Back to School
Judy goes back to school, not to be certified but to be corrected. The lessons are unwelcome and better than she expected: friends in rooms she had written off, colleagues doing her own work more rigorously than she has, coaches whose corrections she had not stayed long enough to hear. She comes out heavier, harder to knock down, more attached to boxing than when she thought she already knew it. What she wants now is a fight she could lose.
Most chapters in Parts One to Three, and Chapter 25 in Part Four, close their charge with a receipt: the smallest piece of work that would settle the matter, what it would cost, and what a negative result would establish. The convention has held since Chapter 1, so its form here will surprise no one. Written one at a time, the receipts could be read as a list of complaints with prices attached. Part Five collects them, orders them by what they would tell us per unit of effort, and makes them into a program.
It is short, dry and priced. It contains only research earned by charges actually made in this book. Nothing is proposed because it would be interesting; the wider program of work this tradition needs stays in the companion volume, where it can be argued on its own terms.
Chapter 28, What Would Count Against It?, states what observation would make the model give something up. Chapter 9 says that repair is available to us and that we have not made it, and until it is made, every other proposal in this part is decoration. Chapter 29, The Two Cheapest Decisive Experiments, sets out the two decisive tests that could be run now, with existing data and existing access, by people already in the field. Chapter 30, The Bill: A Book Plus a Live Register, sets out what the whole program costs, and why no total is struck.
The standard is Beer’s. “So remember: a model is neither true nor false: it is more or less useful,” he wrote in 1985, and in 1966 he had said how usefulness is judged, by “testing of the predictive value” of the model: “If this predictive value is low, we had better start again.”1 The tradition kept the first sentence and dropped the test.
We designed the experiments so that we could lose them, and that is the specification. Each is set up so that a result running against the model is as interpretable as one running for it, and we have said in advance which is which. A negative outcome is then a publishable finding rather than a failure. It is also the only kind of outcome that would change what any of us tells a client next year. Advocacy can be written in the most measured prose available and still leave nothing at stake, so the difference between it and an agenda of scrutiny is not one of tone. What is proposed here leaves something at stake, and the something is ours.
Notes
- Stafford Beer, Diagnosing the System for Organizations (John Wiley & Sons, 1985), 2, where the sentence is set as three broken lines; Stafford Beer, Decision and Control (John Wiley & Sons, 1966), 113. ↩