A model of changeover behavior in two‐alternative choice
Switching between two response options follows a simple leaving-rate equation drawn straight from the matching law.
01Research in Context
What this study did
Avellaneda (2025) built a math model that predicts when animals switch between two levers. The model uses simple leaving rates tied to the matching law.
He tested the model against old data from pigeons pecking two keys. Each key paid off on its own variable-interval schedule.
What they found
The Markov-chain fit matched the real changeovers almost perfectly. You can guess switch timing just by knowing how fast each option stops paying off.
No extra rules or fudge factors were needed. The matching law alone told the story.
How this fits with other research
Hanley et al. (2003) catalog hundreds of functional analyses that use two-choice setups. Their review never mentions a clean predictor for switching; Avellaneda fills that gap.
Debert et al. (2009) show emergent relations in Go/No-Go tasks. Both papers deal with concurrent cues, but Avellaneda gives the numbers behind the jumps.
Preston (1994) warns that FA methods vary too much. A steady switch model like this could tighten those designs by setting clearer switch expectations.
Why it matters
If you run concurrent operant assessments, you now have a quick way to forecast client switching. Plug the reinforcer rates into the leaving-rate formula and see when a jump is likely. That lets you spot abnormal persistence or flip-flop before it clouds your functional hypothesis.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Count reinforcers per minute on each alternative during your next concurrent FA, calculate the predicted leaving rate, and see if the client switches when the model says they should.
02At a glance
03Original abstract
The amount of time that organisms spend on a variable-interval schedule of a concurrent pair before departing to the other one (i.e., the dwell time on the schedule) follows an exponential distribution, meaning that the probability of switching to the other schedule does not increase or decrease throughout the visit. This appears to reflect an innate behavioral pattern and implies that concurrent-schedule performance can be modeled using continuous-time Markov chains. In the two-alternative case, the behavior of a Markov chain is completely determined by the leaving rates from each alternative (i.e., the number of departures per unit of time), so finding expressions for these leaving rates should suffice to completely characterize changeover behavior in concurrent schedules. Such expressions can be derived from the matching law in combination with either the mathematical principles of reinforcement or Baum's laws of allocation, induction, and covariance. The resulting equations are assessed in the particular case of concurrent variable-interval schedules using a large data set from a published study that systematically manipulated both the relative and the overall rates of reinforcement, resulting in excellent fits. The performance of the model is also assessed against that of competing models, proving to be superior in most cases.
Journal of the Experimental Analysis of Behavior, 2025 · doi:10.1002/jeab.70025