Context-Content Systems of Random Variables: The Contextuality-by-Default Theory

Abstract

This paper provides a systematic yet accessible presentation of the Contextuality-by-Default theory. The consideration is confined to finite systems of categorical random variables, which allows us to focus on the basics of the theory without using full-scale measure-theoretic language. Contextuality-by-Default is a theory of random variables identified by their contents and their contexts, so that two variables have a joint distribution if and only if they share a context. Intuitively, the content of a random variable is the entity the random variable measures or responds to, while the context is formed by the conditions under which these measurements or responses are obtained. A system of random variables consists of stochastically unrelated "bunches," each of which is a set of jointly distributed random variables sharing a context. The variables that have the same content in different contexts form "connections" between the bunches. A probabilistic coupling of this system is a set of random variables obtained by imposing a joint distribution on the stochastically unrelated bunches. A system is considered noncontextual or contextual according to whether it can or cannot be coupled so that the joint distributions imposed on its connections possess a certain property (in the present version of the theory, "maximality"). We present a criterion of contextuality for a special class of systems of random variables, called cyclic systems. We also introduce a general measure of contextuality that makes use of (quasi-)couplings whose distributions may involve negative numbers or numbers greater than 1 in place of probabilities.