From Backwaters to Major Policymakers: Policy Polarization in the States, 1970–2014

Abstract

Political scientists often characterize state and local governments as marginal and highly constrained in policymaking. However, I suggest that in recent decades state governments have moved from the margins to the center of partisan battles over the direction of U.S. public policy. Across 16 issue areas, I investigate interstate policy variation, policy differences across states, and policy polarization, the changing relationship between party control of state government and policy outcomes. Since the 1970s, interstate variation has increased such that an individual’s tax burden, right to obtain an abortion, and other relationships to government are increasingly determined by her state of residence. Policy polarization increases dramatically after 2000 in 14 of the 16 areas. I show that party control increasingly predicts socioeconomic outcomes in the polarized area of health care, but not in the nonpolarized area of criminal justice.

In a 2011 phone call with a radio host impersonating David Koch, Wisconsin Governor Scott Walker explained that he was part of a national movement of conservative governors who “got elected to do something big” across their states.¹ Democratic governors have similarly called for coordinated efforts by Democratic state governments to oppose initiatives by the Trump administration and Republican Congress. If their rhetoric is to be believed, politicians at the state level believe they are engaged in major struggles over the direction of public policy in the United States.

Despite the contentious rhetoric, political scientists have suggested that state governments are relatively marginal policymakers. Researchers have long seen the states as “the runt in the American governmental litter,”² with policy agendas that are highly constrained by economic realities and low legislative professionalism.³ Recent research largely continues this characterization. While some studies report important changes in state policy in the polarized era,⁴ the most comprehensive recent studies in this area conclude that state policy outcomes have been generally “stable” over the years,⁵ and that party control of government still plays only a “modest” role in policy differences between states.⁶

Yet there are reasons to expect that the role of states in American federalism has expanded since the 1970s. Although there are significant obstacles to electoral accountability in the states,⁷ mass polarization and residential sorting may increase the distance between the attitudes of voters across states.⁸ Well-resourced partisan interest groups have developed new strategies in coordination and lobbying that have led to significant policy changes in the states.⁹ Moreover, polarization and gridlock at the federal level both increases incentives for policy demander groups to venue shift to the states and limits the ability of Congress to create policy to preempt or standardize state laws. Indeed, journalists describe a strengthening relationship between an individual’s state of residence and her legal right to obtain an abortion, own a firearm, join a labor union, or use drugs, as well as her tax burden, environmental regulatory regime, and generosity of the welfare state.¹⁰

Little research, however, has investigated shifts in the substance of state policy over time.¹¹ I investigate here two dynamics in policy in the U.S. states: increased policy variation (the substantive differences between states) and policy polarization (the relationship between party control and policy outcomes).¹² Importantly, I investigate the substance of policy change across 16 distinct issue areas such as abortion or tax policy.

The analyses show a large increase in policy variation and a tightening relationship between party control and policy change in recent years. Across each issue area, the range of state policies has increased. For instance, the difference between the most restrictive states for abortion and the least restrictive states has expanded since Roe v. Wade (1973). This variation is increasingly related to party control of government; prior to 2000, whether a state was controlled by Democrats or Republicans said little about the policies it would adopt, but the parties have implemented highly divergent policy agendas after 2000.

Issue area analysis shows two important areas of exception, however, where policy outcomes have not polarized: education and criminal justice. I corroborate this finding with analysis of its socioeconomic consequences. Health and welfare policy has sharply polarized in recent years, and I find that party control of state government increasingly predicts rates of health insurance coverage. However, in the non-polarized area of criminal justice, I find no change in the relationship between party control and incarceration rates.

The implications of this study suggest a growing need for research on American federalism in the age of hyper-polarization. Rather than a decentralized federalist system with vertical differences across levels and horizontal differences across regions, American governmental institutions look increasingly like a single arena of partisan combat over public policy.

The Minimalist View of States

Nearly three decades after Elazar predicted resurgent states in an emerging “neo-dualist” era of federalism,¹³ observers point to intensifying battles over public policy at the state level. However, there has been little empirical investigation of systemic policy changes in the states over time. Though scholars are now less likely to call them the “backwaters” of American politics,¹⁴ recent literature may only focus on the states as a means to increase one’s N to 50 in order to “address a domain of questions with greater statistical rigor because of the large number of states.”¹⁵

Institutional, developmental, and historical research, in contrast, engages directly with temporal dynamics in federalism and public policy.¹⁶ Overwhelmingly, however, this research takes a minimalist view of state policymaking. John Kincaid, former director of the U.S. Advisory Commission on Intergovernmental Relations and a prominent scholar of federalism, describes a twentieth century in which the role of states shrank and U.S. federalism became “more adaptable to policy preferences defined increasingly by the national government”—where the federal government moved from “senior partner” to “commanding partner.”¹⁷ By 1975, even the predominant federalism scholar William Riker suggested that the existence of lower levels of government “makes no particular difference for public policy.”¹⁸

Additional research lends credence to the minimalist view by highlighting the constraints that face lower levels of government in federalism. Fiscal federalism implies that the threat of exit from businesses and wealthy residents exerts downward pressure on taxation, redistribution, and regulation,¹⁹ which reduces the potential for variation across states.²⁰ Fiscal federalism implies that state governments have little policy discretion compared to the federal government. They face a greater threat of exit, and with no ability to manipulate a floating currency, they face economic forces beyond their control and greater pressure to balance budgets.²¹

State legislators also lack the policymaking resources of members of Congress.²² Lower salaries increase the incentive to spend time earning money outside of their political offices, and fewer staff limits the ability to research and draft legislation. Even if state legislators face equivalent pressures from voters and interest groups as members of Congress, we would expect those in state capitals to be less productive due to these resource constraints.

Despite these constraints, however, roll-call voting in state legislatures has polarized in recent years.²³ Whether the prior cause of polarization stems from voters, interest groups, or politicians themselves, greater polarization implies greater distance between the policy preferences of Democrats and Republicans,²⁴ and thus increasing polarization of policy outcomes in the states. Yet the most comprehensive studies of state policy polarization over time, those of Caughey and Warshaw and Caughey, Xu, and Warshaw , again conclude in favor of the minimalist view of state policy. While “Democrats and Republicans may disagree consistently and even violently,” Caughey, Xu, and Warshaw conclude that “the actual policy consequences of these disagreements are far less dramatic.”²⁵ The increasingly partisan and ideologically consistent rhetoric of Democratic and Republican governors and state legislators is just that—talk, with little consequence for public policy.²⁶

My empirical analysis challenges this line of research. I turn to measurement in a later section, but here I describe theoretical reasons that we might expect, contra the minimalist view, substantial policy polarization in the states.

Polarized Federalism

The minimalist view highlights the constraints on state policymaking and leads us to expect, at most, modest increases in state policy variation and polarization in recent years. But there are a number of theories that predict major changes in state policy—and major growth of state policy polarization. Popular explanations for congressional polarization, voters and interest groups may drive polarization in the states. There are also institutional reasons why polarization at the upper level of a federalist system can generate polarization at lower levels. This section briefly outlines these sets of theories.

Theories of voters and public opinion. One set of explanations focuses on changes in the preferences of the median voter. Policy differences across states are likely to increase if the preferences of states’ median voters are diverging. Some research finds evidence of polarizing ideological and policy attitudes in the mass public.²⁷ An additional plausible mechanism for diverging median voters is geographic sorting. Bishop argues that Americans have increasingly opted to live in communities that tend to share their political views.²⁸

However, other research suggests that elected officials in the states are unlikely to be responsive to mass opinion. The electoral connection in the states may be weak because voters pay little attention to state politics,²⁹ and the precipitous decline of state politics journalism may make policy even less “traceable” for voters.³⁰ As a whole, there is little electoral accountability for state legislators who engage in behavior that is “out of step” with their constituents.³¹

Theories of interest groups. There are significant barriers to translating mass preferences into state policy, but more concentrated and well-resourced actors may be more influential. A second set of theories involves groups of intense policy demanders. A classic literature argued that concentrated and elite interests are advantaged at lower levels of government,³² which diffuse and mass interests can counter by “extending conflict” to higher levels.³³ Relative to voters, who are often cross-pressured and inconsistent,³⁴ concentrated interests and organized activist groups are likely to have intense and consistent preferences³⁵—which, to the extent they are implemented, increase interstate policy variation.

Arguing that well-resourced organizational networks have increased their investments in state politics, recent studies harken back to this classic literature. Organizations’ investments in lobbying and the provision of “model bills” to state legislators appear highly effective in shaping state policy outcomes.³⁶ These organized investments in state politics may be especially influential because of groups’ political advantages over voters. Voters are mostly immobile, but well-resourced organizations can make coordinated political investments across many states, and more strategically “venue shift” to favorable institutional arenas.³⁷ Moreover, state legislators face considerable informational and human resource constraints relative to members of Congress,³⁸ which potentially increases the effectiveness of groups’ investments in informational lobbying.³⁹

Not only do groups appear more involved in state politics generally, but the groups themselves are more partisan than a generation ago. Midcentury pluralists described groups that maintained running partnerships with both parties and party-group coalitions that reshuffled depending on the agenda item;⁴⁰ anti-pluralists discussed “iron triangles” and other relationships in which powerful groups captured bipartisan committees and agencies in order to extract rents.⁴¹ Although some groups continue to partner with both major parties, contemporary research suggests that groups now tend to be much more aligned with a single party.⁴² The growing partisanship of the interest group environment may be especially extreme at the state level: Many of the groups that remain bipartisan are concentrated in the foreign policy and national security arenas, a policy area that is largely confined to national level.

Theories of institutional incentives. Research in “new institutionalism” and formal theory has shown that institutions structure and influence preferences and incentives, and that this is especially true of institutions associated with federalism.⁴³ Specifically, it is plausible that polarization at the national level can generate polarization at lower levels of government in a federalist system.

This theory also gives a prominent role to policy demander groups. National level polarization may increase state policy polarization because federalism may serve as a “safety valve” for policy demanders who are stymied in Washington, and this safety valve grows more valuable as polarization increases. Polarization reduces the ability of the minority party and their aligned policy demanders to influence or extract compromises from the majority party, increasing the relative benefit of shifting their focus to the states. Frustrated climate activists may turn their hopes to the states,⁴⁴ as might organized labor,⁴⁵ LGBT rights activists,⁴⁶ or antistatist and business interests.⁴⁷

Polarization and divided government have produced gridlock and a slowdown of national policy production,⁴⁸ again generating incentives for groups to venue shift to the states.⁴⁹ Since the 1970s, polarization has increased in Congress and divided federal government has become a more frequent occurrence.⁵⁰

This has led to policy gridlock⁵¹ and “drift”⁵²—and higher costs of national policy change for policy demanders. Faced with federal gridlock, policy demanders may increasingly turn to states to implement their agendas. Federal gridlock also means that these policy demanding groups can be more confident than in earlier periods that their state policy victories will not soon be reversed by federal legislation or court rulings.⁵³

In the rarer moments when important federal policy does pass, polarization and divided government may increase incentives for members of Congress to delegate authority to the states.⁵⁴ A legislator who would ideally implement his or her ideal policy across all 50 states may accept a decentralized policy as a second-best option if it moves the average outcome (such as the policy regime for the average state or average individual) toward his or her ideal. Moreover, the district-based electoral connection in Congress can improve the relative appeal of the second-best option because “representatives know that when they delegate to state and local agents, policy for their constituents will be set by representatives elected by those same constituents.”⁵⁵ Indeed, the rise of polarization in Congress has coincided with what scholars call a “devolution revolution.”⁵⁶ In a similar fashion, the federal judiciary has undergone a “federalism revolution” in which the courts are an increasingly “state friendly arena”⁵⁷ precisely during an era of increasingly partisan and narrow (5–4 split) decisions.⁵⁸

Measuring Policy Outcomes

There are strong reasons to expect increased policy variation and polarization in the states, but to what extent does it occur in recent decades? Do policy outcomes diverge over time, and is this divergence related to party control of government? In this section, I describe my strategy to measure policy outcomes and estimate the changing relationship between party control and policy.

This study employs the most comprehensive dataset of state policy outcomes since 1970. ⁵⁹ To build it, I collect data on 35 policies, to which I add data from Jordan and Grossmann, Caughey and Warshaw, and Boehmke and Skinner to create a dataset of 135 policies.⁶⁰ (I also extend years of coverage for 16 policies from the other datasets.) Caughey and Warshaw provide a detailed description of many of the policies, which can be binary (e.g., Right to Work laws), ordinal (e.g., mandatory parental notification or consent for a minor’s abortion), or continuous (e.g., marginal tax rate on high incomes).⁶¹ Table 1 lists the policies, and I provide descriptions and sources for each policy in the online appendix.

Table 1 State policies (1970–2014)

Note: Issue area categories in bold.

The data I collect covers policies of considerable importance. They include voter ID laws, state capital gains taxes, as well as various regulations related to public sector unions, abortion rights and coverage, campaign finance, and immigrant workers.⁶² Of particular note is my data collection of criminal justice policies. Although some research focuses specifically on dynamics in criminal justice policy,⁶³ research that summarizes policy across issue areas has largely neglected incarceration.⁶⁴ I collect data on laws that criminal justice research considers central to the rise of mass incarceration: truth-in-sentencing laws, which require individuals to serve a minimum percentage of their original sentence; three strikes laws, which increase penalties for an individual’s third felony; and determinate sentencing laws, which specify mandatory minimum sentences.⁶⁵

To measure party control of government, I use variables that indicate whether a state is under unified Democratic control, unified Republican control, or divided control.⁶⁶ While control of the executive branch or one or more legislative chamber may have an independent or partial effect on policy outcomes,⁶⁷ the polarized federalism theory focuses on unified control because polarization and divided government interact to produce gridlock.⁶⁸ I provide additional models with measures of control of the governorship, lower house, and upper house in the online appendix.

Key to the analyses is the comparison of the party-policy relationship across time. Because policy change is rare compared to other political dynamics, estimating a completely dynamic party effect (i.e., by year) is difficult. Precision and clarity are greatly improved by estimating an average party effect for different eras that span multiple years.⁶⁹ I compare the association between party control and policy change during two eras: the 1970–1999 period and the 2000–2014 period.⁷⁰ In practice, this entails interacting the party control variable with a dummy variable for the 2000–2014 period to estimate the marginal effect of party control on policy change during the different eras.

Unidimensional Measures

Political scientists often summarize public opinion, legislative votes, and more recently, policy outcomes on a unidimensional left-right dimension. Recent unidimensional policy measures provide a summary of the ideological content of policy on a dimension typically described as “policy liberalism” or “the role of government.”⁷¹

As a first cut at the data, I estimate policy variation and polarization with four unidimensional left-right measures of policy outcomes. The first is the State Policy Liberalism (SPL) measure from Caughey and Warshaw (2016), a set of state-year policy ideal points generated from a dynamic Bayesian IRT model. Second, I estimate the same ideal point model with my expanded policy dataset to produce an Expanded SPL measure. The third and fourth measures are Substantive Scales, simple additive indices (averages) that are the sum of a state’s liberal policies minus its conservative policies in a given year. These measures serve as expert-coded alternatives to the Bayesian IRT latent dimension estimates and are analogous to the “Policy” measure from Erikson, MacKuen, and Stimson.⁷² One of the additive indices weights policies equally, while the other is the average of issue area-specific indices. (Subsequent sections address how the ideological direction of policies is determined.) All measures are normalized to a range between 0 and 1.

I calculate two measures of policy variation with these unidimensional scales: the range and the standard deviation of policy ideal points across states in each year. I plot yearly estimates from 1970 to 2014 in appendix figure 5. The spread of ideal points widens greatly since the 1970s. The range and standard deviation estimates are remarkably similar across the measures. The range of ideal points is at least a third larger in the 2010s than in the 1970s and 80s, and the standard deviation is at least two-thirds larger.⁷³

These measures suggest that policy polarization has similarly increased. Using dynamic panel regressions, appendix figure 6 plots the marginal effect of unified party control of government on change in ideal points for the 1970–1999 period and the 2000–2014 period. All of the estimates show at least a twofold increase in the magnitude of the relationship between party control and policy ideal points.⁷⁴

The expanding variation and polarization evident in the unidimensional analysis motivates the investigation of issue-specific policy dynamics. Unidimensional ideal points serve as strong summary measures, but generally, they may create obstacles to inference by obscuring multidimensional variation or conflating extremism and consistency,⁷⁵ and they rely on relatively strong assumptions about the comparability of policies across issue domains. More importantly, it is difficult to draw conclusions about the substantive content of policy—its effect on members of the polity—from unidimensional ideal point estimates. Policy scholars may be interested in more specific temporal dynamics in residents’ relationship to government. Are state abortion laws more or less restrictive? In which direction have state tax rates, restrictions on campaign contributions, and the generosity of welfare benefits moved in recent decades?

Additionally, although they may be advantageous in the study of roll-call votes,⁷⁶ there are two reasons to prefer straightforward additive indices over latent dimension estimates (e.g., factor analysis or Bayesian IRT) for the measurement of policy outcomes. First, historical, normative, and policy scholarship provides clear priors about the ideological content of policy. Empirically deriving model parameters (the ideological content of policy) from the data rests on the joint assumption that (a) liberal states are liberal because they pass liberal policies, and (b) that liberal policies are liberal because liberal states pass them. When this assumption is violated historically (e.g., during the 1960s and 1970s conservative Southern states were early adopters of liberal abortion laws), the model may produce parameters that do not conform to substantive understandings about the ideological content of policy. In a separate analysis, I find that parameters for some relatively inconsequential policies (e.g., mandatory registration of beer keg rentals) are larger than those of more important policies, and that some similar laws have parameters that point in opposite directions.⁷⁷ Second, the real world consequences of substantive policy is, for the most part, additive. Whereas latent dimension estimates rely on the correlations between policy items to provide “relative” measures of policy outcomes,⁷⁸ averages can provide absolute measures of policy outcomes.

Policy Indices by Issue Area

Issue area measures provide a clearer picture of historical changes in policy substance. Although many studies have employed summary measures of policy outcomes in a single issue area,⁷⁹ mine is the first to compare across many issue area indices. I group the policies into 16 discrete issue areas: abortion, campaign finance, civil rights and liberties, criminal justice, drug policy, education, environment, gun control, health and welfare, housing and transportation, immigration, labor (private sector), labor (public sector), LGBT rights, taxes, and voting.

In each area, I calculate a simple substantive measure of average policy outcomes: the number of liberal policies minus the number of conservative policies.⁸⁰ Because policies can be binary (e.g., medical marijuana laws), ordinal (e.g., voter ID laws, which can be strict or non-strict), or continuous (e.g., minimum wage level), I normalize each policy to range from 0 to 1. A binary policy, which a state either has or does not have, takes on the values of 0 or 1, whereas an ordinal or continuous policy, such as a tax or minimum wage, is transformed to the [0, 1] scale. A state’s score in an issue area index is therefore the sum of the liberal policies minus the sum of the conservative policies.

This kind of measure relies on three assumptions: first, the ideological “direction” of policy (whether it is liberal, conservative, or neither); second, that policies are of equal substantive importance; and third, that the direction and importance remain constant over time. These assumptions are unlikely to be satisfied in practice, especially equality of substantive importance.⁸¹ However, I argue that these simple index measures strike a balance between agnosticism, precision, transparency, risk of bias, and substantive interpretability.

Determining the ideological direction of more than 130 policies is a difficult task. The primary left-right ideological dimension, or “what goes with what” has changed over time, but for the most part political observers characterize policies on the left to be those that 1) expand the use of state power for economic regulation and redistribution,⁸² or to increase or protect the rights of historically marginalized groups in society (black Americans and other nonwhite racial groups, women, LGBT individuals, immigrants, and religious minorities);⁸³ and 2) restrict the use of state power for the punishment of deviant social behavior.⁸⁴ Policies on the right do the opposite.⁸⁵ Although there is considerable nuance throughout political and intellectual history, in short, left policies promote social libertarianism and economic interventionism, while right policies promote traditional (incumbent) social values and oppose state intervention in markets.

Yet even with this large body of historical and normative scholarship, there is still no objective, unifying test of whether a certain moral principle, political action, or legal statute is on the left or right. Many scholars argue that the first dimension of politics represents the “size of government,”⁸⁶ but this is not always the case. For instance, policies that expand rights and protections for black Americans, which are understood to be liberal, can involve expansions of state power (e.g., anti-lynching laws) or restrictions on state power (e.g., laws that reduce prison sentences). The same is true of abortion laws, where Medicaid coverage of abortion and bans on “partial birth abortion” both involve greater state intervention, but are quite ideologically distinct. It is thus no surprise that there is an ongoing debate about whether the clustering of policies along partisan and ideological lines is due to “natural” ideological or psychological principles,⁸⁷ or whether they are the products of idiosyncratic historical coalition partnerships between interests in society that over time became path-dependent.⁸⁸

I argue that an issue-specific left-right conceptualization can improve inference for studies of policy dynamics. Rather than assuming that issues “go together” in unidimensional space, table 2 shows conceptual dimensions that determine the ideological direction of policies within each issue area.⁸⁹ The left-right dimension for abortion policy, for example, represents the legality and costs (broadly defined) of obtaining an abortion. Other issue areas represent multiple related concepts. Tax policy, for example, is comprised of two concepts: absolute rates and progressivity (the distribution of marginal rates across income levels), and health and welfare policy is comprised of both benefit levels and the strictness of eligibility. I base a policy’s direction—left, right, or, in a small number of cases, neither—on its expected effect on the issue-specific dimension. This issue-specific conceptualization also helps to avoid the problem of sorting and shifts over time regarding which issues “go together” on a single left-right dimension. While the cluster of issues on the left and right has shifted over the nineteenth and twentieth centuries,⁹⁰ issue-specific assessments (e.g., whether a policy restricts or broadens access to abortion) have largely remained constant.⁹¹

Table 2 Ideological content of policy issue areas

Interstate Policy Variation

In this section I estimate change in state policy since 1970. Figure 1 plots each issue area policy index. The grey lines represent the policy outcomes for each individual state over time.

Figure 1 Issue area scales by party

States’ policy outcomes within each issue area (the grey lines) diverge greatly over time; this represents increased overall variation in state policy outcomes in each area. Compared to the 1970s, the policy regime under which an individual lives is increasingly determined by her state of residence. For instance:

• • Abortion: In 1973, states only differed in Medicaid coverage for abortion and other minor regulations. By 2014, the most restrictive states mandate waiting periods, parental notification, counseling, licensed physicians, a 20-week gestation limit, and restricted insurance coverage for abortion.

• • Environment: In 1970, the greenest states had state EPAs and endangered-species laws. By 2014, they had strict regulations of greenhouse gas emissions for cars and utilities, solar tax credits, and a plethora of recycling programs.

• • Gun Control: In 1970, the least strict states allowed open carry and the strictest states required dealer licenses and purchaser background checks. By 2014, the least strict states had added Stand Your Ground laws, while the strictest states banned assault weapons and mandated registration and waiting periods for purchases.

• • Health and Welfare: In 1970,states varied in AFDC benefits and Medicaid adoption. By 2014, Massachusetts offered generous Temporary Assistance for Needy Families (TANF) and SCHIP benefits and had expanded Medicaid, while Alabama did not expand Medicaid, requires drug tests for public benefits, and requires a monthly income below $268 for a family of three to qualify for TANF.⁹²

• • Immigration: In 1970, states mostly varied in laws establishing English as official state language, and all legal immigrants were eligible for public welfare and health programs. By 2014, only some states provide public benefits to new legal immigrants.⁹³ Some states provided in-state tuition for undocumented college students, driver’s licenses for undocumented immigrants, and banned the use of e-verify for employment, while other states require all employers to use it.

• • Taxes: In 1970, some states had no income or capital gains taxes, while the highest tax state, Vermont, had a 5.54% top capital gains rate and 14.88% top income rate. By 2014, many states continued to collect no income or investment taxes, but California had a 14.1% top capital gains rate and a 14.1% top income rate.

Some areas, such as environmental policy, become more liberal over time on average. All of the major policies in this area increase environmental regulation or public spending in pursuit of environmental quality, and the most conservative states on the environment simply do not pass the major environmental laws that the “green” states do. Abortion policy, in contrast, tracks more conservatively since Roe v. Wade (1973). A few states become more liberal on abortion over time as they pass laws to provide Medicaid coverage for abortion and over-the-counter emergency contraception. This liberal trend, however, is swamped by the spread of abortion restrictions in states, such as mandatory parental notice for minors and bans on “partial birth abortion.” Though not included in this analysis, prior research finds similar dynamics for Targeted Regulation of Abortion Provider (TRAP) laws, which “single out abortion providers and impose on them requirements and regulations that are excessive and more stringent than those imposed on other medical practitioners.”⁹⁴ A third set of issue areas, such as immigration and labor, sees similar growth in variation, but does not become more liberal or conservative on average since the 1970s.

Each issue area shows growing policy variation across states, but they also show partisan policy polarization: Policy outcomes in Republican states are more distant from those in Democratic states. In particular, figure 1 shows the correlation between party control and policy outcomes in each area (with the blue lines representing unified Democratic states, the red lines representing unified Republican states, and the green line representing divided states). There are two issue areas that do not fit this pattern, where increased overall variation appears nonpartisan: criminal justice and education. The averages of Republican, Democratic, and divided states in figure 1, however, are simple correlations, so the growing policy divergence by party control could be simple sorting—states with conservative policies becoming Republican and states with liberal policies becoming Democratic. To test the changing relationship between party control and policy change, in contrast, I estimate dynamic panel regressions and compare the marginal effect of party control on policy outcomes for the 1970–1999 period and the 2000–2014 period. Figure 2 plots these results.

Figure 2 Party effect on issue area scales

Partisan Policy Polarization

Figure 2, which tests the relationship between party control and policy change, corroborates the correlations shown in figure 1.⁹⁵ Again, in 14 of the 16 issue areas, the party effect polarizes after 1999: There is a greater difference in the effect of unified Democratic control relative to that of unified Republican control in the 2000–2014 period than in the 1970–1999 period. The amount of polarization depends on the partisanship of policy—that is, whether, for instance, Democratic states increase taxes relative to Republican states. But the overall amount of policy activity in a given area matters. For example, states become less active on civil rights and liberties as time progresses, but more active in areas like drug policy, LGBT rights, and voting rights (refer to figure 9 in the online appendix for counts of policy changes).

Figure 2 shows that party control is no better at predicting policy change in criminal justice or education in recent years. Both before and after 2000, party control does not predict change in criminal justice policies. States controlled by Democrats pass punitive and liberal criminal justice policies at similar rates to divided and Republican states. In both eras, states controlled by Democrats are slightly more likely to pass liberal education policies (e.g., increase spending in K–12 or higher education) and less likely to pass school choice, voucher, and charter laws. However, party control becomes slightly less predictive of education policy changes after 2000. In both of these issue areas, the static or decreasing predictiveness of party control stands in contrast to the other 14 issue areas in which party control increasingly explains policy change.

But does this policy polarization matter for the lives of these states’ residents? Does it matter for socioeconomic outcomes that there is polarization in 14 issue areas, such as tax and health policy, but non-polarization in criminal justice and education? In the next section I provide evidence that it does. In the polarized area of health policy, party control of state government increasingly predicts rates of health coverage. In the non-polarized area of criminal justice, however, party control does not increasingly predict rates of incarceration (overall or among black residents).

The Socioeconomic Consequences of Policy Polarization

The polarization of policy carries major socioeconomic consequences for residents. In the polarized areas of health and environmental policy, party control of state government increasingly predicts rates of health coverage and carbon intensity of a state’s energy supply, respectively. In the non-polarized areas of criminal justice and education, however, party control does not increasingly predict rates of incarceration (overall or among black residents) or graduation rates, respectively. In this section, I focus in depth on health and criminal justice policy. Analysis of socioeconomic outcomes in education and environmental policy is provided in the online appendix.

The health policy agendas of the national Democratic and Republican parties have been distinct since at least the 1930s. Health policy in the states has been similarly polarized for decades, as Democratic states tended to have more generous Medicaid eligibility and benefits. As the role of states in health policy expanded with the development of state prescription drug benefits for seniors, as well as federal grants for the State Children’s Health Program (1998) and Medicaid expansion under the Affordable Care Act (2014), state health policies increasingly varied—and this variation was increasingly related to party control of government.

Socioeconomic outcomes related to health policy polarized accordingly. Figure 3 shows the relationship between party control and the uninsured rate. Plot A displays state uninsured rates (the grey lines) and the average Republican (red), Democratic (blue), and divided (green) state from 1987 through 2014. Plot B shows the marginal effect of party control for the 1987–1999 and 2000–2014 periods from different time-series regression models.

Figure 3 Party control and health insurance coverage

A. Correlation

B. Marginal effect of party control

In both the correlation and the regressions, party control of government is increasingly associated with health insurance coverage in more recent years. Whereas prior to 2000, party control does not predict change in the uninsured rate, after 2000 unified Republican control is associated with a 0.75 percentage-point increase in the uninsured rate and unified Democratic control is associated with a 0.75 percentage-point decrease in the uninsured rate. These differences in coverage are of considerable social consequence. Health policy scholars, for instance, “estimate the number of deaths attributable to the lack of Medicaid expansion in opt-out states at between 7,115 and 17,104.”⁹⁶

In contrast, education and criminal justice policies are—uniquely—non-polarized. In education, Democratic state governments pass school choice and charter school laws, and spend at similar rates to Republican state governments.⁹⁷ In criminal justice, Democratic and Republican states both instituted “tough on crime” laws that led to mass incarceration. The lack of polarization in these areas relative to others has largely reflected the positions of the national Democratic and Republican parties, and a substantial literature describes the bipartisan history of policymaking in these areas.⁹⁸

Mass incarceration—the internationally unprecedented number and proportion of Americans, disproportionately black, under correctional control—has drawn increasing scholarly attention with respect to its origins⁹⁹ and consequences.¹⁰⁰ Mass incarceration is in large part the result of changes in law and bureaucracy in the U.S. states. Of the powers reserved to the states in the 10th Amendment of the U.S. Constitution, police powers are the most prominent and likely the most socially consequential. State and local agencies account for the overwhelming majority of law enforcement, and the federal prison system houses less than six percent of the U.S. incarcerated population. There is new but limited research focusing on the interaction of mass incarceration and federalism.¹⁰¹

Despite the social importance and comparative punitiveness of American criminal justice policy, its politics has been mostly bipartisan as the parties compete to be perceived as “tough on crime.” Weaver discusses how after 1968 “even liberal Democrats did not talk about civil rights without deploring crime.”¹⁰² Alexander places responsibility on not only the Republican Party, but also on Democrats, for adopting “tough on crime” policies, especially during the 1990s.¹⁰³ As shown in table 3, punitive criminal justice policy has not polarized in the states. An exception is the repeal of the death penalty; five Democratic states repealed the death penalty between 2000 and 2014.¹⁰⁴

Table 3 Criminal justice policies by party control

Note: Democratic and divided state governments passed more punitive criminal justice policies than did Republican governments, though removals of the death penalty mostly occurred in Democratic states. Numbers in parentheses represent repeals.

Yates and Fording find a significant association between Republican control of government and incarceration rates for white and especially for black people between 1978 and 1995,¹⁰⁵ and I similarly find a statistically significant effect of unified Republican government for the 1978–1999 period.¹⁰⁶ The substantive effect, however, is modest and inconsistent across models: The two-way fixed-effect model (the least strict test) shows an increased incarceration rate of about 30 people per 100,000 residents, but the other models show no effect (refer to Plot B in figure 4). An increase in a state incarceration rate of 30 individuals per 100,000 residents is substantively minuscule in a society in which one in 36 adults is under correctional jurisdiction.¹⁰⁷

Figure 4 Party control and incarceration

A. Correlation

B. Marginal effect of party control

More importantly, there is no evidence of a polarization of incarceration rates by party across time. This decreased effect of Republican control in the post-2000 period is a stark contrast to the rapid polarization in other policy areas.

I also provide estimates of the relationship between party control and the incarceration rate for black people in the online appendix, figure 10. Even more than for the overall incarceration rate, the black incarceration rate becomes less polarized after 2000. For most models prior to 2000, Republican control is associated with an increase in the black incarceration rate of about 100 per 100,000 residents, but the party differences decrease after 2000.

Overall and black incarceration rates do not appear to polarize in the states, but recent years have seen growing partisan conflict over the use of private prisons.¹⁰⁸ The use of private prisons may be more polarized than overall incarceration because it may generate conflict not only over crime and punishment concerns, but also over profit incentives for punitiveness, reports of inhumane conditions, and the fundamental role of the state and the social contract.¹⁰⁹ I estimate the relationship between party control and the percent of inmates who are housed in privately owned facilities in the online appendix, figure 11, but only beginning in 1999 due to a lack of available data. Analogous bivariate and panel regression analyses suggest a modest relationship between party control and private prisons. After 2010, Democratic states have significantly lower proportions of inmates in private facilities. However, the panel regressions show at most a small effect of party control (less than 1%), which is only statistically significant in the model employing the first differenced dependent variable (not the two-way fixed effects or lagged models).

Health policy and criminal justice are substantively important and illustrative cases in which major socioeconomic outcomes are polarized to the extent that relevant policies are polarized. This pattern generalizes further. Like diverging tax rates, state governmental revenue and spending have polarized over time, with Democratic control predicting greater increases relative to Republican and divided states in recent years. In the polarized environmental policy area, carbon efficiency is also predicted by party control of government (refer to the online appendix, figure 12). In contrast, non-polarized policy in the issue area of education, like criminal justice, appears to be associated with non-polarized socioeconomic outcomes: Party control does not predict high school graduation rates any more in recent years than it does in earlier years (refer to the online appendix, figure 13),¹¹⁰ and it barely predicts charter school enrollment (refer to the online appendix, figure 14).

State Resurgence

There are strong historical and theoretical reasons to expect state governments to be marginal players in American policymaking. Compared to the federal government, states face greater threat of exit from business and wealthy residents. Their legislatures are poorer in terms of the time, money, and information required to change policy. Major interstate differences in policy, such as the legality of racial segregation or gender discrimination in employment, have been washed away by landmark federal policies. Yet this minimalist characterization of states has grown antiquated.

While the federal government grew more gridlocked, states implemented major policies that shape the lives of their residents. Federal laws from the 1930s through 1970s decreased interstate variation in many issue areas.¹¹¹ Since 1970, in contrast, interstate variation increased as some states implemented restrictions on guns, abortion, labor unions, welfare, and voter eligibility, while others loosened restrictions. Moreover, some of the most significant recent federal policies have served to increase interstate variation rather than decrease it. In addition to welfare devolution in 1996,¹¹² the Supreme Court ruling in NFIB v. Sebelius (2012) gave states great discretion in the implementation of the Affordable Care Act, the choice of whether to expand Medicaid and create a state-run health insurance marketplace.¹¹³

There are notable exceptions where Congress and the federal courts have decreased variation in state law, however. In a famous example of “coercive federalism,” the National Minimum Legal Drinking Act of 1984 threatened to withhold federal highway grants from states that did not increase their drinking age to 21. The area of LBGT rights is also prominent. Lawrence v. Texas (2003) invalidated state sodomy bans. Though not included here because it occurred after 2014, the Obergefell v. Hodges (2015) case legalized same-sex marriage by invalidating state marriage bans.

The upward trend in interstate policy variation is not inevitable, however. In 2017, the federal government came under unified Republican control. The Trump administration has signaled a desire to act against state and local immigration and marijuana policies. Recent decades saw the buildup of considerable interstate policy variation, but an aggressive federal government may move the center of policymaking in American federalism back to the national government. Further research should investigate interbranch conflict in the polarized era.

Supplementary Materials

The below supplementary materials are available at https://doi.org/10.1017/S153759271700425X:

• • Polarized Federalism and the Size of Government

• • Party Control of Government

• • Policy Productivity

• • Party Control and Criminal Justice Outcomes

• • Party Control and Environmental Outcomes

• • Party Control and Education Outcomes

• • Additional Regression Specifications

• • Alternative Temporal Breakpoints

• • Policy Data