between Public Opinion and Social Policy

Nate Breznau

This paper pushes forward political research from across disciplines seeking to understand the

linkages between public opinion and social policy in democracies. It considers the thermostatic and

the increasing returns perspectives as pointing toward a potentially stable set of effects running

between opinion and policy. Both theoretical perspectives argue that opinion and policy are

reciprocally causal, feeding back on one another. This is a general argument found in opinion-policy

literatures. However, much empirical research claims to model “feedback” effects when actually using

separate unidirectional models of opinion and policy. Only a small body of research addresses

opinion-policy endogeneity directly. In this paper I consider an opinion-policy system with

simultaneous feedback and without lags. I argue that there is a theoretical equilibrium in the

relationship of opinion and policy underlying the otherwise cyclical processes that link them. Given

that available cross-national data are cross-sectional and provide limited degrees of freedom, an ideal

theoretical model must be somewhat constrained in order to arrive at empirically meaningful results.

In this challenging and exploratory undertaking I hope to open up the possibility of a general system

of effects between public opinion and social policy and how to model them in future research. I focus

on social welfare policy as it is highly salient to public interests and a costly area of government

budgets, making it an area of contentious policymaking. Social policy is also a major part of the

thermostatic model of opinion and policy, which was recently extended to the cross-national

comparative context (Wlezien & Soroka, 2012) providing a critical predecessor to this paper because

identification of equilibrium between public opinion and social policy in any given society is greatly

enhanced through comparison with other societies. This counterfactual approach helps to identify

opinion-policy patterns that may not change much within societies, but can be seen as taking on

discrete trajectories between societies.

KEY WORDS: public opinion, social policy, welfare states, simultaneous feedback

本文从跨学科视角推动政治研究以力求理解民主体制下公众舆论和社会政策的关系。本文

从舆论和政策之间可能存在的一系列稳定作用的角度重新审视了温度调节器和递增反馈这两个

观点。这两个理论视角都认为舆论和政策互为因果, 相互反馈。这是舆论-政策的相关文献的普

遍观点。然而, 大多数号称建立了“反馈”作用模型的实证研究实质上仅仅对舆论和政策分别作

了一个单向的模型。只有一小部分研究直接处理了舆论-政策的内生性问题。在本文中, 我考虑

建立一个同步反馈没有延时的舆论-政策系统。我提出在舆论和政策循环往复的过程中潜藏着一

个理论上的均衡将他们联系起来。由于现有的跨国数据为自由度有限的截面数据, 一个理想的理

论模型为得到有效的实证结果不得不在一定程度上受限。通过这项充满探索和挑战的工作, 我希

583

doi: 10.1111/psj.12171

VC 2016 Policy Studies Organization

The Policy Studies Journal, Vol. 45, No. 4, 2017

bs_bs_banner

望可以为以后的研究提供一种对民意和社会政策的总体作用进行分析建模的可能性。我集中于

社会政策是因为它既与公众利益高度相关, 而对于政府而言又是一个高投入的领域。这使得它

成为一个非常具有争议性的政策制定领域。此外, 社会政策也是舆论和政策“温度调节器模型”

的一个主要应用领域。这一模型最近被扩展到跨国比较中（Wlezien & Soroka, 2012）。这一

应用是本文的一个重要先驱, 因其通过与其他社会的比较提高了公众舆论和社会政策均衡在任

一社会中的识别。这一反事实研究路径帮助我们来认识那些在一个社会内部可能保持不变而在

不同社会间有着不同发展轨迹的民意-政策模式。

Public Opinion and Social Policy: What Relationship in Democracies?

The thermostatic model of public preferences by Wlezien (1995) and its later theo-

retical developments discussing feedback running between opinion and policy

(Soroka & Wlezien, 2010), follows in a tradition of research suggesting that the public

are responsive to spending in the form of proposed or actual changes (Durr, 1993;

Wlezien & Soroka, 2007), and policymakers and policymaking processes are respon-

sive to public demands both real and imagined (Converse, 1987; Hill & Hinton-

Andersson, 1995; Page & Shapiro, 1983), both of which should move opinion and

policy toward congruence (see also Campbell, 2012). In the responsive democratic

polity, the political denizen1 has allegiance to the system as well as opportunities to

influence system outcomes, for example through enfranchisement, minority veto,

and regular rotation of offices (Lipset, 1960). Policies should move toward the prefer-

ences of the median or average voter (Downs, 1957). If a policy moves away from

the median, there will be more members of the public dissatisfied with it leading to

an eventual change in the policy back toward the median preference, in this way the

public acts as a thermostat partially regulating policy spending (Wlezien, 1995). The

same applies when spending moves toward the median-voter because public prefer-

ences to increase policy will be allayed and thus reduced. In both cases the public

moves negatively against the direction of a spending change. However, this negative

response is a product of public opinion relative to policy and policy changes, and

does not necessarily reflect a change in opinion as to what the absolute or ideal level

of overall policy should be. As I will show in the next section, formal equations are

available to express the thermostatic model.

Not necessarily incompatible with the thermostatic model, the increasing returns

model of Pierson (1998, 2000) suggests that legally defined policy entitlements gener-

ate positive public opinion responses. The materials and services given to individu-

als via welfare state policies lead to public reliance on the state and public opinion in

support of these returns from the state. The public tend to enjoy whatever level of

benefits they receive and tend to demand more over time, leading political actors to

want to offer more as a way to gain power. The nature of returns and public support

are linked to particular historical “happenstance” configurations of social policies

and the institutions that sprang up around them; some examples being Bismarck’s

worker protections to ally them with the state against socialism in Germany (Lidtke,

1966), the initially racially coded and controversial “Old-Age Retirement, Survivors

and Disability Insurance” in the United States (Skocpol, 1988) or care for Civil War

584 Policy Studies Journal, 45:4

veterans also in the United States (Skocpol, 1992), and subsidies for agricultural sales

in Sweden linking to large cross-class coalitions (Weir & Skocpol, 1985). Recipients,

workers in agencies of government, and individuals along economies of scale associ-

ated with providing state goods and services have a material self-interest in main-

taining and expanding given policies. Therefore, they tend to form or support

special interest groups, and this mobilizes public opinion through informal social

interaction, media, and party politics.

The increasing returns model is an institutionalist account of opinion-policy

where each reinforces the other over periods longer than annual cycles. Borrowed

from economics, the idea of increasing returns explains political behaviors as struc-

tured by incentives and especially disincentives generated by historical institutional

trajectories (e.g., North, 1990). Once political institutions are in place it is inefficient

to change them and more rational to adjust individual or group political strategies to

work within them (Pierson, 1993). The values, ceremonies, norms (informal rules),

laws (formal rules), myths, and habits that are built into social policy institutions

play key roles in shaping political behaviors and collective problem solving. These

embedded features come to define what constitutes “rational” policymaking behav-

iors or responses to collective problems, including attitude formation and heuristic

shortcuts (Homans, 1974; Jepperson, 1991; Ostrom, 2011; Weber, 1946, p. 229). The

process of institutionalization means that what was before comes in time to be seen

as what ought to be (Hall & Taylor, 1996; Meyer & Rowan, 1977; Thelen, 1999), an

argument found in the institutional analysis and development framework and ratio-

nal choice theories (see review in Nowlin, 2011). The results are national systems

that expand along discrete institutional trajectories; what policy scholars might call

“worlds of welfare” (Esping-Andersen, 1985), and institutional theorists might call

“institutional isomorphism” (DiMaggio & Powell, 1983). In short, increasing returns

means path dependency (Ebbinghaus, 2005a; Lijphart, 1999).

In the increasing returns perspective policy moves positively in response to public

opinion similarly to the thermostatic model, because policymakers and parties pick

up on public shifts in “mood” and then make corrections to policies (Stimson, Mack-

uen, & Erikson, 1995). Policymakers tend to alter their policymaking behaviors to

reflect what they perceive to be public preferences (a product of historical experien-

ces and expectations), and these perceptions reflect actual public preferences—with

some room for error due to asymmetric information (Butler & Nickerson, 2011; Jones

& Baumgartner, 2004). The thermostatic model points toward cyclical processes usu-

ally considered as happening over budgetary or election cycles. In this perspective

there is an opinion-policy relationship where policy outcomes are constantly pulled

toward opinion leading to a system of equilibrium, other policy shocks notwith-

standing; similar to a punctuated equilibrium linkage where the impact of opinion

on policy is stronger closer to more intense policymaking periods, such as elections

or budgetary deadlines (Jones & Baumgartner, 2012). Thus, the thermostatic model

likely feeds into increasing returns trajectories by helping the system stay on the

path. Unlike the thermostatic model, the increasing returns model has no formal

equations. Quantitative models have been used only rarely to test for increasing

returns, probably because measuring institutional embeddedness is a great

Breznau: Positive Returns and Equilibrium 585

challenge. The bulk of the argument is rational-theoretical or based on historical

comparative, qualitative, and descriptive data (Pierson, 1993, 1994, 2000).

Although usually applied to expansions, the mechanism of increasing returns

could also explain declining returns.2 Globalization, economic stagnation, and ageing

demographics may reduce social benefits without policymakers taking any actions of

policy cuts (Hacker, 2004; Pierson, 2002). Also, when policymakers do retrench social

policy the public may not notice, as these cuts are purposely hidden or too complicat-

ed to understand by the public (Korpi & Palme, 2003; Pierson, 1996; Schumacher, Vis,

& van Kersbergen, 2013; Thelen, 2014). Regardless of how it happens, social policy

retrenchment has taken on many forms since the 1980s in most rich democratic wel-

fare states (Starke, 2006). As a result the public should expect less from the state and

expect to compete more in a private marketplace for social insurance and welfare. As

these new rules of the game become embedded, individuals who must compete more

in private may become less supportive of state provided social provisions. Thus, it is

plausible that opinion and policy mutually decrease each other, which means they

retain a positive relationship but move simultaneously in a negative direction. I adopt

the label positive returns to apply Pierson’s theoretical mechanisms to both directions.

The feedback perspectives of the thermostatic and positive returns theses sug-

gest a systems perspective (e.g., Durkheim, 1964; Easton, 1965; Luhmann, 1982).

From this perspective the system follows a trajectory determined by its current state

and the environment, e.g., disasters, media frenzies, crime, and any other external

stimuli to opinion-policy. In the system, policymaking has implications for the entire

system (Anderies & Janssen, 2013), and public opinion is both outcome of and input

into policymaking, and therefore the entire system as well (Mettler & Soss, 2004).

However, without equilibrium there is no system, and all effects between opinion

and policy are event-specific. For example, others have shown how a shock to either

opinion or policy leads to changing, reverberating, and decaying effects over years

around the event (e.g., Jennings, 2009). However, here I consider that shocks are con-

stantly impacting opinion-policy systems and at the core there is a stable relationship

that exists to absorb all of the shocks. This is an argument for a general theory of

opinion-policy applicable across events and context-specificities.

Simultaneous Feedback: From Theory to Model Specification

The thermostatic model may be expressed as two equations (Wlezien & Soroka,

2012, p. 1409–10):

Rt5a1b1Pt1b3Wt1et (1)

DPt5q1y1Rt211y2Gt211ut (2)

Feedback takes place via equation (1) as public responsiveness and equation (2) as

policy representation. In equation (1), relative public preferences Rt for a policy are

equal to the slope b1 (i.e., effect) of policy output Pt (they measure it as spending) plus the slope b3 of absolute public preferences Wt (unobserved but estimated by proxy); plus intercept a and error et. Equation (1) assumes that relative preferences

586 Policy Studies Journal, 45:4

are observed in a survey and represent the difference between existing levels of

spending and absolute preferences for spending. In equation (2), a change in Pt equals the slope y1 of relative public preferences in the previous year Rt21 plus the

slope y2 of partisan control of government in the previous year Gt21 (relative rates of

left and right parties); and intercept q and error ut. The slope of public responsiveness (b1) is always negative because the public

want spending (Pt) to move toward their more stable absolute preferences (Wt) (see

also Roosma, van Oorschot, & Gelissen, 2014; Steele, 2015, for alternative approaches

to relative and absolute preferences). Meanwhile, the slope of policy representation

(y1) is positive because it follows what the public wants. Taken together, spending

(and spending changes) will move public opinion to react negatively immediately

and then in the next year, policy will change positively in response.

These two equations suggest a feedback loop because any change in spending or

opinion triggers subsequent changes in the other. Of course these effects trigger sub-

sequent effects, and the feedback process swings like a pendulum until the effects

caused by any single change in the system return to zero. This process is estimated

by Soroka and Wlezien (2010) to take about 10 years. But given that opinion and pol-

icy are regularly changing in response to economic, demographic, and media persua-

sion factors, the pendulum of effects is in fact perpetually in motion. Reconsidering

the positive returns model here, absolute preferences may change over many years

as policy provisions change. Thus, opinion may feed back on itself through policy

and the system can be at equilibrium while moving in a certain direction, with or

without shocks.

Although Soroka and Wlezien (2010) estimate feedback, they estimate equations

(1) and (2) separately and then combine the results post hoc. This generates a poten-

tial unobserved endogeneity problem because the result of each equation is depen-

dent on the other. No scholar to date explicitly employed endogeneity correcting

models in the case of social welfare policy and specific policy preferences, but there

is blossoming analogous research using various lagged variable models to deal with

autocorrelation (i.e., endogeneity), most often a form of vector autoregression (VAR)

(Freeman, Williams, & Lin, 1989). These models account for different short-term and

long-term effects of opinion and policy on each other. For example, Jennings (2009)

considers a thermostatic model of border control and asylum policy with public

opinion finding that as preferences for less immigrants and asylum seekers increase,

the applications and case numbers processed decreases. Jennings also demonstrates

that shocks to the system in terms of changes in asylum applications have effects

that persist at a decaying rate across years, similar to the predictions about moving

toward equilibrium in the theoretical model of Wlezien and Soroka. Others have

looked at left-right subjective placement and the content of policy speeches, and the

linkages between media messages and public opinion using similar methods (Habel,

2012; Hakhverdian, 2012; Lee, 2014). The lagged variables models are powerful tools

to assess opinion-policy, but their greatest strength lies in identifying opinion-policy

feedback at specific moments before, during, and after policy change.

Perhaps there is a perpetual relationship of opinion and policy, not specific to

any moment. This would mean underlying thermostatic feedback and increasing

Breznau: Positive Returns and Equilibrium 587

returns together captures a stable set of effects between opinion and policy character-

ized by equilibrium. If present, these effects would weather exogenous shocks to the

opinion-policy system. It is difficult if not impossible to capture all shocks to opinion

and social policy stemming from culture, constitutional structure, corporatism, reli-

gion, or other normative forces (Breznau, 2015; Hall & Lamont, 2013; Huber, Ragin,

& Stephens, 1993; Larsen, 2008). Most poignantly, media diffusion and persuasion is

ever jolting public and policymaker responsiveness alike (Miller & Stokes, 1963;

Mullinix, 2011). Such specific mechanisms are very difficult to quantify and when

omitted from an empirical model they potentially confound estimation of the causal

paths between opinion and policy (Knight & Winship, 2013; Pearl, 2009). I argue that

estimating simultaneous effects despite these potential omitted confounders is

acceptable because each context has fixed effects that define the unique features of

sociopolitical relations in each society but should be randomly distributed across

societies, leaving findings representative of the total effects of opinion or policy on

each other. The question is whether opinion and policy operate as a system at equilibrium

despite context-specific effects.

To estimate simultaneous feedback I first theorize in more detail about equilibri-

um, then I use theoretical steps to specify a structural model from the Wlezien and

Soroka equations into a nonrecursive system, and finally I discuss instrumental vari-

ables necessary to identify the effects. All of these stages require interplay of theory

and methods as models cannot be estimated without certain components and these

components cannot be identified without theoretical arguments.

Equilibrium

The reciprocal effects of opinion and policy on each other must be theoretically

causal forces (or total effects of various causal forces) that exist at the same time and

occur perpetually. Extreme shocks or inefficiencies caused by corruption, war, or tran-

sitioning away from a planned economy characterize disequilibrium. Thus, I focus

on rich democracies which did not face extreme political shocks after WWII. There

are small disruptions such as election cycles, media frenzies, new special interest

groups, and economic booms and busts causing opinion-policy shifts (Papadakis,

1992). But the overall political institutions and rule of law persist without spiraling

into revolution or chaos.

As argued in both the thermostatic and positive returns literatures, policymakers

are constantly gauging public opinion and attempting to predict how the public will

respond to their actions (Miller & Stokes, 1963; Stimson et al., 1995). This means that

public opinion has an effect on policymaking that is constant existing in the minds of

policymakers. Even if there are shocks to public opinion, policymakers still expect

the public to react to policymaking practices in certain, predictable ways and adjust

their behaviors accordingly. Policy also has a stable effect on opinion from an institu-

tional perspective. The resources distributed by policies in cash and in kind are slow

to change, thus the public develop normative expectations that guide attitudes and

behaviors. The returns from the state shape public opinion to favor the

588 Policy Studies Journal, 45:4

institutionalized welfare status quo (Pierson, 2000) and the public tend toward

uncertainty avoidance making them likely to oppose changes whose outcomes seem

unclear (Bartels, 1986). As social policies are exceedingly complex, the public are

unlikely to understand what policy changes mean, and be averse to any changes

(Pierson, 1996).

Wlezien and Soroka state that opinion and policy are constantly moving toward

equilibrium. This movement takes place in cycles where effects are feeding back

between opinion and policy. Soroka and Wlezien (2010, p. 179) estimate that a distur-

bance takes about a decade to return to equilibrium. I assume that disturbances to

the opinion-policy relationship are perpetually taking place. Thus, every potential

observation window includes feedback effects that are coming to a close, in the mid-

dle, or just started. Thus, from a longer-term perspective the system is perpetually

moving, but moving in a steady way; for example public opinion over the course of

a year (as opposed to daily, weekly, etc.) smooths out erratic fluctuations, leaving a

more consistent pattern. With policy, periods less than a year yield small changes if

any, because of annual budget-making and the slow process of policy-change (Tepe

& Vanhuysse, 2010). There are budgetary changes within years and these are rele-

vant for policymakers and the public, but are small in comparison to policy viewed

over many years. Except for large-scale legal policy changes, the replacement rates

for things like pensions and unemployment rarely change, and when they do they

sometimes take a long time to implement. Even recent liberalizations and retrench-

ments of social policy after the 1980s have small impacts on current pensioners for

example because cuts fall more onto future cohorts (Thelen, 2014), leading to what

Pierson (1993, 2002) refers to as a “sticky” institutional process.

Model Specification

Figure 1 represents the Wlezien and Soroka thermostatic model drawn from

equations (1) and (2), expressed as a path diagram (Wright, 1934). I deconstruct this

path diagram with the process of equilibrium in mind.

1

1

1

Figure 1. The Thermostatic Model as a Path Diagram

Breznau: Positive Returns and Equilibrium 589

In Figure 1, solid arrows are effects and labeled to match equations (1) and (2),

while dotted lines are correlations. I assume mean-centered variables (no mean struc-

ture). The error terms are some disturbance from the predicted values linked by an

effect size of one. The lagged measure of policy Pt21 is not in equation (2), but is part

of the dependent variable DPt (5Pt2Pt21) as it appears after adding Pt21 to both sides of equation (2).3

There are sources of bias in Figure 1. The first is that previous policy shapes pre-

sent public opinion. Relative preferences change based on how policy presently dif-

fers, or not, from before. Thus, Pt21 should have an arrow toward Rt. Also, Pt21 likely shapes Wt because previous policy has an institutional or normative impact on

what individuals come to expect and prefer. Second, partisan control of government

Gt21 is likely correlated with Wt, because a lagged measure of Wt at the time that

seats were distributed in the last election should have shaped public voting behav-

iors at that time. As absolute preferences are stable over short periods of time, any

previous version of W will then be highly correlated with its later self. Therefore,

any measure of G is not independent of Rt because W is a general cause of R. This

argues for a current measure of G as opposed to a lagged measure.

A second bias is the impact of current opinion on current policy. For example,

public opinion may motivate executive orders that cause immediate spending or

even policy changes, it may influence current judicial decisions, and might proxy the

level of need for welfare among the public which in turn predicts the level of spend-

ing that year. The former cases are more exceptional, but the latter conjecture sug-

gests that factors such as unemployment, labor market vitality, public health, old age

quality of life, and other social welfare needs in a society cause opinion change and

simultaneously cause policy (spending) change. Need is a strong predictor of sup-

port for social welfare policies, and of course need also reflects spending due to take-

up of social benefits (Andreß & Heien, 2001; Gelissen, 2000; Meltzer & Richard,

1981). As all sources of need cannot be measured an empirical space exists for opin-

ion to have a measurable impact on policy, as indirect through needs. Moreover, as

the experience of needing welfare can be a profound life course shaping event it

should shape absolute preferences which directly determine relative preferences Rt.

So, causal mechanisms aside, taking up a policy leads to a change in attitudes and a

change in spending on that policy, and theoretically leads to a measurable effect of

opinion on policy in the same year.

In Figure 2, thicker grey lines represent biasing paths that should be present in a

theoretical model of opinion and policy. Although biasing paths can be controlled

for with independent variables, the feedback loop now present between Rt and Pt creates endogeneity that undermines estimates in independent equations (1) and (2).

The models in Figures 1 and 2 cannot be estimated using comparative survey

data because of the lagged opinion effect Rt21. Opinion is not measured yearly in

any comparative survey covering a large number of countries. Instead, opinion is

measured in intervals of 2–6 years depending on the survey and country. However,

I question the utility of using only one year lagged variables. Recent research sug-

gests that the impact of opinion and policy on each other changes depending on the

length of a lag (Jennings, 2009). Therefore, I would potentially need a decade of lags,

590 Policy Studies Journal, 45:4

measured yearly to capture an adequate array of effects. Moreover, as I, and Wlezien

and Soroka argue: the relationship between opinion and policy perpetually moves

toward equilibrium, and effects are constantly starting, progressing, and coming to a

close. This constant movement of opinion and policy toward each other suggests

that regardless of when they are observed they should be found in perpetual motion

on a shared path. This renders the lagged variables unnecessary if not biasing for the

purpose of estimating simultaneous effects at equilibrium. Assuming this movement

is similar across time and space is the key proposition of the general simultaneous

feedback theory I advance here, as argued in the Equilibrium section. Relying on this

theory of equilibrium I drop lagged effects.

In equation (1), a measure of policy is necessary to estimate R rendering endoge-

neity into the estimate (as shown by the feedback loop in Figure 2). Practically, W is

not observed in questions about the government providing more or less on a given

policy. Thus, the question arises how to estimate W and R. There is a solution in

questions regarding the role of government. Such questions found in several cross-

national surveys probe the degree of agreement that respondents express regarding

the government’s role to provide various social policies. Implicitly these questions

contain both W and R as they are correlates of both the preferred role (i.e., absolute

ideal-level of provisions) and the perceived performance (i.e., relative preference for

provisions) (Roosma et al., 2014). Therefore I estimate O, which is simply a measure

of both W and R simultaneously. These changes plus the addition of X to represent a

set of all potential independent variables that cause both opinion and policy other

than partisan control of government are reflected in my final feedback model in Fig-

ure 3.

For identification this model requires that: (a) there are as many model observa-

tions as unknown statistical parameters (i.e., the counting rule that model degrees of

freedom must be � 0), and (b) at least one more case exists in the observed data than free parameters (Kline, 2011). I am interested only in effects and not mean structures,

thus for (a): Figure 3,4 has 11 parameters or model moments (2 variances and 1

covariance of eO and eP; 4 covariances of O; and 4 covariances P each with IVO, IVP,

Figure 2. Potential Sources of Bias

Breznau: Positive Returns and Equilibrium 591

Gt, X, and 1 covariance between O and P. In Figure 3 there are 11 effects (4 arrows

pointing at Ot and 4 pointing at Pt, and residuals with 2 variances and 1 covariance

between them). In the most basic version of the model there are 11 parameters – 11

effects 5 0 degrees of freedom leaving a just identified model. However, I also con-

sider an alternative model where the intercepts of latent O and P are constrained to

zero. Models with and without the means constrained to zero are identical in that

they capture covariance structures without a theoretical mean or a mean with no

numerical value. The latter model has the benefit of 13 parameters and thus 2

degrees of freedom useful for model comparison (Paxton, Hipp, & Marquat-Pyatt,

2011, p. 6). The addition of more X variables does not change the model degrees of

freedom; however, it increases the number of free parameters.

For (b): The number of free parameters in the model determines the minimum

number of cases needed in the data. Free parameters are those which are freely esti-

mated, not observed or fixed. Thus, with 11 free parameters at least 12 cases are need-

ed in the data to achieve overidentification. While sampling theory suggests 10

observations in the data per free parameter or variable are necessary to have

trustworthy effects when working with population data (Pedhazur, 1997),

country-level data are arguably not a sample from a population, for sure not a

random sample (Ebbinghaus, 2005b). Although this leads to an imperfect statisti-

cal reality, estimations are still possible and when coupled with counterfactual or

historical analysis, researchers routinely rely on less than 10 cases per parameter,

if not using the minimum in deterministic studies (Bollen, Entwisle, & Alderson,

1993). I do not claim to have determined all causes of opinion and policy, that is

why I have the residual correlation eO,eP in Figure 3, this leaves the somewhat

ambiguous goal of getting as many cases as possible per parameter.

Public Opinion Measure. The International Social Survey Program (ISSP) is the only com-

parative public opinion data covering a globally broad sample of rich democratic

countries;5 this was likely the motivation for Wlezien and Soroka’s (2012) cross-

national work. The ISSP Role of the Government and Religion modules each have

two questions on social policy related to income redistribution and employment.6

This makes usage of 70 data points in 19 countries possible. Having only one X

Figure 3. Path Diagram of Simultaneous Opinion-Policy Feedback

592 Policy Studies Journal, 45:4

variable in Figure 3 requires 12 or more cases in the data to have at least 1 or more

degrees of freedom and each additional X variable adds 2 free parameters; thus I set

the maximum at 3 X variables in addition to G which yields 15 free parameters leav-

ing only (19–155) 4 degrees of freedom for maximum likelihood identification. I

would call this a bare minimum.

Policy Measures. The theoretical arguments and empirical evidence underlying the

thermostatic model and the positive returns model diverge with respect to poli-

cy. The thermostatic model is focused on spending and budgeting while the

returns model is about replacement rates and the generosity of other welfare

entitlements—also known as decommodification. These are highly correlated.

However, budgets are also flexible and spending can change without an explicit

policy change. Therefore, I seek to test models of both spending and legal entitle-

ments. I measure spending from the OECD social expenditures variable

(“SOCX”). This measure is the amount of public social welfare spending

expressed as a percentage of GDP (OECD, 2012).7 Mostly it comes from pensions

and health care and then to a lesser degree (un)employment, family policies,

housing, and a few others. Decommodification is from the dataset compiled by

Scruggs (2004) and measures the replacement rates and the generosity of welfare

policies. Decommodification measures more precisely how social policy impacts

individual welfare; however, the variable only extends to 2002 and is not avail-

able for two countries, thus reducing the sample size. Both spending and decom-

modification fluctuate slightly from year to year within countries but vary

widely between countries making them suited for observing comparative longer-

term trends without the need for longitudinal analyses.

Instrumental Variables

When there are two endogenous feedback outcomes, a statistical model is nonre-

cursive and cannot be identified (Kline, 2011, p. 134). There is a vast literature on this

subject, with much technical detail (Bollen, 1989, 2012; Duncan, Haller, & Portes,

1968; Kline, 2011, 2013; Pedhazur, 1997, p. 292). One critical rule is that at least one

instrumental variable must be in the model for each endogenous (dependent) variable. These

instruments must have a causal effect on either opinion or policy while having no

causal effect on the other, i.e., an effect of zero. Reviewing opinion-policy research

reveals two options for instrumental variables.8

Veto points is an instrument for social policy. The constitutional structure

and political institutions of a country create a setting which enables or con-

strains social policymaking. More competitive veto points means more opportu-

nities for blocking legislation, what Lijphart (1999) refers to as the “federal-

unitary dimension” of national political systems (see also Crepaz & Moser,

2004). Thus the electoral and judicial systems of a country have a direct impact

on social policy based on how the policies are formed into law (D€aubler, 2008).

In rich democracies, the institutions of veto have not changed since before

1985.9 Thus, veto points represent stable political institutions that should not

Breznau: Positive Returns and Equilibrium 593

impact public attitudes which are otherwise changing in response to current

political events. One attractive argument in favor of using veto points is that it

has a successful track record in previous research (Matsusaka, 2005; Poterba,

1995); in particular Gabel and Scheve (2007) argue that electoral institutions

(part of how veto points are measured) are exogenous to public opinion.

Female labor force participation (female LFP) should be an instrument for pub-

lic opinion. Employment leads individuals to be less supportive of social wel-

fare, all else equal (Gelissen, 2000), and presumably more female LFP means

more women employed which leads to a reduction in overall support of social

welfare among women for reasons of material security. Employed women or

women who have been or expect to be employed (as is true of all persons) are

less likely to need public or partner-based provisions for their welfare because

they can privately provide their own, and they may be less willing to use their

earned income to fund welfare via taxation; these are classic arguments of mate-

rial self-interest (Andreß & Heien, 2001). As male labor force participation varies

far less than female LFP across rich democracies, this instrument captures some-

thing unique in public opinion. However, there is a clear association with the

rise of female employment and social policy. It appears that the expanding role

of women working in the public sector coincided, if not interacted with, the

political development of the welfare state and thus shaped policy spending

(Huber & Stephens, 2000). This causal argument could rule out female LFP as a

valid instrument; however, this argument applies to early, post-war welfare

state expansions prior to the rise of comparative public opinion surveys in the

1980s.

Sometime during the 1980s, female LFP in the United States and Canada for

example caught up with that of Scandinavian social democracies and today women

work at similarly high rates. In the United States and Canada this is presumably a

result of a highly commodified labor market where working is necessary for survival

(Esping-Andersen, 1990). Incidentally these two countries spend nearly the least on

social welfare of all rich democracies, while Sweden and the Netherlands for exam-

ple have similarly high female LFP rates but spend some of the most on social wel-

fare policy. Thus, nearly opposite types of welfare states in terms of policy have

similarly high female LFP. Moreover, the familial tradition of the Mediterranean

countries leads to relatively low female LFP, but these countries have moderate

spending levels, much higher than the United States but lower than Northern

Europe. Thus, female LFP does not appear to cause social policy in a comparative

context since the 1980s. Moreover, in many countries female LFP continues to rise

while spending decreases.

Nil Policy Feedback Hypotheses

When policies are initiated or reformed, a variety of interests meet to see

them through to their street-level implementations. This process may lead to

what Patashnik and Zelizer (2013) refer to as “incomplete displacement” and

594 Policy Studies Journal, 45:4

“inadequate state capacity” to bring about major and stable policy changes at the

level of individual recipients. This harks to Pierson’s “sticky” conceptualization

of the new politics of welfare where institutionalized policymaking gets in the

way of efforts at lasting policy change (see also Soss & Schram, 2007). Without

effectual material changes resulting from policy change, long term public opinion

should not change as a response. At the same time, these varieties of interests

that make and implement policy change may collectively ignore public opinion

as each potentially represents a different segment of the public, or represents no

public interest. Therefore, it is possible that increasing returns only applied to the

historical rise of welfare states policies, and these mature and institutionalized

polices no longer bring about sustainable opinion change (Raven, Achterberg,

van der Veen, & Yerkes, 2011), or only lead to opinion change in subsets of the

population (Bendz, 2015), or that opinion does not cause policy change and is

instead just manufactured by private interests (Habermas, 1989). Therefore, it is

possible that there is no system-stable effect of policy on opinion and especially

opinion on policy. Given the decisive effect of policy, it may drive the entire sys-

tem and constrain or enable its own future via policymakers despite public opin-

ion (see also Morgan & Campbell, 2011).

Variables and Models

Individual Predictors of Opinion

Much previous research identifies socioeconomic and demographic deter-

minants of public opinion (attitudes, preferences, etc.). For example, women on

average are more supportive of government-provided social services such as

pensions, health care, general (re)distribution, and (un)employment support

across countries (Bean & Papadakis, 1998; Evans, Kelley, & Peoples, 2010;

Hasenfeld & Rafferty, 1989; Taylor-Gooby, 2001). Also, older persons tend to be

more supportive, although age, period, and cohort effects may explain inconsis-

tencies (Jæger, 2006; Svallfors, 1997). These findings point toward material inter-

ests as a key factor in public opinion: aged persons more likely to support

pensions, unemployed more supportive of unemployment benefits, and more

materially secure persons are generally less supportive (Blekesaune & Qua-

dagno, 2003; Breznau, 2010; Breznau & Eger, 2016; Eger, 2010).

Weaker economies encourage public opinion to be more supportive of social pol-

icy also out of need, thus lower levels of GDP per capita and higher rates of unem-

ployment should predict lower support for social policy (Blekesaune, 2007; G€erxhani

& Koster, 2012). Finally, as argued by Soroka and Wlezien, the percentage of the total

seats held by right parties (variable G) as opposed to left or other parties should

reflect less support of social welfare policy (and spending), but also lead to less pro-

vision of welfare through their policymaking agendas (Huber et al., 1993; Jensen,

2011). Thus, partisan control links both to opinion and policy.

Breznau: Positive Returns and Equilibrium 595

Predictors of Policy

In addition to partisan control of government, the distribution of welfare needs

in a given society is the largest predictor of social spending. Where there are more

sick, unemployed, and especially pensioners there are higher levels of social spend-

ing. As pensioners are also at a point in the life course that has higher medical costs,

the percentage of the population over age 64 is a critical predictor variable (Pampel

& Williamson, 1988). Brooks and Manza (2007) provide a lengthy discussion of fac-

tors impacting social spending. In addition to the above, they also discuss the theo-

retical impact of GDP per capita (see also Wilensky, 1975), and growth over time.

Dependent variables for public opinion, spending, and decommodification were already

described in the Model Specification section.

Table 1 summarizes all variable definitions (correlations in online appendix10).

Models

Table 2 summarizes all models. The letters “A&B” denote models with social

spending as the policy measure, and “D&C” for decommodification. I start with only

one independent variable. I tried right party control, but many of the models had

identification problems, so instead I use aged first. For each configuration of varia-

bles, I specify three models. The first allows b1 and b2 to be freely estimated. The sec- ond follows Wlezien and Soroka’s (2012) thermostatic model and constrains b1 to be negative and b2 to be positive similar to their estimates. The third assigns equal, moderate-sized effects for b1 and b2 to represent positive returns.

Results

Table 3 reports the main results from 12 specifications of Models A–D. For each

set of specifications I placed the preferred model in bold, and Figure 4 diagrams

these preferred models.

In Figure 4, gray represents nonsignificant paths11 and numbers are standard-

ized coefficients plus a residual correlation between eO and eP. All four preferred

models support the positive returns perspective in both policy spending and decom-

modification. Preferred Model 6A (N 5 19) has a moderate-sized positive effect of

policy on opinion b1 5 0.15 and a similar effect of opinion on policy b2 5 0.15, and this model fits better than the thermostatic and freely estimated models, and better

than models with more independent variables. Model 12B (N 5 70) has similar

effects as 6A with b1 5 0.15 and b2 5 0.15. In Model 12C and 12D, b1 5 0.16 and 0.15, and b2 5 0.14 and 0.15 respectively.

To select preferred models in Table 3, I looked at a variety of fit statistics. The first

criterion is a chi-square exact fit test that the difference between the observed and

implied moment matrices is zero reported in column H0, where a star indicates that at

p> .05 the model is consistent with the data. As the freely estimated b1 and b2 models are just-identified, they do not have chi-square tests and I must rely on LL, AIC, and

596 Policy Studies Journal, 45:4

BIC. Then I used an equal fit test that the nested constrained model was better fitting

than the larger model with freely estimated b1 and b2, for each set of models employing the same independent variables. I borrow this test from the mean structure models in

Supporting Information Table 6 where a star indicates that at p> .01 the larger model

does not fit better as shown in column H1 in Table 3. Next I looked across the remaining

fit statistics and aimed to find the best combination. Many fit statistics are within just a

few points of each other making it plausible that the preferred model is not obviously

better than another. This suggests that in the case of spending, the thermostatic model

may fit equally well as the positive returns model; however, in all cases the chi-square

and fit statistics pointed in the direction of the positive returns model. In the case of

Table 1. Variable Names and Definitions

Name Type Measurementa Source

Public Opinion Endogenous depen- dent variable

Two-item scale from respondents level of agreement with the responsibility of gov- ernment to provide jobs and reduce income differences.

ISSP Role of Government (I,II, III,IV) and Reli- gion (I&II) modules

Social Spending Endogenous depen- dent variable measuring Social Policy

The amount of spending on social policy provi- sions, mostly pensions, employment, unem- ployment, and health care expressed as a percentage of GDP in the same year.

OECD (2012); also known as "SOCX"

Decommodification Endogenous depen- dent variable measuring Social Policy

An aggregate score of generosity and replace- ment rate levels. Intended to measure how much or how lit- tle an individual must rely on the labor mar- ket or private funds for prosperity.

Scruggs (2004)

Aged Independent variable

Percent of the population over age 64.

OECD Social Indicators Data

Right Independent variable

Percent of national gov- ernment seats held by right parties.

Svennson et al. (2012); Quality of Government Data

Unemp. Independent variable

Percent of the labor force that is unemployed

OECD Social Indicators Data

GDP Independent variable

Gross Domestic Product at PPP.

OECD Social Indicators Data

Female LFP Instrument for Public Opinion

Percent of the total female population in the labor force.

OECD Social Indicators Data

Veto Points Instrument for both Social Policy variables

A scale of institutional measures for the amount of chances a policy has to be vetoed. Based on the work of Lijphart (1999).

Svennson et al. (2012); Quality of Government Data

aAll variables are measured simultaneously at the current year of the endogenous variables.

Breznau: Positive Returns and Equilibrium 597

Table 2. Specifications for Simultaneous Feedback Models

Modela

# Letters Unit-Case Numbers Free Parameters

Independent Variables Constraintsb

1 A & B Country 19; Country-time 70

9 Aged None

2 A & B Country 19; Country-time 70

7 Aged Thermostatic model

3 A & B Country 19; Country-time 70

7 Aged Positive returns

4 A & B Country 19; Country-time 70

11 Right; aged None

5 A & B Country 19; Country-time 70

9 Right; aged Thermostatic model

6 A & B Country 19; Country-time 70

9 Right; aged Positive returns

7 A & B Country 19; Country-time 70

13 Right; aged; unemp.

None

8 A & B Country 19; Country-time 70

11 Right; aged; unemp.

Thermostatic model

9 A & B Country 19; Country-time 70

11 Right; aged; unemp.

Positive returns

10 A & B Country 19; Country-time 70

17 Right; aged; unemp.; GDP

None

11 A & B Country 19; Country-time 70

15 Right; aged; unemp.; GDP

Thermostatic model

12 A & B Country 19; Country-time 70

15 Right; aged; unemp.; GDP

Positive returns

1 C & D Country 17; Country-time 65

9 Aged None

2 C & D Country 17; Country-time 65

7 Aged Thermostatic model

3 C & D Country 17; Country-time 65

7 Aged Positive returns

4 C & D Country 17; Country-time 65

11 Right; aged None

5 C & D Country 17; Country-time 65

9 Right; aged Thermostatic model

6 C & D Country 17; Country-time 65

9 Right; aged Positive returns

7 C & D Country 17; Country-time 65

13 Right; aged; unemp.

None

8 C & D Country 17; Country-time 65

11 Right; aged; unemp.

Thermostatic model

9 C & D Country 17; Country-time 65

11 Right; aged; unemp.

Positive returns

10 C & D Country 17; Country-time 65

13 Right; aged; GDPc

None

11 C & D Country 17; Country-time 65

11 Right; aged; GDPc

Thermostatic model

12 C & D Country 17; Country-time 65

11 Right; aged; GDPc

Positive returns

aAll models have female labor force participation as instrumental variable for opinion and veto points as instrument for policy. For each numbered model, letters A & B have spending and C & D have decommo- dification as the policy measure. See Figure 4. bThermostatic model has fixed metric effect sizes based on Wlezien and Soroka’s (2012) results, and posi- tive returns model has moderate and similar sized positive effects for both b1 and b2. cModels 10, 11, and 12 C & D have unemployment omitted because it is highly endogenous with GDP; also, a second set of models were estimated with the mean of both dependent variables set at zero gener- ating two less free parameters and two more model degrees of freedom; see Supporting Information Tables 5, 6, and 7 in the online appendix https:/sites.google.com/site/nbreznau/publications.

598 Policy Studies Journal, 45:4

T a b le

3 . S im

u lt a n e o u s O p in io n P o li cy

F e e d b a ck

M o d e ls a

F e e d b a ck

E ff e ct s

F it S ta ti st ic s

In d e p e n d e n t v a ri a b le

e ff e ct s

M o d e l

S td .- Y X . C o e ff

R e si d .

C h i- S q u a re

b R -S q u a re

b A g e d

b R ig h t

b U n e m p

b G D P

b F L P

b V e to

# b 1

b 2

co v (e

O ,e P ) V a lu e

d f H

0 H

1 R M S E A

L L

C F I

A IC

B IC

P O

S P

P O

S P

P O

S P

P O

S P

P O

S P

P O

S P

1 A

0 .7 8

2 0 .0 2

2 0 .6 1

0 .0

0 –

– 0 .0 0

2 6 9

1 .0 0

1 5 6

1 6 4

0 .3 6

0 .3 0

0 .0 4

0 .5 8

2 0 .5 6

2 0 .2 6

2 A

2 0 .0 5 *

0 .3 0 * 2 0 .0 6

2 .7

2 *

* 0 .1 4

2 7 0

0 .9 6

1 5 4

1 6 1

0 .5 0

0 .3 6

0 .5 0

0 .4 0

2 0 .5 1

2 0 .2 3

3 A

0 .1 5 *

0 .1 5 * 2 0 .1 6

1 .4

2 *

* 0 .0 0

2 7 0

1 .0 0

1 5 3

1 6 0

0 .5 4

0 .3 8

0 .4 0

0 .4 8

2 0 .5 1

2 0 .2 7

4 A

0 .8 0

0 .0 9

2 0 .8 0

0 .0

0 –

– 0 .0 0

2 6 4

1 .0 0

1 4 9

1 6 0

0 .2 9

0 .4 7

0 .0 3

0 .5 1

2 0 .0 8

2 0 .3 2

2 0 .5 7

2 0 .2 6

5 A

2 0 .0 6 *

0 .2 4 * 2 0 .2 4

3 .6

2 *

* 0 .2 1

2 6 6

0 .9 4

1 4 9

1 5 8

0 .3 8

0 .5 7

0 .5 0

0 .4 0

2 0 .3 6

2 0 .2 7

2 0 .5 6

2 0 .3 2

6 A

0 .1 5 *

0 .1 5 * 2 0 .4 1

1 .8

2 *

* 0 .0 0

2 6 5

1 .0 0

1 4 7

1 5 6

0 .6 2

0 .4 8

0 .3 9

0 .4 7

2 0 .3 0

2 0 .3 1

2 0 .5 7

2 0 .3 4

7 A

0 .9 1

0 .1 6

2 0 .8 7

0 .0

0 –

– 0 .0 0

2 6 3

1 .0 0

1 5 1

1 6 4

0 .2 0

0 .4 6

2 0 .0 3

0 .4 8

2 0 .0 8

2 0 .3 2

2 0 .1 4

2 0 .0 6

2 0 .6 5

2 0 .2 6

8 A

2 0 .0 5 *

0 .3 0 * 2 0 .3 3

4 .6

2 *

* 0 .2 6

2 6 5

0 .9 0

1 5 2

1 6 2

0 .6 4

0 .4 3

0 .4 9

0 .4 1

2 0 .4 0

2 0 .2 9

2 0 .1 2

2 0 .1 1

2 0 .6 4

2 0 .3 5

9 A

0 .1 5 *

0 .1 5 * 2 0 .4 5

2 .8

2 *

* 0 .1 4

2 6 4

0 .9 7

1 5 0

1 6 1

0 .6 4

0 .4 8

0 .3 8

0 .4 8

2 0 .3 4

2 0 .3 2

2 0 .1 4

2 0 .0 7

2 0 .6 8

2 0 .3 7

1 0 A

0 .5 6

0 .2 8

2 0 .8 0

0 .0

0 –

– 0 .0 0

2 6 1

1 .0 0

1 5 1

1 6 6

0 .5 8

0 .4 7

0 .1 7

0 .4 2

2 0 .2 6

2 0 .2 4

2 0 .2 6

0 .0 2

2 0 .3 8

0 .2 3

2 0 .5 4

2 0 .2 9

1 1 A

2 0 .0 5 *

0 .3 1 * 2 0 .2 3

2 .6

2 *

* 0 .1 3

2 6 2

0 .9 8

1 5 0

1 6 2

0 .7 4

0 .4 6

0 .4 9

0 .4 1

2 0 .4 6

2 0 .2 3

2 0 .2 6

0 .0 2

2 0 .3 7

0 .2 6

2 0 .5 4

2 0 .3 5

1 2 A

0 .1 5 *

0 .1 5 * 2 0 .3 9

1 .4

2 *

* 0 .0 0

2 6 1

1 .0 0

1 4 9

1 6 1

0 .7 3

0 .4 9

0 .3 9

0 .4 8

2 0 .4 1

2 0 .2 8

2 0 .2 7

0 .0 3

2 0 .3 7

0 .1 8

2 0 .5 7

2 0 .3 6

IB 0 .8 4 †

0 .0 5

2 0 .6 5

0 .0

0 –

– 0 .0 0

2 2 6 8

1 .0 0

5 5 9

5 8 4

0 .1 7

0 .3 4

2 0 .0 5

0 .5 3

2 0 .5 3

2 0 .2 2

2 B

2 0 .0 5 *

0 .3 1 * 2 0 .0 8

7 .2

2 0 .1 9

2 2 7 2

0 .9 2

5 6 2

5 8 2

0 .4 1

0 .3 4

0 .4 3

0 .4 1

2 0 .5 1

2 0 .2 1

3 B

0 .1 5 *

0 .1 5 * 2 0 .1 5

3 .6

2 *

* 0 .1 1

2 2 7 0

0 .9 8

5 5 8

5 7 9

0 .4 5

0 .3 5

0 .3 2

0 .4 8

2 0 .5 1

2 0 .2 4

4 B

0 .8 1

2 0 .1 3

2 0 .5 0

0 .0

0 –

– 0 .0 0

2 2 6 1

1 .0 0

5 4 9

5 7 8

0 .1 6

0 .3 9

2 0 .0 5

0 .5 1

2 0 .0 6

2 0 .2 1

2 0 .5 4

2 0 .2 2

5 B

2 0 .0 5 *

0 .3 1 * 2 0 .1 6

7 .9

2 0 .2 1

2 2 6 5

0 .9 3

5 5 3

5 7 8

0 .4 8

0 .3 7

0 .4 4

0 .4 2

2 0 .2 6

2 0 .1 7

2 0 .5 5

2 0 .2 4

6 B

0 .1 5 *

0 .1 5 * 2 0 .2 4

4 .2

2 *

* 0 .1 3

2 2 6 4

0 .9 7

5 4 9

5 7 4

0 .5 0

0 .4 0

0 .3 3

0 .4 9

2 0 .2 2

2 0 .2 0

2 0 .5 5

2 0 .2 6

7 B

1 .0 3 †

2 0 .0 0

2 0 .7 4

0 .0

0 –

– 0 .0 0

2 2 6 4

1 .0 0

5 5 1

5 8 4

0 .0 0

0 .4 0

2 0 .1 5

0 .5 6

2 0 .0 5

2 0 .2 3

2 0 .2 3

0 .1 1

2 0 .6 8

2 0 .2 3

8 B

2 0 .0 5 *

0 .3 1 * 2 0 .1 6

1 0 .1

2 0 .2 4

2 2 6 5

0 .9 0

5 5 7

5 8 6

0 .4 8

0 .3 7

0 .4 4

0 .4 2

2 0 .2 7

2 0 .1 7

2 0 .0 3

0 .0 1

2 0 .5 7

2 0 .2 4

9 B

0 .1 5 *

0 .1 5 * 2 0 .2 3

5 .8

2 *

* 0 .1 7

2 2 6 3

0 .9 5

5 5 2

5 8 2

0 .5 0

0 .4 0

0 .3 3

0 .4 9

2 0 .2 3

2 0 .2 0

2 0 .0 7

0 .0 6

2 0 .5 9

2 0 .2 6

1 0 B

0 .7 1 †

0 .0 9

2 0 .6 5

0 .0

0 –

– 0 .0 0

2 2 5 8

1 .0 0

5 5 0

5 8 8

0 .3 5

0 .4 1

0 .0 5

0 .5 1

2 0 .1 3

2 0 .1 9

2 0 .2 6

0 .1 5

2 0 .2 9

0 .1 3

2 0 .5 4

2 0 .2 4

1 1 B

2 0 .0 5 *

0 .3 1 * 2 0 .1 1

5 .4

2 *

* 0 .1 6

2 2 6 1

0 .9 6

5 5 1

5 8 5

0 .5 3

0 .3 9

0 .4 7

0 .4 0

2 0 .2 9

2 0 .1 4

2 0 .1 3

0 .1 3

2 0 .2 8

0 .2 3

2 0 .4 7

2 0 .2 5

1 2 B

0 .1 5 *

0 .1 5 * 2 0 .2 0

2 .7

2 *

* 0 .0 7

2 2 5 9

0 .9 9

5 4 8

5 8 2

0 .5 6

0 .4 1

0 .3 6

0 .4 7

2 0 .2 5

2 0 .1 8

2 0 .1 7

0 .1 4

2 0 .2 8

0 .1 6

2 0 .4 9

2 0 .2 6

1 C

0 .5 0

2 0 .1 2

0 .1 4

0 .0

0 –

– 0 .0 0

2 5 7

1 .0 0

1 3 2

1 4 0

0 .7 8

0 .2 4

0 .2 3

0 .5 1

2 0 .6 5

2 0 .4 1

2 C

2 0 .0 6 *

0 .2 7 *

0 .6 3

3 .4

2 *

* 0 .2 1

2 5 9

0 .9 4

1 3 2

1 3 7

0 .5 5

0 .4 2

0 .4 3

0 .3 2

2 0 .6 6

2 0 .2 2

3 C

0 .1 5 *

0 .1 5 *

0 .4 9

1 .7

2 *

* 0 .0 0

2 5 8

1 .0 0

1 3 4

1 4 1

0 .6 9

0 .3 7

0 .3 5

0 .3 9

2 0 .6 3

2 0 .3 0

4 C

0 .5 1 †

0 .0 6

2 0 .0 6

0 .0

0 –

– 0 .0 0

2 5 2

1 .0 0

1 3 1

1 4 2

0 .7 8

0 .6 3

0 .2 3

0 .4 6

2 0 .0 5

2 0 .5 2

2 0 .6 7

2 0 .4 3

5 C

2 0 .0 6 *

0 .2 7 *

0 .4 5

3 .9

2 *

* 0 .2 4

2 5 4

0 .9 4

1 3 1

1 4 0

0 .6 2

0 .6 4

0 .4 6

0 .3 6

2 0 .3 3

2 0 .4 9

2 0 .6 8

2 0 .3 2

6 C

0 .1 5 *

0 .1 4 *

0 .3 2

1 .6

2 *

* 0 .0 0

2 5 3

1 .0 0

1 2 9

1 3 8

0 .7 1

0 .6 3

0 .3 8

0 .4 2

2 0 .2 2

2 0 .5 1

2 0 .6 6

2 0 .3 8

7 C

0 .5 3 *

0 .3 7 * 2 0 .4 6

0 .0

0 –

– 0 .0 0

2 4 9

1 .0 0

1 2 4

1 3 5

0 .7 8

0 .7 8

0 .2 2

0 .2 8

2 0 .0 6

2 0 .5 7

2 0 .0 4

2 0 .3 6

2 0 .6 9

2 0 .4 3

8 C

2 0 .0 6 *

0 .2 6 *

0 .3 1

4 .9

2 *

* 0 .2 9

2 5 1

0 .9 2

1 2 5

1 3 4

0 .6 2

0 .7 4

0 .4 6

0 .3 5

2 0 .4 1

2 0 .6 1

2 0 .2 1

2 0 .3 5

2 0 .7 3

2 0 .3 9

9 C

0 .1 6 *

0 .1 4 *

0 .2 2

3 .7

2 *

* 0 .2 2

2 5 1

0 .9 6

1 2 4

1 3 3

0 .7 0

0 .7 2

0 .3 8

0 .4 0

2 0 .2 8

2 0 .6 2

2 0 .1 4

2 0 .3 2

2 0 .7 2

2 0 .4 3

Breznau: Positive Returns and Equilibrium 599

T a b le

3 . co n t. F e e d b a ck

E ff e ct s

F it S ta ti st ic s

In d e p e n d e n t v a ri a b le

e ff e ct s

M o d e l

S td .- Y X . C o e ff

R e si d .

C h i- S q u a re

b R -S q u a re

b A g e d

b R ig h t

b U n e m p

b G D P

b F L P

b V e to

# b 1

b 2

co v (e

O ,e P ) V a lu e

d f H

0 H

1 R M S E A

L L

C F I

A IC

B IC

P O

S P

P O

S P

P O

S P

P O

S P

P O

S P

P O

S P

1 0 C

0 .5 3 *

0 .2 7

2 0 .2 8

0 .0

0 –

– 0 .0 0

2 5 2

1 .0 0

1 3 0

1 4 0

0 .7 8

0 .7 1

0 .2 2

0 .3 2

2 0 .0 4

2 0 .4 0

0 .0 2

0 .2 3

2 0 .6 4

2 0 .4 2

1 1 C

2 0 .0 6 *

0 .2 7 *

0 .4 2

3 .5

2 *

* 0 .2 1

2 5 4

0 .9 5

1 2 9

1 3 8

0 .6 0

0 .6 7

0 .4 7

0 .3 4

2 0 .3 3

2 0 .4 3

0 .0 1

0 .2 1

2 0 .6 4

2 0 .3 8

1 2 C c

0 .1 6 *

0 .1 4 *

0 .3 1

1 .9

2 *

* 0 .0 0

2 5 3

1 .0 0

1 2 8

1 3 7

0 .7 0

0 .6 5

0 .3 9

0 .4 0

2 0 .2 3

2 0 .4 6

0 .0 0

0 .1 7

2 0 .6 5

2 0 .4 2

1 D

0 .0 4 †

2 0 .1 4

0 .2 8

0 .0

0 –

– 0 .0 0

2 2 4 0

1 .0 0

5 0 2

5 2 6

0 .6 5

0 .1 4

0 .2 2

0 .4 5

2 0 .6 2

2 0 .3 5

2 D

2 0 .0 6 *

0 .2 8 *

0 .5 2

9 .1

2 0 .2 3

2 2 4 5

0 .9 2

5 0 7

5 2 7

0 .4 5

0 .3 9

0 .3 6

0 .2 9

2 0 .6 3

2 0 .2 5

3 D

0 .1 4 *

0 .1 5 *

0 .4 1

4 .0

2 *

0 .1 3

2 2 4 2

0 .9 8

5 0 2

5 2 2

0 .5 9

0 .3 4

0 .3 0

0 .3 4

2 0 .6 1

2 0 .3 0

4 D

0 .4 1 *

2 0 .0 3

0 .1 7

0 .0

0 –

– 0 .0 0

2 2 3 2

1 .0 0

4 9 2

5 2 0

0 .6 6

0 .3 9

0 .2 2

0 .4 5

2 0 .0 6

2 0 .3 9

2 0 .6 3

2 0 .3 6

5 D

2 0 .0 6 *

0 .2 8 *

0 .4 4

8 .0

2 0 .2 2

2 2 3 7

0 .9 4

4 9 6

5 2 0

0 .4 8

0 .5 1

0 .3 9

0 .3 3

2 0 .2 3

2 0 .3 6

2 0 .6 4

2 0 .2 8

6 D

0 .1 4 *

0 .1 5 *

0 .3 4

2 .9

2 *

0 .0 8

2 2 3 4

0 .9 9

4 9 1

5 1 4

0 .5 9

0 .4 7

0 .3 2

0 .3 8

2 0 .1 5

2 0 .3 8

2 0 .6 2

2 0 .3 2

7 D

0 .4 4 *

0 .1 2

0 .5 7

0 .0

0 –

– 0 .0 0

2 2 3 1

1 .0 0

4 9 3

5 2 5

0 .6 7

0 .4 9

0 .2 0

0 .3 8

2 0 .0 5

2 0 .3 8

2 0 .0 7

2 0 .1 6

2 0 .6 7

2 0 .3 6

8 D

2 0 .0 6 *

0 .2 8 *

0 .4 2

5 .7

2 *

0 .1 7

2 2 3 4

0 .9 6

4 9 4

5 2 3

0 .4 7

0 .5 2

0 .3 9

0 .3 2

2 0 .2 5

2 0 .3 7

2 0 .1 5

2 0 .1 9

2 0 .6 9

2 0 .2 9

9 D

0 .1 5 *

0 .1 5 *

0 .3 2

1 .9

2 *

0 .0 0

2 2 3 2

1 .0 0

4 9 1

5 1 9

0 .5 8

0 .4 9

0 .3 2

0 .3 7

2 0 .1 6

2 0 .3 8

2 0 .1 1

2 0 .1 7

2 0 .6 7

2 0 .3 3

1 0 D

0 .4 5 *

0 .1 4

2 0 .0 3

0 .0

0 –

– 0 .0 0

2 2 3 1

1 .0 0

4 9 4

5 2 6

0 .6 7

0 .5 0

0 .1 9

0 .3 4

2 0 .0 3

2 0 .3 3

0 .0 6

0 .1 6

2 0 .6 6

2 0 .3 8

1 1 D

2 0 .0 6 *

0 .2 8 *

0 .4 2

5 .6

2 *

0 .1 7

2 2 3 5

0 .9 6

4 9 5

5 2 3

0 .4 6

0 .5 2

0 .3 8

0 .2 9

2 0 .2 1

2 0 .3 1

0 .1 0

0 .2 0

2 0 .6 6

2 0 .3 1

1 2 D

0 .1 5 *

0 .1 5 *

0 .3 3

2 .1

2 *

* 0 .0 2

2 2 3 3

0 .9 9

4 9 2

5 2 0

0 .5 7

0 .4 8

0 .3 1

0 .3 5

2 0 .1 4

2 0 .3 4

0 .0 7

0 .1 6

2 0 .6 5

2 0 .3 4

a b 1 is e ff e ct

o f p o li cy

o n o p in io n a n d b 2 is o p in io n o n p o li cy . P O

is sh

o rt fo r p u b li c o p in io n a n d S P fo r so ci a l p o li cy . S ig n if ic a n ce

v a lu e s *p

< .0 5 , † p < .1 0 o n ly

re p o rt e d

fo r fe e d b a ck

e ff e ct s. S e e T a b le

2 fo r m o re

d e ta il s.

b “ d f”

is m o d e l d e g re e s o f fr e e d o m ; H

0 te st s ex ac t- fi t h y p ot h es is , * p > 0 5 ; H

1 is eq u al -f it h y p ot h es is th a t th e cu

rr e n t m o d e l is b e tt e r fi tt in g th a n th e la rg e r m o d e l in

e a ch

g ro u p w it h id e n ti ca l in d e p e n d e n t v a ri a b le s ta k e n fr o m

m e a n st ru ct u re

m o d e ls in

S u p p o rt in g In fo rm

a ti o n A p p e n d ix

T a b le

6 , *p

> .0 1 (s e e K li n e , 2 0 1 1 , C h . 8 H y p o th e -

si s T e st in g ).

c In

a n e q u a l- fi t te st

b e tw

e e n m e a n st ru ct u re

m o d e l o f 7 C

a n d 1 2 C , M o d e l 1 2 C

is p re fe ra b le , se e S u p p o rt in g In fo rm

a ti o n A p p e n d ix

T a b le

6 a t h tt p s: / / si te s. g o o g le .

co m / si te / n b re z n a u / p u b li ca ti o n s.

600 Policy Studies Journal, 45:4

decommodification, evidence points toward the positive returns model as favorable in

C Models, but the freely estimated model could fit better in the D Models, except for

12D the preferred model whose equal-fit test rejects the freely estimated model.

Supporting Information Table 5 in the online appendix provides a number of

alternative models as sensitivity checks: including a known scale measurement error

for public opinion, unemployment in Models 10–12 C&D, constrained nil effects for

either policy on opinion (b1 5 0) or opinion on policy (b2 5 0), and two equally plau- sible alternatives to Models 11&12. These models do not alter the main findings that

the positive returns model, estimated in a variety of ways, is superior. The nil effect

models provide an interesting alternative, in that when freely estimated, the effect of

policy on opinion could be zero in one out of four models (Model X1A). The effect of

opinion on policy could be zero in all four models (X2A-D). The possibility of a zero

effect is problematic. The coefficient that represents a potentially zero effect of policy

on opinion in X1A is 0.78. This is a massive-sized standardized effect. The only rea-

son it includes zero is because it has very broad confidence bands, i.e., this is not evi-

dence that it is zero. A similar story can be told about opinion on policy. Although at

between 0.11 and 0.24 (Models X2A–D) the effects are closer to zero, they still have

very broad confidence intervals. Moreover, these models do not fit better than the

preferred models, leading to evidence against the nil hypothesis.

As an exploratory exercise I find an argument for more ideal sizes of effects

in both the thermostatic and positive returns theoretical perspectives (based on

the preferred models) by tweaking the size of b1 and b2 until I achieve the best possible model fit (all indices considered). In Model 11BT (Supporting

Figure 4. Results of Simultaneous Feedback Models of Public Opinion and Social Policy Note: Coefficients are YX-standardized and *p < .05 and †p < .10.

Breznau: Positive Returns and Equilibrium 601

Information Table 5), I show that reducing the size of the negative policy

responsiveness to about half of what Wlezien and Soroka recommend (down to

b1 5 20.03), in addition to reducing the size of the opinion representation effect slightly (down to b2 5 0.13) produces a more optimal fit for both spending and decommodification measures. In the positive returns model—which fits even

better—I find that having a very large impact of spending (b1 5 0.33) and an even larger effect of decommodification (b1 5 0.45) on opinion is optimal when combined with a much smaller impact of opinion on spending (b1 5 0.11) and on decommodification (b1 5 0.09). Figure 5 positive returns models (lower two models) present what I believe to be ideal empirical approximations of the sys-

tems perspective theory given these data. I report the ideal thermostatic models

out of interest, but the positive returns models fit better here.

Discussions

Implication of Findings

The thermostatic model has some empirical support, but the evidence favors the

positive returns model. But these are not competing theories. They both represent how

inputs arrive into a system of equilibrium. If negative reactions constitute relative

preferences, and relative preferences are driven by absolute preferences, then it is the

absolute preferences that are more important to the long-term policy linkages as

Figure 5. Opinion-Policy Feedback Model Results After Optimization Note: Coefficients are YX-standardized and *p < .05 and †p < .10.

[Correction added on 29 July, after first online publication: Model labels have been amended in Figure 5.]

602 Policy Studies Journal, 45:4

modeled here (“absolute” and “relative” used in reference to Soroka & Wlezien,

2010). As the public adjusts its absolute preferences in a positive way to policy

spending and legal policy provision changes, then by default their relative preferen-

ces adjust; and if relative public opinion has a moderate to large impact on policy

then the public could drive its own absolute preferences to change, leading to the

positive effect of policy on opinion observed in these results. Thus, the theories are

compatible and the research reported in this paper does not argue in any way

against the thermostatic perspective.

Given these data and models, a theory of positive returns in a system is simply

the better explanation for what is taking place. The thermostatic model explains

what takes place over shorter periods of time and helps keep a system on a path of

longer term positive returns. Therefore, both can be seen as theoretical processes that

lead to positive feedback with differences in theoretical explanation bounded by

length of timespans. However, neither theoretical perspective says anything about

the direction that opinion and policy move, this can be increasing or declining

returns. Given recent neoliberal policy agendas since the 1970s and 80s the expecta-

tion is that declining returns are underway. Thus, in order to extend Pierson’s theo-

retical description of the golden age of welfare state growth to recent neoliberal

turns, I adopted the label positive returns. More research should probe this theoretical

extension and attempt to look outside the period investigated herein from 1985

through 2008, because it could represent a unique historical moment. Although Pier-

son’s increasing returns work suggests that the process was in place throughout the

history of welfare states, a simultaneous feedback model similar to the one employed

here is difficult to test due to lack of historical public opinion data.

I propose a general theory of opinion-policy as a system where positive returns

leads public opinion to partly be a product of policy shaping opinion over long peri-

ods of time and partly a product of opinion itself transmitted through policy thermo-

statically, the only requirement for this to be empirically feasible is that the impact of

opinion on policy have a larger absolute value than policy on opinion. This is exactly

what Wlezien and Soroka find and what I have found in the ideal Figure 5, and in

various sensitivity tests. These results apply to both spending and decommodifica-

tion measures of policy and a general system theory of opinion-policy should grap-

ple with various measures of policy as all types of policy could impact public

opinion through social interaction, political rhetoric, media messages, and the goods

and services received (or not received) via the government; and all could have slight-

ly different effects. Moreover, I have considered national government policy here,

but more decentralized systems provide greater amounts of state and local-level

social policy and this should be an area for future research.

A side finding emerged in this undertaking. I did not focus on the effects of

independent variables specifically, other than to make the model as close of an

approximation of their theoretically causal relationships as possible. However, the

consistently strong positive effect of GDP on both policy spending and decommodifi-

cation supports something argued long ago by Wilensky (1975): that richer countries

(i.e., more developed) spend more on social policy. However, this empirical phe-

nomenon only appears to be true when comparing the poorer countries of the

Breznau: Positive Returns and Equilibrium 603

world to the richer countries. It does not appear to be true within the group of

rich welfare states as borne out in much of the worlds of welfare literature, and

the fact that for example Sweden and the United States. have similar GDPs (corre-

lation of GDP and social policy measures is near zero, see Supporting Information

Table 4). When modeling simultaneous feedback, I find a moderate positive effect

of GDP on spending and decommodification in almost all the models, and this

suggests that Wilensky’s theory may still apply within rich democracies, but that

the effect is suppressed in unidirectional models because of un-modeled, con-

founding endogeneity.

I considered large, rich democratic welfare state societies. However, the model

could apply, if only to a lesser degree, to all societies. Without the consent of the gov-

erned or without successful coercion of the governed, the public are likely to revolt

as suggested long ago by Machiavelli in Chapter 15 of The Prince. This threat of revo-

lution may impact policy similar to democratic electoral processes, thus future

research might look into more or less democratic states as well. Moreover, the results

of models from these 19 democratic states are only a sliver of the full story of policy

in each. There is no question that every policymaking and public opinion moment is

produced by unique historical, socioeconomic, political, and other features in geo-

graphic space and time. This paper argues that there may be a common baseline

relationship of opinion and policy that is ever-present, and which is complemented

by the powerful forces peculiar to each context.

Institutions and Equilibrium

The relationships identified here—after correcting for endogeneity bias and apply-

ing instrumental variables—provide an empirical and theoretical starting point for fur-

ther understanding and developing theories of the common causal linkages between

opinion and policy. They suggest that an institutional positive returns perspective

should be part of this development. This work provides theoretical and empirical evi-

dence that equilibrium is part of the story of positive returns, or what is more generally

known as institutional path dependency. There is an institutional component to the

opinion-policy feedback relationship where the goods and services transferred by the

government during the inception, or history of social policy become embedded norms

on which individuals form their policy attitudes and expectations, and within which

policymakers must operate. The norms are perpetuated through the social and political

institutions of a given society. It is very costly for political actors to deviate from these

norms because they define what range of actions are possible, and changing the norms

would require overthrowing parts of the institutions—literally destroying legal and

brick-and-mortar pieces of these institutions in many cases. These embedded institu-

tions, and the expectations of policymakers and the public lead to an equilibrium

between public opinion and social policy. Equilibrium is a product of normative institu-

tions that exists despite shocks and cycles in the effects running between opinion and

policy. The argument herein is that although punctuations likely exist in opinion-policy

dynamics as discussed in the thermostatic perspective and punctuated equilibrium

604 Policy Studies Journal, 45:4

theories (Jones & Baumgartner, 2012), they are so frequent and regularly occurring in

stable democratic welfare states that they tend to smooth themselves out, leaving a core

relationship of opinion and policy that is stable over time and across societies. Punctua-

tions in the opinion-policy links are constantly starting, proceeding, and coming to a

close; therefore opinion and policy are at equilibrium within them—policy and opinion

neither reach a stasis nor do they engender revolution.

Also, the looming question of what exactly is social policy challenges any theory to

describe the relationship of public opinion and social policy. Policy spending may

influence public opinion mostly through the media, whereas policy provisions of

goods and services influence the public materially. Spending and the generosity of

state-provided goods and services (i.e., decommodification) are highly correlated but

retain distinct qualities. Although the positive returns model is always favored, rul-

ing out the thermostatic model, or the freely estimated model which points toward

positive returns, would be statistical overkill. Given a system at equilibrium, the

effects between opinion and policy are the result of processes taking place across

many years, cycles, budgets, elections, and other punctuations. The Wlezien and Sor-

oka thermostatic model explicitly applies to a one-year cycle. Therefore, seeing even

slight evidence of it in a stable systems perspective spanning decades is unexpected,

and suggests that policy could have a negative force acting on public opinion over

long periods of time. An empirical explanation for what is discussed above is that

there is both a negative and positive causal effect of policy impacting opinion and

that these represent distinct effects of policy; effects that alternative modeling and

better measures would be needed to identity. This is evidenced by the ideal-type

model in Figure 5, where opinion has a small effect on policy vis-�a-vis a large effect

of policy on opinion. Perhaps the small policy effect only appears small because

there is a simultaneous negative effect suppressing it. Further research and theoriz-

ing should prove worthy in sorting out two types of simultaneous effects within the

feedback loop, or one set of simultaneous effects and one set of cyclical effects.

Methodological Discussion

Theory is crucial to the methodological specification of a simultaneous feedback

model. This requires at minimum a lay theory that opinion and policy are reciprocal-

ly related, yet models improve significantly when drawing on well-developed theo-

ries from Wlezien and Soroka, and Pierson, among others. Without imposing

theoretical properties the simultaneous feedback model does not converge to mean-

ingful results. Without theory, the feedback effects of the freely estimated model are

often not significant, sometimes seemingly implausible in size, and leave indepen-

dent variables with also seemingly implausible effects. It is only after carefully apply-

ing opinion-policy theories that the models begin to function. Also, theory is required

to derive the instrumental variables necessary for identification. As female labor

force participation and veto points have some limitations, future research should the-

orize about other instruments. Methods, at least in the case of a simultaneous feed-

back models, are a theoretical undertaking as much as a statistical one. This is a stark

Breznau: Positive Returns and Equilibrium 605

reminder that without theory a researcher could blindly arrive at any number of sta-

tistical models, many of which are meaningless, and thus draw uninformed and per-

haps even theoretically impossible empirical conclusions.

Another methodological issue stems from the measurement of public opinion.

The ISSP data has at least two items by which to measure public opinion. Here I

averaged them to create the dependent variable. However, a measurement model

would be preferable as public opinion toward social policy in general (i.e., the entire

welfare state system) is something latent, not observed in individual questions on

specific policies. But this is not possible given the limited degrees of freedom. Sensi-

tivity tests in Supporting Information Table 5 model a one-minus the opinion alpha

reliability to try and account for the known measurement error in the scale, results

are similar but the models are clumsy. To assign a specific measurement error in the

model forces ignorance of other potential sources of error and may falsely achieve

identification or fit, and thus may be a risky methodological choice without a strong

argument why the error path is exactly a given value.12

The simultaneous feedback model is especially designed to estimate recipro-

cal (total) causality in cross-sectional data. This is a blessing and a burden. It is a

tool to unlock the limitations of sporadically fielded cross-national surveys in

both timing and country-coverage. However, it would be plausible to adapt the

feedback model to longitudinal data looking at each year as a case and attempt-

ing to determine if the general opinion-policy relationship, i.e., equilibrium, is

present over many decades. The trick here would be to find instrumental varia-

bles, as female labor force participation and veto points are only suited to be

cross-national instruments. If scholars manage to construct within-country feed-

back models, their results might be compared to those of Soroka and Wlezien

and to the vector auto-regression studies that find that the effects of opinion

and policy on each other change over the years before and after a policy

moment. Given the theoretical argument here, I would expect that across the

variation in effect sizes by year, there would be a common core effect. This

would be a fruitful application of the simultaneous feedback theory where there

are b1 and b2 which have constant feedback effects each year (fixed-coefficients), and then b1i and b2i that have varying effects by year (random-coefficients). At this stage, researchers have not identified within-country instruments, thus the

longitudinal simultaneous feedback model remains purely theoretical.

The partisan component of these models calls for further research. The partisan

make-up of government threatens arguments of equilibrium. There are constant

efforts by parties to make policy; however, their capacities to influence policy (and

perhaps public opinion) are greatly enhanced based on the number of seats they

hold. The distribution of seats is erratic over time. It is a product of public opinion,

media, and how well policy serves public interests. Therefore, partisan control is also

endogenous with public opinion and social policy and a theoretical model should

have reciprocal causality running between opinion, policy, and partisan control of

government. As modeling more endogenous dependent variables is beyond the scope

of this paper and the data, I settled for right party control as an independent variable,

despite knowing that opinion and policy both influence it as well. However, I ran

606 Policy Studies Journal, 45:4

sensitivity models omitting right party seats as a variable, with the assumption that

the reciprocal impact goes in to the total effects running between opinion and policy

and the error covariance. When doing this all models tend to fit well (see Models

10Ar-12Dr9) pointing toward the strained role of right party seats as a variable in a

simultaneous feedback model, but given its prominent theoretical role in opinion and

policy I keep it in the main models. Also, the sensitivity results still largely favor the

positive returns model, and in the case of the freely estimated Model 10Cr, results

point toward the modified positive returns model (Figure 5) where policy has a larger

impact on opinion than opinion on policy; i.e., b1 larger than b2.

Nate Breznau is a postdoctoral fellow at the Mannheim Centre for European Social

Research. He received his PhD in Sociology from the Bremen International Gradu-

ate School of Social Sciences in 2013. Before coming to Germany in 2009, he did a

BA in Sociology and African-American Studies at Bates College in Lewiston,

Maine; and an MA in Sociology at the University of Nevada, Reno. His research

and teaching revolves around public opinion, social welfare attitudes, welfare

states, social policy, and changing social inequalities. He is passionate about open

science and is committed to sharing all code and technical information on his aca-

demic website.

Notes

I am grateful to those active on the SEMNET listserve and the anonymous reviewers. Many thanks also go

to feedback from Alexi Gugshvili, Liza Steele, Anthony Sealey, Nadine Sch€oneck-Voss, Christoph Bur-

khardt, Judith Offerhaus, Christopher Wlezien, Bernhard Ebbinghaus, Tobias Wolbring, Sergi Vidal, J.

Timo Weishaupt, Ralf G€otze, and Dwayne Woods at various stages in this research.

1. A reference to Lipset’s (gendered) conception of the “political man.”

2. I opt for “declining returns” here to avoid confusion because “decreasing returns” typically refers to

some kind of diminishing marginal utility in economics.

3. This is a purposeful oversimplification of change-scores to move forward with model construction

(Allison, 1990).

4. That is, treated as exogenous in a structural equation model—also the default estimation procedure in

MPlus software.

5. The European Social Survey could provide a source but this would be a restricted sample.

6. The ISSP “Role of Government” (’85,’90,’96,’06) and “Religion” (’91,’98) modules provide a 2-item

scale which is part of a larger 6-item scale asking about the welfare state more generally with pensions

and health care (Cronbach’s alphas are 0.62 and 0.71, respectively); unfortunately the 6-items are

fielded in far less surveys (43 as opposed to 70 cases).

7. This is the standard variable in cross-nationally comparative social welfare policy research (Brooks &

Manza, 2007; Green-Pedersen, 2004; Kenworthy, 1999; Pampel & Williamson, 1988), despite its

known problems (Castles, 2009). Like Wlezien and Soroka (2012) I use this measure for its utility and

comparability across societies.

8. There are alternatives such as military spending, tax systems, and post-materialist development, but

these are difficult to measure or may have more serious endogeneity problems. Although not exhaus-

tive, the two selected are derived from 4 years of dissertational and postdoc research.

9. At least for the countries I analyze in this paper. Two small exceptions are New Zealand (Vowles,

1995), and Italy in 1994 (Lijphart, 1999).

Breznau: Positive Returns and Equilibrium 607

10. https://sites.google.com/site/nbreznau/publications.

11. Note that the gray paths are significant at p < .2 and have substantial mathematically sized effects. I

do not wish to engage in a debate about p-hacking, but alert the reader to the fact that they matter in

size.

12. I conclude this based on expert discussions on the SEMNET listserve.

References

Allison, Paul D. 1990. “Change Scores as Dependent Variables in Regression Analysis.” Sociological Meth-

odology 20: 93–114.

Anderies, John M., and Marco A. Janssen. 2013. “Robustness of Social-Ecological Systems: Implications

for Public Policy.” Policy Studies Journal 41 (3): 513–36.

Andreß, Hans-J€urgen, and Thorsten Heien. 2001. “Four Worlds of Welfare State Attitudes? A Compari-

son of Germany, Norway and the United States.” European Sociological Review 17 (4): 337–56.

Bartels, Larry M. 1986. “Issue Voting Under Uncertainty: An Empirical Test.” American Journal of Political

Science 30 (4): 709–28.

Bean, Clive, and Elim Papadakis. 1998. “A Comparison of Mass Attitudes Towards the Welfare State in Dif-

ferent Institutional Regimes, 1985-1990.” International Journal of Public Opinion Research 10 (3): 211–36.

Bendz, Anna. 2015. “Paying Attention to Politics: Public Responsiveness and Welfare Policy Change.”

Policy Studies Journal 43 (3): 309–32.

Blekesaune, Morten. 2007. “Economic Conditions and Public Attitudes to Welfare Policies.” European

Sociological Review 23 (3): 393–403.

Blekesaune, Morten, and Jill Quadagno. 2003. “Public Attitudes Toward Welfare State Policies: A Com-

parative Analysis of 24 Nations.” European Sociological Review 19 (5): 415–27.

Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New York, NY: Wiley.

———. 2012. “Instrumental Variables in Sociology and the Social Sciences.” Annual Review of Sociology 38

(1): 37–72.

Bollen, Kenneth A., Barbara Entwisle, and Arthur S. Alderson. 1993. “Macrocomparative Research

Methods.” Annual Review of Sociology 19: 321–51.

Breznau, Nate. 2010. “Economic Equality and Social Welfare: Policy Preferences in Five Nations.” Interna-

tional Journal of Public Opinion Research 22 (4): 458–84.

———. 2015. “The Missing Main Effect of Welfare State Regimes: A Replication of ‘Social Policy Respon-

siveness in Developed Democracies’ by Brooks and Manza.” Sociological Science 2: 420–41.

Breznau, Nate, and Maureen A. Eger. 2016. “Immigrant Presence, Group Boundaries, and Support for

the Welfare State in Western European Societies.” Acta Sociologica, doi:10.1177/0001699316645168.

Brooks, Clem, and Jeff Manza. 2007. Why Welfare States Persist: The Importance of Public Opinion in Democ-

racies. Chicago: University of Chicago Press.

Butler, Daniel M., and David W. Nickerson. 2011. “Can Learning Constituency Opinion Affect How

Legislators Vote? Results from a Field Experiment.” Quarterly Journal of Political Science 6 (1): 55–83.

Campbell, Andrea Louise. 2012. “Policy Makes Mass Politics.” Annual Review of Political Science 15 (1):

333–51.

Castles, Francis G. 2009. “What Welfare States Do: A Disaggregated Expenditure Approach.” Journal of

Social Policy 38 (1): 45–62.

Converse, Philip E. 1987. “Changing Conceptions of Public Opinion in the Political Process.” The Public

Opinion Quarterly 51: S12–24.

Crepaz, Markus M. L., and Ann W. Moser. 2004. “The Impact of Collective and Competitive Veto Points

on Public Expenditures in the Global Age.” Comparative Political Studies 37 (3): 259–85.

D€aubler, Thomas. 2008. “Veto Players and Welfare State Change: What Delays Social Entitlement Bills?”

Journal of Social Policy 37 (4): 683–706.

608 Policy Studies Journal, 45:4

DiMaggio, Paul J., and Walter W. Powell. 1983. “The Iron Cage Revisited: Institutional Isomorphism and

Collective Rationality in Organizational Fields.” American Sociological Review 48 (2): 147–60.

Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row.

Duncan, Otis Dudley, Archibald O. Haller, and Alejandro Portes. 1968. “Peer Influences on Aspirations:

A Reinterpretation.” American Journal of Sociology 74 (2): 119–37.

Durkheim, Emile. 1964. The Division of Labor in Society, ed. George Simpson (Translator). New York: Free

Press.

Durr, Robert H. 1993. “What Moves Policy Sentiment?” American Political Science Review 87 (1): 158–70.

Easton, David. 1965. A Framework for Political Analysis. Englewood Cliffs, NJ: Prentice-Hall.

Ebbinghaus, Bernhard. 2005a. “Can Path Dependence Explain Institutional Change? Two Approaches

Applied to Welfare State Reform.” MPlfG Discussion Paper (05/2). Cologne, Germany: Max Planck

Institute for the Study of Societies.

———. 2005b. “When Less Is More: Selection Problems in Large-N and Small-N Cross-National

Comparisons.” International Sociology 20 (2): 133–52.

Eger, Maureen A. 2010. “Even in Sweden: The Effect of Immigration on Support for Welfare State

Spending.” European Sociological Review 26 (2): 203–17.

Esping-Andersen, Gøsta. 1985. “Power and Distributional Regimes.” Politics & Society 14 (2): 223–56.

———. 1990. The Three Worlds of Welfare Capitalism. Oxford: Oxford University Press.

Evans, M. D. R., Jonathan Kelley, and Clayton D. Peoples. 2010. “Justifications of Inequality: The Norma-

tive Basis of Pay Differentials in 31 Nations.” Social Science Quarterly 91 (5): 1405–31.

Freeman, John R., John T. Williams, and Tse-min Lin. 1989. “Vector Autoregression and the Study of

Politics.” American Journal of Political Science 33 (4): 842–77.

Gabel, Matthew, and Kenneth Scheve. 2007. “Estimating the Effect of Elite Communications on Public

Opinion Using Instrumental Variables.” American Journal of Political Science 51 (4): 1013–28.

Gelissen, John. 2000. “Popular Support for Institutionalised Solidarity: A Comparison between European

Welfare States.” International Journal of Social Welfare 9: 285–300.

G€erxhani, Klarita, and Ferry Koster. 2012. “‘I Am Not Alone’: Understanding Public Support for the Wel-

fare State.” International Sociology 27 (6): 768–87.

Green-Pedersen, Christoffer. 2004. “The Dependent Variable Problem within the Study of Welfare State

Retrenchment: Defining the Problem and Looking for Solutions.” Journal of Comparative Policy Analy-

sis: Research and Practice 6 (1): 3–14.

Habel, Philip D. 2012. “Following the Opinion Leaders? The Dynamics of Influence Among Media Opin-

ion, the Public, and Politicians.” Political Communication 29 (3): 257–77.

Habermas, J€urgen. 1989. The Structural Transformation of the Public Sphere: An Inquiry into a Category of

Bourgeois Society, English Tr. ed. Thomas Burger, and Frederick Lawrence. Cambridge, MA: MIT

Press.

Hacker, Jacob S. 2004. “Privatizing Risk without Privatizing the Welfare State: The Hidden Politics of

Social Policy Retrenchment in the United States.” American Political Science Review 98 (2): 243260.

Hakhverdian, Armen. 2012. “The Causal Flow between Public Opinion and Policy: Government Respon-

siveness, Leadership, or Counter Movement?” West European Politics 35 (6): 1386–406.

Hall, Peter A., and Michèle Lamont. 2013. “Why Social Relations Matter for Politics and Successful Societ-

ies.” Annual Review of Political Science 16 (1): 49–71.

Hall, Peter A., and Rosemary C. R. Taylor. 1996. “Political Science and the Three New Institutionalisms.”

Political Studies 44 (5): 936–57.

Hasenfeld, Yeheskel, and Jane A. Rafferty. 1989. “The Determinants of Public Attitudes toward the

Welfare State.” Social Forces 67 (4): 1027–48.

Hill, Kim Quaile, and Angela Hinton-Andersson. 1995. “Pathways of Representation: A Causal Analysis

of Public Opinion-Policy Linkages.” American Journal of Political Science 39 (4): 924–35.

Homans, George C. 1974. Social Behavior: Its Elementary Forms. New York: Harcourt, Brace, Jovanovich.

Breznau: Positive Returns and Equilibrium 609

Huber, Evelyne, Charles Ragin, and John D. Stephens. 1993. “Social Democracy, Christian Democracy,

Constitutional Structure, and the Welfare State.” The American Journal of Sociology 99 (3): 711–49.

Huber, Evelyne, and John D. Stephens. 2000. “Partisan Governance, Women’s Employment, and the

Social Democratic Service State.” American Sociological Review 65 (3): 323–42.

Jæger, Mads Meier. 2006. “What Makes People Support Public Responsibility for Welfare Provision: Self-

Interest or Political Ideology? A Longitudinal Approach.” Acta Sociologica 49 (3): 321–38.

Jennings, Will. 2009. “The Public Thermostat, Political Responsiveness and Error-Correction: Border

Control and Asylum in Britain, 1994–2007.” British Journal of Political Science 39 (4): 847–70.

Jensen, Carsten. 2011. “Two Sides of the Same Coin? Left-Wing Governments and Labour Unions as

Determinants of Public Spending.” Socio-Economic Review 10 (2): 217–40.

Jepperson, Ronald. 1991. “Institutions, Institutional Effects, and Institutionalism.” In The New Institutional-

ism in Organizational Analysis, ed. Walter W. Powell, and Paul J. DiMaggio. Chicago: University of

Chicago Press,143–63.

Jones, Bryan D., and Frank R. Baumgartner. 2004. “Representation and Agenda Setting.” Policy Studies

Journal 32 (1): 1–24.

———. 2012. “From There to Here: Punctuated Equilibrium to the General Punctuation Thesis to a Theo-

ry of Government Information Processing.” Policy Studies Journal 40 (1): 1–20.

Kenworthy, Lane. 1999. “Do Social-Welfare Policies Reduce Poverty? A Cross-National Assessment.”

Social Forces 77 (3): 1119–39.

Kline, Rex B. 2011. Principles and Practice of Structural Equation Modeling, 3rd ed. New York, NY: Guilford

Press.

———. 2013. “Reverse Arrow Dynamics: Feedback Loops and Formative Measurement.” In Structural

Equation Modeling: A Second Course, ed. Gregory R. Hancock, and Ralph O. Mueller. Charlotte, NC:

Information Age Publishing, 41–79.

Knight, Carly R., and Christopher Winship. 2013. “The Causal Implications of Mechanistic Thinking:

Identification Using Directed Acyclic Graphs (DAGs).” In Handbook of Causal Analysis for Social

Research, Handbooks of Sociology and Social Research, ed. Stephen L. Morgan. Dordrecht, the Nether-

lands: Springer, 275–99.

Korpi, Walter, and Joakim Palme. 2003. “New Politics and Class Politics in the Context of Austerity and

Globalization: Welfare State Regress in 18 Countries.” American Political Science Review 97 (3): 425–

46.

Larsen, Christian Albrekt. 2008. “The Institutional Logic of Welfare Attitudes: How Welfare Regimes

Influence Public Support.” Comparative Political Studies 41 (2): 145–68.

Lee, Han Soo. 2014. “Analyzing the Multidirectional Relationships Between the President, News Media,

and the Public: Who Affects Whom?” Political Communication 31 (2): 259–81.

Lidtke, Vernon L. 1966. The Outlawed Party: Social Democracy in Germany 1878-1890. English Tr. Princeton,

NJ: Princeton University Press.

Lijphart, Arend. 1999. Patterns of Democracy: Government Forms and Performance in Thirty-Six Countries.

New Haven, CT: Yale University Press.

Lipset, Seymour Martin. 1960. Political Man: The Social Bases of Politics. Garden City, NY: Doubleday and

Company.

Luhmann, Niklas. 1982. “The World Society as a Social System.” International Journal of General Systems

8 (3): 131–38.

Matsusaka, John G. 2005. “The Endogeneity of the Initiative: A Comment on Marschall and Ruhil.” State

Politics & Policy Quarterly 5 (4): 356–63.

Meltzer, Allan H., and Scott F. Richard. 1981. “A Rational Theory of the Size of Government.” The Journal

of Political Economy 89 (5): 914–27.

Mettler, Suzanne, and Joe Soss. 2004. “The Consequences of Public Policy for Democratic Citizenship:

Bridging Policy Studies and Mass Politics.” Perspectives on Politics 2 (1): 55–73.

Meyer, John W., and Brian Rowan. 1977. “Institutionalized Organizations: Formal Structure as Myth and

Ceremony.” American Journal of Sociology 83 (2): 340–63.

610 Policy Studies Journal, 45:4

Miller, Wakken E., and Donald E. Stokes. 1963. “Constituency Influence in Congress.” American Political Science Review 57 (1): 45–56.

Morgan, Kimberly J., and Andrea Louise Campbell. 2011. The Delegated Welfare State: Medicare, Markets, and the Governance of Social Policy. New York: Oxford University Press.

Mullinix, Kevin J. 2011. “Lingering Debates and Innovative Advances: The State of Public Opinion Research.” Policy Studies Journal 39 (s1): 61–76.

North, Douglass C. 1990. Institutions, Institutional Change and Economic Performance. Cambridge: Cam- bridge University Press.

Nowlin, Matthew C. 2011. “Theories of the Policy Process: State of the Research and Emerging Trends.” Policy Studies Journal 39 (s1): 41–60.

OECD. 2012. Social Expenditures Database [Online]. Organization for Economic Cooperation and Devel- opment. http://www.oecd.org. Accessed October 1, 2012.

Ostrom, Elinor. 2011. “Background on the Institutional Analysis and Development Framework.” Policy Studies Journal 39 (1): 7–27.

Page, Benjamin I., and Robert Y. Shapiro. 1983. “Effects of Public Opinion on Policy.” The American Politi- cal Science Review 77 (1): 175–90.

Pampel, Fred C., and John B. Williamson. 1988. “Welfare Spending in Advanced Industrial Democracies, 1950-1980.” American Journal of Sociology 93 (6): 1424–56.

Papadakis, Elim. 1992. “Public Opinion, Public Policy and the Welfare State.” Political Studies 40 (1): 21–37.

Patashnik, Eric M., and Julian E. Zelizer. 2013. “The Struggle to Remake Politics: Liberal Reform and the Limits of Policy Feedback in the Contemporary American State.” Perspectives on Politics 11 (4): 1071– 87.

Paxton, Pamela, John R. Hipp, and Sandra Marquat-Pyatt. 2011. Nonrecursive Models: Endogeneity, Recipro- cal Relationships and Feedback Loops. London: Sage Publications, Ltd.

Pearl, Judea. 2009. Causality: Models, Reasoning and Inference, 2nd ed. New York, NY: Cambridge University Press.

Pedhazur, Elazar J. 1997. Multiple Regression in Behavioral Research: Explanation and Prediction. New York: Thomson Learning.

Pierson, Paul. 1993. “When Effect Becomes Cause: Policy Feedback and Political Change.” World Politics 45 (4): 595–628.

———. 1994. Dismantling the Welfare State? Reagan, Thatcher and the Politics of Retrenchment. Cambridge: Cambridge University Press.

———. 1996. “The New Politics of the Welfare State.” World Politics 48 (2): 143–79.

———. 1998. “Irresistible Forces, Immovable Objects: Post-Industrial Welfare States Confront Permanent Austerity.” Journal of European Public Policy 5 (4): 539–60.

———. 2000. “Increasing Returns, Path Dependence, and the Study of Politics.” The American Political Sci- ence Review 94 (2): 251–67.

———. 2002. “Coping with Permanent Austerity: Welfare State Restructuring in Affluent Democracies.” Revue française de sociologie 43 (2): 369–406.

Poterba, James M. 1995. “Capital Budgets, Borrowing Rules, and State Capital Spending.” Journal of Public Economics 56 (2): 165–87.

Raven, Judith, Peter Achterberg, Romke van der Veen, and Mara Yerkes. 2011. “An Institutional Embeddedness of Welfare Opinions? The Link Between Public Opinion and Social Policy in the Netherlands (1970-2004).” Journal of Social Policy 40 (2): 369–86.

Roosma, Femke, Wim van Oorschot, and John Gelissen. 2014. “The Preferred Role and Perceived Perfor- mance of the Welfare State: European Welfare Attitudes from a Multidimensional Perspective.” Social Science Research 44C: 200–10.

Schumacher, Gijs, Barbara Vis, and Kees van Kersbergen. 2013. “Political Parties’ Welfare Image, Elector- al Punishment and Welfare State Retrenchment.” Comparative European Politics 11 (1): 1–21.

Scruggs, Lyle. 2004. Welfare State Entitlements Data Set: A Comparative Institutional Analysis of Eigh- teen Welfare States [Online]. Version 1.2. http://sp.uconn.edu/~scruggs/#cwed. Accessed April 20, 2013.

Breznau: Positive Returns and Equilibrium 611

Skocpol, Theda. 1988. “The Limits of the New Deal System and the Roots of Contemporary Welfare Dilemmas.” In The Politics of Social Policy in the United States, ed. Margaret Weir, Ann Shola Orloff, and Theda Skocpol. Princenton, NJ: Princeton University Press, 293–311.

———. 1992. Protecting Soldiers and Mothers: The Political Origins of Social Policy in the United States. Cam- bridge, MA: Belknap Press.

Soroka, Stuart N., and Christopher Wlezien. 2010. Degrees of Democracy: Politics, Public Opinion and Policy. Cambridge: Cambridge University Press.

Soss, Joe, and Sanford F. Schram. 2007. “A Public Transformed? Welfare Reform as Policy Feedback.” The American Political Science Review 101 (1): 111–27.

Starke, Peter. 2006. “The Politics of Welfare State Retrenchment: A Literature Review.” Social Policy & Administration 40 (1): 104–20.

Steele, Liza G. 2015. “Income Inequality, Equal Opportunity, and Attitudes About Redistribution.” Social Science Quarterly 96 (2): 444–64.

Stimson, James A., Michael B. Mackuen, and Robert S. Erikson. 1995. “Dynamic Representation.” The American Political Science Review 89 (3): 543–65.

Svallfors, Stefan. 1997. “Worlds of Welfare and Attitudes to Redistribution: A Comparison of Eight West- ern Nations.” European Sociological Review 13 (3): 283–304.

Taylor-Gooby, Peter. 2001. “Sustaining State Welfare in Hard Times: Who Will Foot the Bill?” Journal of European Social Policy 11 (2): 133–47.

Tepe, Markus, and Pieter Vanhuysse. 2010. “Elderly Bias, New Social Risks and Social Spending: Change and Timing in Eight Programmes across Four Worlds of Welfare, 1980-2003.” Journal of European Social Policy 20 (3): 217–34.

Thelen, Kathleen. 1999. “Historical Institutionalism in Comparative Politics.” Annual Review of Political Science 2 (1): 369–404.

———. 2014. Varieties of Liberalization and the New Politics of Social Solidarity. Cambridge: Cambridge Uni- versity Press.

Vowles, Jack. 1995. “The Politics of Electoral Reform in New Zealand.” International Political Science Review 16 (1): 95–115.

Weber, Max. 1946. From Max Weber: Essays in Sociology, ed. Hans Heinrich Gerth, and C. Wright Mills. New York: Oxford University Press.

Weir, Margaret, and Theda Skocpol. 1985. “State Structures and the Possibilities for ‘Keynesian’ Responses to the Great Depression in Sweden, Britain, and the United States.” In Bringing the State Back In, ed. Peter B. Evans, Dietrich Rueschemeyer, and Theda Skocpol. Cambridge: Cambridge University Press, 107–63.

Wilensky, Harold L. 1975. The Welfare State and Equality: Structural and Ideological Roots of Public Expendi- ture. Berkeley, CA: University of California Press.

Wlezien, Christopher. 1995. “The Public as Thermostat: Dynamics of Preferences for Spending.” American Journal of Political Science 39 (4): 981–1000.

Wlezien, Christopher, and Stuart N. Soroka. 2007. “Relationships between Public Opinion and Policy.” In Oxford Handbook of Political Behavior, ed. Russell Dalton, and Hans-Deiter Klingemann. Oxford: Oxford University Press, 799–817.

———. 2012. “Political Institutions and the Opinion–Policy Link.” West European Politics 35 (6): 1407–32.

Wright, Sewall. 1934. “The Method of Path Coefficients.” The Annals of Mathematical Statistics 5 (3): 161– 215.

Supporting Information

Additional Supporting Information may be found online in the supporting

information tab for this article.

612 Policy Studies Journal, 45:4

This document is a scanned copy of a printed document. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material.