CHAPTER 7: Determinants of Polarization

In Chapter 5 I took a narrow focus and examined the breadth and extent of polarization in public opinion on gay rights. In Chapter 6 I broadened that focus to include the entire social issue dimension as well as other types of issues such as economic issues and issues on foreign policy and examined aggregate linear polarization trends for these issues. In both of these prior analyses I focused on the questions of whether and to what extent polarization had occurred using several measures of polarization to answer those questions. In this chapter, I continue to ask whether polarization has or has not occurred over the “culture wars” time period from the 1970’s to the present. However, rather than look at the overall trend in polarization, I take a closer look at the trends in the public opinion on political issues by examining the trends in detail rather than summary form. The linear models employed in the previous chapters may have masked non-linear trends that actually reflect polarization on issue dimensions. Furthermore, there may be early or late-breaking trends that is lost in the summary findings from the full time series. Secondly, I take an initial step at assessing the causes of polarization. What are the factors that cause public opinion to polarize or depolarize? Are these random, chaotic events with no rhyme or reason? Or, is political polarization rational in the aggregate: a response by society and groups within society to real world political events, policy promises made by candidates and parties, and the actual implemented policies of the government?

A causal model of public opinion is, in many ways, the Great White Whale, of political behavior analysis. The degree to which perturbations in public opinion are responsive to actual events, the manipulations of elite actors, perceptions of politics screened by a changing media landscape, implemented policies, the economy, personal circumstances, cultural and social group factors, etc. has been the subject of a great deal of theorizing and analysis in the discipline since Walter Lippmann pontificated on public opinion in the 1920’s and the empirical study of public opinion that began in earnest in the 1950’s. Studies of the relationship between public opinion and public policy have mostly shown that policy-makers are responsive to public opinion and that policy elites drive public opinion towards their views on policy (Shapiro et al. 1990; Page 1979; Stimson 2004; Zaller 1992). As Harwood Childs noted in an early study linking public policy and public opinion, “the relationship between public opinion and public policy varies greatly from issue to issue. The influence of public opinion varies from virtually no influence to enormous influence. Influence may be exerted quickly or slowly; it may change over time or remain constant, and its impact may be direct or indirect” (Childs 1965). As Childs argued, the extent of the influence of public opinion on policy depends on factors including the degree of agreement within the public, the intensity of that opinion, and the extent of organized support for and against the public position. To that I can add the polarization of opinion on an issue.

Identifying causes of public opinion trends is a complex and difficult undertaking, plagued by poor and inconsistent measures of attitudes, interrelated and highly correlated causal variables, and significant omitted-variable bias. As Childs argued, the relationship between public opinion and public policy is cyclic and dynamic. In other words, public opinion not only influences public policy but, as I will show here, public policy also influences public opinion. However, contrary to Childs’ supposition that the public becomes more accepting of a policy change once it is made, I will show that public policy opinion can move in the opposite direction: signaling disapproval of the adopted policy rather than acceptance. Indeed, this formulation is consistent with Sharp’s “thermostatic sequence” where public policy responds to public opinion and then public opinion provides feed back where public opinion adjusts to the policy change. Changes in policy responding to public opinion can move the policy too far, leading to the public moving in the opposite direction on the issue (Sharp 1999). Thus the importance of answering not merely the ‘what’ of polarization but also the ‘why’ of polarization is clear. While definitive answers may be allusive, analyses which identify important determinants of public attitudes on political issues, however contingent these models may be, are of substantive significance. Theoretically, we can define the potential relationships as follows:

Note, these models are not mutually exclusive. In other words, the acceptance of one does not necessitate the rejection of the other models of public opinion influence. Rather, each could be a factor influencing public opinion. I will examine three potential explanations of political polarization: partisanship, exogenous events or ‘shocks’ to the political system, and implemented public policy on the issues themselves.

Partisanship as an exogenous influence on public opinion can take on many forms. We can assess the partisan identification of the citizenry itself, public opinion on the relationship between the parties and the issues, the parties’ activities as an organization, and the parties as represented by elected and appointed officials in the government. It is this later conceptualization of ‘partisanship’ that I use as an independent causal influence on public attitudes on the issue dimensions. Specifically, I use as a proxy for the partisanship in the government the party affiliation of the presidential administration. While the federal government has often been split between the Republican and Democratic Parties (divided government), the president is the most visible individual within the government, and the public expectations that presidents will be decisive during political crises and in implementing public policy is well documented. There is thus a natural implication that presidents can influence public opinion through their policy advocacy and implementation of policy through executive administration and leadership on a legislative agenda. Using the party affiliation of the president as an independent factor causing shifts in public opinion trends serves two ends, one methodological the other substantive. As a measure, it permits a statistical analysis of presidential impact on public opinion while avoiding the classic n=1 problem associated with empirical studies of and including the president. In addition to assessing presidential administrations in terms of partisanship, using survey year rather than the individual president as the unit of analysis is useful in resolving the small sample problem that presidents present. Substantively this decision is justifiable given that presidents tend to have policy positions that mostly comport with that of their parties, and even when they diverge, from the perspective of the mass public, the president and the president’s party are strongly associated. Given the well-established positions of parties on major political issues, it is reasonable to expect that citizens generally view that policy does or will reflect the policy preferences of the party that controls the White House. This, in turn, could affect the public’s preferences for policy or policy changes on that issue dimension.

The second cause I consider in this analysis is exogenous political events. I define these as “exogenous shocks” to the political environment in the sense that they are events that are not directly the product of government policy. The stock market crash in 1989, the fall of the Berlin Wall, the September 11^th attacks, these are events that can be distinguished from overt and direct policy decisions such as the 1986 tax reform bill, the Reagan Cold War defense build-up, and President Bush’s decision to invade Iraq in 2003. While policy decisions may have created a situation which precipitated an exogenous event (e.g. the limited policy response of the Clinton administration to Al Qaeda attacks in the 1990’s), or policy may be made in response to an exogenous event (e.g. the U.S. invasion of Afghanistan in the wake of the 9-11 terrorist attacks), the events are clearly separate from and thus independent of the decision making and thus direct control of the policymakers themselves. These happenings can cause dramatic responses in the average opinion and opinion distributions of the American citizenry on one or more issue dimensions. Indeed, sometimes these events precipitate an extended and relatively stable time-period where public opinion finds a new equilibrium quite distinct from the previous period.

The third independent variable in this analysis of polarization determinants is public policy itself. I test, where possible, if public opinion polarization is responsive to the actual policies on the issues. Now, for some policies this isn’t practical, as the policy itself hasn’t changed over the period, data on the policy is not available, or what is available fails to capture the essence of the policy itself. For example, in examining the abortion issue, we could look at the number of abortions each year, but is that a measure of abortion policy? Perhaps, in some indirect sense, it is. However, it could just as easily be argued that public responsiveness to abortion policy involves Supreme Court decisions, abortion legislation, and the continued aftershocks from the landmark decision in Roe v. Wade. It may be possible to measure ‘abortion policy’ and include it in an analysis, but I do not do so here, except to assess abortion-related events in the context of polarization tends. However, on other issues, there is a clearly relevant and central policy that is measurable over the time period and thus I can assess public opinion and whether it varies with actual changes in the policy. For example, consider the opinion on levels of defense spending. The actual level of defense spending is clearly the central policy output relevant to public opinion on this issue. As such, I include a measure of this actual policy output in my analysis of polarization causes.

By taking into account multiple potential causes of polarization in addition to assessing the polarization trends in a detailed rather than summary fashion, a more nuanced and substantively rich depiction of polarization in American public opinion on political issue dimensions emerges. This analysis suggests polarization is comprehensible as a rational, aggregate reaction to prime movers in public policy such as the presidential administrations, the exogenous events directly related to the relevant political issues, and to actual policy outputs produced by the federal government since the 1970’s. From the broad, ideological dimension to the relatively narrow issue of abortion, I examine polarization in the context of the policy actors taking positions, advocating for policies, and implementing policy changes. I examine polarization in relation to real-world events that have structured and shifted American public opinion—sometimes in dramatic fashion. And I look at the programs, policy-proscriptions, and declared positions offered by administrations and perceived by the public both in terms of the policies themselves and as screened through the lense of partisanship. I present a sophisticated, multivariate analysis of the determinants of political polarization.

Ideology Dispersion & Bimodality: Consistent Ideological Polarization of the American Public

As one polarization scholar argued, a key test of political polarization is the degree to which the American public ideology has become more dispersed and more coherent in the ‘two opposing camps’ sense. The increasing trend in the dispersion of the distribution of mass ideology is depicted in Figure

7.1 divided by presidential administrations over the course of the data period. There was a slight increase in dispersion through the Nixon and into the Carter years, however this declined or held steady through all of the 1980’s into George H.W. Bush’s term. With the election of Clinton the dispersion in ideology reached a new and higher plateau, but it remained steady throughout his presidency until the last two years (the run-up to the 2000 presidential election). Whether a consequence of the divisive and controversial 2000 presidential election of George W. Bush, the attacks on September 11^th, or the 2002 War in Iraq, the dispersion in ideology for the mass public increased substantially in the first decade of the 21^st century. By 2008 the dispersion in mass ideology was as high as it had ever been during the entire time series. On average the distribution of ideology at the mass level has become more spread out (more polarized), suggesting a worsening political environment for compromise and a target-rich environment for the purveyors of political and partisan conflict .The trend of increasing dispersion in the ideological distribution of the American public is consistent with the trend of increasing bimodality as shown in Figure 7.2. While the 1990’s and the Clinton administration coincided with an increasingly polarized ideological distribution relative to the previous two decades, the election of George W. Bush in

2000 was proximate to the most significant playtkurtic spike in the time series. Between 1998 and 2000 the bimodality (kurtosis) of ideology goes from the closest it ever was to the normal distribution at any point in time in the series (-0.24) to a new low, indicating that the mass public had increasingly shifted into two ideological camps. Indeed, the bimodality of ideology for the entire Bush presidency is lower than any of the points for any of the other years in the time series. Hence on two major indicators of polarization—dispersion and bimodality—the American public has become significantly and substantively more polarized on ideology. Relative to the 1970’s, the average distance between ideological identifiers is greater and the distributional shape of ideology indicates conservatives grouping with fellow conservatives and likewise for liberals. Either through a reduction in the number of moderate and slightly ideological identifiers, or a spike in the tails of the ideological distribution, there is considerably more ideological conflict in America. The war in Iraq further polarized the country, as 2002 set the high watermark of bimodality in the ideological distribution. The change in bimodality between 1998 and 2002 is quite remarkable. In the span of four years, bimodality on ideology set its global maximum and global minimum for the time series.

The most apparent intervening events between those two survey years are the election of President Bush in 2000 and the war in Iraq in 2002. The 2000 election was a highly controversial and thus potentially polarizing election. The dispute centered mostly over who, between George W. Bush and Al Gore, won the state of Florida. Initially reported on election eve as going to Gore, the final count showed George W. Bush with a narrow lead, and hence the decisive electoral votes necessary to win the presidency. Subsequent recount and challenges narrowed Bush’s lead, but it was certified by the Florida Secretary of State at 537 votes. Gore appealed to the Florida Supreme Court, which ordered a statewide recount of 70,000 rejected ballots. This recount process was ruled unconstitutional by the United States Supreme Court in violation of the equal protection clause in the constitution in a 5-4 vote. While Gore conceded following the court decision, many Democrats in and out of elected office believed

Figure 7.3: Bimodality Trend in Public Opinion on Government Spending & Services, 1972-2008

the result to be illegitimate. Adding to the potentially polarizing nature of the event was the fact that Gore won the popular vote (+543,895). The bitterness over such a hotly contested presidential election and the subsequent controversy and public spectacle of an election contest had a significant impact on the polarization of the American public on ideology.

While 2004 and subsequent survey years have shown less bimodality, each year of the Bush presidency reflects a more bimodal distribution of ideology than in any previous year since 1972. To the extent that ideology serves as a proxy for the political agenda for issues and the conflict of social groups, the evidence on dispersion and bimodality in ideology suggests a more conflictual political space and a policy arena less friendly to compromise. The evidence of increasing political polarization at the mass level is strong and robust.

Government Spending Bimodality & Dispersion: Mixed Evidence of Political Polarization

The evidence for polarization on government spending and services is not as clear as that for the ideological dimension. There is no apparent trend in the dispersion of opinion on the government role in helping citizens acquire and keep jobs. However, there is some evidence that this issue has become the subject of more political conflict within the mass electorate. Figure 7.3 depicts the trend in the bimodality of opinion on government spending since 1982. The trend is relatively stable, though there is significant volatility in the measure in the 1990’s. While 1998 was a low point in the ebb and flow if ideological kurtosis, it is the most bimodal distribution of opinion on government spending in the time series. While it recovers somewhat for the 2000 survey year, there is a fairly steady decline in kurtosis for the subsequent years of the Bush administration. However, even at its most leptokurtic point, opinion on government spending is non-normal with bimodal characteristics. The absence of a trend here should not be surprising. Whether or not and to what extent the government should provide services or the market is one of the fundamental questions of democratic politics and has likely been a

subject of division between social groups, parties, and ideological camps for both the mass public and elites since the ancients invented the word “government.”

As I noted at the beginning of the chapter, some issues lend themselves to a direct assessment of the relationship between opinion polarization and the actual public policy on the issue. The nature of government spending does provide an opportunity to assess the degree to which the bimodality of mass public opinion on government services and spending is related to actual (as opposed to perceived) levels of spending and services as well as assess the trends in opinion on government spending relative to real spending levels. It further can be suggestive as to what is causing variation in opinion on government spending as well as what spurs changes in the polarization of opinion on government spending. Figure 7.4 shows government spending as a percentage of GDP (Gross Domestic Product) from 1970 to 2008, with the fiscal years of 2009 and 2010 estimated according to budget projections.⁵⁰ The U.S. Government spending has oscillated around 35% of GDP since the 1980’s. However, note the significant uptick in 2008 followed by a substantial jump in the projected figures for the first year of the Obama administration.

Figure 7.5 converts government spending as a percentage of GDP, the average opinion on the level of government spending and services from the ANES, and the bimodality measure (kurtosis) for government spending to Z-scores based on the means and standard deviations for the data series between 1982 and 2008. This permits an analysis of changes in the polarization trends for government spending opinion and actual government spending on the same scale. Thus changes in the opinion variables are directly comparable to changes in the actual levels of government spending. The Z-Score calculation can be found in Equation 7.1:

Equation 7.1: Z-Score for Government Spending

Z_GS =

Where:

= the i^th observed value of government spending for year.

= the i^th mean value of government spending for year.

= the i^th standard deviation of government spending for year.

This allows for a direct comparison of the variables despite the fact that they are scaled using disparate units. As such the values for government spending (%GDP), average opinion on government spending, and the bimodality of government spending are translated into differences from their respective means and standardized relative to the Z-distribution (Z-scores) centered on a mean of zero and a standard deviation equal to 1. Hence the 1.14 Z-score on government spending bimodality in 1990 indicates that the kurtosis for government spending in that year was 1.14 standard deviations above the mean of government spending for the full time series. In order to match the year values of government spending with the bi-yearly ANES data on government spending opinion, I interpolate intervening values for opinion on government spending and services for the intervening years (see footnote 43 for the model of linear interpolation used in this analysis).

The trend lines in Figure 7.5 appear to be fairly independent of one another on the whole. Through the 1980’s, public opinion on government spending, both in terms of the average opinion and the bimodality of the opinion distribution, track closely with actual levels of government spending as a percentage of GDP. Interestingly, conflict in the late 1980’s over government spending was more unusual than the actual levels of spending or the opinion on levels of spending would indicate. While it is just speculation, this rise in conflict is coincident with Black Monday, the 1987 world-wide stock market crash that saw the Dow Jones Industrial Average plunge over 500 points and lose over twenty

TABLE 7.1: Deviation Models Regressing Defense Spending Levels (%GDP) on Def. Spending Public Opinion.

Model:	Intercept (S.E.)	Parameter Estimate (S.E.)		R2	N
SPENDING (MEAN)	-0.001 (0.196)	-0.049 (0.200)		.002	27
SPENDING (KURTOSIS)	-0.001 (0.195)	0.088 (0.199)		.008	27

* significant at .10 level

** significant at .05 level

***significant at .01 level

percent of its value (22.61%). The decade of the 1990’s has the most volatile shifts in opinion on government spending in addition to the trend lines moving more independently than at any other time. In terms of the average opinion on government spending, it peaks in the early part of the 1990’s but takes a precipitous drop proximate to the 1994 “Republican Revolution” and the historic shift of control of both Houses of Congress to the Republican Party. However, this trend reverses itself and by 2000, the average opinion on government spending is right at the mean (4.25) for the time series.

Opinion shifts substantially in 2008 towards increased government spending, undoubtedly in part a reaction to the financial collapse and stock market crash in the fall of 2008 in the lead-up to the November elections. The bimodality in government spending for the 1990’s reflects extreme oscillations relative to the mean, perhaps in reaction to, at first, the Clinton administration’s unpopular health care proposal in 1993 and then to the equally unpopular government shut down in 1995 that Clinton successfully (from the standpoint of public opinion) depicted as the Republicans’ responsibility. This volatility gives way to a steady bimodal trend in the Bush administration. While public opinion on government spending is apparently sensitive to partisan manipulation and exogenous events, actual government spending increased from 1994 to the present day independent of partisan control of the White House, Congress, or external events (that is, external events do not seem to have resulted in any spikes or troughs in government spending outside of the most recent economic collapse).

Table 7.1 employs the real government budget outlays (% of GDP) as an independent variable in regression models of public opinion on government spending and services. The absence of any apparent linear relationship in the three variables on government spending nor any covariance between actual spending and opinion on spending evident in Figure 7.5 is consistent with the findings (or lack thereof) in Table 7.1. The models using government spending to predict the average public opinion on spending levels and the bimodality in the distribution of opinion on government spending and services fail to explain the variance in either of those measures given standard levels of significance (.10, .05, & .01). Indeed, these linear models contribute almost no improvement over the mean model of government spending (.002 and .008 R²). While opinion on government spending does appear responsive to partisan control of government and economic issues, there is no evidence it is responsive to the actual levels of government spending itself as a percentage of GDP.