Last lesson you learned the distinction: first-order thinking stops at the first answer, second-order thinking keeps going. That’s the philosophy. This lesson is the machine — the actual repeatable steps you run on a decision so you stop being impressed by your own first answer.
Because here’s the uncomfortable truth: “think about second-order effects” is useless advice on its own. It’s like telling someone to “just be funnier.” How? The answer is a procedure with four moving parts — a chain, a tree, a sign-flip you’re hunting for, and a stopping rule. By the end you’ll have a checklist you can run on any decision in under two minutes.
Before you read — take a guess
You're about to evaluate a policy. Which of these is the actual *work* of second-order thinking — the thing you physically do?
The consequence chain: “and then what?” on repeat
Picture a row of dominoes. You flick the first one. The satisfying part isn’t that one tile falls — it’s that the first one knocks the second, which knocks the third, and the energy travels somewhere you weren’t looking when you flicked. A decision is a flick. The first domino is loud and obvious; the interesting tiles are three down the line.
The consequence chain is what you get when you take any effect and repeatedly ask “and then what?” — feeding each answer back in as the next question. The discipline it comes from is systems thinking (the study of how parts of a system feed back on each other over time), and it’s the same logic economists use under the name “and then what?” — most famously Henry Hazlitt’s rule that the good economist accounts not just for the seen effects but the unseen ones that follow.
The positions on that chain have names you’ll use constantly:
| Order | Question it answers | Example (you raise prices) |
|---|---|---|
| First-order | What happens immediately? | More profit per sale |
| Second-order | And then what? | Some customers leave; rivals notice |
| Third-order | And then what to that? | A rival undercuts you; you lose share long-term |
Each order is just a position — how many “and then what?“s deep you are. First-order is one link from the decision. Third-order is three links out. That’s the whole vocabulary.
The chain is directional
Every link is caused by the one before it. “Customers leave” only exists because “prices rose.” If an effect doesn’t trace back to the decision through a chain of because, it’s not on your tree — it’s just background noise.
Chains branch into a tree
Here’s where the domino picture breaks — and where it gets interesting. Real dominoes fall in a tidy line. Consequences don’t. One effect provokes several reactions at once, like a single ancestor with a sprawling family tree of descendants. Raise prices and — simultaneously — some customers leave, and rivals notice, and your brand looks more premium, and your sales team gets harder conversations. Four children from one node, each with children of its own.
So the chain is really a consequence tree: a branching map where the decision is the root, first-order effects are the first generation, and every effect can sprout multiple next-order effects. This is the mental object this lesson is teaching you to build.
Let’s grow one. A government decides to cap city rents at a low fixed price — a textbook case beloved by economists precisely because the first-order effect is so obviously good and everything after it is so obviously not. Open each node and follow it down. Watch the color of the markers.
And then what?
Rent control, three orders deep
Start at the decision. Open each “and then what?” to follow the consequences another order deeper — watch where the obvious first move leads:
- Decision
Cap city rents at a low fixed price
Goal: make housing affordable for current tenants
Look at the leftmost branch. It starts green — tenants pay less, exactly as intended — and by the time you’re two links down it’s solid red. That’s not a coincidence; that’s the entire reason economists keep using this example. The policy is judged on its green root and condemned by its red leaves.
One number, so it's not just vibes
This isn’t hand-waving. Studies of San Francisco’s 1994 rent-control expansion (Diamond, McQuade & Qian, 2019) found it cut the supply of rental housing from affected buildings by 15% as owners converted or redeveloped to escape the cap — pushing up citywide rents by about 5.1%. The first-order winners (sitting tenants) were real; so were the third-order losers (everyone trying to rent later).
Reading the tree: hunt for the sign-flip
A treasure map isn’t useful because it shows every grain of sand — it’s useful because of one X. Reading a consequence tree works the same way. You are not trying to forecast every leaf. You’re scanning for one specific thing: the sign-flip — the branch where a good first-order effect turns bad one or two orders downstream (or, occasionally, the reverse).
The sign-flip is where the decision’s real verdict lives, because that’s the effect the first-order thinker never sees. Formally: a sign-flip is any path along the tree where the valence changes from good/mixed to bad (or vice-versa) as you move outward. In the rent-control tree, every important insight sits at a flip — “tenants pay less” (good) → “building becomes unprofitable” (bad).
Here’s the scanning routine, in prose, on one chain at a time:
Take the plastic-bag ban. Walk it one link at a time and name the valence as you go:
- Decision: ban thin single-use plastic bags.
- First-order: fewer thin plastic bags in circulation. Valence: good. ✅ This is where most arguments stop.
- And then what? Shoppers still need to carry groceries, so they buy thicker reusable totes — often cotton. Valence: mixed. A cotton tote has a much bigger manufacturing footprint: a UK Environment Agency study found an organic-cotton tote must be reused ~131 times to beat the per-use climate impact of one lightweight plastic bag (and far more if the plastic bag is reused once as a bin liner).
- And then what? People forget their totes, or treat them as nearly free, and accumulate dozens. If your average tote gets used 20 times, not 131, its per-use footprint is worse than the bag it replaced. Valence: bad. ⛔ ← the sign-flip.
The flip is between step 2 (good) and step 4 (bad). Notice you didn’t need to forecast landfill tonnage to the decimal — you just needed to find the link where the color changed.
On a consequence tree, what exactly are you scanning for? Spot the trap.
Branching factor: how deep do you go?
If each effect spawns roughly three reactions, the math is brutal. One decision → 3 first-order effects → 9 second-order → 27 third-order → 81 at the fourth order. That’s the branching factor (the average number of children per node, a term from search trees in computer science), and it means the tree explodes exponentially while each new leaf gets fuzzier and less certain. Forecast far enough out and you’re not analyzing — you’re writing fan fiction.
So you don’t trace infinitely. Two practical stopping signals:
| Stop when… | Because… |
|---|---|
| You’ve found the sign-flip | You’ve got the insight the first-order thinker missed; deeper nodes rarely change the verdict |
| Effects shrink or blur | When each branch’s impact is small or its probability is a coin-flip, more depth adds noise, not signal |
A workable default for an everyday decision: go about 2–3 orders deep, and stop the moment effects get smaller than the decision itself or too uncertain to bet on.
Don't mistake more depth for more wisdom
The failure mode here is analysis paralysis dressed as rigor — tracing a tree to the sixth order and feeling smart while the leaves are pure speculation. Depth past the sign-flip usually buys false confidence, not insight. Knowing when to stop is itself a skill, and it’s so important it gets its own lesson (05, “When to Stop”). For now: deep enough to catch the flip, then quit.
Fill in the stopping rule:
Pick the right option for each blank, then check.
Because the tree's makes nodes multiply exponentially while each leaf grows more uncertain, you trace roughly orders deep — stopping once you've found the sign-flip or the effects get too to matter.
A second tree: drive this one yourself
Policy is the easy case — the textbook loves it. But the procedure is domain-blind; it works just as well on a Tuesday-afternoon engineering call. Suppose a manager decides to crunch the team — mandatory overtime — to ship a feature two weeks early. First-order, it works: the feature ships early. Then the tree does what trees do.
This time you drive. Predict the valence of each node before you open it, then check yourself.
And then what?
Crunch to ship early, three orders deep
Start at the decision. Open each “and then what?” to follow the consequences another order deeper — watch where the obvious first move leads:
- Decision
Crunch the team to ship the feature 2 weeks early
Mandatory overtime for a hard deadline
Notice the pattern repeats exactly. One green root (“ships early”), and every branch flips red within two links. The first-order thinker high-fives the team on launch day. The second-order thinker already knows they’ll be down two engineers and a quarter of velocity by next quarter — and that the data they use to plan is now quietly poisoned by padding.
Sort each effect from the crunch decision by its position on the tree.
Place each item in the right group.
- Your best engineers quit
- Estimates inflate to dodge the next crunch
- The team is exhausted and resentful
- Future velocity drops for months
- Rushed work ships with more bugs
- The feature ships two weeks early
The procedure, as a checklist
Strip away the analogies and the whole method fits on an index card. This is the thing to actually run.
The 'and then what?' procedure
- Name the decision. State it plainly — that’s your root node.
- List the first-order effects. What happens immediately and directly? Usually 2–4 of them. Mark each one’s valence: good, bad, mixed, or neutral.
- For each, ask “and then what?” Trace 2–3 orders out, letting each effect branch into the several reactions it provokes. You’re growing a tree, not a line.
- Mark where the sign flips. Scan for the branch where a good first-order effect turns bad downstream (or vice-versa). That flip is the insight.
- Decide on the whole chain, not the first link. Judge the decision by where its branches end up, not by how shiny the root looks.
If you only remember one line: the first link is the bait; the verdict is downstream.
A teammate says: 'Free overdraft protection will make customers love us — let's ship it.' Which response actually runs the procedure?
Recap
Big picture
The and-then-what procedure
- And-Then-What Chain
- Chain
- Ask "and then what?" on repeat
- Order = links from the decision
- From systems thinking / Hazlitt
- Tree
- Each effect → several reactions
- Decision = root; effects = generations
- Branching factor explodes (~3ⁿ)
- Sign-flip
- Where good → bad downstream
- The insight the 1st-order thinker misses
- Don't forecast every leaf — find the X
- Stop
- ~2–3 orders deep
- Stop at the flip, or when effects blur
- Full treatment in lesson 05
- Checklist
- Name → list → "and then what?"
- Mark the flip
- Judge the whole chain
- Chain
Check yourself
Why is a chain of consequences "really a tree"?
Check your answer to continue.
Where this goes next
You can now grow a consequence tree and find its sign-flip. But there’s a deeper question lurking: why do so many trees flip from good to bad in the same predictable way? It’s rarely random. It’s usually because the decision changed someone’s incentives — and people, like the toad and the over-drafting customer, quietly optimize for whatever you actually reward. Lesson 03, “Incentives, Feedback & the Cobra Effect,” is about the engine that drives the sign-flip: when a reward you set up teaches everyone to do the opposite of what you meant.