Sage Journals: Discover world-class research

Abstract

French

Qualitative political analysis has made substantial methodological progress in the last 25 years. This article examines the contributions to this progress made by the work of three American social scientists (King, Keohane, and Verba, 2021 [1994], hereafter KKV) and the responses that their work provoked. The article identifies a recurring ambiguity in this methodological literature. In the quantitative tradition to which KKV want to hold qualitative methods endogeneity is a methodological problem that induces a search for methodological workarounds. Yet in qualitative work, endogeneity is often more a basic feature of the social and political world that needs to be modeled directly. While there can be substantial theoretical differences in how these features are modeled, the presumption is that endogeneity is more an ontological claim than a methodological problem. The article identifies how this ambiguity first arises in the work of KKV and then traces out the implications through a discussion of a range of methodological options, from process tracing to instrumental variables.

Keywords

endogeneity latent structure of cases qualitative analysis realism social and political ontology

Introduction

Qualitative political analysis has made substantial methodological progress over the past 20 years. This article critically assesses a subset of this literature, examining the challenges to qualitative methodology posed by King, Keohane, and Verba (2021 [1994], hereafter KKV) and responses from qualitatively oriented scholars in the discipline of political science. While the focus is thus specific to a notional discipline, the issues that arise have wider implications given the vagueness of disciplinary boundaries in the human sciences and the interest in cross-disciplinarity.

The article focuses on the question of endogeneity, in relation to both methodology and ontology. Endogeneity has become a much talked about methodological issue. Much of the methodological discussion of endogeneity, however, does not recognize the ontological implications of observables generated by intentional action in settings of social interdependence, or what I will term the implications of ‘endogenous action’. This remains the case, even after KKV, particularly for that literature on qualitative methodology that has taken to heart the (relatively) recent emphasis on the latent structure of cases. Endogenous action sets out an explicit position about how the social world is generated and, in doing so, reconfigures common binary oppositions that have structured our thinking about our methodological choices. The complex concept of endogenous action ‘turns a methodological vice into a theoretical virtue’ (Meadwell, 2005: 88, 2011: 7–8, 10, compare Fourcade and Healey, 2017: 18; Morgan, 2016: 490).

The article is organized in four sections. It begins in the next section by summarizing the basic elements of endogenous action. In the second section, I begin to examine an ambiguity about endogeneity – whether endogeneity is a methodological problem to be corrected or an ontological assumption or position. I show how this problem arises in the arguments in KKV, arguably the source that has most shaped (whether as foil or as exemplar) recent discussion about qualitative methods in political science. I show why it is necessary to distinguish their methodological thesis from a common ontological position consistent with endogenous action that they and, as I illustrate, others share. In the third section, I discuss a standard quantitative solution to the methodological problem of endogeneity – the instrumental variable – to show that this ‘solution’ replicates the conflation between methodology and ontology. In this part of the discussion, I use some of the later work of James Heckman against the literature on instrumental variables. In the fourth section, I briefly discuss recent methodological innovations in qualitative political analysis, innovations that are associated with the latent structure of cases, such as process tracing and causal mechanisms. The article then concludes.

The ontology of endogenous action: A combination of realism and constructivism

I argue in this section that the ontological structure of endogenous action is distinctive. First, this discussion of endogenous action is more compatible with some form of philosophical realism than with positivism (Meadwell, 2011. On realism’s relation to qualitative methods, compare Bennett and Checkel (2015: 10–13) and Morgan and Winship (2017: 445–446). Endogenous action is latent and unobservable and this violates the positivist rejection of unobserved entities. Moreover, positivism is ambivalent about causation. ‘Positivists tend to be non-realists, not only because they restrict reality to the observable but also because they are against causes and are dubious about explanations’ (Hacking, 1983: 42).¹ Endogenous action identifies a process of social generation of observable data that is incompatible with these features of positivism. For these reasons, endogenous action is not fully compatible with this further feature of positivism: its commitment to a ‘physical thing language’ – the thesis that every meaningful sentence, whether true or false, can be translated into physical language (Gates, 2001: 251) – into the ‘physical thing language’ of Carnap (1934: 166–167).

The virtues of much of the qualitative turn to ‘depth’ in explanation (Strevens, 2008), if put in terms of philosophy of science, thus is not rooted in positivism. The general interest in moving beyond and ‘below’ statistical associations and empirical regularities in constructing depth explanations has the earmarks of a form of realism, and this interest can be found in both quantitative and qualitative traditions.

But, second, endogenous action can not only incorporate elements of philosophical realism, it is also constructivist. The facts of constructivism are institutional facts. The process of endogenous generation that underlies social constructions thus is dependent on intentional action in social contexts. Social constructions, by definition, are within the purposive reach of human beings and that is enough to make social constructions endogenous. This gives endogenous action an unusual position in the philosophy of social science – it has elements of both realism and constructivism.²

Inside the discipline of political science, the organizing ideas of endogenous action encourage institutionalists to specify their particulars. It is difficult to imagine how institutions and organizations can be accounted for independent of endogenous action. To do so would amount to showing that institutions are not in principle endogenous to human action and interaction. My own view is that this would be a (bizarrely) self-defeating position to take.

Objections

Some possible objections to the organizing ideas of endogenous action need to be noted. The objections turn in particular on questions related to consequentialism, to voluntarism, to the relevance of dispositions and social practices, to the importance of political construction, the weight of history and to determinism, all of which might be said to limit the coherence and scope of endogenous action. I briefly summarize each of these objections and provide short responses.

Unintended consequences

The existence of unintended consequences of intentional action does not undercut its importance. First, unintended consequences logically entail the ontological existence of their contrast – intended consequences, that is, intentions. Second, empirically, in all likelihood the identification of unintended consequences requires the prior or simultaneous identification of intended consequences. Unintended consequences are byproducts or correlates of intentional action and choice, and they are part of the social world. They are no challenge to the importance of endogenous action.

Voluntarism

Endogenous action implies voluntarism, but this is a conceptual rather than empirical implication. The conceptual contrast is between different kinds, a contrast between voluntary acts and choices and involuntary reflexes or responses. A voluntary intentional act may be empirically constrained in various ways without being a contradiction in terms but an involuntary intentional act is conceptually puzzling.

Dispositions and practices

A practical disposition is demonstrated in multiple acts of a single agent, all of which share a common description, and a social practice gathers together as instances of the same kind ‘indefinitely many acts of indefinitely many agents’ (Thompson, 2008: 158–159). Dispositions and practices may be ‘given’ in a sense, but they are retrievable and potentially malleable because of their relation to human intentions and endogenous acts. Once brought to reflexive light, neither dispositions nor practices turn out to have either the weight of instincts, or the automaticity of involuntary reflexes, nor do they bear the imprint of efficient causes. None of this is a challenge to the importance of endogenous action.

The political construction of the social world

Endogenous action does not mean that the social world can be changed at whim. Social constructions that distribute life chances (and even equality-preserving social constructions distribute life chances) are stabilized (or not) by the distribution of social power. Questions about the endogeneity of power that then arise, particularly in relation to political equilibrium, are not taken up in this article, however.

The weight of the past

The ‘burden of history’ is not an effective challenge to endogenous action since the distinction between ‘past’ and ‘present’ is a cognitive construction placed on an ongoing process of social generation to make both human experience and the analysis of the social world, and of time, more tractable. Every ‘past’ was at one time a ‘present’ (Kosselleck, 2004), keeping in mind that the distinction between past and present is ordinarily relative to some kind of human association so that the ‘past’ typically is taken to condition the ‘present’ in some way – as a legacy, or ‘inheritance’ for example. These features of endogenous action – intentionality, political construction and social interdependence – thus are consistent with what I term the cultural transmission of non-genetic inheritances. An ‘inheritance’ of the kind I have in mind is not an invariant structure that has been reproduced through time nor is it a fixed parameter. In this treatment, an inheritance is pseudo-exogenous in the short-run (i.e. ‘inherited’ and not shaped by an actor), but endogenous in the long run.

The limits of endogenous action

Endogenous action is patterned and ordered, but its order is not immediately determined by physical laws, or laws of nature, or laws of the universe, however one wants to set out what determinism entails. Endogenous action cannot violate physical laws and must in some sense be compatible with them, but this is not to argue for determinism. Compatibilism does not eliminate the relevance of endogenous action but it does set external limits to it. These limits are not explored in this article but likely are related to human finitude, reflecting our status as one of many finite biological creatures.

The question of endogeneity

The influential treatment in KKV presented endogeneity as a methodological problem (King et al., 2021 (1994): 105–205). I term this their Methodological Thesis (MT). The authors then made a second claim – that endogeneity is a naturally occurring feature of social and political processes, rather than an artifact of a particular research procedure: ‘[E]ndogeneity is not always a problem to be fixed but is often an integral part of the process by which the world produces our observations’ (King et al., 2021 [1994]: 198). I treat this as an ontological claim and, accordingly, call it their ‘General Ontological Thesis’ (GOT).

This combination is potentially in tension. The tension is not immediately apparent because their GOT gets no discussion or development, as if this one sentence speaks for itself. Rather than represent or model this fundamental substantive feature of social life, ‘the way of the world’ we might say, the authors introduce corrections that treat the social world as if it was generated according to a research protocol.³ But the methodological solutions that are proposed do not match the ontological position. The solutions are variable-centered, yet endogeneity is not a variable. By assumption, according to KKV, endogeneity is a constitutive (‘integral’) feature of the social and political world. Endogeneity does not go on and off.

The problem may run deeper. Their General Ontological Thesis is actually methodologically neutral. It supports methodological pluralism. The GOT does not directly entail or imply their MT, nor, therefore, does it entail or imply the methodological solutions that KKV introduce, because the ontological thesis is consistent with different methodological positions. I illustrate briefly below. These illustrations, I believe, are consistent with KKV’s ontological position, although ‘consistency’ might not be a high threshold, given the thinness of the discussion in KKV. Indeed, the discussion of the ontology of endogeneity in the previous section is meant to provide more substance, in anticipation of the need to provide some analytic structure for this part of my argument. KKV’s GOT and the examples below share a commitment to endogenous action, and this is what motivated the discussion in the earlier section.

When Przeworski makes the provocative statement that ‘endogeneity is the motor of history’ (Przeworski, 2004a, 2004b, 2007), he is invoking a version of endogenous action when he argues the following: ‘The conditions under which we live are somehow created by people in pursuit of their ends . . . We thus must treat the observable world as having been produced by ‘us’, that is, as having been generated endogenously’ (Przeworski, 1995: 17). Endogeneity is indeed the motor of history.

Similarly, when it is argued that ‘if we accept the premise that politics is the result of making choices, then the entire history of political interaction is one of repeated selection. Put another way, then, history is one grand selection model’ (Signorino, 2002: 94), a version of the ontological thesis is at work, notwithstanding the specialized theoretical language of the research program within which that thesis is embedded in this particular illustration. And when Dvora Yanow (2014: 105), a specialist in interpretive sciences, associates ‘interpretive philosophies’ with the view that ‘social realities and human knowledge of them are created by human actors through our actions and interactions, inter-subjectively’, she is drawing on a version of endogenous action. All of these examples have prominent places in their respective fields. All are social scientists, not philosophers of history, but their statements read like (unelaborated) philosophies of history. And I take this to be an ontological virtue, not a vice or error or mistake to be corrected.

A further illustration: Dunning (2010) makes an important observation in his discussion of field and natural experiments when he argues that ‘[M]any of the interventions that might provide the basis for plausible natural experiments are the product of the interaction of actors in the social and political world’. This argument informally relies on the ontology of endogenous action identified earlier. Under conditions of social and political interaction, so fundamental a condition that we can say this is the way the world is, there is not enough exogenous purchase to support plausible ‘natural’ interventions.⁴ He continued,

It can strain credulity to think that these interventions are independent of the characteristics of the actors involved, or that they do not encourage actors to ‘self-select’ into treatment and control groups in ways that are correlated with the outcome in question. (Dunning, 2010: 295)

On the face of it, then, endogenous action (and KKV’s rather thin GOT) supports methodological pluralism. There is no ontological reason to rule out interpretivism, for example, although that is what the recent methodological turn in qualitative analysis, post-KKV, attempts to do (Goertz and Mahoney, 2012; King et al., 2021 [1994]). I do not account in detail in this article for this combination of methodological difference apparent in the illustrations above, in the face of a loosely shared ontology, but I offer a formal interpretation: In each instance, other auxiliary assumptions or premises are in play. These auxiliary assumptions take the form of strong theoretical priors that lead to methodological divergence rather than methodological convergence even in those arguments that share a general ontological position. My goal here is not to identify these theoretical priors but instead to examine the claim that has led to this point, notably the recognition that endogeneity is not simply a methodological problem to be corrected. I believe I can go further: This ontological position – endogenous action – is influential in the social sciences. I would conjecture that another measure of its importance is the way in which many arguments draw implicitly on this ontology without acknowledging doing so.

I do not treat endogenous action as the only available ontological position. Contemporary social metaphysics allows for other ontological possibilities (e.g. Epstein, 2015), either as complements or substitutes for endogenous action. However, questions such as these cannot really be raised, let alone addressed, in much of the contemporary literature on ‘methods’. That is unfortunate since questions about the ontological implications of observables generated by intentional action in settings of social interdependence obviously did not come to the fore only recently. So framed, a large part of social and political theory, not to mention the history of modern philosophy, has been concerned in various ways with this problem. This broader context and history, illustrated for example by Gordon (2010), is written out of the discussion when ‘method’ is narrowly construed.

The methodological problem of endogeneity that KKV recognize arises when a dependent variable causes changes in the explanatory variable (King et al., 2021 [1994]: 185). They deal extensively with this issue. One solution draws on the early work of Heckman (1976, 1979): Transform the problem of endogeneity into a problem of omitted variable bias. Yet endogeneity is not a variable, whether omitted or not. Nor is it a source of potential bias to be controlled or eliminated. Rather, it is by their assumption (and mine) the way the world is. As I also have illustrated above, versions of this assumption are far more common that has been recognized and can be linked to different methods than those privileged by KKV. ‘Correcting’ for endogeneity is not sensible methodological advice if endogeneity is a constitutive feature of social generation. Moreover, there is nothing in their General Ontological Thesis that entails KKV’s methodological solutions.

Endogeneity instead needs to be modeled. The variety of arguments, as above, that draw on endogenous action (action, not processes, which is a rather vague term) imply that it could be modeled in different ways, as long as these general features are incorporated: intentional action, social interdependence, and political construction.

Another solution KKV provide, beside assimilating the problem of endogeneity to omitted variables, mimics one that Philip Wright (1928, Appendix 1) introduced to address the simultaneity of supply and demand in macroeconomic models. His work is now treated as an early statement of the identification problem (Angrist and Kreuger, 2001): Disaggregate both dependent and explanatory variables into two components. In the case of the explanatory variable, this means one part that is clearly exogenous and the other part that is endogenous. This is effectively to introduce an instrumental variable, a solution which I address in more detail in the next section.

Brady and Collier (2010 [2004]) wrote a direct response to KKV, which became an important part of the pushback against KKV from qualitatively oriented scholars. They also provide a treatment of the problem of endogeneity. Their collections include a discussion of KKV’s ‘confusion’ between the ‘independence assumption’ and ‘conditional independence’ (Brady, 2010 [2004] 74–76) and the conflation of explanation and causal inference (Brady, 2010 [2004]: 77). Collier, Seawright, and Munck recognize the importance of endogeneity in historical data, although endogeneity is still treated as methodological vice or problem (Collier et al., 2010 [2004b]). On the other hand, the problem of selection bias that most bothered them was the problem of investigator-induced bias (Collier et al., 2004), although they move beyond this focus in their later discussion (Collier et al., 2010 [2004a]: 140–145). There, they recognize the problem of self-selection but only in passing. When they do, they do not trace out in detail the consequences for case-based qualitative research. It is revealing, however, that their takeaway point about self-selection in qualitative research requires treating cases (‘countries’) like persons.

The updated version of Brady and Collier (2010 [2004]) includes criticism of the instrumental variable approach to the problem of endogeneity, although only Dunning addresses the kinds of problems raised by Heckman in his later work (and then only partially; Dunning, 2010: 288–289; Seawright, 2010: 266–268). This can be seen in the absence of a discussion of heterogeneity of responses and in the limited recognition of the question of self-selection (Dunning, 2010: 289, 295).⁵ Brady and Collier do not identify the conflation of ontology and methodology characteristic of KKV and, by default, reproduce it.

Endogenous action and methodological options: The instrumental variable

A standard solution for problems of endogeneity in mean-dependent models of estimation has been the instrumental variable (IV). The IV is discussed in more detail in this section, drawing on the work of James Heckman. We assume an observed relationship between an explanatory variable and an outcome variable (X, Y, X ⇒ Y). Endogeneity becomes a problem in a variable-centered, regression-based system of variables, for example, if there is an unobserved antecedent condition (Z) causally related to the explanatory variable X and to the outcome Y. In this case – in the instance of an unobserved relationship between Z and Y when X is endogenous to Z – the estimate of the relationship between X and Y is biased and the error term of the stochastic equation which represents this system of variables contains systematic rather than random error.

An instrumental variable, which is not an explanatory variable but a methodological workaround, is an exogenous variable correlated with an endogenous regressor, but uncorrelated with the outcome.⁶ An IV allows the relationship between X and Y to be estimated without bias. More specifically, as Signorino (2002: 95ff.) shows, an instrumental variable is equivalent to a selection mechanism rooted in exogenous nature.⁷ Practically speaking, this means the instrumental variables method can allow the relationship of interest between X and Y to be estimated without actually investigating Z (Angrist and Kreuger, 2001: 73). But that can look like a severe limit rather than a strength, indeed it can appear to be an evasion of the question as to how the social and political world is generated. Theoretically speaking, this points to the tension between method and ontology raised earlier, namely, in this instance, the oddity of research protocols, rooted in assumptions of random assignment or selection, when applied to social and political worlds that are non-randomly generated.

Heckman (1992, 1997, 1999) has been critical of the use of instrumental variables. Some of his work identifies an ontology of endogeneity consistent with endogenous action. Most political scientists might measure Heckman’s contribution as methodological rather than theoretical. In essential ways, however, his arguments are theoretical. His arguments systematically draw our attention to the importance of heterogeneity and the complexity of selection effects, once we allow human action and choice, particularly in the general context of social interdependence. And this is accomplished through his theoretical interrogation of a methodological device – the instrumental variable – that has widespread currency in the social sciences.

Heckman argues that the use of instrumental variables depends on implicit and limiting assumptions about behavior. There are several related ways he makes this point: When individuals possess private information and use it, instrumental variables will produce inconsistent estimations of parameters; the use of instrumental variables assumes that individuals do not have better information than the econometrician studying them or they have that information but do not use it; the use of instrumental variables assumes homogeneity of individual responses, the use of instrumental variables assumes irrationality or ignorance.

Much discussion on the use of instrumental variables and their limits thus is to be found in micro-econometrics. The data in play in this literature is generated by choice, and the general application is to the problem of program evaluation and treatment effects. If this application seems narrow, it is well to recall that the questions of evaluation and treatment effects motivate a considerable part of the contemporary literature on causal analysis and causal inference and experimental designs.

Heckman’s main conceptual points can be summarized as follows: (1) Responses are defined as homogeneous when persons with a given set of characteristics respond identically to treatment (Heckman, 1997: 460). When responses to treatment are homogeneous, the case for IVs is the most persuasive (Heckman, 1999: 828). (2) If responses to treatment vary, IVs identify the mean effect of treatment on the treated or the effect of treatment on randomly selected persons only when persons do not select into the program on the basis of the idiosyncratic component of their responses to treatment (Heckman [1997: 442]). That is, in the general case of heterogeneous response the IV does not estimate the effect of treatment on the treated unless agents do not select into the program on the basis of the unobserved (to the analyst) heterogeneity in their response (Heckman, 1996a: 461, 1999: 829). (3) The basic point: When responses to treatment are heterogeneous, standard econometric methods have to be modified (Heckman, 1999: 831). (4) If persons do not select in on the basis of their heterogeneity, when responses are heterogeneous, then agents are either irrational or ignorant (Heckman, 1997: 460, 1999), either ignorant of information that is relevant to their choice or not acting on their private information. (5) If, in the case of heterogeneous responses, persons are neither irrational nor ignorant, then the IV is an inappropriate approach to the problem of endogeneity.⁸ Heckman’s arguments thus imply a limit or idealization under which the instrumental variable is appropriate for data-generating processes that involve human choice. Otherwise put, the IV and the two-stage ‘Heckman correction’ can converge under a condition of full information (or the slightly more complicated condition of private information that is not acted upon).

There is little doubt that the specialized language of these arguments narrows its range for some social scientists, who do not trade intellectually in notions of information or choice theory of the kinds associated with micro-econometrics. Nevertheless, there is a robust core of this work that is consistent with endogenous action. I have used his discussion to establish a result that bears on KKV’s arguments: Their methodological correction to endogeneity – the IV – violates their ontological thesis. In other words, I have now made good my earlier claim in this article that there is tension between what I called their methodological thesis and their ontological thesis. Otherwise, one thesis would not contradict the other. There is a form of incoherence in their argument, which I have been able to identify by holding KKV to the full implications of their GOT.

The latency of endogenous action

While it is the case that ‘early’ Heckman (1976, 1979) argued that selection bias can be thought of as a form of omitted variable bias (and that self-selection and investigator-induced selection were functionally the same problem), his subsequent work moved away from the intuition that the problem of evaluation and treatment effects is primarily one of omitted variables. The latent (not omitted) variable in the models that he sets out is not so much a variable as it is itself a model of choice.

It is a model of the choices of agents who are exposed to treatment(s). Persons can select in or out of treatments. In the general case, there is heterogeneity in both the rewards of participation and in preferences, or costs of participation. Persons do not face the same ‘price’ of participation, unlike a competitive market for example (Björklund and Moffitt, 1987). Responses to treatments also vary in the general case, even when persons share a given set of characteristics, and both kinds of heterogeneity have consequences for the estimation of coefficients on characteristics. In contrast, assignment of treatment provides a clean IV when it is random, notably in a randomized clinical trial (Heckman, 1996b). The analogy to an experiment is, of course, clear. Agents are potential ‘subjects’ of a (‘experimental’) treatment. The ‘planner’ (see below) is much like the experimenter.

In the observable data generated in the social world, however, treatments themselves, or programs, are the result of the intentions and actions of other agents, although not explicitly modeled in the latent variable equation. What do appear in the equation are the resulting treatments. But behind the existence of these states lies another agent who herself is facing a problem of selection, known in this literature as the ‘planning problem’ – choosing treatments (programs) for the members of a heterogeneous population (e.g. Manski, 2005: 3).

Treatments, then, in the most general case, are not exogenous in either of two senses. (1) The assignment of treatment is not made by exogenous nature. (2) The choice of treatments made by the planner is endogenous to her ends. Given what was argued above about the general case where heterogeneity of both costs and reward is allowed, and once the stylized planner is herself made fully endogenous, then the latent structure of the situation has changed considerably.

Now imagine a social setting in which persons influence the ostensible planner, so that the choice of program or treatment is an endogenous social choice. It is here that the implications of private information seem to escape the experimental setup of the model of treatment effects. Heckman’s treatment concerns choices conditioned on private information but his setting is not yet fully interdependent because the ‘planner’ is exogenous. We are motivated here by a different kind of question: not the classic question that interests much of this literature – the question of program evaluation and thus of responses to treatments – but rather the question of how choices of programs are made. This will follow quickly once we make the planner endogenous. In this case, programs are political acts rather than ‘treatments’.

And consider a ‘program’ which is to hold over some reference group but which is to be decided without a planner (a social situation in which every person plans for herself and outcomes are jointly produced). In these kinds of settings, observable data are generated endogenously by interdependent choice. Full-blown endogenous choice can be modeled (such as by using game theory, the basic assumptions of which are consistent with endogenous action)⁹ but is harder to fit into a mean-dependent model for purposes of statistical estimation. The most advanced work in this area, however, strives to take into account both the consequences of interdependence and private information and provides both models of action and estimates of effects where relevant.¹⁰

Causal process observations and data set observations

The discussion in this article escapes many of the binary oppositions that have structured our thinking about methodological options: quantitative or qualitative, cases versus variables (Ragin, 1997), large n versus small n, thick versus thin analysis (e.g. Brady and Collier, 2010 (2004)). Endogenous social and political generation – endogenous action in my terms – cuts across such stylized oppositions. That, too, I take to be a virtue.

One of the consequences of its distinctive features is that endogenous action has implications both for the estimation of average-effects in mean-dependent models and for the analysis of (non-averaged) ‘real’ effects in case-driven research. As a further consequence, some of the methodological treatments of mean-dependent models that consider choice as a latent generator of observable data have implications for case-based social research. These implications are particularly important for qualitative research that assumes the existence of latent structures. This kind of assumption is not unusual or idiosyncratic. On the contrary, it is now pervasive and characterizes an influential methodological literature on mechanisms, on process tracing, and on causal process observations more generally.

An illustration: the latency of choice has implications for a distinction that is at the heart of contemporary qualitative political methodology, and this distinction has been used as a linchpin in some of the criticisms of regression-based models of explanation. Qualitative methodologists make a distinction between ‘data set observations’ [DSOs] and ‘causal process observations’ [CPOs].¹¹ The former are taken to be typical of regression-based research. Think of a data set organized by variables and cases. CPOs, however, go ‘deeper’ – think of process tracing (e.g. Bennett and Checkel, 2015) – and they are a central element in much of the discussion of qualitative methods.

The distinction between data set observations and causal process observations implies latent structure consistent with philosophical realism. The reference to ‘mechanisms’ and to the ‘tracing’ of causal processes suggests a ‘deeper’ level, for example, when Collier et al. (2010 [2004a]: 186–187) cite the work of Goldthorpe on ‘generative processes’ and the ‘microscopic level’ at which mechanisms operate, a level ‘perhaps not directly observable’ (Goldthorpe, 2001: 9).¹² Understood in this way, in the way in which CPOs in fact were presented, it is understandable why Beck (2006) argued that the concept of a causal process observation was an oxymoron.

The distinction between DSOs and CPOs, however, seems related to a theoretically specified model that assumes a data-generating process (DGP), since such a model already implies, in effect, two levels. There are observables in a DGP, but there is also the generative process itself. The qualitative distinction between CPOs and DSOs looks to be similar to a two-level model of a data-generating process. A model of the latter kind is not merely a set of data observations. It is also, at least potentially, a model of how observables are generated.¹³

Conclusion

I have been able to identify and address the ambiguity that has arisen between methodology and ontology in some of the literature on qualitative political analysis that discusses endogeneity. I have argued that endogeneity is not so much a methodological vice to be corrected as it is an ontological position to be modeled.

In developing its arguments, this article draws on the increasing interest in ontology in the context of methodological discussions in social science. I offer, in effect, an ontology of endogeneity and then consider the ensuing implications. This growing attention to ontology is laudable because it encourages expanded discussion of methodology in larger intellectual contexts, at a time when ‘method’ is in danger of becoming fetishized. Methods can be improved by considering and improving ontology (Epstein, 2015: 278). Methodological workarounds do not solve ontological problems and they are unlikely to yield satisfactory results. Attention to ontology has the virtue of raising the possibility that some methodological problems are instead philosophical problems – that some methodological problems can be artifacts of philosophical positions. Of course, the more expansive perspective on method developed here is likely to leave some methodologists uncomfortable, particularly those with positivist philosophical priors, because the article denies the positivist position that science is metaphysics-free. The context of this article, though, is methods, not philosophy proper, and, ultimately, it is grounded in the demands of political analysis.

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Hudson Meadwell

Notes

Author biography

Hudson Meadwell is an Associate Professor in the Department of Political Science at McGill University. He has written in the general areas of comparative politics, European political development, political methodology, and philosophy of social science.

References

Aldrich

Alt

Lupia

(2008) The EITM approach: Origins and interpretations. In: Box-Steffensmeier

Brady

Collier

(eds) Oxford Handbook of Political Methodology. Oxford; New York: Oxford University Press, 828–845.

Angrist

Kreuger

(2001) Instrumental variables and the search for identification: From supply and demand to natural experiments. Journal of Economic Perspectives 15(4): 69–85.

Angrist

Pischke

(2013) Mostly Harmless Econometrics. Princeton, NJ: Princeton University Press.

Archer

Decoteau

Gorski

, et al. (2016) What is critical realism? American Sociological Association, 23 December. Available at: http://www.asatheory.org/current-newsletter-online/what-is-critical-realism (accessed 25 August 2022).

Beck

(2006) Is causal-process observation an oxymoron? Political Analysis 14(3): 347–352.

Bennett

Checkel

(eds) (2015) Process Tracing: From Metaphor to Analytical Tool. Cambridge: Cambridge University Press.

Björklund

Moffitt

(1987) The estimation of wage gains and welfare gains in self-selection models. Review of Economics and Statistics 69(1): 42–49.

Brady

(2010 [2004]) Doing good and doing better: How far does the quantitative template get us? In: Brady

Collier (eds) Rethinking Social Inquiry. Diverse Tools, Shared Standards. Lanham, MD: Rowman & Littlefield, 67–82.

Brady

Collier

(eds) (2010 [2004]) Rethinking Social Inquiry: Diverse Tools, Shared Standards. 2nd ed. Lanham, MD: Rowman & Littlefield.

10.

Carnap

(1934) The Unity of Science (trans Black

). London: Kegan Paul, Trench, Trübner & Company.

11.

Collier

Brady

Seawright

(2010 [2004a]) Critiques, responses and trade-offs: drawing together the debate. In: Brady

Collier

(eds) Rethinking Social Inquiry. Diverse Tools, Shared Standards. Lanham, MD: Rowman & Littlefield, 125–159.

12.

Collier

Mahoney

Seawright

(2004) Claiming too much: Warnings about selection bias. In: Brady

Collier

(eds) Rethinking Social Inquiry: Diverse Tools, Shared Standards. Lanham, MD: Rowman & Littlefield, 85–102.

13.

Collier

Seawright

Munck

(2010 [2004b]) The quest for standards: King, Keohane and Verba’s Designing Social Inquiry. In: Brady

Collier

(eds) Rethinking Social Inquiry: Diverse Tools, Shared Standards. Lanham, MD: Rowman & Littlefield, 33–64.

14.

De Marchi

(2005) Computational and Mathematical Modeling in the Social Sciences. Cambridge: Cambridge University Press.

15.

Dunning

(2010) Design-based inference: Beyond the pitfalls of regression analysis? In: Brady

Collier

(eds) Rethinking Social Inquiry: Diverse Tools, Shared Standards. Lanham, MD: Rowman & Littlefield, 273–312.

16.

Dunning

(2012) Natural Experiments in the Social Sciences: A Design-Based Approach. New York: Cambridge University Press.

17.

Elster

(2015) Explaining Social Behavior: More Nuts and Bolts in the Social Sciences. Rev. ed. Cambridge: Cambridge University Press.

18.

Epstein

(2015) The Ant Trap: Rebuilding the Foundations of the Social Sciences. New York: Oxford University Press.

19.

Fourcade

Healey

(2017) Seeing like a market. Socio-Economic Review 15: 9–29.

20.

Gates

(2001) Physicalism, empiricism and positivism. In: Gillett

Loewer

(eds) Physicalism and its Discontents. Cambridge: Cambridge University Press, 251–270.

21.

George

Bennett

(2005) Case Studies and Theory Development in the Social Sciences. Cambridge, MA: MIT Press.

22.

Goertz

Mahoney

(2012) A Tale of Two Cultures: Qualitative and Quantitative Research in the Social Sciences. Princeton, NJ: Princeton University Press.

23.

Goldthorpe

(2001) Causation, statistics and sociology. European Sociological Review 17(1): 1–20.

24.

Gordon

(2010) Continental Divide: Heidegger, Cassirer, Davos. Cambridge, MA: Harvard University Press.

25.

Hacker

Pierson

Thelen

(2015) Drift and conversion: Hidden faces of institutional change. In: Mahoney

Thelen

(eds) Advances in Comparative Historical Analysis. Cambridge: Cambridge University Press, 180–208.

26.

Hacking

(1983) Representing and Intervening. Cambridge: Cambridge University Press.

27.

Haslanger

(2012) Resisting Reality: Social Construction and Social Critique. New York: Oxford University Press.

28.

Heckman

(1976) The common structure of statistical models of truncation, sample selection, and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement 5(4): 475–492.

29.

Heckman

(1979) Sample selection bias as a specification error. Econometrica 47(1): 153–161.

30.

Heckman

(1992) Randomization and social policy evaluation. In: Manski

Garfinkel

(eds) Evaluating Welfare Training Programs. Cambridge, MA: Harvard University Press, 201–230.

31.

Heckman

(1996a) Identification of causal effects using instrumental variables: Comment. Journal of the American Statistical Association 91(434): 459–462.

32.

Heckman

(1996b) Randomization as an instrumental variable. Review of Economics and Statistics 78(2): 336–341.

33.

Heckman

(1997) Instrumental variables: A study of implicit behavioral assumptions used in making program evaluations. Journal of Human Resources 32(3): 441–462. Addendum published 33 (1998).

34.

Heckman

(1999) Instrumental variables: A response to Angrist and Imbens. Journal of Human Resources 34(4): 828–836.

35.

Hedström

(2008) Studying mechanisms to strengthen causal inferences in quantitative research. In: Box-Steffensmeier

Brady

Collier

(eds) Oxford Handbook of Political Methodology. Oxford and New York: Oxford University Press, 319–336.

36.

Hedström

Swedberg

(1998) Introduction. In: Hedström

Swedberg

(eds) Social Mechanisms: An Analytical Approach to Social Theory. Cambridge; New York: Cambridge University Press, 1–31.

37.

Hempel

Oppenheim

(1948) Studies in the logic of explanation. Philosophy of Science 15(2): 135–175.

38.

Jackson

(2008) Endogeneity and structural equation estimation in political science. In: Box-Steffensmeier

Brady

Collier

(eds) Oxford Handbook of Political Methodology. Oxford; New York: Oxford University Press, 404–431.

39.

Keele

Titiunik

(2016) Natural experiments based on geography. Political Science Research and Methods 4(1): 65–95.

40.

King

Keohane

Verba

(2021 [1994]) Designing Social Inquiry: Scientific Inference in Qualitative Research. Princeton, NJ: Princeton University Press.

41.

Kosselleck

(2004) On the disposability of history. In: Kosselleck

(ed.) Futures Past (trans Tribe

). New York: Columbia University press, 192–204.

42.

Mahoney

(2001) Path dependence in historical sociology. Theory & Society 29(4): 507–548.

43.

Mahoney

(2010) After KKV: The new methodology of qualitative research. World Politics 62: 120–147.

44.

Mahoney

Thelen

(eds) (2009) Explaining Institutions: Ambiguity, Agency and Power. Cambridge: Cambridge University Press.

45.

Mallon

(2016) The Construction of Human Kinds. New York: Oxford University Press.

46.

Manski

(2005) Social Choice with Partial Knowledge of Treatment Response. Princeton, NJ: Princeton University Press.

47.

Meadwell

(2005) Institutions and political rationality. In: Lecours

(ed.) The New Institutionalism: Theory and Analysis. Toronto, ON, Canada: University of Toronto Press, 80–96.

48.

Meadwell

(2011) Faith and heresy in social science. In: CPSA meetings. Available at: https://cpsa-acsp.ca/papers-2011/Meadwell.pdf (accessed 26 October 2022).

49.

Morgan

(2016) Process tracing and the causal identification revolution. New Political Economy 21(5): 489–492.

50.

Morgan

Winship

(2017) Counterfactuals and Causal Inference: Methods and Principles for Social Research. New York: Cambridge University Press.

51.

Murray

(2006) Avoiding invalid instruments and coping with weak instruments. Journal of Economic Perspectives 20(4): 111–132.

52.

Przeworski

(1995) The role of theory in comparative politics. World Politics 48(1): 16–21.

53.

Przeworski

(2004a) Institutions matter? Government and Opposition 39(4): 527–540.

54.

Przeworski

(2004b) The last instance: Are institutions the primary cause of economic development? European Journal of Sociology 45(2): 165–188.

55.

Przeworski

(2007) Is the science of comparative politics possible? In: Boix

Stokes

(eds) Oxford Handbook of Comparative Politics. New York; Oxford: Oxford University Press, 147–171.

56.

Putnam

(1994) Making Democracy Work: Civic Traditions in Modern Italy. Princeton, NJ: Princeton University Press.

57.

Ragin

(1997) Turning the tables: How case-oriented research challenges variable-centered research. Comparative Social Research 16: 27–42.

58.

Seawright

(2010) Regression-based inference: A case study in failed causal assessment. In: Brady

Collier

(eds) Rethinking Social Inquiry. Lanham, MD: Rowman & Littlefield, 247–272.

59.

Sekhon

Titiunik

(2012) When natural experiments are neither natural nor experiments. American Political Science Review 106(1): 35–57.

60.

Signorino

(1999) Strategic interaction and the statistical analysis of international conflict. American Political Science Review 93(2): 279–297.

61.

Signorino

(2002) Strategy and selection in international relations. International Interactions 28(1): 93–115.

62.

Signorino

Yilmaz

(2003) Strategic misspecification in discrete choice models. American Journal of Political Science 47(3): 551–556.

63.

Sovey

Green

(2011) Instrumental variables estimation in political science: A readers’ guide. American Journal of Political Science 55(1): 188–200.

64.

Strevens

(2008) Depth: An Account of Scientific Explanation. Cambridge, MA: Harvard University Press.

65.

Thompson

(2008) Life and Action: Elementary Structures of Practice and Practical Thought. Cambridge, MA; London: Harvard University Press.

66.

Wright

(1928) The Tariff on Animal and Vegetable Oils. New York: Macmillan.

67.

Yanow

(2014) Neither rigorous or objective? Interrogating criteria for knowledge claims in interpretive science. In: Yanow

Schwartz-Shea

(eds) Interpretation and Method: Empirical Methods and the Interpretive Turn. 2nd ed. Armonk, NY: M. E. Sharpe, 97–119.

Endogeneity and qualitative political analysis: Debates about method or debates about ontology?

Abstract

Keywords

Introduction

The ontology of endogenous action: A combination of realism and constructivism

Objections

Unintended consequences

Voluntarism

Dispositions and practices

The political construction of the social world

The weight of the past

The limits of endogenous action

The question of endogeneity

Endogenous action and methodological options: The instrumental variable

The latency of endogenous action

Causal process observations and data set observations

Conclusion

Footnotes

Funding

ORCID iD

Notes

Author biography

References