Trust Game

What you see

The board shows one round as four amounts in a row: what you send, that amount tripled, what the partner returned, and what you hold from the round. The last two show ? until you send. The question names the partner by letter (Partner A, B, C or D), and the pills above the board show the round and your points so far.

Below them are the partner's past returns: one entry for each earlier round with the same partner, showing the percentage of the tripled amount they sent back, such as 48%. If you sent nothing that round, the entry reads None sent, because a partner who receives nothing reveals nothing.

At the bottom, choose how much to send with the row of coins marked 0, 10, 20 and so on up to 100 (on a phone, a slider in steps of 10), then press the send button, which names the amount, for example Send 60 (or press Space). The four amounts fill in, and the You hold box turns green when you end the round with more than 100 and red when you end it with less. The same button then reads Next round; after the last round it reads Finish.

One run of the game is one trial of 8 to 24 rounds. Every round starts with a fresh 100 points. There is no time limit, and speed has no effect on the score.

How the partners behave

Partners take turns in a fixed rotation: with three partners, rounds go A, B, C, A, B, C and so on. Each partner returns a share of the tripled amount, and each follows one of four patterns, set before the run begins. Which letter gets which pattern is drawn afresh every run, so Partner A is simply the first partner you meet: fair in one run, stingy in the next. They do not react to how much you send; your task is to read them, not to win them over.

What changes with level

With four partners in rotation, the late reformer meets you only three times in a 15-round run, so it never reforms at level 6. At level 7 it reforms on its fourth and last meeting; at level 10, with 24 rounds, it is fair on three of its six meetings. Longer runs therefore reward noticing change.

What it measures

The Trust Game feeds the Negotiation and trust facet and carries 15 per cent of your Decision Making score.

It is the investment game introduced by Joyce Berg, John Dickhaut and Kevin McCabe in 1995. In their version, a sender with $10 could transfer any amount to an anonymous receiver; the transfer was tripled, and the receiver chose how much to return. Standard self-interest predicts that receivers return nothing, so senders should send nothing. People do not behave that way: as Alós-Ferrer and Farolfi note in their review of trust games, real people are far more trusting and more trustworthy in the laboratory than selfishness would imply.

Look at those two numbers together. An average return of 37 per cent is only just above the one-third that makes sending worthwhile. In the real world, trusting a random stranger roughly breaks even; the money is in telling the trustworthy from the rest. That is precisely the skill our version isolates, by giving you partners who differ sharply and consistently.

The research also shows the two sides mean different things. In a microfinance study, Dean Karlan found that people who were trustworthy in the game (returned more) were less likely to default on real loans, while people who were trusting (sent more) were not. In our game you are always the sender, so your score reflects judgement about others, not your own trustworthiness.

How it is scored here

Every round, what you hold is: 100 − sent + returned. Returned is the partner's share × 3 × sent. Rearranged:

The one-third rule

Hold = 100 + sent × (3 × share − 1). If a partner returns more than a third of the tripled amount, every point you send makes you money; if less, every point you send loses money. At exactly one third you break even.

The benchmark for your run is what you would have held by sending everything in every round where that partner's share exceeds one third, and nothing where it does not. Your score is your total divided by that benchmark, capped at 100 per cent.

Three consequences follow.

1. Half-sends never beat the better extreme. Because the payoff is a straight line in the amount sent, once you believe a partner is above or below one third, the all-or-nothing choice beats any middle amount. A half-send only reduces the swing.
2. The floor is high. You keep whatever you do not send, so even sending nothing all run captured roughly 67 to 81 per cent of the benchmark in our simulations of the scoring. Sending everything to everyone captured about 70 to 77 per cent once exploiters appear. Sharp reading moves you to around 87 to 93 per cent. The differences look small; they are the whole game.
3. 98 per cent is close to out of reach after level 1. The benchmark knows each partner's share before you do, including the first meeting. In our simulations, even strong learning strategies essentially never reached 98 per cent once two or more partners were present, so expect to hold your level. The score already adjusts for difficulty.

Level changes follow the usual rule: 98 per cent or more moves you up, zero moves you down (in practice impossible here, because you always keep what you do not send), anything else holds. New sessions start at level 3.

Your report adds descriptive measures: Distance from the best amount to send (how far each send was from the ideal all-or-nothing amount, as a share of the stake; lower is better), Sent to partners who paid back more, minus the rest (how much more of your stake you sent, on average, in rounds where the partner's share was above one third than in the others), Change in sending after a losing round (after a round where a partner sent back less than you sent, how your next send to that same partner compared, in percentage points of the stake; it appears from level 2, once you have sent something to a partner who returned less and met them again), Sent in the first four rounds (your average send over the first four rounds, as a share of the stake) and Value captured (your score).

Across sessions your Trust Game score sits on our 0–100 scale, where 50 means what the difficulty you work at predicts.

Where you will meet it

Candidate-preparation sites describe the first of two pymetrics Money Exchange games as this exact design: you start with $10, can transfer any amount to a computer-generated partner, the amount is tripled, and the partner decides how much to return, after which you rate how fair they were. That description is candidate-reported, not vendor documentation. Pymetrics was the first-stage screen in Unilever's graduate hiring, according to Business Insider, though the report does not list the individual games.

In commercial versions like this there is usually one round and no right answer: your transfer is a trait reading, compared with a role profile. Our version is repeated and scored, which trains the part that transfers: reading behaviour and acting on it.

The method

1. Open every new partner with a send

   You cannot read a partner who receives nothing. The first meeting with a new partner is roughly a coin flip, so choose any opening you are comfortable with. A small send such as 10 reveals the share just as precisely as 100; a larger one earns more if the partner turns out fair. In our simulations, the opening size moved the final score by only a few points either way.

2. Apply the one-third rule to the latest past return

   If the partner's latest share is above 33 per cent, send 100 next time. If it is below, send nothing, or keep a small probe going (next step). Half-sends are for nerves, not for points.

3. Look for a pattern before a trend

   Two different shares in a row may be a pattern, not a change. High, low, high means the next is probably low. A partner who returns about half every time will almost certainly keep doing so.

4. Keep a cheap probe on stingy partners

   Sending nothing shows None sent and hides any change. A send of 10 to a partner returning 12 per cent costs you about 6 points; missing a partner who has turned fair costs you about 50 points per round. In long runs, the probe pays for itself.

5. Track each partner separately

   The rotation means consecutive rounds are with different partners. A betrayal by Partner B says nothing about Partner C. Keep a one-word label per partner in your head: fair, stingy, alternating, changing.

Worked example 1: two histories

Round 9 of a 12-round run at level 4. Partner A's past returns read 11%, 9%, 14%. Partner B's past returns read 48%, 52%, 46%. How much should you send to each, and what will you hold?

Show the answer

Partner B returns about half. Sending 100 turns into 300 on their side; at roughly 50 per cent they return about 150, so you hold about 150 instead of 100. Send 100.

Partner A returns about a tenth. Sending 100 would bring back about 30, leaving you 30 instead of 100. Send 0, or a 10-point probe if the run is long enough for a change to matter (at level 4 there is no late reformer, so 0 is fine).

Partner's share	Hold if you send 100	Hold if you send 50
10%	30	65
20%	60	80
33%	100	100
50%	150	125
60%	180	140

Worked example 2: the alternator

Level 5. Partner C's past returns read 60%, 20%, 62%. What do you send on the fourth meeting, and on the fifth?

Show the answer

The pattern is high, low, high, so the fourth meeting is probably low: send 0 or a 10 probe. If the fourth return confirms a low share, send 100 on the fifth meeting, which should be high again.

A candidate who treats Partner C by averaging (the average of 60, 20 and 62 is about 47 per cent, above one third) would send 100 every time and lose about 40 points on each low meeting (a 20 per cent share returns 60 of 300). Reading the alternation captures the high meetings and sidesteps the low ones.

Worked example 3: the reformer

Level 10, 24 rounds, four partners. Partner D's past returns after three meetings read 11%, None sent, None sent: you sent 100 on the first meeting, then nothing on the next two. Round 16 is your fourth meeting with D. What should you do?

Show the answer

Send a probe of 10. The two None sent entries mean you have no information about meetings two and three; a partner that has not been tested can change unseen. If D is still stingy, the probe costs about 6 points. If D has turned fair and returns about 50 per cent, the probe tells you so, and you can send 100 on the remaining meetings, holding about 150 instead of 100 each time.

Sending nothing all the way would be safe but blind. In our simulations of level 9 and 10 runs, keeping a small probe going with stingy partners scored about 2 to 3 points higher than cutting them off entirely.

Tips and traps

Levelling up

As levels rise, more partners join the rotation, patterns get less obvious, and runs get longer. From level 6 all four patterns are present, and from level 7 the late reformer actually reforms within the run. What changes for you:

• More to track. With four partners you see each one only every fourth round. Name each partner after two meetings.
• Probes matter more. Longer runs give a changed partner more rounds to reward you, so the 10-point probe becomes worth it.
• Levels rarely move. Because the benchmark is clairvoyant, expect to hold your level and judge progress by Value captured and Sent to partners who paid back more, minus the rest instead.

1. Week 1
   • Two runs; before each send, say the partner's last share and whether it is above one third
   • Check Distance from the best amount to send after each run
   • Stop hedging: only 0, 10 or 100
2. Week 2
   • Two runs; label each partner after two meetings
   • Spot the alternator and follow its pattern
   • Check Sent to partners who paid back more, minus the rest: aim well above 20 points
3. Week 3
   • Two runs of 18 rounds or more if your level allows; keep a probe on every stingy partner
   • Note the round where any partner changed
   • Compare Value captured with week 1

Check yourself

5 practice questions

1. A partner returns 40 per cent of the tripled amount. If you send 100, what do you hold from the round?
2. Above what share of the tripled amount does sending money pay?
3. You are confident a partner returns about 50 per cent. Which send maximises what you hold?

and 2 more

Create a free account to answerSign in

Sources

1. Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10(1), 122–142. doi.org/10.1006/game.1995.1027
2. Johnson, N. D., & Mislin, A. A. (2011). Trust games: A meta-analysis. Journal of Economic Psychology, 32(5), 865–889. doi.org/10.1016/j.joep.2011.05.007
3. Alós-Ferrer, C., & Farolfi, F. (2019). Trust games and beyond. Frontiers in Neuroscience, 13, 887. doi.org/10.3389/fnins.2019.00887
4. Karlan, D. S. (2005). Using experimental economics to measure social capital and predict financial decisions. American Economic Review, 95(5), 1688–1699. doi.org/10.1257/000282805775014407
5. JobTestPrep. Free pymetrics games online practice and guide (candidate-preparation site; descriptions are unofficial). jobtestprep.com
6. Feloni, R. (2017, June 28). Consumer-goods giant Unilever has been hiring employees using brain games and artificial intelligence. Business Insider. businessinsider.com

Put it into practice: Trust GameSign in to try it