Evol Ecol Res 16: 153-164 (2014)     Full PDF if your library subscribes.

The limitations of inferring decision rule use from individuals’ sampling behaviour: a computational test of old and new algorithms

Bart J. Kensinger and Barney Luttbeg

Department of Zoology, Oklahoma State University, Stillwater, Oklahoma, USA

Correspondence: B.J. Kensinger, Department of Zoology, Oklahoma State University, Stillwater, OK 74078, USA.
e-mail: bart.kensinger@okstate.edu

ABSTRACT

Background: The foraging and mate choices of individuals depend on how individuals gather information about available options and use that information to make decisions. Several alternative decision rules have been proposed, but little progress has been made in using observations of behaviour to discover what decision rule is being used. An important approach has been to combine the rules into an algorithm that chooses between them (‘the prevailing algorithm’).

Question: How accurately does the prevailing algorithm classify which decision rule is most likely to produce observed sequences of individuals choosing one item from a pool of many? Can we create better algorithms?

Methods: Using computer simulations, we construct sampling sequences that follow the threshold, best-of-n, and comparative Bayes rules. We vary sampling cost, pool size, signal variance, and population variance. We mimic sequences seen in nature and avoid parameter values in which individuals sample infinitely or not at all. We apply the prevailing algorithm to determine which rule created each sequence, and measure its probability of success. We use a recursive partitioning exercise to produce two new algorithms: one based on individuals’ sampling sequences, and one based on the population’s averaged metrics. We compare the performances of these two new algorithms and the prevailing algorithm. Finally, we apply all three algorithms to previously published empirical data sets.

Conclusions: The prevailing algorithm did no better than a random classification, but the individual algorithm had 64% success and the population algorithm had 97% success. The latter also succeeded in identifying the predominant rule even if different individuals of the group had different rules. The population algorithm can reliably infer what rule the majority of a population is using. However, using observed sequences to infer the decision rule of an individual is unreliable.

Keywords: decision rules, mate choice, foraging.