A comparison of different measures of response strength in the study of stimulus generalization.
Plot response rate after the first response; latency and trial count hardly change the gradient.
01Research in Context
What this study did
The team worked with two pigeons in a Skinner box.
They tested many wavelengths of light to see how responding spread.
They counted three things: how fast the bird pecked, how long it waited to start, and how many trials it began.
Then they asked which of these numbers really shapes the generalization curve.
What they found
After they removed the effect of first-peck latency and trial count, the gradient stayed almost the same.
The curve was driven by the rate after the bird already started pecking, not by how quickly it began.
How this fits with other research
Howard (1979) extends this idea to people.
That study showed humans make gradients based on words like "high class" and "low class," not just color or brightness.
Together the papers show the same gradient rule holds for birds and for verbal humans.
Mello (1966) used the same pigeon setup but asked a different question.
That paper proved you must give discrimination training during punishment or the gradient flattens.
The two 1966 studies fit side-by-side: one tells us what to measure, the other tells us what training to give.
Why it matters
When you graph how a client spreads responding across new pictures, sounds, or people, plot the rate after the first response.
Ignore how long they take to start or how many trials they try.
This gives you the cleanest picture of true stimulus control and keeps data nights short.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Record rate only after the first correct response when you graph generalization probes.
02At a glance
03Original abstract
In Experiment 1, three pigeons were given variable interval training to peck at a light of 550 mmu and then were tested for stimulus generalization in extinction to several different wavelengths. A gradient was obtained for latency of the first response in each test period, for the number of test periods in which responding occurred, and for the measure of response rate. When the response rate gradient was corrected for differences in initial latency and in number of responded trials, the change was minimal, indicating that the major component of response rate as usually measured is rate of responding having once responded. In Experiment 2, three other pigeons were trained to respond to 550 mmu (for variable interval reinforcement) and not to 570 mmu (extinguished). Analysis of generalization gradients dictated the same conclusion as that reported for generalization following single stimulus training.
Journal of the experimental analysis of behavior, 1966 · doi:10.1901/jeab.1966.9-239