Judgment Playbook

Second-order thinking

Follow the decision past its obvious effect by repeatedly asking: and then what?

5 minute readtaught in course chapter 7
AlternativesUncertainty

What it is

Second-order thinking is the practice of following a decision past its first effect. You state what the decision does immediately, then ask what the people affected will do in response, and what their response does to the goal you started with. It is for choices whose first effect is easy to see and whose reactions arrive later, when the decision has already been made.

Where it comes from

The idea has no single author. Its clearest early statement is Frédéric Bastiat's 1850 pamphlet Ce qu'on voit et ce qu'on ne voit pas: "an act, a habit, an institution, a law produces not only one effect, but a series of effects. Of these effects, the first alone is immediate; it appears simultaneously with its cause; it is seen. The other effects emerge only subsequently; they are not seen." Robert Merton named the failure in 1936, in "The Unanticipated Consequences of Purposive Social Action" in the American Sociological Review, and traced it to ignorance, error and the pressure of immediate interest. Jay Forrester supplied the machinery in 1971, in "Counterintuitive Behavior of Social Systems": social systems run on multiple feedback loops, and intuition trained on short chains of cause and effect misreads them. The phrasing most business readers meet is Howard Marks's "second-level thinking", the first chapter of The Most Important Thing (Columbia University Press, 2011). Ray Dalio's Principles (2017) calls the same move weighing second- and third-order consequences.

What it corrects

The dangerous situation is a decision whose direct effect can be measured now and whose reactions cannot. A competent analyst builds the case on the measurable part, because that is where the evidence sits, and the model of a price cut or a channel change is usually right about the first step. Ordinary care then makes that model more precise rather than longer: better numbers about the same single move. The reactions go missing for two structural reasons. They belong to people outside the decision, who act on their own interest rather than yours, and they land after the review date, when the decision has already been credited with its first-order win.

How it works

  1. Write the direct effect in one sentence, with the number you expect it to move.
  2. List the actors who feel that effect: customers, competitors, distributors, staff, regulators.
  3. For each actor, name the response their own interest now recommends.
  4. Take the strongest of those responses and follow it one step further, to what it does to your number.
  5. Say when the loop closes: weeks, quarters or years.
  6. Re-test the recommendation against the number after the loop, and name one signal, measured outside your own systems, that would show the response beginning.

Worked example

On 20 November 1989 Mexico City introduced Hoy No Circula, which barred each car from the roads one weekday a week according to the last digit of its plate. The first-order arithmetic was clean: the rules covered 2.3 million vehicles, about 460,000 of them off the road on any weekday, and enforcement was strict, so compliance was near universal. The second-order response was just as simple. A household with two cars whose plates end in different digits drives every day.

Lucas Davis measured what followed, using monitoring-station readings, in the Journal of Political Economy in 2008. Across pollutants and specifications he found no evidence that air quality improved. Registered vehicles rose by about 325,000 around the start of the programme, with a confidence interval of 51,000 to 597,000. New car sales rose about 15 percent, which was under 2 percent of the vehicle stock, so the additional cars were overwhelmingly used ones brought in from elsewhere in Mexico and from the United States. Older cars emit more. Subway and bus ridership did not rise, and petrol sales did not fall. The ban succeeded against every car it touched and lost to the fleet it created.

In a Business Case Weekly case

In the Nike case, the June 2020 fork is Consumer Direct Acceleration: cut wholesale accounts and sell to the customer directly. The first-order case is on the slide, in margin per pair and pandemic digital growth, and it is credible. Second-order thinking asks what a retailer does with the shelf space Nike hands back, and what a shopper does when the shoe is missing from the store they walk into. That is why the case asks you for tripwires measured outside Nike's own channels: a dashboard that management controls cannot report a reaction happening in someone else's shop. The method sets up the question. It does not settle whether the plan was right.

In your answer

  • "The direct effect is X, and I expect it to hold for about N months."
  • "The actor with most to gain from responding is Y, whose cheapest response is Z."
  • "If Y does Z, the benefit falls from A to B, which changes my recommendation as follows."
  • "The signal I would watch sits outside our own channel, and it is …"

Common misuse

The counterfeit is a long chain of speculation: five links out, each one plausible, none of them sized, ending in a scenario nobody will ever check. Such a chain cannot be wrong, so it never overturns a recommendation and never commits to one. The test is a single question per link: can you name the actor, their incentive, and the observation that would show the response starting? A link without all three is decoration. Stop at the last link that has an owner and a signal, and size that one.

References

Steel-man

Next method

State the strongest opposing case so well that its advocate would recognise it.

Open next tool