People spread evenly across whatever buckets you give them

Give people a set of options and ask them to divide something across it — money, attention, a rating, their time — and a lazy, powerful default kicks in: spread it roughly evenly across whatever buckets are in front of them. Not evenly across the real space of possibilities. Evenly across the buckets you happened to draw. Redraw the buckets and the “even” split moves with them, even though nothing about the actual options changed. Behavioral researchers call this partition dependence, and once you’ve seen it you notice you’ve been pulling this lever by accident your whole building life.

The cleanest demonstration comes from retirement investing. Offer people a menu of funds and, faced with a choice they don’t fully understand, a lot of them reach for the 1/n rule: put an equal slice into each option on the menu. Sounds reasonable — until you notice it means the menu’s composition silently decides their portfolio. Put one bond fund and four stock funds on the list and the average person ends up heavily in stocks. Put four bond funds and one stock fund in front of a different group — same underlying choice, “how much risk do you want?” — and they end up conservative. Nobody chose a risk level. They chose “a bit of each,” and you chose what “each” was.

It generalizes far past money. Ask people to estimate the chance tomorrow’s weather falls into “warmer than today” versus “colder than today” and you get one answer; split “colder” into “a little colder” and “much colder” and the total probability they assign to cold goes up — because there are now two cold buckets pulling their fair-share instinct. The mind doesn’t hold a stable distribution and then report it. It builds the distribution out of the categories you offer, at the moment you ask.

For anyone shipping an interface, this means the way you partition a choice is a design decision as loud as any default — and it’s usually made without thinking. Your pricing page groups plans into tiers: the number of tiers, and which features you lump under each, tilts where the “sensible middle” lands. Your onboarding asks users to allocate a budget, a schedule, or a set of priorities across categories you invented: those categories are the answer more than the user is. Your survey offers a 1–5 scale; collapse it to three points or stretch it to seven and the mush-in-the-middle moves. A resource slider divided into “storage / compute / bandwidth” produces a different average setup than the same resources split five ways. You think you’re presenting the options neutrally. There is no neutral partition.

So the build move is to treat your categories as a lever you’re choosing to pull, not scenery. Before you draw three buckets or five, ask: if a distracted user just spreads themselves evenly across these, where do they land — and is that where their actual interest would have put them? If splitting one real option into two sub-options quietly doubles its pull, that’s either an honest reflection of its importance or a thumb on the scale, and you should know which one you’re doing. The most defensible partitions are the ones where “a fair share to each bucket” happens to be close to a genuinely good outcome for the user — you can design for that.

The uncomfortable part is that you can’t opt out. You cannot present a set of choices without grouping them somehow, and any grouping tilts the result. Partition dependence isn’t a bias you can design away; it’s a force you’re always applying, the only question being whether on purpose. The builders who take this seriously don’t try to be neutral. They pick the partition that makes the honest, fair-share choice also the good one — and they resist the cheaper temptation to slice whatever they most want to sell into the most buckets.

The science, to look up: the 1/n or “naïve diversification” heuristic (Benartzi & Thaler’s work on 401(k) allocation); partition dependence in choice and in probability judgment (Fox, Ratner & Lieb, and related studies showing description-dependent — not just option-dependent — allocation).