In writing up some subsequent experiments, we had considerable trouble relating our results to a published paper that reported avoidance conditioning but not any subjects that failed to learn. When I later saw one of the authors at a meeting, he said they actually had thrown out a lot of rats that had failed to learn. He seemed unaware that their failure to mention this subject elimination in their article might mislead others.
rice, lime, tuna, beef, fish, pear, cake, veal
They were then asked various questions that had some, none, or all of the words as answers. Among these questions were:
Which have exactly two vowels?
Which have exactly two consonants?
Their task was to process the memorized list serially, speaking correct answers aloud.
The sequence of response times followed an additive model that demonstrated the independence of the two primary processes: memory retrieval; and the decision whether a given word was an answer. The decision processing times could be quantified with this model. As one interesting result, decision time for the listed consonant question was twice as long as for the vowel question although the two have identical answers.
Occasionally, a subject failed to speak a correct answer. In order to complete the data curve, the mean latency for the other subjects was substituted in these cases. Such missing responses occurred about 3% of the time in the later experiments, a frequency that seems not unreasonable for this particular task. Various checks indicated that this substitution procedure did not bias the results. Extremely long response times occurred about 1 time in 200 and were replaced in a similar manner.
This experiment is one of many in which within subject replication would be desirable. This would have eased the missing data problem and, more important, allowed assessment of individual differences through single subject analysis.