A flexible model for generalization gradients.
A 1959 machine showed you can draw a generalization curve without animals, but later work says schedule choice and observing behavior fine-tune that curve in real life.
01Research in Context
What this study did
AZRIN (1959) built a machine that acts like a generalization gradient. No birds. No rats. Just gears and levers.
The goal was to remove living-organism noise so researchers could see a clean curve.
What they found
The paper does not give new animal data. It shows how a mechanical device can draw the same curve we usually get from pigeons.
The author argues the machine curve proves you can study stimulus control without messy biology.
How this fits with other research
Zeiler (1969) later ran real pigeons and found that different reinforcement schedules make the curve steeper or flatter. The live data match the mechanical idea, but add the rule: FI, VI, and VR keep the curve stable longer.
Crossman et al. (1973) gave a practical tip: end each session with a long, stretching interval schedule. This keeps the gradient from collapsing during extinction probes. Their method makes the mechanical dream work in actual feathers.
Lyons (1995) updated the story. Watching the stimulus, not just reinforcing it, drives discrimination. The 1959 gears ignored observing; the 1995 theory puts it center stage.
Why it matters
You still run generalization probes. Use K et al.’s long-adjusting-interval wrap-up to keep the curve steady. Remember that schedule type, not just reinforcement, shapes the slope. And watch where your client looks—observing may matter more than reinforcer size.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →End your next generalization probe session with one reinforcer on a long, stretching interval to keep stimulus control alive.
02At a glance
03Original abstract
In much current research in Learning, the task of E has been cut drastically: He merely programs the experiment, and works up the results. It is true that he still has certain non- psychological duties, for he must make sure that the apparatus works, and he usually puts the S into the experimental situation and removes him at the close of the session. This isn't unduly onerous, for S is often reduced to an N of one. The present paper suggests that even this N is excessive-variability can be reduced by substituting a mechanical model for the S, at least in studying generalization. The model has the further advantage that it, unlike the rat or pigeon, can be made to behave in terms of certain preconceived principles of generalization.
Journal of the experimental analysis of behavior, 1959 · doi:10.1901/jeab.1959.2-319