The 7th Video Lecture is here: Building Agent Based Models: A Realist Methodology The material was broken up into four parts: Chapters 7A, 7B, 7C, and 7D, described and linked below. This lecture provided feedback on model building and methodology based on student homeworks which attempted to model the Ultimatum Game. In the revised writeup, this will become Chapter 3, as that is where it belongs logically - but this was the actual sequence followed in the course.
Chapter 7A · What Is a Model For?
The chapter starts from the class’s first models of the ultimatum game: what they did well, and what they lacked. Lewis Carroll’s map on the scale of a mile to the mile shows that a complete description is perfectly accurate and useless. We learn from experience only when similar situations recur, so a model must keep the details that drive the result and drop the scaffolding. The chapter then sets out economics’ standard for a good model, optimisation and equilibrium, and shows how it was saved by tuning a weight on fairness. A fit that was built in cannot test the model, so fit is a design requirement, not a test. Against this it sets the book’s alternative, rules that state reasons and are judged by whether people actually reason that way, and introduces nominal and real models. It then asks why economists will not ask people their reasons, from Friedman’s billiard player to the marginalist controversy. The last section shows a performative theory at work: students learn to distrust their own hearts, business schools teach the frame, and the opioid epidemic shows that frame at scale, confirmed in court.
Chapter 7B · Puzzles and Hidden Causes
This chapter asks where a model comes from: from a puzzle, through a hypothesis about causes we cannot see. It sets out the four puzzles in the class’s results, and shows that stating a puzzle already looks beyond the case, since “unusual” means unlike similar cases. Peirce’s abduction leads from a surprising fact to a hidden cause. The Qur’an asks the same reasoning of us in reading the world as signs, from the Bedouin’s “the camel’s dung points to the camel” to the modern argument from design. A model claims hidden drivers, adds scaffolding, and claims the scaffolding is robust. An imaginary bayramlık shop, whose busiest month moves with the festival, shows a nominal and a real reading of the same pattern; the Machiguenga and the Lamalera show the same contrast in the game. The chapter then works through the 59 failed pairings: four rival explanations, and how one’s role shapes judgements of fairness, from the class’s split by role to Qur’an 83:1–3. It closes with answers that travel, underdetermination and the fitting machine.
Chapter 7C · Stating and Testing a Hypothesis
This chapter shows how to state a hypothesis precisely enough to compute, and where to test it. A rule card joins reason, belief and situation, and the test of a rule is to change the situation and see whether it gives a new answer. Ar’s rule shows reasons dropping out one at a time in the dictator game and in secrecy, and how a dominant reason hides the others. Se’s rule shows that beliefs inside a rule can be wrong, and wrong in one direction, which helps explain the failed pairings. The chapter shows how to check drivers against scaffolding, noting that the same assumption can be either, depending on the result asked about. It then finds the situations in which rival hypotheses disagree: Ab’s two models fit the ultimatum results equally well, and secrecy decides between them. Opening with Ibrahim’s star, moon and sun as the form of a test, the last section sets the standards for a test beyond the case. A test means a new situation, a comparison of whole distributions, honesty about when a prediction was made, and testing the weakest assumption first. A legitimate scope names a variable; a shield names the case.
Chapter 7D · Checking Against People, and Two Kinds of Model
This chapter checks hypotheses against the people whose motives they describe: ourselves first, then a partner, the literature, and the answers people give when asked. Social consensus is defined as agreement that a motive is plausible, not that everyone shares it, and a validated model is valid only for now. The chapter brings three accounts of testing into one, from Chapter 1, the homework and these chapters. It then sets out what to do when validation fails, and how to use AI as a tutor rather than an author. Checking by hand gets its own worked example, the 108 pairings in a spreadsheet grid. The method of Chapters 7A to 7D is gathered into seven steps, each tied to an example from the class’s work. Nominal and real models are set side by side, with the instrumentalist defence answered: every reform aims at a mechanism. The chapter closes all four with wallāhu aʿlam: strive to get the mechanism right, offer it for others to test, and hold it with humility.
Supplementary Materials:
Zaman, Asad, Scarcity: East and West (March 31, 2010). Journal of Islamic Economics, Banking and Finance, Volume 6, Number 1, January-March 2010. p. 87-104, Available at SSRN: https://ssrn.com/abstract=3061271