Published online by Cambridge University Press: 01 January 2022
Models that fail to satisfy the Markov condition are unstable because changes in state variable values may cause changes in the values of background variables, and these changes in background lead to predictive error. Such error arises because non-Markovian models fail to track the causal relations generating the values of response variables. This has implications for discussions of the level of selection: under certain plausible conditoins most standard models of group selection will not satisfy the Markov condition when fit to data from real populations. These models neither correctly represent the causal structure generating nor correctly explain the phenomena of interest.
Thanks to Chris French and Maria Glymour for conversations and aid in coding. Work on the simulations was supported by a grant from Kansas State University's Center for the Understanding of Origins, whose support I gratefully acknowledge.