Charles Ragin and Howard Becker, What Is a Case? (1992), Chapters 3, 4.

Chapter 3: White (1992) in What is a Case?

Key point prediction is myth; its replacement should be a discrimination into prefiguring, projecting, and interdiction.

  • three species of case studies, which relate to religion (creating an identity situated within a collectivity), law (the label for the process by which instn’s tidy up messes into cases by telling contentious parties how, and about what, it is they have been wrangling), and politics (applied to social fixing to maintain control – analogous to engineering, which seeks to fix and gain control of the physical) (86)
  • the experimental method can be a guide to finding and fixing an identity, or interlocking set of identities, in a region or phenomena. And so experiment can be a guide to features of the search for identity in general: it triggers recognition that intense concern for what is thrown out characterizes any search for identity. (88)
  • the design of case studies should depend on mission. Prefigure, explain, interdict: these are proposed as three possible basic missions. (91)
    • prefigure: limit the environment so much as to elicit the required, the correct response, which is to say the identity (91)
    • explain: the law, and comparative/statistical studies more generally, are about explanation. They aim for spread and balance in accountings. Their task is to provide sets of stories through which most happenings can be explained away, explained simultaneously from different points of view (92)
    • interdict: Intervention takes the active mode toward the environment. Fixing-for-control studies, the third species of case, thus have an entirely different relation to ‘environment’: rather than seeking a stable environment (second species) or a neutered environment (first species), a political engineer is endlessly changing and renegotiating and restructuring what might be environment, and with particular emphasis on time boundaries. (93)

Chapter 4: Lieberson (1992) in What is a Case?

Main point: application of Mill’s method to small-N situations does not allow for probabilistic theories, interaction effects, measurement errors, or even the presence of more than one cause (117)

  • a probabilistic approach is often necessary to evaluate the evidence for a given theoretical perspective, even if we think in deterministic terms due to: measurement errors; complex multivariate causal patterns (e.g. multicollinearity or the existence of other variables pushing in the other direction); omitted variables. (106)
  • a deterministic theory has deterministic outcomes, but often we can measure it only in probabilistic terms
  • despite these facts, small-N studies operate in a deterministic manner, avoiding probabilistic thinking either in their theory or in their empirical applications… partially since: except for probabilistic situations which approach 1 or 0, studies based on a small number of cases have difficulty evaluating probabilistic theories (107, 108)
  • the method of difference does not empirically or logically eliminate interaction effects. Rather, it arbitrarily assumes that they do not operate and that therefore constants cannot influence the dependent variable (111)
    • also assumes that only one variable causes the phenomenon under study
  • researchers have to guard against using broad categories so as to make it relatively easy for cases to fall under the same rubric (115)
  • rigor is mandatory when locating the variables if they are nominal, and even more so when they are ordinal
  • the method of agreement cannot work when more than one causal variable is a determinant and there is a small number of cases (112)
  • since the method of agreement will only work if all the cases for one causal variable fall in the same category and if no other variable has such uniformity, cutoffs are critical (115)
  • neither of Mill’s methods will work in cases of multicausal probabilistic statements (113)
  • the small-N application of Mill’s methods cannot be casually used with all macrosocietal data sets. The method requires very strong assumptions: a deterministic set of forces; the existence of only one cause; the absence of interaction effects; confidence that all possible causes are measured; the absence of measurement errors; and the assumption that the same ‘clean’ pattern would occur if data were obtained for all cases in the universe of relevant cases (114)
  • Ragin’s Boolean method is a step in the right direction
  • because of the subtle pressure to obtain only one variable that is homogeneous (in the case of agreement the dv) or only one variable that is heterogeneous (in the case of disagreement in the dv), one must guard against the bracketing of attributes in the former case, and decomposition in the latter. For this method to work at all, researchers must introduce formal criteria for these decisions which can be followed in advance of a given research project. To my knowledge, the do not exist at this time. (116)
  • because of the small N’s and the reasoning this method requires, it is vital to include all possible causal variables. Yet this will tend to lead to inconclusive results if carried out in a serious way, since the method of agreement will probably turn up with more than one variable that is constant, and likewise, the method of difference will have more than one IV associated with the difference in DVs
  • the relationship b/w the IVs and the DV is distorted if the cases are selected so as to have agreement or disagreement with respect to the dv (rather than simply sampling from all of the cases). It can be shown that sampling in order to obtain a certain distribution with respect to the dv ends up distorting the explandandum’s association with the ivs (unless the ration of odds is used)