Circular Analysis in Complex Stochastic Systems
Authors
Affiliations
Ruling out observations can lead to wrong models. This danger occurs unwillingly when one selects observations, experiments, simulations or time-series based on their outcome. In stochastic processes, conditioning on the future outcome biases all local transition probabilities and makes them consistent with the selected outcome. This circular self-consistency leads to models that are inconsistent with physical reality. It is also the reason why models built solely on macroscopic observations are prone to this fallacy.
A conditional likelihood is required to estimate the selection coefficient in ancient DNA.
Valleriani A Sci Rep. 2016; 6:31561.
PMID: 27527811 PMC: 4985692. DOI: 10.1038/srep31561.
Predict or classify: The deceptive role of time-locking in brain signal classification.
Rusconi M, Valleriani A Sci Rep. 2016; 6:28236.
PMID: 27320688 PMC: 4913298. DOI: 10.1038/srep28236.