Guess 2/3 of the average: method

What it measures

What it is: a strategic guessing game that shows how many steps a player reasons about what others will do (level-k thinking), and, for the whole group, the spread of guesses. What it isn't: a measure of intelligence or of an individual trait.

Procedure

One round lasts one ISO week in UTC (Monday 00:00 to the next Monday 00:00). A player submits one whole number from 0 to 100. One guess per website visitor id and per internet connection counts in each round; later guesses are stored but not counted. The running average is never shown during the round. After the round ends (plus 5 minutes grace), the result is computed with exact fractions, but only if at least 30 guesses counted: average, target = 2/3 x average, winning guess or guesses (closest to the target; ties share), the full histogram and every number's rank.

Parameters

Scoring

Immediate feedback: the guess is placed on the nearest rung of the reasoning ladder (level k = 50 x (2/3)^k, k = 0 to 4) or on the equilibrium (0); a guess above 66.7 is marked as one that can never win. The guess is also compared with four published contests (distance to each target, and whether it is at least as close as the published winning number): Nagel 1995 lab first round (average 36.73, target 24.49), Financial Times 1997 (18.91, 12.61, winner 13), Spektrum der Wissenschaft 1997 (22.08, 14.72, winner 14.7), Politiken 2005 (32.41, 21.6, winner 21.6). Round result: average, target, winners, histogram, rank of each number.

Valid-run rules

Only whole numbers from 0 to 100 are accepted. A second guess from the same visitor id or the same internet connection in a round is stored as 'already played' and not counted, so people sharing a network count once. Rounds with fewer than 30 counted guesses have no published result. Bot signals and quality flags are stored with each guess.

Differences from the classic task

Lab versions used small paid groups and repeated rounds; newspaper contests used one large round with prizes. Here rounds are weekly, open to anyone, unpaid, and the player pool changes every week. Some published contests allowed decimals; this game allows whole numbers only.

Limitations

The players are self-selected website visitors, and some know the game, so targets vary from week to week. Counting one guess per internet connection merges people on shared networks. A single guess says little about a person. Not a measure of intelligence; no clinical use.

Languages and review status

Numbers only; instructions in all site languages. All languages play in the same weekly round.