Abstract:
In the conducted resource allocation games, «constant» behavior (CB) of the players was found when the players do not change their bids during several steps of the game. A feature of the game data is that the CB occupies a large share in them. A modified design of the games is proposed, where payments are made for each of the five game steps and the interface does not allow the player to easily enter an unaltered bid. The results of games with modified design showed that the proportion of CB has decreased, but remains at a fairly high level. The following are the results of a search for the causes of this behavior using statistical hypotheses and solving problems of classifying decisions of players. Among the statistical hypotheses, hypotheses are used about the random nature of CB and stopping CB in case of decrease payoffs. The classification task allows the selection of informative features (parameters and history of the game) and the rule of decision-making by players (classifier) to stop the CB on the features basis. The results of constructing the classifier gave the idea of complicating the model: the player has not only the principle of stopping/continuing CB, but also the principle of starting CB. It is also shown that the results of statistical research and classification have common features and can further complement each other.
Fulltext:
PDF file (647 kB)