---
title: "Ch28 what would count against content"
description: "Chapter 9 closed Part One with the charge that none of the eight defects has ever cost us anything, and it gave the reason: nothing in the model’s core, the…"
url: "https://wiki.cybernetic-futures.org/punch-and-judy/part-05-put-model-at-risk/ch28-what-would-count-against/ch28-what-would-count-against-content"
modified: "2026-11-12T15:26:25.957Z"
language: "en"
site: "Cybernetic Futures"
---

# What Would Count Against It?

Chapter 9 closed Part One with the charge that none of the eight defects has ever cost us anything, and it gave the reason: nothing in the model's core, the necessity of the five functions and the recursion claim, has been stated in a form that could be disconfirmed. It also named the repair available to us and noted that we have not made it. Here we make it. The accusations of Parts One to Four become claims that can lose, and for each we state the observation that would oblige us to give something up.

The rule for the conversion binds us before anyone else. A failed test cannot be redescribed after the fact as evidence of some other pathology while the original claim stays unchanged.[^ch28-nosek] If the result changes what we say we meant, the registered claim changes with it. A framework that survives by renaming every loss has not survived a test.

The temptation the rule is written against is specific, and it is ours. The model’s vocabulary is rich enough to describe any outcome once the outcome is known, so a disappointing result can always be absorbed: the system in focus was misidentified, the recursion level was drawn in the wrong place, the client did not implement what the diagnosis recommended. Any of those may on occasion be true. They are disqualified as a standing defense because they are available every time, at no cost, and after the fact. The conditions below therefore close their escape routes in advance. Where a boundary judgment could later be invoked to explain a loss, it has to be registered before the result is known, or it does not count.

Three kinds of exposure follow, in ascending order of what a loss would cost us. Claims about what Beer wrote can be settled by reading, and losing one costs a chapter. Claims that a VSM diagnosis outperforms its neighbors can be settled by comparison, and losing one costs the practice its warrant for charging as a theory. The claim that the five systems are necessary can be settled by a single well-documented case, and losing it costs the model its central proposition. The cheap tests come first because they are the ones we have least excuse for not having run.

Only charges traceable to a specific argument in this book belong here. Wider questions, however interesting, stay in the companion register. Normative claims take a different route again: evidence can expose their costs and consequences, but cannot settle what ought to be valued. They return in the Coda.

The three sections share one form. A claim we have made is named; the evidence that bears on it is identified; the result that would count against us is stated, and what we would surrender with it; and the smallest fair test is specified, so that nobody has to fund a decade of fieldwork to hold us to account. The fourth column carries the argument. A condition of disconfirmation that costs nothing when it is met is not a condition at all.

## Textual exposure

The cheapest claims to lose are claims about what Beer wrote. Chapters 1 to 3 accuse the tradition of teaching a neurological mapping Beer did not write, of shrinking System Four to environmental scanning, and of citing a theorem more strongly than its proof permits. Each can be tested against a defined corpus, edition and search protocol, and each can be tested by somebody who thinks we are wrong.

| Claim at risk | Evidence that bears | Result against us | What changes | Smallest fair test |
|---|---|---|---|---|
| Current teaching has displaced Beer’s anatomical mapping | Reliably paginated *Brain of the Firm* and dated teaching sources | Beer’s assignments match the mapping we say is later | Surrender the displacement charge; retain only the modern-neuroscience audit | Two independent coders extract every System Two–Four anatomical assignment from the same edition and record Beer’s conflicting assignments as conflicts |
| System Four in Beer does more than scan | Beer’s System Four passages, including those Chapter 2 reads as a checkable wager on the System 3–System 4 homeostat[^ch28-s4] | The wider modeling and anticipation functions cannot be found in the stated corpus | Revise the chapter to a charge against later reconstruction, not Beer’s text | Register terms and functional criteria before searching the named Beer corpus |
| The good-regulator result is routinely cited beyond what it proves | The 1970 paper[^ch28-conant] and a dated sample of VSM citations, with Beer’s own restatements as a separate stratum[^ch28-restate] | The sample consistently states the paper’s actual criterion and concessions | Withdraw the citation-practice charge while retaining the separate question of organizational modeling | Blind-code a reproducible citation sample against declared overclaim criteria |

These tests are worth nothing without exact editions, dates and search surfaces. A term missing from an index may still be present in an argument, and a gap in one corpus says nothing about every text Beer wrote. The discipline cuts both ways. A coder who cannot find a functional criterion because we wrote the criterion loosely has learned something about our protocol, and nothing about Beer.

## When an analogy bears weight

Chapter 2 asked this chapter for a distinction it needed and did not have: between an analogy that decorates an argument and one that carries load. The neurophysiological charge is the place to draw it, because it has an obvious worthless version, and that is the version most people would write.

"Beer's neuroscience is outdated, therefore the VSM is unfounded" is a bad argument, and it attacks a claim Beer disclaimed in print. In Chapter 10 of *Brain of the Firm* he anticipates exactly the objection that he conflates pain circuits, synaptic biochemistry and sleep-arousal control, concedes it, and states his warrant: "I want to distinguish sharply between the way in which the body achieves results and the logic of the results it achieves. Neurophysiological descriptions in this book are there because they are interesting, and we can learn from them." He gives no glossary of neurophysiological terms because they are "names for bits and pieces inside the body rather than for definable concepts". The training plan drawn up for Chile would teach the cybernetic principles "but not the neurophysiology". Beer also wrote that the more the neurophysiological model is studied in detail, "the more astonishingly accurate does it seem to be in reflecting what happens in the firm." A paper that ignores all this refutes a position Beer disclaimed and only half asserted.[^ch28-beer]

The easy debunking also aims at things Beer never used. *Brain of the Firm* nowhere mentions MacLean, "triune" or "limbic", and it has no Sherrington, no Pavlov, no Moruzzi.[^ch28-search] The triune brain is discredited in comparative neurobiology and still widespread outside it, in psychology textbooks among other places.[^ch28-triune] It is not Beer’s, and attacking it here attacks a straw Beer.

The discriminator therefore has two halves, and it applies beyond this case. An analogy is decorative where the argument survives its removal, and load-bearing where a conclusion is inferred from a property of the source domain. The audit is worth funding only at joints of the second kind, and there are at least three. First, System Four’s core mechanism in *Brain of the Firm* rests substantively, not decoratively, on Kilmer, Blum and McCulloch’s 1969 computational model of the reticular formation as a behavioral-mode selector,[^ch28-kilmer] and that model is obsolete. The modern arousal picture runs parabrachial to basal forebrain, with the monoamines modulating arousal rather than generating it, noradrenaline among them as a gain control, and thalamic lesions have remarkably little effect on wakefulness.[^ch28-arousal] *The Heart of Enterprise* derives System Four without the reticular formation, from the requirement that it contain a model of the whole system, so a result against the 1969 model would reach the System Four of *Brain of the Firm* and leave that derivation standing.[^ch28-s4-hoe] Second, some of Beer’s structural inferences draw their force from anatomical position. To argue that a staff function is really command because the diencephalon sits on the vertical axis is to argue from where a thing is. Third, the cortical apex. The cortical placement of System Five is the element current work challenges most directly, and the specialist objection to locating a self anywhere is precisely that it reproduces the homunculus. Beer’s own account of identity does not rest on the placement. Viable systems, he wrote, "are self-referential. Their logic closes-in on itself," and in that lies "the maintenance of identity"; System Five he called "the multinode," "an elaborately interactive assemblage of elements."[^ch28-closure] The audit should test whether that distributed account survives, not the apex.

The other half constrains the result rather than the target. We must equally resist writing "the VSM anticipated predictive processing". A triumphalist paper would be worse than a debunking: it would be unfalsifiable, and it would let us count a correction as a confirmation. The test is whether the audit produces at least one statement of the form *this VSM claim is wrong, and here is what replaces it*. An audit that produces none has told us only that we are fluent, which we already knew.

## Comparative exposure

The harder question is whether a VSM diagnosis adds anything beyond neighboring instruments. Part Two credited those literatures with finding the same components independently, and that credit has a consequence. A vocabulary may redescribe an organization elegantly without improving prediction, discrimination or intervention, and elegance is not a result we can bank.

| Claim at risk | Evidence that bears | Result against us | What changes | Smallest fair test |
|---|---|---|---|---|
| A VSM diagnosis identifies recursion levels rather than drawing them | Interaction-rate blocks, time-span measures and blind VSM diagnoses on the same organization | Diagnosed levels fail to track either independent measure beyond chance or analyst discretion | Replace the claim of discovered levels with an explicitly negotiated design convention | One organization, one interaction matrix, one prior diagnosis and two blind analysts |
| VSM diagnosis predicts a failure or outcome unavailable to simpler alternatives | Pre-outcome engagement documents, outcomes and matched baseline frameworks | VSM coding adds no predictive or diagnostic value over a simple baseline and a neighbor | Retain it, at most, as a facilitation language rather than a predictive theory | Ten retrospective interventions, failures retained and coded by a non-participant |
| External diagnosis changes the client system in the way the diagnosis recommends | Pre-stated intervention criteria and client as well as consultant outcome accounts | Diagnosis produces no discriminable change, or only changes the consultant’s description | Surrender the intervention claim; retain perturbation as the narrower effect | Register outcomes against criteria written before the intervention |

None of the three rows asks the model to be right about everything. Each asks only that a diagnosis produced under the model be distinguishable, by somebody who did not produce it, from one produced without it. The bar is low, and we have not yet cleared it. Chapter 29 gives the protocols. What matters here is the consequence of losing. The result cannot be absorbed by declaring that the organization had misidentified its system in focus, unless that boundary claim was itself registered in advance. Nor can it be absorbed by observing that the client failed to implement, unless implementation criteria were written down before the engagement closed. The second and third rows are where a negative result costs most, because a facilitation language and a predictive theory are priced differently, and much of our practice has been sold at the second price.

## Structural exposure

Part Four states the narrow claim and the residue that has no other home, and Chapter 24 sets out six kinds of work that nothing else supplies. The strongest version of that position is a necessity claim, and a necessity claim should face the strongest counterexample available, not the most convenient one.

Beer answered the obvious counterexample before anyone raised it. System Two, he wrote in 1985, "is always present, or the organization would shake itself to pieces." Six years earlier: "System Four exists — it has to: but it is disseminated, and lacks integration."[^ch28-beer-present] In *Brain of the Firm* he complained that "our age has produced so many large-scale organizations without a System Two," and in the next sentence reported that most of the successful ones he had observed "have been entirely unofficial and largely unrecognized."[^ch28-beer-unofficial] Where System Four is virtually empty, the metasystem "collapses into System Three," and the institution becomes a decerebrate cat: "It lives on; its viability is ensured by a bogus environment, and is sustained by artificial sustenance."[^ch28-beer-cat] And when someone asks whether an organization that exists is "by definition a viable system," Beer's reply separates the two: "The fact that it is there does not prove that it is effectively there."[^ch28-beer-there]

Read together, these passages close the obvious route to a counterexample. A persisting organization that lacks a function officially is found to perform it unofficially, and one that performs it badly is diagnosed as pathological, not as a system that has survived without it. Nobody added that defense after a failed test. It is in the founder's statement of the claim, and it is why a surviving system that lacks one of the five is so hard to report. The necessity rows below therefore rest on one coding and one outcome, both registered before any case is read. Other chapters refer to them as the coding in Chapter 28.

**The coding.** For a named system at a named level and over a stated observation window, each of the five functions receives one of three codes, by indicators registered in advance:

- *Official*: the function is performed by a role or unit designated to perform it.
- *Disseminated*: the function is performed, but unofficially or by no designated role. This is Beer's disseminated System Four and his unrecognized System Two.
- *Absent*: nobody performs the function in the observation window, as judged by the registered indicators.

The indicators state, for each function, what activity in the records counts as performing it. A coder who finds a function only by reading it into activity after the outcome is known has not coded it. Disseminated is a code, not a refuge: where Beer names the pathology that a weak function produces, as with the "uncontrollable oscillation" he expects without System Two,[^ch28-beer-oscillation] that expectation is registered with the code.

**The outcome.** The outcome is Beer's viability, not bare persistence. He took the definition from the dictionary, "able to maintain a separate existence,"[^ch28-beer-viable] and then refused to let the fact of existence stand for it, as the decerebrate cat shows. The outcome indicators are therefore registered separately from the five codes and must not be computed from them, or the test is circular. We take five from Beer’s own distinction between being there and being viable, each registered for the named system at the named level before the window opens:[^ch28-outcome]

- *O1, separate existence*: at the end of the window the unit is still an identifiable unit with its own decision rights, not dissolved, merged, taken over or run directly by the containing system.
- *O2, effectiveness*: attainment of the purpose the unit itself named before the window, against its own baseline.
- *O3, efficiency*: resources used per unit of that attainment, against the unit’s baseline and comparable units, dependence on subsidy included.
- *O4, recovery*: after a change in its environment named in advance, O2 returns to a stated level within a stated period.
- *O5, no rescue*: the unit continued without extraordinary support from outside or from the containing system.

Loss of viability is registered as O1 failing, or two of O2 to O5 failing, and the five are also reported separately. No indicator may be built from the activities used to code a function: O4 is the restored performance, not whether a planning unit produced a response. Outcome coders do not see the function codes, and function coders do not see the outcomes. Cases without outcome data are excluded, not imputed. O1 is Beer’s minimum; whether he would count efficiency as part of viability rather than of performance is our judgment, not his wording.

| Claim at risk | Evidence that bears | Result against us | What changes | Smallest fair test |
|---|---|---|---|---|
| Viability in a changing environment requires the five functions | Cases coded official, disseminated or absent for each function by registered indicators, and for viability by indicators independent of that coding | A system coded absent for one function keeps its viability through the window, or loses it from a structural cause the model cannot see with all five coded present and none pathological | Revise necessity into a contingent configuration claim and identify the substitution or omitted cause | Publish the coding protocol and outcome indicators, then seek deviant cases rather than exemplary ones |
| The six claimed residues are not supplied elsewhere | Direct conceptual and operational comparison with named neighboring theories | A neighbor specifies the same work with equal or greater diagnostic precision | Remove that uniqueness claim while retaining any useful translation | Adversarial paired reading by specialists from both traditions |
| The five systems are jointly effective even if individually insufficient | A cross-case configuration with outcomes fixed independently of the diagnosis | No configuration containing the five predicts viability better than plausible alternatives | Surrender the configural defense rather than moving the outcome criterion | A preregistered qualitative comparative analysis[^ch28-ragin] on an open case register |
| Regulation fails when a system's relaxation time exceeds the interval between shocks | Relaxation times and shock intervals recorded for named loops before the outcome, and viability indicators independent of them | Systems whose recorded relaxation times exceed the interval between shocks keep their viability as often as those whose do not | Drop the condition as a prediction and keep relaxation time as a descriptive parameter | Register relaxation times and shock intervals for the loops of a few named systems, then follow the viability outcome |

The first row is the home of the two candidates Chapter 9 named, restated in the terms of the coding: a system coded absent for one of the five that keeps its viability in a changing environment, and a system that loses its viability from a structural cause the model cannot see, with all five coded present, intact and healthy. We accept both conditions in advance, and both are met by observation rather than by argument. Neither is protected by the objection that the analyst drew the boundary in the wrong place. The coding protocol is published before the cases are sought, and cases are sought because they look deviant, not because they look instructive. Michael D. Pfiffner operationalized viability as a whole for 135 organizations, which shows the claim can be measured; with a reflective specification no result could have shown a function dispensable (Chapter 29).[^ch28-pfiffner]

Beer wanted a test of this kind for his own work. Of the identity exercise in *The Heart of Enterprise*, which clusters managers' answers about the cohesion of their enterprise, he wrote that "maybe (one day) thresholds will be established, to exceed which will predict collapse."[^ch28-beer-thresholds] Nobody has established them. A registered threshold of that kind would be a structural claim that could lose, and it is the founder's proposal, not ours. The same book supplies a second candidate, the last row of the table above. Once a loop is disturbed, Beer asks how long it takes to return to stability, and answers that "if the system has not returned to stability before another shock drives it once more towards instability, then regulation is impossible."[^ch28-beer-relaxation] Relaxation time and the interval between shocks can be recorded before the outcome is known, which makes the condition a prediction that could fail.

Chapter 27 lists five items of residue that Part Five must turn into claims against which evidence can be registered: the closed-set necessity claim, the transducer, the algedonic architecture together with the justification offered for it, one currency across scales, and freedom as a quantity, which Beer declared computable in principle and nobody has computed, and which we call *residual variety* in the sense defined in Chapter 26. Only the first is converted above in its own right. The transducer has its row in Chapter 25’s receipt, a blind coding of documented regulatory failures that loses if failures do not concentrate at crossings, or if coders cannot tell crossing from channel without VSM labels as their evidence. The algedonic architecture has its row in Chapter 30, where the channel is red-teamed against modern incident and escalation practice. Freedom as a quantity takes the row below. One currency across scales cannot be converted yet: until the measurement defect Chapter 27 names is addressed, there is no quantity whose travel between a person and a planet could fail.

| Claim at risk | Evidence that bears | Result against us | What changes | Smallest fair test |
|---|---|---|---|---|
| Residual variety, in Chapter 26’s sense, can be counted | The distinctions available to a named unit, recorded before and after one deployment by a procedure registered in advance | Two coders cannot agree on the count, or the count does not move where the design says room was returned or taken | Withdraw freedom as a quantity and keep it as the question Chapter 26 asks at diagnosis | Count the distinctions available to one unit before and after one deployment, as Chapter 26 proposes |

The companion register makes configural analysis, structural disconfirmation and the intervention question answerable across cases. Their claims have been converted above; Chapter 29 keeps the protocols in full, and Chapter 30 prices the whole program. The division is deliberate. It stops the same paragraph from being counted once as a test and again as an experiment, which is the arithmetic by which a research agenda comes to look larger than it is.

[^ch28-beer]: Stafford Beer, *Brain of the Firm*, 2nd ed. (John Wiley & Sons, 1981): the objection and the warrant, ch. 10, "The Biggest Switch", 139–40; the glossary note, 401; the Chilean training plan, ch. 18, "The October Watershed", 317; the claim of accuracy, 109, followed there by the example of the sympathetic ganglia.

[^ch28-nosek]: The rule is the one the preregistration literature argues for: Brian A. Nosek, Charles R. Ebersole, Alexander C. DeHaven, and David T. Mellor, "The Preregistration Revolution," *Proceedings of the National Academy of Sciences* 115, no. 11 (2018): 2600–2606, https://doi.org/10.1073/pnas.1708274114, on the difference between prediction and postdiction, and on hindsight bias as the reason the two are confused unless the questions and the analysis are fixed before the outcomes are seen. Read in the authors’ version on OSF (https://osf.io/2dxu5); the published abstract is the same. Chapters 29 and 30 refer to this note.

[^ch28-kilmer]: W. L. Kilmer, J. Blum, and W. S. McCulloch, *The Reticular Formation* (Air Force Office of Scientific Research, 1969), the report Beer lists in *Brain of the Firm*, 2nd ed., 407, under "On 'arousal' mechanisms (System Four)". The journal version gives the authors in a different order: W. L. Kilmer, W. S. McCulloch, and J. Blum, "A Model of the Vertebrate Central Command System," *International Journal of Man-Machine Studies* 1, no. 3 (1969): 279–309, https://doi.org/10.1016/S0020-7373(69)80025-8. Beer calls the ascending reticular formation "the core of System Four" (*Brain of the Firm*, 148).

[^ch28-s4-hoe]: Stafford Beer, *The Heart of Enterprise* (John Wiley & Sons, 1979), 235: "System Four must contain a model of the viable system of which it is the System Four." The book does not mention the reticular formation (full-text search of the OCR copy, 29 September 2026).

[^ch28-closure]: Stafford Beer, *Diagnosing the System for Organizations* (John Wiley & Sons, 1985), 4; Beer, *Heart of Enterprise*, 260, where closure is "a self-referential process"; Beer, *Brain of the Firm*, 2nd ed., 201, for the cortex as System Five, and 205, for the multinode.

[^ch28-restate]: Beer, *Heart of Enterprise*, 235: "any would-be regulator must contain a model of that which needs to be regulated"; and 352: "Every regulator contains a model of whatever is regulated. This model must have requisite variety."

[^ch28-outcome]: Beer, *Diagnosing the System*, 1: an organization "is viable if it can survive in a particular sort of environment," and its existence is separate, "so that it enjoys some kind of autonomy, it cannot survive in a vacuum"; Stafford Beer, *Think Before You Think*, ed. David Whittaker (Wavestone Press, 2009), 147, where viability means "capable of independent existence," and 148, where an organization that is there is not thereby "effectively there" or "efficiently there," and some operate "at such an enormous cost that they are becoming less and less viable in front of everyone's eyes."

[^ch28-pfiffner]: Michael Dominik Pfiffner, "System Viability of Organizations and the Aetiology of Organizational Crisis: A Quantitative Assessment of Stafford Beer's Viable System Model" (PhD diss., Utrecht University, 2017), https://hdl.handle.net/1874/356772; the reflective measurement model, 180. Chapter 18 discusses the design.

[^ch28-beer-present]: Stafford Beer, *Diagnosing the System for Organizations* (John Wiley & Sons, 1985), 66; Stafford Beer, *The Heart of Enterprise* (John Wiley & Sons, 1979), 143.

[^ch28-beer-unofficial]: Beer, *Brain of the Firm*, 2nd ed., 175. See also 183: "there is always a System Four, even if it is not identified in quite the form specified here."

[^ch28-beer-cat]: Beer, *Heart of Enterprise*, 265.

[^ch28-beer-there]: Stafford Beer, *Think Before You Think: Social Complexity and Knowledge of Knowing*, ed. David Whittaker (Wavestone Press, 2009), 148.

[^ch28-beer-oscillation]: Beer, *Heart of Enterprise*, 177: "System Two is logically necessary to any viable system, since without it System One would be unstable — it would go into an uncontrollable oscillation."

[^ch28-beer-viable]: Beer, *Diagnosing the System*, 1, quoting the *Oxford English Dictionary*; the same definition opens chapter 5 of *Heart of Enterprise*, 113, and chapter 8, 199: "The ability of a system to maintain a separate existence, which we call viability, depends on a number of necessary conditions which, in sum, will also be sufficient."

[^ch28-beer-thresholds]: Beer, *Heart of Enterprise*, 432.

[^ch28-beer-relaxation]: Beer, *Heart of Enterprise*, 390. Checked 29 September 2026 in the calibre PDF.

[^ch28-search]: Full-text search of an OCR copy of the 2nd edition (Wiley, 1981), including the bibliography, for "MacLean", "triune", "limbic", "Sherrington", "Pavlov" and "Moruzzi", 23 September 2026: no hits. The 1972 first edition has not been searched.

[^ch28-s4]: Stafford Beer, *Brain of the Firm*, 2nd ed. (John Wiley & Sons, 1981), 182–83 and 189, and Stafford Beer, *The Heart of Enterprise* (John Wiley & Sons, 1979), 240, the passages Chapter 2 relies on. The phrase "prediction channel", which Chapter 2 used for the line to command until 12 October 2026, occurs in none of the eight Beer books searched there; Chapter 2 now reads the same passages as a checkable wager on the System 3–System 4 homeostat.

[^ch28-conant]: Roger C. Conant and W. Ross Ashby, "Every Good Regulator of a System Must Be a Model of That System," *International Journal of Systems Science* 1, no. 2 (1970): 89–97, https://doi.org/10.1080/00207727008920220. Chapter 3 sets out what the paper proves and what its authors conceded.

[^ch28-arousal]: Patrick M. Fuller, David Sherman, Nigel P. Pedersen, Clifford B. Saper, and Jun Lu, "Reassessment of the Structural Basis of the Ascending Arousal System," *Journal of Comparative Neurology* 519, no. 5 (2011): 933–56, https://doi.org/10.1002/cne.22559: extensive thalamic lesions had little effect on EEG or behavioral measures of wakefulness; large basal forebrain lesions, and lesions of the parabrachial nucleus and precoeruleus area that project to it, left animals unresponsive; and cell-specific lesions of the monoaminergic and cholinergic groups have never reproduced that coma. For noradrenaline as a gain control, Gary Aston-Jones and Jonathan D. Cohen, "An Integrative Theory of Locus Coeruleus-Norepinephrine Function: Adaptive Gain and Optimal Performance," *Annual Review of Neuroscience* 28 (2005): 403–50, https://doi.org/10.1146/annurev.neuro.28.061604.135709.

[^ch28-triune]: Joseph Cesario, David J. Johnson, and Heather L. Eisthen, "Your Brain Is Not an Onion With a Tiny Reptile Inside," *Current Directions in Psychological Science* 29, no. 3 (2020): 255–60, https://doi.org/10.1177/0963721420917687, who document the layered-brain misconception in introductory psychology textbooks and in psychological research. We have not surveyed management writing for it.

[^ch28-ragin]: Charles C. Ragin, *The Comparative Method: Moving beyond Qualitative and Quantitative Strategies* (University of California Press, 1987), as in Chapter 15.