Identifying Shifts in Policy Regimes: Cluster and Interrupted Time-Series Analyses of U.S. Income Taxes

One of the mainstays of political sociology and political science for generations has been debate about the conditions under which one policy regime—that is, a distinct mix of public policies—is replaced by another. For instance, a vast literature seeks to identify the factors that caused the shift from orthodox to Keynesian policy regimes in the United States and Europe after the Second World War, as well as the shift during the 1980s to more conservative neoliberal policy regimes (e.g., Campbell 1998; Gourevitch 1986; Hall 1989, 1992; Weir and Skocpol 1985). Similarly, scholars have argued about whether the process through which regime shifts occur is a slow and incremental one, driven by bureaucratic inertia, muddling through, and path-dependent constraints, or a rapid and abrupt one, sparked by cataclysmic events like war that trigger sharp breaks with past policies (e.g., Baumgartner and Jones 1993; Hall 1993; Krasner 1984; Lindblom 1959; Pierson 1993, 1994). Regime shifts have also captured the imagination of historical sociologists who have recognized that history is marked by critical turning points that differentiate among relatively stable time periods and that this requires scholars to carefully identify historically specific patterns among variables (Abbott 1988, 1992, 1997; Isaac 1997).

Unfortunately, such care is not always taken. Often absent from the debates about shifting policy regimes is a sustained methodological discussion about how best to specify the dependent variable—that is, the policy regimes themselves—and, in turn, the shifts from one regime to another. Several problems result. To begin with, the criteria for identifying regimes and regime shifts often remain vague, especially when researchers use intensive case studies.¹ For example, scholars frequently differentiate among tax regimes on the basis of qualitative characteristics such as the economic growth or budgetary strategies upon which policymakers base a set of tax policies (e.g., Martin 1991; Stein 1990). Because of the fluidity of these strategies over time, these researchers admit that it is sometimes hard for them to identify clear boundaries between regimes (Martin 1991: 16–17), and, as a result, even the most astute observers have difficulty deciding when regime shifts have occurred (e.g., Stein 1996: 196–201).

However, problems remain even when scholars specify more precise criteria for differentiating among regimes. Those who adopt relatively simple, unidimensional indicators of a regime or who track indicators over relatively brief periods of time may detect regime shifts, where others with more complex and historically sensitive indicators may not. Hence, observers of the Reagan administration’s welfare reforms concluded from changes in budget allocations during the early 1980s that a new and fundamentally more austere welfare policy regime had begun, whereas those who adopted a more multidimensional approach that examined budget allocations as well as structural changes in welfare programs across the entire decade disagreed (e.g., Piven and Cloward 1982; Pierson 1994: 13–17). This sort of problem may be particularly acute in statistical analyses of time-series data because scholars who use these techniques generally identify shifts from one policy regime to another by focusing on changes in the regression models that explain a single outcome variable—such as fluctuations in average tax rates or tax progressivity, when they are examining shifting tax policy regimes (e.g., Allen and Campbell 1994; Campbell and Allen 1994), or strike rates, when they are investigating shifting labor policy regimes (e.g., Isaac and Griffin 1989). The point is that, when trying to identify shifts in policy regimes, it is important to carefully specify all the important dimensions that characterize a regime and track them simultaneously over as long a period of time as possible.

The purpose of this essay is to demonstrate how this can be done more rigorously by using two quantitative techniques: cluster analysis and interrupted...