Quantifying contingent relations from direct observation data: transitional probability comparisons versus Yule's Q.

Vos et al. (2013) compared two ways to measure how one behavior leads to another. They looked at transitional probability and Yule's Q. Both turn raw observation notes into a single number that says 'when A happens, B follows this often.'

The paper is a guide, not an experiment. No kids, no clients, just math. The goal was to tell analysts which index to trust under which conditions.

Yule's Q works best when you have few data points. Transitional probability wins when you have lots of clear yes-no records. Pick the wrong one and you can miss a real link or see a fake one.

Prain et al. (2012) saw the same problem with inter-rater reliability. They warned that percent agreement inflates trust when behaviors are rare. Their fix—report Cohen's kappa—mirrors P et al.'s advice: pick the index that fits the data shape.

Killeen (1978) compared three rules for calling behavior 'stable.' Like P et al., the paper showed that simpler rules save time but can mislead. Both studies push analysts to match the tool to the data size.

Vassos et al. (2023) found short 'transition states' in single-case graphs. Those fuzzy first treatment points are exactly where Yule's Q, not transitional probability, is safer because the sample is tiny.

Next time you chart session notes, count your data points first. Under twenty pairs? Use Yule's Q in Excel. Over fifty clean 0-1 records? Transitional probability is fine. The right pick keeps your contingency claim believable and saves you from extra sessions.