Positive bias in the perception of covariation

Perception of covariation often differs from statistically normative values: People find order in random series and relationships between uncorrelated values. Theoretical analysis, allowing for working-memory limitations, shows that the degree of covariation in the typical, locally representative series is more negative, whereas that of the atypical series is more positive, than the covariation in the complete set. I assumed that typical series serve as a norm to which other series are compared, and predicted a positive bias in the perception of covariation. This prediction was tested and found to hold across a wide range of actual relationships in 2 experiments involving sequential dependencies and events with co-occurring values. Another analysis revealed positive correlations to be more informative than negative ones when events are not equiprobable. Positive bias may thus be a rational predisposition for early detection of a potentially more informative relationship.