Sheena Iyengar and Mark Lepper set a tasting table up in a supermarket in Menlo Park. Some days it held twenty-four jams and other days six. The big table pulled more people over. It then sold to about three in a hundred of them, against roughly thirty in a hundred at the small one.
Below you are picking a time for a call. Some rounds show two free slots and some show twenty-four. The clock starts when the grid appears and stops when you click. The question is on screen before the grid is, so reading it is not part of what gets timed.
The order of the sizes is shuffled on every visit. Without that the twenty-fours would always land late in the run, and being tired would look exactly like choice overload.
There is still a hole in timing somebody looking at twenty-four things and calling the result deciding. A grid of twenty-four takes longer to read than a grid of two whatever you are doing with it. So four of the rounds have nothing to decide: the question names one of the slots and you click that slot. Same grid, same sizes, same distance for your hand. Whatever those cost is looking, and it comes off the top.
That control is not clean and it errs in a direction I can name. The slots are in time order, so hunting for a named one can stop early while deciding means reading all of them. It under-counts the looking. Whatever survives the subtraction is the most that deciding could have cost, not what it cost.
This one needs JavaScript. It times you deciding, and there is nothing to time without it.
What twelve picks can hold up
Two timings at each size. The median of two numbers is the average of two numbers, and one distracted round moves it by half of whatever the distraction was worth. That is the instrument. It is not a good one.
So the page is built to refuse. If twenty-four is not slower than two it says so. If it is slower by an amount two trials cannot separate from nothing, it says that instead. If the rounds with nothing to decide grew just as fast as the rounds with a decision in them, then the run measured your eyes rather than your judgement, and there is a branch for that too.
A direction is not a size. Twenty-four slots coming in 40 ms behind two would be a real direction and a worthless finding. Rounds under 250 ms are thrown out as clicks that started before the grid did, and rounds over 15 seconds are thrown out as somebody looking away.
What it costs you
The cost lands on the person, not on the product. Adding an option is one row in a config file and it is free at your end. The reading and the weighing happen on the other side of the screen where you cannot see them, and neither can your analytics.
The worst outcome is not a slow choice. It is no choice. The jam table that pulled the crowd is the one that sold almost nothing. Walking away does not show up as a bad pick. It shows up as nothing at all, which is why a page with eleven plans on it looks healthy in any funnel that only counts the people who got to the end of it.
A default spends options without deleting any. One sensible pre-selection turns twenty-four into one for most people and leaves the other twenty-three exactly where they were. The option count in the database does not change. The option count in the head does.
Sorting is not reducing. Time order makes a named slot easier to find and does nothing at all for somebody who has no name in mind. The list is still twenty-four long. It is just tidier.