Haute Lumière

Commerce · V.09 · MMXXVI · daylight

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

WORKBOOK — THE STUDENT

Chapter V.09 · Health as a Return on Capital

For the person studying this alone, or in a seminar, with no claim run to read and no budget to defend. You are not too early. The skill this chapter teaches — telling a measurement from a selection effect — is the most portable thing in the entire volume, and you will use it on your own life first.

---

WHY THIS WORKBOOK IS DIFFERENT

The chapter was written for someone who can pull three years of claims data. You probably cannot. It would be easy to conclude that the method has to wait.

It does not, and here is why. Every claim you will ever hear about what is good for you has the same structure as the wellness ROI, and almost all of them have the same flaw. The people who take cold showers are different from the people who do not, before the cold shower. The students who use the library are different from the students who do not, before the library. The runners were healthier first. Learn to ask what was the gap before and you have acquired a lifetime instrument in one afternoon.

You will also do something the executive cannot: you will run a genuine randomised trial, on yourself, at zero cost, because you control the allocation. That is a rare position and it is worth using while you have it.

---

PART ONE — DISCOVERY

Weeks 1–3: find the claims you already believe

Exercise 1.1 — The belief inventory (75 minutes)

Write down ten things you currently believe are good for your health, your concentration or your mood. Beside each, write the source in one line. Not "it's well known" — the actual source: a friend, a video, a paper you have read, a paper you have heard about.

Now mark each one with a letter.

• R — you have seen a randomised trial.
• O — you have seen a comparison of people who do it and people who do not.
• A — somebody told you, or it is simply in the air.

Most lists come back with one R, three Os and six As. That distribution is the finding. It is not a reason for cynicism; it is a map of where to spend your reading.

Exercise 1.2 — Reconstruct a selection effect (60 minutes)

Take one of your O items and write the two groups out explicitly. Who does this thing? Who does not? Then answer one question in writing: name three ways those two groups differed before anybody did anything.

If you cannot find three, you have not looked. Income, time, prior health, family, a job that lets you leave at five — there are always three.

Exercise 1.3 — The five cases (2 hours, reading)

Read the five Discovery cases in the chapter and, for each, write one sentence naming who could not choose who was in the programme. Volpp's randomisation. California's inspection lottery. PepsiCo's split by diagnosis. STAR's cluster randomisation. Oregon's actual lottery. That single property is what all five have in common, and it is the whole reason they are usable.

---

PART TWO — THE ARITHMETIC

Weeks 4–7: compute before you argue

Exercise 2.1 — Reproduce the chapter's decomposition by hand (2 hours)

On paper, without the module:

  5,000 employees x 150.00 USD                   =    750,000 USD
  participation 56.0 %                           ->     2,800 people
  after:  6,600 - 5,100                          =      1,500 USD
  naive:  1,500 x 2,800 / 750,000                =       5.60 : 1
  before: 6,100 - 4,900                          =      1,200 USD
  effect: 1,500 - 1,200                          =        300 USD
  DiD:    300 x 2,800 / 750,000                  =       1.12 : 1
  ratio                                          =       5.00 x

Then do it again with different invented numbers until the shape is automatic. The shape is the skill; the figures are a demonstration of it.

Exercise 2.2 — The regression-to-the-mean simulation (90 minutes)

In a spreadsheet, generate 2,000 rows with two correlated columns at a correlation of 0.35, a mean of 6,000 and a standard deviation of 12,000. Sort by the first column, take the top decile, and average the second.

You should land near 13,371 against a selected mean near 27,060 — an apparent saving of about 13,689, or 50.6 per cent, with no intervention in the model at all. Seeing it happen in your own spreadsheet is worth more than reading it ten times. Keep the file.

Exercise 2.3 — The power calculation, twice (60 minutes)

Compute, by hand:

  n per arm = 2 (1.96 + 0.84)^2 x SD^2 / effect^2

For spending: SD 12,000, effect 300, giving 25,088 per arm and 50,176 in total. For a proportion, use p(1-p) in place of the variance: a twenty per cent turnover rate and a four-point effect gives 1,568 per arm and 3,136 in total. Write one sentence: which of these could a firm of 5,000 run?

The bracketed term is (1.96 + 0.84)^2 = 7.8400: the first is the standard normal quantile for a five per cent two-sided test, the second for eighty per cent power. Memorise the 7.8400 and you can do this calculation in a meeting.

Exercise 2.4 — The QALY, worked (45 minutes)

Compute 100 × 0.07 × 4.5797 = 32.06 QALYs, and 600,000 / 32.06 = 18,716 USD per QALY. Then place it against 25,400 to 38,100 USD and against 16,429 USD, and write two sentences saying why both comparisons are legitimate.

Exercise 2.4b — The DALY, worked, and the ratio inside it (30 minutes)

Compute the injury side: 12 workers, disability weight 0.20, duration 0.50 years, giving YLD = 1.20 DALYs. Then the fatality side: a death at 45 against a reference life expectancy at birth of 88.87 years gives 43.87 years of life lost. Total 45.07.

Now divide 43.87 by 1.20 and sit with the answer for a moment. One death outweighs twelve serious injuries by more than thirty to one in this arithmetic. Write two sentences on what that means for a safety case built on injury frequency alone — and one sentence on what the DALY method is assuming when it says so, because it is assuming something.

Exercise 2.5 — Your own capture ratio (30 minutes)

You have one too. Estimate how long you will stay in your current situation — course, city, job, relationship to your own future — and compute 0.03 / (0.03 + ln 2 / that number). At 3.90 years it is 14.44 per cent and the hurdle multiple is 6.92 x; at 12.00 years it is 34.18 per cent and 2.93 x.

Then ask the only question that matters: which investments in yourself are you declining because you are pricing them against a horizon shorter than your life? Most people find at least one.

---

PART THREE — DREAM AND DESIGN

Weeks 8–11: build the apparatus

Exercise 3.1 — Run a real randomised trial on yourself (4 weeks)

Pick one thing with a plausible short-run effect on something you can measure daily: sleep duration, morning concentration, resting heart rate, time to start work. Then:

1. Write the outcome and how it is measured before you start.
2. Assign each of the next 28 days to treatment or control at random — a coin, a random number, anything you do not choose.
3. Record the outcome every day, including the days you forget the treatment.
4. At the end, compute the two means and the difference.

This is a genuine n-of-1 randomised trial and it is the one experimental design a student can afford. Its weakness is real and should be written in your notes: you are not blinded, so expectation is in the effect.

Exercise 3.2 — The baseline document (60 minutes)

Whatever you intend to change this term, write down its current value first, dated, in a file you will not edit. This is the single most valuable habit in the chapter and it takes ten minutes. The 1,200 USD that was always there gets counted as the 1,500 USD that was earned only when nobody wrote the baseline down.

Exercise 3.2b — The presenteeism band (45 minutes)

Take your own situation. Estimate the fraction of your working or studying time lost to feeling unwell, tired or scattered — be honest, most people land near 5.0 per cent. Now price it three ways against a notional 60,000 USD of annual value: at a multiplier of 0.5 it is 1,500.00 USD, at 1.0 it is 3,000.00 USD, at 1.5 it is 4,500.00 USD. A ten per cent improvement is worth 150.00, 300.00 or 450.00 USD respectively, which against a 150.00 USD intervention is a return of 1.00 : 1, 2.00 : 1 or 3.00 : 1.

Write the sentence you would have to be able to defend in order to pick one of those three multipliers. Then notice that almost nobody writing a business case has written that sentence, and that the spread between the ends is 3.00 x.

Exercise 3.3 — Read one paper properly (3 hours)

Take Jones, Molitor and Reif (2019) and read it with three questions in hand: what was randomised, what was measured, and what does the confidence interval exclude. Write half a page. You are looking for the sentence that says the intervals exclude 78.0 per cent of prior published estimates, and you are looking for what the authors say the study cannot rule out.

---

PART FOUR — DESTINY AND DELIGHT

Weeks 12–14: make it hold

Exercise 4.1 — The nine-second question, in public (ongoing)

Ask "what was it before?" out loud, once a week, in a seminar or a conversation, about somebody else's claim. Do it neutrally and with real interest — it is not a gotcha and it does not work as one. Note what happens to the conversation.

Exercise 4.2 — The generosity clause (30 minutes)

List three things you do for other people that you have been quietly justifying with a return. Cross the justification out. Write instead: I do this because it is good and it costs little. Notice whether it becomes easier or harder to keep. In the chapter's experience, easier.

Exercise 4.3 — Delight, on purpose (ongoing)

Spend one afternoon with a source that cannot flatter you — a bank statement, a sleep log, a git history, a reading record. Not to audit yourself. To enjoy the rare experience of reading a document with no opinion about you. It is restful, and most people are surprised by that.

---

Exercise 4.4 — The counter-list (45 minutes)

Scepticism is the easy half and it is not the assignment. Write down the five things in this chapter you would now be willing to fund on the evidence, with the figure attached to each: cessation paid in real money at 14.7 per cent against 5.0 per cent; disease management for the already diagnosed at 3.78 : 1; randomised safety inspection at 26.0 per cent lower compensation cost and 355,000 USD a firm; schedule control and supervisor support on burnout and turnover; and insurance itself, on a 30.0 per cent relative reduction in depression and the elimination of catastrophic out-of-pocket spending.

Keep that list somewhere you will find it. A person who can only say what is wrong with a field is not yet useful to it.

---

THE TERM PROJECT

One piece of work, carried the whole way

Take one widely believed health or performance claim and take it apart.

Deliverable: 2,500 words and one figure.

1. The claim, and its best source. Find the strongest version, not the weakest. Quote the number and the paper.
2. The comparison it rests on. Randomised, observational, or neither?
3. The selection story. Who joins, and how did they differ beforehand? Try to quantify the before-gap, even crudely.
4. The decomposition. Apply naive / true = 1 + (baseline gap / causal effect) and state what the true effect would have to be for the claim to hold.
5. The endpoint and the power. What was measured, how many people, could the study have detected the effect it reports?
6. What survives. Every one of these claims has something true in it. Say what. A demolition is not the assignment; an honest accounting is.

Mark yourself down for any sentence that is merely sceptical. The chapter's five Discovery cases exist to show that the good version of this field is excellent, and your project should end where they end: with what is real.

---

SELF-ASSESSMENT

Score yourself 0 to 3 on each. Honest threes are rarer than you think.

Under 12: reread the Arithmetic and redo 2.1. Over 20: teach 2.2 to somebody else, which is the fastest way to find the part you have not understood.

---

CARRYING IT FORWARD

The instrument you have built this term is not about wellness programmes. It is about any claim of the form people who do X are better off. You now have four questions that fit in your pocket:

1. What was the gap before?
2. Who chose who was in?
3. Could this study have detected the effect it reports?
4. Who keeps the benefit, and for how long?

Four questions, nine seconds each, and they will serve you for thirty years.

---

APPRECIATIVE QUESTIONS FOR YOUR SEMINAR

1. When has one of us changed something and genuinely known it worked? What made that knowable?
2. Which claim in this field do we find most credible, and what specifically makes it so?
3. If we could randomise one thing about how this course runs, what would we learn?
4. What are we each already doing that a good trial would probably support?
5. Whose health is improved by something we do, that none of us has ever counted?
6. What would we want a future employer to invest in on our behalf, knowing they keep 14.44 per cent of it?