Diagrams and predictions of the different models tested in the experiment
With respect to the birds= decision strategy, our starting point was based on prior work examining delayed presence-vs.-absence discriminations in pigeons. Consider, for example, pigeons' delayed discrimination of food vs. no-food samples (e.g., Grant, 1991). In this procedure, a sample consists of the brief presentation of food or the brief presentation of an empty food hopper. Following a delay, red and green choice stimuli are presented. A response to one choice stimulus is reinforced on food trials (with more food) while a response to the other choice stimulus is reinforced on no-food trials. In most cases, choice performance declines as a function of delay interval following food samples, but declines little, if at all, following no-food samples (Colwill, 1984; Colwill & Dickinson, 1980; Grant, 1991; Sherburne & Zentall, 1993a, 1993b; Wilson & Boakes, 1985). Exactly the same pattern of results occurs when the procedure involves a delayed discrimination between the presence-vs.-absence of a keylight instead of a food sample (Grant, 1991; Dougherty & Wixted, 1996; Wixted, 1993).
One popular account of this asymmetrical "forgetting" in presence-vs.-absence discriminations is the default response hypothesis. According to this account, the presentation of a food sample produces a memory code of that event, but no-food samples are basically treated as nonevents (and, as such, produce no memory code). Whenever memory for a food sample exists, the pigeon correctly responds to the appropriate choice alternative. In the absence of a memory code, the other alternative is selected by default. Note that a memory code is absent on all no-food trials (leading to correct choices) and on long-delay food trials in which the sample has been forgotten (leading to incorrect choices). Thus, performance on food sample trials declines with an increasing retention interval as the memory trace fades away, but performance on no-food sample trials remains constant because memory is not involved. Instead, the no-food choice alternative is selected by default on no-food trials regardless of the size of the retention interval.
Wixted (1993) pointed out the formal equivalence of the default response/asymmetrical coding account to an earlier theory of human psychophysical discrimination known as high threshold theory (HTT; Blackwell, 1963; Swets, Tanner, & Birdsall, 1961; see Link 1992 for a review). HTT proposes that when detecting any dimensional signal from a background of noise, subjects respond yes (signal) if the strength of the signal exceeds a fixed threshold or criterion. On all other trials (noise alone trials & subthreshold signal+noise trials), HTT proposes that observers make a biased response among the yes and no alternatives. As such, in terms of this model a bird's default response is a manifestation of a strong bias to choose the "no-food" alternative when no memory for a food sample is present. The state diagram in figure below illustrates the different choice pathways of this particular theory.
The major competitor to HTT models of performance has been signal detection theory (SDT; Peterson, Birdsall, & Fox, 1954; Tanner & Swets, 1954). According to signal detection theory, the subject's decision is guided by information derived from the stimulus and the relative placement of a response or decision criterion. In conjunction, these two factors determine whether the subject will respond "yes" or "no" on signal+noise and noise trials. The diagram belows illustrates the different choice pathways of this particular theory and how they contrast with those of HTT.
The two figures show that an important difference between HTT and SDT concerns the nature of the errors made on noise-only trials. Because in HTT no internal representation of a signal is possible on noise trials (i.e., there is no path between the No signal stimulus and the internal state of a signal), this model says that false alarms (choosing yes on noise trials) come from an incomplete bias on the part of the observer to choose one of the response alternatives. Returning to our original example, the birds on no-food trials have a very strong, but not complete, bias to choose the no-food alternative, resulting in the occasional choice of the food alternative. On the other hand, SDT proposes that false alarms are the product of the variability associated with the representation of the stimulus and the subject=s criterion. Because the perceptual system is imperfect, such variability exists even on noise trials. As such, when the strength of evidence due to this noise exceeds the decision criterion, the subject mistakenly reports that a signal occurred. This important theoretical difference between HTT and SDT is empirically manifested in their different predictions about the form of the receiver operating characteristic (ROC). The ROC curve is a standard psychophysical graph showing the hit rate as a function of the false alarm rate over conditions in which the subject's response bias has been manipulated. The predicted shape of the ROC according to the HTT model is a linear function, while in contrast, the ROC function predicted from SDT is curvilinear. It is these contrasting predictions which were examined in this experiment.
Dynamic
Texture Stimuli 
Same-Different
& SDT 
Same-Different
w/ Multiple Stimuli
Dr.
Robert Cook's Home Page 
Home
Page for all 3 articles
Give us Feedback 
Search the
Site
08/25/99