Second-Order Thinking

The problem it solves

A decision usually gets judged on whether it produced the effect it was aimed at, and that effect is rarely where the story ends. The referral bonus raises referrals. The response-time target lowers response times. The office mandate fills the office. Each of those is a real result, arriving on schedule, and each one is the point at which most organisations stop paying attention.

What happens afterwards is that people adjust. Somebody now has a reason to behave differently, they find the cheapest way of doing so, and the gain that was meant to land in one place turns up somewhere else while a cost settles on whoever wasn't in the room when the decision was made.

Those later effects tend to be slower, harder to attribute and easier to blame on something unrelated, so a decision can be written up as a success several weeks before its interesting consequences arrive.

The idea

Second-order thinking is the habit of carrying on past the intended effect and asking what that effect then causes. It's a small question, asked once more than feels necessary, and its whole value is in being asked before the commitment rather than during the post-mortem.

Enlarge: Everything above the line gets watched. Everything below it happens anyway.

A decision at the top, connected downwards to what it directly does. A dashed line runs across beneath that, labelled decisions are judged above this line. Below the dashed line hang three boxes, reading people adapt to whatever you set, the gain moves to whoever adapted, and the cost lands on somebody else.

Everything above the line gets watched. Everything below it happens anyway.Close

Everything above the line gets watched. Everything below it happens anyway.

The second row is mostly people responding, which is what makes it tractable. Almost nothing in it is a physical consequence, and almost all of it is somebody adjusting to a new arrangement in the cheapest way that works for them. So the question has a practical form. Who now has a reason to behave differently, and what is the least effortful thing they can do about it?

Economics has run on this since Bastiat, whose argument is that the effect you can see is one of many and the ones you can't see are the same size. It transfers to organisations with a caveat worth stating, because an economy has millions of adjustments averaging out into something predictable, and a team of nine has a handful of adjustments made by particular people with names, which is less predictable and much easier to ask about directly.

How to use it

Ask it twice, in that form, and write the answers down where the decision is recorded rather than in a separate document.

Start with who is affected who wasn't consulted, because the second row lands on them first, and their adjustment is the one nobody in the room will predict unaided.

Where the decision involves a number that judges people, assume it will be managed rather than measured, and decide in advance what second number moves in the opposite direction when the first is being gamed. Publishing both, and saying that both are watched, does most of the work.

Then stop. Agree the depth before starting, and two rows past the intended effect is enough for almost everything. Anything further out gets written down as a question to revisit rather than argued to a conclusion, because past that point the group is doing fiction with a whiteboard.

Where it breaks down

Every concept here has one. It is the section most summaries leave out.

There's no natural stopping rule, and the method supplies none. Consequences carry on indefinitely, each row is more speculative than the last, and a group with a whiteboard will keep going for as long as the session lasts. The discipline is knowing when to cut the chain, which comes from somewhere other than this idea.

Predicted second-order effects are cheap and unfalsifiable. Anyone can generate a plausible chain of consequences for any decision, and there's no way to check one in advance, so the technique is easily used to argue against anything a person already dislikes. A stated chain that would change the decision if it were wrong is worth having, and the rest is decoration.

It's an excellent instrument for delay. Because the questions sound responsible and the answers can always be extended by one more step, a sufficiently anxious group can use this to avoid deciding for weeks while appearing rigorous throughout. Two-way door decisions should be taken quickly and revisited, and running this exercise on one is a waste.

The evidence is illustrative rather than experimental. Campbell's law has substantial field support and the general practice has none, in the sense that no study shows groups that ask the question outperforming groups that don't. What's established is the mechanism, which is that people respond to incentives in ways the incentive-setter didn't intend, and that is about as well documented as anything in the social sciences.

Hindsight makes it look easy and it isn't. Read afterwards, the second-order effect of a referral bonus is obvious. Sat in the room before the decision, it's one of perhaps thirty candidate consequences, most of which won't happen, and the method offers no help at all in ranking them.

In one line

Ask what happens after the thing you wanted happens, and ask it about the people who weren't in the room.