Haute Lumière

Commerce · II.09 · MMXXVI · daylight

La Bourse  /  Volume II  /  Nº II.09  /  Workbook — the student

WORKBOOK — THE STUDENT

Chapter II.09 · Evolution and Economic Selection

For the person learning this alone or in a seminar. A term of practice, one project carried the whole way, and a self-assessment you can actually mark. The subject is a population you can see with your own eyes.

---

WHY THIS WORKBOOK IS DIFFERENT

Most teaching of evolutionary economics stops at the metaphor, and a metaphor cannot be practised. This workbook has you compute a real selection term on a real population, predict a real rate, and then find out whether you were right.

You will need a population of at least fifteen members that you can observe over time and about which two numbers are recordable: a trait and something that determines who persists. Candidates that work well:

• The cafés, shops or restaurants on one street, over one academic year.
• The student societies at your institution, with membership as fitness.
• Open-source repositories in one ecosystem, with forks or contributors.
• Recipes, methods or citation practices inside one research field.
• The teams in one league, with a tactical trait and league position.

Choose one in the first week and do not change it. The discipline of a fixed population is most of the learning, because a population you can swap is a population you can fit a story to.

---

PART ONE — DISCOVERY

Weeks 1–4: find a population and see the variation

Exercise 1.1 — The census (3 hours)

Record every member of your population. For each one, write down:

1. An identifier.
2. The trait z, a number. It must be observable without asking anyone's permission and without a survey — you need to be able to re-measure it later.
3. Its size or weight now.
4. The date.

That table is the parent generation and everything else depends on it. Date it. Photograph it if it is a street.

Exercise 1.2 — Is there anything to select on? (1 hour)

Compute the variance of z across your population, and the range. Then answer in writing: if selection ran hard on this population for ten generations, could it move anywhere? A population where every member has nearly the same z cannot respond, no matter how strong the pressure. If yours is like that, you have found something real — write it up, and either widen the definition of the population or change the trait.

Exercise 1.3 — Name what is already working (1 hour, out loud)

Before analysing anything, find the member of your population doing the thing best, and write two hundred words on how they came to do it. Who did they learn it from? Where did it come from before that? You are tracing an inheritance line, and it is the most interesting part of the term.

Exercise 1.4 — The three legs (30 minutes)

For your population, answer in one sentence each:

• Where does variation come from? Who tries new things and why?
• What does selection actually consist of here — customers, funders, votes, attention, survival?
• How does anything inherit? Name the mechanism: people moving, copying, training, documents, imitation.

If you cannot name the inheritance mechanism, your population may not be evolving at all. That is a legitimate finding and a good essay.

---

PART TWO — THE ARITHMETIC

Weeks 5–8: compute the selection term before you argue

Exercise 2.1 — Offspring per parent (2 hours)

Define w for your population and defend the definition in writing. w is offspring per parent — weight carried into the next period per unit of this period's weight. For shops it might be still open, and floor space; for societies, membership next term over membership this term; for repositories, contributors next quarter over contributors this quarter.

The rule that matters: w must be measurable independently of the trait. If your fitness measure contains the trait, you have built a tautology and everything after it is decoration.

Exercise 2.2 — Compute Cov(w,z)/w̄ by hand, once (2 hours)

Do this once on paper before you ever do it in a spreadsheet, with the chapter's five-firm table as your worked model:

  z̄ = 6.000 %      w̄ = 1.100      E(w·z) = 7.320
  Cov(w,z) = 7.320 − (1.100 × 6.000) = 0.720
  selection term = 0.720 / 1.100 = 0.6545 pp per generation

Now do yours. Write out the w·z column, take its mean, subtract w̄·z̄, divide by w̄. State the units. Units are not a formality here — the answer is in trait-units per generation, and if you cannot name the generation length you do not yet have a model.

Exercise 2.3 — The second fitness function (2 hours)

Now choose a different plausible measure of who persists in your population — one a different observer would reasonably use. Recompute the selection term.

In the chapter's population, revenue-weighting gives +0.6545 and margin-weighting gives −0.5688: a swing of 1.2234 points a generation, or 12.23 points over thirty years, on the same firms doing the same things.

Write five hundred words on what your own two numbers mean. This exercise is the chapter.

Exercise 2.4 — Put a clock on it (1 hour)

Take the best variant in your population and estimate its selection advantage s per year. Compute the time to half the population from its current share:

   t = ln[ (x₁/(1−x₁)) · ((1−x₀)/x₀) ] / s

From 2 percent to half, that is ln(49.00)/s = 3.8918/s — 129.7 years at s = 0.03, 25.9 years at s = 0.15. Write down your number and then ask the question that matters: is anybody deliberately copying it? If yes, the clock is irrelevant, and that is the most useful thing you will learn all term.

Exercise 2.5 — The null (1 hour)

Before you claim a trend, compute the null. If a generation's response is R and the direction reverses, n generations produce an expected displacement of zero with a standard deviation of R·√n. At R = 1.40 and n = 10 that is 1.40 × 3.1623 = 4.43 points. Any observed move smaller than that is indistinguishable from nothing.

Compute your own band. Write it at the top of your project notebook in ink.

---

PART THREE — DREAM AND DESIGN

Weeks 9–12: the boundary, and the prediction

Exercise 3.1 — Draw the boundary two ways (2 hours)

Every population sits inside a larger one. Find the group level above your members — the street, the federation, the ecosystem, the league — and answer:

1. Does the trait cost the individual and benefit the group?
2. Estimate V_within and V_between.
3. Compute the threshold b/c > 1 + V_within/V_between. In the chapter's example that is 1 + 2.6667/9.0000 = 1.2963, with b/c = 2.0000 against a critical b* of 0.03889.
4. State whether your trait is plausibly above or below its threshold, and what would move it.

Exercise 3.2 — Check the five criteria (1 hour)

Go through Wilson and Sober's five, in order, and mark your population pass or fail on each with one sentence of evidence. Most real populations fail criterion four — group-level heritability — and finding out which one yours fails is the result.

Exercise 3.3 — The prediction, sealed (1 hour)

Write on one page, dated and signed:

• The trait, the population, and the generation length.
• z̄ now, to three decimals.
• The predicted Δz̄ for the next generation, from your selection term.
• An interval, stated as a number, e.g. ±0.25.
• The one thing that, if it happened, would make you wrong for a reason other than the model being wrong.

Seal it. Give a copy to someone else. This page is the difference between the subject and its imitation, and it is the only exercise in the workbook that cannot be done retrospectively.

---

PART FOUR — DESTINY AND DELIGHT

Weeks 13–16: score yourself, and watch the inheritance

Exercise 4.1 — Re-census (3 hours)

Repeat Exercise 1.1 exactly. Same trait, same definition, same method. Compute the realised Δz̄. Open the sealed page.

Then write, in this order: what you predicted, what happened, and which of the two Price terms accounted for the difference — was the mean moved by who persisted, or by the persisters changing? In the chapter's population selection did 88.5 percent of the work and transmission 11.5 percent.

Exercise 4.2 — Follow one routine (2 hours)

Pick one practice you saw in Exercise 1.3 and find out where else it now is. Ask three people. Trace it forward rather than backward this time. The pleasure of this exercise is the point of the chapter: you are watching something inherit, in public, without anyone managing it.

Exercise 4.3 — The wrong answer, written up (1 hour)

If your prediction failed, write six hundred words on why — and be strict about the difference between the model was wrong and the measurement was wrong. A failed prediction honestly diagnosed is worth more marks in this workbook than a lucky one, and it is the more common professional experience.

Exercise 3.4 — Design a better selection environment (2 hours)

You now know what your population is being selected on. Write a one-page criterion that you would propose instead, and make it specific enough to be scored: the trait, the boundary it is measured at, the period, who verifies it, and how much notice a change requires. Then write the honest paragraph underneath — what would this criterion select for that you did not intend? Every measure produces a population fit for itself, including yours, and the paragraph where you name your own criterion's blind spot is the one your supervisor will read first.

---

THE FIVE MISTAKES, SO YOU CAN SEE THEM COMING

Every cohort makes these, and they are worth naming before the term rather than after it.

Fitness that contains the trait. You define w as growth and z as something that is part of how growth is recorded. The covariance is then guaranteed and means nothing. Test it by asking whether someone who had never heard of your trait could compute your w.

A generation that is never stated. A response of 1.40 points per generation is meaningless until you say how long a generation is. For shops it may be a lease cycle; for societies, an academic year; for repositories, a release. Write it down in week one.

Retro-fitting the population. Members get quietly dropped because they are hard to measure, and the ones that are hard to measure are usually the ones that failed. This is the single most common way a student project turns into a just-so story without anyone deciding to.

Reading a trend off a small move. Compute the null band before the claim. An observed change inside R·√n is consistent with no selection at all, and at R = 1.40 over ten generations that band is 4.43 points wide.

Mistaking deliberate copying for selection. If a practice spread faster than the replicator clock allows, somebody taught it. That is a better finding than the one you were looking for, and it belongs in the project.

---

THE TERM PROJECT

One population, carried the whole way

Deliverable: eight pages.

1. The population, with the census table and its date. One page.
2. The three legs — variation, selection, inheritance — named and evidenced for your case. One page.
3. The arithmetic — z̄, w̄, E(w·z), Cov(w,z), both selection terms, with units and working shown. Two pages.
4. The boundary analysis — V_within, V_between, the threshold, and the five criteria marked. One page.
5. The sealed prediction, reproduced as written, with its date. One page.
6. The outcome, the realised Δz̄, and the two-term decomposition. One page.
7. What you would measure differently, and what the null band was. One page.

The standard. A project that predicts badly and diagnoses honestly passes. A project that predicts nothing and explains everything afterwards does not, at any level of polish — that is the just-so story, and identifying it is half of what this chapter teaches.

---

SELF-ASSESSMENT

Mark yourself honestly. Three points each.

Under 12 — you have learned the vocabulary. Redo Part Two. 12 to 18 — you can compute a selection term, which most economists cannot. 19 to 24 — you can falsify your own claim. That is the whole subject.

---

CARRYING IT FORWARD

Three things worth keeping beyond the term.

The habit of asking what is being selected on. Of any organisation you join, any field you enter, any market you study: what number determines who carries weight into the next period? It is rarely the stated objective and it is always findable.

The habit of sealing a prediction. One page, a number, an interval, a date. It costs an hour and it is the only reliable defence against believing your own explanations.

The habit of tracing inheritance. When you see something working, ask where it came from and who carried it. Most of what an economy knows is held this way, in nobody's records, and learning to see it is a permanent advantage.

---

APPRECIATIVE QUESTIONS FOR YOUR SEMINAR

1. Whose population showed the widest variation, and what has protected that variety from being tidied away?
2. Which sealed prediction came closest, and what did that person measure that the rest of us did not?
3. What is the best example anyone found of a routine that outlived the organisation that invented it — and how did it travel?
4. Where in this institution is something being selected on that nobody chose deliberately, and what would it take to write that criterion down?