At this year’s GenCon, I discovered a thought-provoking game: “Are You a Werewolf?” – a.k.a. “Mafia.”  The game’s a brilliant exploration of cheating, cheater detection, and democratic information aggregation.  The basic set-up:

1. There are 15 players and a moderator. 

2. Each player gets one card, which you may not reveal until you leave the game.

3. Twelve of the cards say “Villager,” two say “Werewolf,” one says “Seer.”

4. Each turn, the moderator tells the whole village to “go to sleep.”  You’re on your honor to keep your eyes shut until the moderator tells you to open them.

5. Next, the moderator tells the werewolves to wake up and silently choose a single victim to eat.  The werewolves then go back to sleep.

6. Then the moderator wakes up the seer.  The seer points to one other player and the moderator signals whether that one person is a werewolf.  The seer then goes back to sleep.

7. The moderator wakes up the village, revealing the latest victim.  The victim shows his card. 

8. The villagers then argue about “Who’s a werewolf?” until a majority of the survivors pick a victim to lynch. The victim reveals his card.

9. Return to step 4.

The game ends once (a) both werewolves are dead, or (b) the number of werewolves equals the number of villagers.

With random elimination, villagers win 39.5% of the time.  In practice, though, the villagers won all six of the games I played.  My sense is that the villagers win about 90% of the time.  The fundamental reason is that villagers have a greater-than-chance ability to detect cheating – even when the cheaters are struggling to conceal their cheating.  How?  From demeanor, words, and behavior.  My werewolf status was almost immediately uncovered in my first game because I incompetently concealed my smug delight. 

Playing Werewolf engages basic human skills at a high level.  What exactly do cheaters look like?  Excessive happiness and unnatural poker forces are both strong signs.  But so is flattened – or extremely bimodal – affect.  Honest villagers tend to be moderately emotionally variable.  Of course, if you repeatedly play with the same people, you should add “fixed effects” – look for deviations from the person‘s normal demeanor rather than deviations from people‘s normal demeanor.

As I played, I thought that my cheater detection skills improved sharply.  My deception skills, not so much.  My knee-jerk conclusion was that detection is easier to improve than deception.  But this article points out a plausible alternative hypothesis: Practice matters.  You practice deception in 2/15 of games, and detection in the remaining 13/15 of games.

There are already small academic literatures on Werewolf in mathematics and psychology – see here and here for starters.  But a few hours of play convince me that it deserves far more attention. 

Has anyone else played?  Any social scientific insights you’d care to share?