Charles C. Ragin
Everything that happens once can never happen again. But everything that happens twice will surely happen a third time.
Paulo Coelho, The Alchemist, p. 157
Paraphrasing Coelho: Every circumscribable instance is unique in its specificity; when we see two instances as more or less ‘the same’, we typically must invoke some sort of categorization, which, in turn, opens up the possibility of multiple instances. In social science, the most fundamental and consequential categorization is the ‘case’. The willingness of social scientists to invoke cases and to focus on patterns across cases as the key to generalizing about them distinguishes much of social science from much of the humanities, where the focus is often on the challenge of representing the specificity of each case. Casing is fundamental to the practice of social science; it could be argued that casing is the quintessential social scientific research act. Researchers ‘case’ their evidence to bring closure to difficult issues in conceptualization and research design and thereby allow cross-case analysis to proceed. Once evidence is packaged in the form of multiple cases, they can be compared and contrasted, which in turn enables empirical generalization – a key goal of social scientific inquiry (Ragin 1992).
Empirical evidence is infinite in its complexity, specificity and contextuality. Casing focuses attention on specific, limited aspects of that infinity, highlighting some aspects as relevant and obscuring many others. In short, casing provides much-needed blinders, making it possible for researchers to see through, or past, complexity. Different casings provide different blinders, different findings and different connections to theory, research literatures and research communities. Casing locates research in the vast domain of social science, linking it to the efforts of some researchers and severing its ties with others (Ragin 2009).
Quantitative and qualitative researchers alike invoke cases. However, the two discourses are very different, despite several formal similarities. The remainder of this essay sketches these different invocations, with a focus on the distinctiveness of each approach.
Qualitative research often begins with an interest in specific phenomena, outcomes or settings. At first, the casing of the phenomenon is fluid and open to revision and reformulation. The usual expectation is that the casing of the phenomenon will become more completely specified as more is learned, usually through in-depth research at the case level. Thus, the initial focus is often on ‘good’ instances of the phenomenon in question, and there is a back-and-forth between the identification of ‘good’ instances and the specification of the casing of the phenomenon.
Qualitative researchers construct diverse casings. At a formal level, the research focus is often on a specific category of phenomena, its constitutive features, and relevant antecedent conditions and processes. In other words, after establishing ‘what it is’, researchers focus on ‘how does it come about?’ Similarities across instances of the phenomenon in question are a key focus in research of this type. If the search for similarities proves to be unproductive, researchers may ‘re-case’ their instances, often with an eye toward differentiating types of cases (George and Bennett 2005).
From the perspective of conventional quantitative research, the qualitative approach just sketched might seem ludicrous. First of all, the explanandum is more or less the same across all instances. Thus, the ‘dependent variable’ does not vary substantially and, accordingly, there is little ‘variation’ to explain. Second, because the qualitative researcher has selected cases that have a limited range of values on the outcome (that is, on the ‘dependent variable’), correlations between antecedent conditions and the outcome are necessarily attenuated (see King, Keohane and Verba 1994), which leads, in turn, to abundant type II errors (that is, accepting the null hypothesis and concluding erroneously that antecedent conditions are irrelevant to the outcome). Third, and more generally, because of its focus on in-depth case analysis, the qualitative approach is necessarily limited to small Ns, which, from the perspective of conventional quantitative research, poses real obstacles to the use of probabilistic assessment and to the utilization of sophisticated multivariate techniques – both essential inferential tools.
In response, the qualitative researcher would reply that there are many analytical tools available to social scientists in addition to those based on correlation, and that there are many modes of empirical analysis that are not focused on the problem of accounting for variation in a dependent variable via some form of correlational analysis. For example, identifying an antecedent condition shared by instances of an outcome might signal the existence of a necessary condition for that outcome. Likewise, an absence of shared antecedent conditions could signal that there are different outcome types and that the researcher’s next step should be to ‘re-case’ the evidence, based on the identification of key differences between types.
Quantitative research tends to be more deductive than qualitative research. The back-and-forth between cases and concepts that is central to qualitative inquiry is almost completely foreign to quantitative research, with its emphasis on theory testing. For quantitative research to proceed, cases must be abundant, independent of each other, and homogeneous with respect to the operation of causal variables. Additionally, they should be drawn from a well-defined population. The populations of quantitative social science are often given or taken for granted. The key is that the population of observations (that is, cases) must be circumscribable; otherwise, sampling bias cannot be evaluated.
Often, the definition of the relevant population in quantitative research is contestable. Consider research on the causes of mass protest in Third World countries against austerity measures mandated by the International Monetary Fund (IMF) as conditions for debt restructuring. While it is possible to identify positive cases (that is, countries with protest), the set of relevant negative cases is somewhat arbitrary. Should the study include all less-developed countries as candidates for IMF protest? Less-developed countries with high levels of debt? Less-developed, debtor countries with recent debt negotiations? Less-developed, debtor countries subjected to severe IMF conditionality? Each narrowing of the set of relevant cases reduces the N of cases available for quantitative analysis, which in turn undermines the possible utilization of inferential technique. Understandably, quantitative researchers generally avoid narrowly circumscribed populations. When Ns are small, standard errors are large, and it is more difficult to generate findings that are statistically significant. For this reason, quantitative researchers tend to err on the side of being over-inclusive. In the example just presented, for instance, the typical solution might be to use all less-developed countries and to include debt level and extent of IMF renegotiations as ‘independent’ variables.
While this solution seems plausible, at least on the surface, there is a world of difference between using debt level and extent of IMF renegotiations as independent variables, on the one hand, and using these same variables to circumscribe the population of relevant candidates for protest against the IMF, on the other. Not only are these two uses very different from a mathematical perspective, they are also very different from a casing perspective. Using them as independent variables entails a casing that embraces all less-developed countries as candidates for IMF protest; using them to circumscribe the relevant population shifts the casing to a relatively small but clearly delineated subset of less-developed countries – those that satisfy certain antecedent conditions.
It is not widely recognized that boosting the N of cases carries with it an increased danger of type I errors – erroneously rejecting the null hypothesis of no relationship. If the N of cases is artificially enlarged by including irrelevant negative cases (that is, cases that are not plausible candidates for the outcome in question), then the correlations between causal and outcome variables are likely to be spuriously inflated (Mahoney and Goertz 2004). This artificial inflation occurs because irrelevant negative cases are very likely to have low scores on both the independent variables and the outcome and thus will appear to be theory-confirming, when in fact they are simply irrelevant. Correlational analysis is completely symmetrical in its calculation; therefore, a case with low (or null) values on both the causal and outcome variables is just as theory-confirming as a case with high values on both. It is important to note as well that an artificially inflated N of cases also increases the danger of type I errors by reducing the size of estimated standard errors, which, in turn, makes statistical significance easier to achieve.
The quantitative researcher would respond by arguing that a central goal of social science is general knowledge and that studying wider, more inclusive populations serves this goal more directly than studying narrowly circumscribed populations. Furthermore, the concern about type I errors can be addressed by using a more stringent significance level or by applying other technical fixes that safeguard against spuriousness. Finally, models incorporating statistical interaction can be used to address causally relevant conditions that enable the impact of other causal conditions.
The fact that qualitative and quantitative researchers alike invoke cases is an important commonality, uniting different approaches to research under the banner of social science. In both arenas, casings enable analysis. Still, the differences between the two discourses are striking, and the opportunities for miscommunication and misunderstanding are many.
Coelho, P. (1993). The Alchemist (trans. Alan R. Clarke). New York, NY: HarperPerennial.
George, A. and Bennett, A. (2005). Case Studies and Theory Development. Cambridge, MA: MIT Press.
King, G., Keohane, R. and Verba, S. (1994). Designing Social Inquiry: Scientific Inference in Qualitative Research. Princeton, NJ: Princeton University Press.
Mahoney, J. and Goertz, G. (2004). The possibility principle: choosing negative cases in comparative research. American Political Science Review, 98: 653–669.
Ragin, C. (1992). Casing and the process of social inquiry. In C. Ragin and H. Becker (Eds.) What Is a Case? (pp. 217–226). New York, NY: Cambridge University Press.
Ragin, C. (2009). Reflections on casing and case-oriented research. In D. Byrne and C. Ragin (Eds.) The Sage Handbook of Case-based Methods (pp. 522–534). Los Angeles, CA: Sage Publications.