“Hybrid regimes” – part democracy, part autocracy – have had something of a moment in political science in recent years. The popularity of the concept is understandable, given the multiple ways that democracies seem to be backsliding. Nevertheless, the hybrid category presents some real conceptual challenges, since two such regimes can exhibit an entirely different mix of attributes from democracy and autocracy. As it happens, this mismatch is a common problem with multidimensional concepts such as democracy. The simple solution to the problem – disaggregating into component dimensions – solves a number of analytic problems and reinforces the meaning of the higher-order concept. Disaggregating is perhaps implicit in some of the strategies stressed elsewhere in this volume regarding conceptualization across contexts and the use of broader and narrower variants of concepts.Footnote ¹ Still, the particular problem of working with multidimensional concepts deserves special focus.

It merits emphasis that building complex concepts is a common and valuable strategy of conventional social science. An interesting concept, such as “populism,” might be interesting precisely because it collects a colorful “syndrome” of multiple, uncorrelated characteristics. Aggregation is likewise seen in standard scaling methods that bring together a number of component attributes to measure a given concept. Integrating disparate components is an understandable measurement approach, since scholars want to represent the full meaning of the concept in their instrument. But if these components are orthogonal to one another, cases can be co-classified but still be quite different from one another. The intervention here is to highlight the value gained from moving in the opposite direction – disaggregation.

Awkward Cases

We have more good concepts in political science than is sometimes appreciated. Yet good concepts do not always yield good analysis. What do I mean by “good” concepts? These are concepts for which there is substantial consensus over the key underlying dimensions and for which serious efforts at measurement and validation of measures have been undertaken. But at the same time there are signs that the concepts are not quite ready for theory-building prime time. The problem may be the lumping of dissimilar cases into shared categories or that good measurement may occur over only part of the distribution of cases.

A traditional reaction to this state of affairs might be to abandon the concept as hopelessly multifaceted or to reconceptualize it in some foundational way. But there is often a better strategy, especially keeping in mind we are focused on concepts around which there is substantial consensus and resonance – which likely exists for good reason. However, here is where disaggregation can come in – a conceptual tool strikingly exemplified with the publication of Ruth Berins Collier and David Collier’s (Reference Collier and Collier1979) work on corporatism.

How does this work? The Colliers did not begin with a disagreement about what corporatism was: “a non-pluralist system of group representation … [through] a limited number of officially recognized, non-competing, state-supervised groups” (Reference Collier and Collier1979: 968). The problem, however, was that to one degree or another, corporative-style institutions had been imposed in myriad, often politically very dissimilar, contexts (D. Collier and R. B. Collier Reference Collier, Collier and Malloy1977). The consequence, they contended, was “that the concept of corporatism may apply to so many different cases that it often tells one little or nothing” (Reference Collier and Collier1979: 968).

Their response was to disaggregate the concept along two critical lines: inducements to participation in the institutional structure, and constraints on the permitted range of actions, demands, or representational claims. This took a useful concept – corporatism – and allowed the sorting of empirical cases into causally similar categories. Corporative systems such as Argentina during Juan Perón’s postwar government and Mexico during Lázaro Cárdenas’ presidency in the 1930s provided powerful inducements to organized labor’s participation in the representation system, and these were useful for the stabilization of the regime through the mechanism of mass political support. By contrast, constraints were essential to operation of corporatist institutions (often in structure similar to those noted earlier) in, for example, the Brazilian bureaucratic authoritarian regime. Here the goal was to use corporatist institutions to demobilize workers rather than engage them so as to build support. Subsequently, in the post-authoritarian era in Brazil the same “constraining” institutions paradoxically laid the foundation for mobilization, once the government reduced their repressive function (see Houtzager and Kurtz Reference Houtzager and Kurtz2000).

Of course, the point here is not about corporatism but instead about disaggregation. The absolutely crucial theoretical move was to recognize that corporatism was a vivid and useful concept as it had been defined, but to be analytically useful, it would have to be understood in terms of its disaggregated components. This is a lesson that more scholars should learn.

“Middle Cases”

Consider the political science usage of the term “democracy.” This concept has been subject to extensive scrutiny at a minimum since Robert Dahl (Reference Dahl1971), and quite probably since Aristotle and Plato. Yet in modern usage, we find widespread agreement that two dimensions crucially define this concept: participation and contestation – or, for some, accountability (Dahl Reference Dahl1971: 2–5).Footnote ² Since that time, these twin dimensions have come to dominate mainstream understandings of the concept, and indeed they form a benchmark against which various “diminished” subtypes of democracy could be defined (Collier and Levitsky Reference Collier and Levitsky1997).Footnote ³

This basic definition has also spawned rigorous efforts to measure the concept. These have included the long-standing Polity approach (Marshall Reference Marshall2020), the dichotomous coding of Przeworski (Reference Przeworski2000), and the more recent metrics of V-Dem (Coppedge et al. Reference Coppedge, Gerring and Knutsen2024). The first and last have, quite laudably, also emphasized the measurement of democracy, giving scholars the ability to work directly with distinct dimensions or to pursue their own aggregation strategies.

These metrics have generally worked quite well. Each of the antipodes they define – democracy and authoritarianism (the latter is also referred to as autocracy) – generally produces coding that has face validity and is uncontroversial. Yet this is not true of the entire distribution of cases.

Unfortunately, as I noted earlier, cases in the middle are challenging. Scholars of partial democracies have been far less attuned to the dimensional character of democracy/authoritarianism, and this has, as a consequence, produced a potentially confusing literature on what have come to be called, variously, “hybrid regimes,” “anocracies,” or various “diminished” subtypes of democracy. In practice, as scholars recognized that these middle cases were distinct from both autocracy and democracy, instead of giving them a firm conceptual footing, they used this measurement positioning to mark out the new conceptual category. For example, anocracy is defined (Vreeland Reference Vreeland2008) in many instances as the middle range (usually -5 to +5) on the Polity measure of political regime (e.g., Fearon and Laitin Reference Fearon and Laitin2003). They are not wrong that these cases are different from democracy and autocracy, but they err when they understand this simply in terms of being in the middle of a continuum.

However well defined the endpoints of democracy and autocracy are, the implicit assumption is that there is a set of shared characteristics about regimes “in the middle.” But this is a coding based not on a positive conceptual foundation but rather on a double residual (not too democratic, not too autocratic). In practice, however, in a multidimensional conceptual space there are quite different ways to come to this middle score. This leads to problems of awkward conceptual fit for this part of the distribution of political regimes – in a way similar to the earlier example of corporatism.

How could disaggregation help? As we saw with corporatism, the key insight is that the dimensions underlying a concept may well not covary. This lack of covariance in the “middle cases” can be the cause of seemingly awkward categorization. Consider the so-called anocracies. These could in principle be equally composed of regimes with relatively strong and broad participation but constrained contestation. One could think of the long-lived Mexican regime under PRI dominance and contemporary Iran as examples. Similarly, a regime that had quite limited suffrage but relatively strong contestation among parties in the enfranchised group might be similarly scored. This grouping would include European states before suffrage expansions, the US of the Jim Crow era, and perhaps late-stage apartheid South Africa.

Yet crucially, the political dynamics characterizing such regimes are likely different, and when treated as homogeneous because of similar aggregate scores, this may defeat attempts to properly assess causal theory. Disaggregation is called for.

Measurement Validity

A further place where the benefits of disaggregation have been underappreciated is in the world of measurement validity. A lot of effort has gone into measuring important concepts, with much of the focus concerning strategies for aggregating a large number of alternative metrics (often containing substantial error) into a composite of better quality and lower error.

However, these measures are often provided as public goods – which is a fundamental boon to scientific advance but also raises the inevitable challenge of trying to use measures developed for one context in a way that is appropriate for another context. My contention here is that disaggregation can be a useful check on whether such metrics are appropriate for the particular question under examination. Too often, I would suggest, they are not entirely appropriate, leading to conceptual overlap with outcomes of interest or measurement biases that are unwittingly compounded in aggregation.

Tautology

Here I return to the concept of democracy, as measured through the Polity dataset, and to anocracy. The issue is once again with the troubling “middle cases” – only now the problem is one of measurement, not of conceptualization. This is best seen by examining the on-the-ground usage of the data. As an example, a large literature has developed, contending that these anocracies (middle political regimes on the Polity scale) are more prone to civil conflict than either democracies or autocracies (for an overview, see Jones and Lupu Reference Jones and Lupu2018). There are plenty of prima facie reasons to expect that such regimes might indeed face more civil conflict, insofar as they cannot avoid conflict, either through democracy’s conflict-mitigation mechanisms or by fully employing autocratic tools of repressive control.

But here, a prominent effort to disaggregate the concept shows how measurement choices that are perfectly understandable in principle become problematic in practice when used without sufficient caution. James Vreeland shows how a definition of anocracy, based on the Polity scale, contains a problematic contribution to its measure of one dimension: contestation. Vreeland (Reference Vreeland2008: 402) shows that key components of the aggregate measure of democracy include inputs based on the degree of factionalism, and he notes that “for these variables, however, observations are coded in the middle when political participation is factional, [defined as] a situation where political competition between groups is ‘intense, hostile, and frequently violent.’” But, of course, this becomes all but tautological when the aggregate measure is used to explain civil conflict. Vreeland demonstrates this by showing the absence of a relationship between anocracy and conflict, once the problematic dimension is purged of tautology.

Concluding Observation: Granularity

This discussion has suggested that valuable analytic leverage can be gained through careful use of disaggregation. The backdrop was the observation that established modalities of concept formation and scaling often push scholars in the direction of higher levels of aggregation. Moving in the opposite direction can be valuable.

I conclude by returning to the V-Dem Project. This major research initiative pulls scholars in a direction closely related to the idea of disaggregation. The project is well known for its effort to achieve a greater degree of “granularity” in measures of democracy. These scholars have adopted a highly disaggregated approach, on the grounds that (1) this kind of fine-grained, granular measurement is essential, given the complex subject matter; and (2) users of the dataset produced by the project should make their own choices about the degree and form of disaggregation that is useful for their own research. Granularity, along with disaggregation, should be a key idea in the literature.