The cover-stimulus effect: role of similarity in durations of the CS and cover cues.
In order to test theoretical accounts of the cover-stimulus effect, two groups of investigators have manipulated the duration of the cover stimulus. Cooper et al. (1990, Experiment 1) exposed two groups of pigeons to a signaled random autoshaping procedure. Duration of the CS was 12 s, and the cover stimulus was either 12 s (short group) or 48 s (long group) in duration. Birds in the short group showed pecking to both stimuli, whereas pecking did not emerge for the CS in the long group. Similar results were obtained in Experiment 2. The authors identified duration of a signaled period of nonreinforcement, relative to duration of the CS, as the determinant of the cover stimulus duration effect in their study. In another study, Williams (1994) manipulated the duration of the cover stimulus using a sign-tracking preparation with rat subjects. Unlike Cooper et al., who presented cover-stimulus durations that were equal to or longer than the CS, Williams shortened the cover stimulus relative to the 15-s CS. The cover stimulus was either 15 s (long signal group) or 5 s (short signal group) in duration. In comparison to a no signal group for which no cover stimulus was presented, rats in the long signal group showed a reliable increase in sign tracking over 30 sessions. Performance of rats in the short signal and no signal groups did not differ reliably, and showed little effect of training. Williams focused on the absolute duration of the cover stimulus as a determinant of the cover-stimulus effect, suggesting that there may be an optimal duration. Alternatively, he pointed out, the optimal duration may vary with duration of the interreinforcer interval.
A different approach is taken in the present report in which CS duration in comparison to cover-stimulus duration is identified as a determinant of the cover-stimulus duration effect. In the three relevant prior experiments (Cooper et al., 1990, Experiments 1 and 2, and Williams, 1994), higher rates of CS-directed responding were observed when the cover stimulus was equal in duration to the CS, as opposed to when it was longer or shorter. These data raise the interesting possibility that duration of the cover stimulus relative to the CS is a determinant of the cover-stimulus effect. Though there are data that do not follow this pattern, those data were obtained under procedures in which the programmed relation between the CS and US was not random (e.g., Cooper et al., Experiment 3). Because no other published studies using the signaled random procedure have manipulated cover and conditioned stimulus durations relative to each other within the same experiment, the present study tested the reliability of the stimulus-duration-equality effect apparent in the data of Cooper et al. and Williams. Experiment 1 was a systematic replication of Williams' long signal and short signal groups. The purpose was to determine if the effects reported by Williams also occur in pigeons, as opposed to rats, using the stimulus and response parameters available for this species. Experiment 2 provides a stronger test of the stimulus-duration-equality effect by exposing different groups of pigeons to cover cue durations that were shorter than, equal to, or longer than the CS duration. In addition, the study was designed to determine if CS duration per se contributes to the stimulus-duration-equality effect.
Research on the signaled random procedure has been guided by two families of associative-learning models--competition and comparator theories. For example, Durlach (1983) showed that the competition model of Rescorla and Wagner (1972) provided a good account for her data. In contrast, Cooper et al. (1990) presented data that were not readily accounted for by the Rescorla-Wagner model, but that could be accommodated by the deletion comparator model, a version of Scalar Expectancy Theory (Gibbon & Balsam, 1981). In the General Discussion of this report, we evaluate the performance of competition and comparator models in accounting for the effects of relative durations of CS and cover cues. We also consider the possibility that the stimulus-duration-equality effect is mediated by stimulus generalization between the cover and conditioned stimuli.
Experiment 1 is a systematic replication of Williams' (1994) study in which cover cues were equal to or shorter than the CS. While Williams measured sign tracking during the CS in rat subjects, the present study employed pigeon subjects and measured sign tracking during both the CS and the cover stimulus. By evaluating responding during both cues, the present study revealed the extent to which responding to the CS could represent stimulus generalization from the cover cue. The present experiment also extended the acquisition phase from 40 sessions (Williams, 1994) to 70 sessions.
Sixteen experimentally naive pigeons maintained at 75% of their free feeding weights were used in this experiment. Eight birds were Silver Kings and eight were White Carneaux. All birds were allowed free access to grit and water in their home cages, and were fed enough grain after each experimental session to maintain their weight at the specified level. The birds were housed individually in a vivarium that maintained a 12-hr light/dark cycle.
All training and testing was conducted in four Lehigh Valley Electronics 3-key experimental chambers for pigeon subjects (Model SEC-002) measuring 37 cm H X 31 cm W X 34 cm D. The front panel contained a houselight, three response keys, and a food hopper. The houselight was centered on the panel, 1.5 cm from the top. It was shielded with a metal covering that directed its light toward the ceiling.
Only the 2.5-cm center key was operational for recording responses and presenting stimuli during the experiment. It was centered on the front panel, 8 cm below the top of the panel. The hopper aperture (5.5 cm W X 5 cm H) was centered 6 cm below the center key. Infrared emitters and detectors were mounted behind the hopper opening, and were used to detect head entries into the hopper. All food deliveries were 1 s in duration from the moment the hopper photobeam was intercepted. Stimulus presentations and data collection were controlled by an IBM AT computer operating Spyder Systems CONMAN[R] software. Data were collected with a resolution of 100 ms. Visual stimuli were presented with an IEE in-line projector (Model no. 10-OK21-1820-L) mounted behind the center key of each chamber. The stimuli used were a blue keylight and a white "X" on a black surround. Stimulus function was counterbalanced across birds and groups.
Magazine training. During magazine training sessions only, the experimental chambers were fitted with black wall liners and the white noise was turned off. Over successive trials, the hopper was activated according to an increasing variable time schedule, ranging from a mean of 15 s to 60 s, and access time to grain was reduced from 20 s to 1 s (Brown, Hemmes, Cabeza de Vaca, & Pagano, 1993). The final criterion required that birds have a median eat-latency of < 2 s across 51 trials during which a 1-s food US was presented. After all birds successfully met this criterion, an extra session of magazine training was conducted for all subjects on the day prior to the next experimental phase. Birds were then randomly assigned to two groups with the restriction that the strain of birds (Silver King, White Carneaux) be balanced across group and chambers. One bird in the CS15-Cov5 group (see below) died after Session 40 and its data were discarded.
Signaled random procedure. The contingencies employed in the present study were identical to those of Williams (1994). The two groups of subjects were exposed to a signaled random procedure in which the CS was 15 s in duration, and the cover stimulus was either 5 (CS15-Cov5 group) or 15 s long (CS15-Cov15 group). Each 30-min session contained 120 15-s periods, of which 18 contained a CS and 102 did not. The US was presented at the end of 50% of the CS and no-CS periods, so that p(US|CS) = p(US|[begin strikethrough]CS[end strikethrough]) = .50. The 51 US presentations during no-CS periods were all preceded by the cover stimulus; that is, p(US|cover stimulus) = 1.0. Accordingly, four types of 15-s intervals were programmed: 9 CS-US trials (a 15-s CS presentation was followed by a presentation of the US); 9 CS-alone trials; 51 cover stimulus-US trials (a 5- or 15-s cover stimulus was immediately followed by a US); and 51 empty intervals (no stimuli presented). The CS and the 15-s cover stimulus commenced at the beginning of each 15-s period in which they were programmed. The 5-s cover stimulus filled only the last 5 s of any 15-s period in which it was scheduled to occur. See Table 1 for total durations of CS and cover cues and of stimulus-free periods. Birds were exposed to their respective conditions for 70 sessions.
Two measures of responding during the CS and cover cues were analyzed--proportion of trials with a response and rate of responding. Effects of the independent variable on both measures were assessed using the Kruskal-Wallis H test. A nonparametric test was selected owing to the presence of significant skewness in CS and cover cue responding as measured by the mean of 70 sessions from each phase (Pearson's Third Moment test for skewness cited by Ramsey & Ramsey, 1990, Zs [greater than or equal to] 1.96, ps < .05). Wilcoxon tests were used for comparisons across phases (Inman, 1974).
Figure 1 (left panel) shows the data for responding to the CS across the seven 10-session blocks of the experiment for both groups. The top and bottom panels show the group mean proportion of trials with a peck and rate of responding, respectively. Birds in the CS15-Cov15 group responded at higher levels than those in the CS15-Cov5 group throughout training. Kruskal-Wallis H tests applied to mean performance across the seven blocks supported this observation for both dependent measures, H(1) = 9.61 and 7.70, for proportion of trials with a peck and rate of responding, respectively, p < .05. A Kruskal-Wallis H test was also applied to the data for each block. In comparison with the CS15-Cov5 group, the CS15-Cov15 group produced a significantly higher proportion of trials with a response in Blocks 4-7, and higher rates of responding in Blocks 5-7, Hs(1) > 3.84, ps < .05.
[FIGURE 1 OMITTED]
Cover Cue-Directed Responding
The right panel of Figure 1 represents performance during the cover cue. As assessed by 70-session means, birds in the CS15-Cov5 group produced a higher proportion of trials with a peck than did those in the CS15-Cov15 group, H(1) = 4.83, p < .05. Tests applied to each block revealed significant differences in Blocks 1-6, Hs(1) > 3.84, ps < .05, but not in Block 7, H(1) = 3.42, p = .064. Mean rates of responding did not differ between groups during session Blocks 1-7, H(1) = 1.33, p > .05. Tests applied to each block did not reveal any significant difference except for Block 5 in which the CS15-Cov5 group responded at a higher rate than the CS15-Cov15 group, H(1) = 3.87, p = .049.
The results from this study support the hypothesis that responding to the CS under the signaled random procedure is affected by CS duration relative to cover-stimulus duration. Virtually no CS-directed responding was observed when CS and cover-stimulus durations were unequal (CS15-Cov5 group), while the equal duration condition (CS15-Cov15 group) did support responding to the CS. The present study provides a fourth demonstration of the stimulus-duration-equality effect, and extends the finding to pigeon subjects under the temporal parameters used by Williams (1994) with rat subjects. By exposing subjects to an extended acquisition phase (70 sessions), it also indicates that the failure to observe responding under the unequal condition is not attributable merely to a slow rate of acquisition. In comparison to Williams' (1994) data, differences in CS-directed responding between equal and unequal groups emerged somewhat more quickly, and overall level of responding was lower. It is possible that this difference can be attributed to species differences and/or the relatively more nonspecific response measure (proportion of time spent on the side of a shuttle box containing the CS) used by Williams. The present results are quite similar to those of Cooper et al. (1990) who studied pigeon autoshaped responding, as in the present study. Durlach (1983) also studied autoshaped behavior and reported comparatively high levels of responding; however, unique to that study was cover-cue pretraining.
The effect of the independent variable on cover-cue performance differed from that observed for CS performance. Whereas the unequal duration group (CS15-Cov5 group) showed a low level of responding to the CS in comparison to the equal condition (CS15-Cov15 group), the opposite was true for the cover stimulus (for the proportion of trials measure only). This result could be anticipated on the basis differences in cover-cue duration (5 versus 15 s in the unequal and equal conditions, respectively), as well as by the ratio of interreinforcer interval to cover-cue duration (Gibbon, Baldock, Locurto, Gold, & Terrace, 1977; Lattal, 1999). The data provide no support for mediation of CS responding by stimulus generalization from the cover cue.
Collectively, four experiments indicate that the cover-stimulus effect is diminished when cover-stimulus duration is either shorter (Williams, 1994; the present Experiment 1) or longer (Cooper et al., 1990, Experiments 1 & 2) than the CS. This generalization would be enhanced if degradation of the cover-stimulus effect were obtained when both unequal conditions (cover stimulus shorter or longer than CS) are presented in the same experiment. Accordingly, in Experiment 2, two groups of pigeons were exposed to a 12-s CS and to a cover stimulus that was either shorter (4 s) or longer (36 s) than the CS. Performance under these two unequal conditions was compared to that of groups with equal CS and cover durations of 4, 12, or 36 s. The design of this experiment permitted assessment of Williams' (1994) suggestions that the cover-stimulus effect may depend upon the absolute duration of the cover stimulus or on that duration relative to the average interreinforcer interval. It also tested the generalization reported by Williams that there is no evidence of a cover stimulus effect for cover stimulus durations of less than 10 s. If relative, rather than absolute, duration of CS and cover cues controls the level of CS responding, then groups exposed to the conditions with equal CS and cover-cue durations should produce higher levels of responding than groups exposed to unequal durations.
Although the cover effect was obtained in Experiment 1, absolute rates of CS responding were relatively low. In conformity to previous studies, a cover cue pretraining phase similar to that of Durlach (1983) was instituted in Experiment 2 to establish the cover cue as an effective conditioned stimulus which might, in turn, permit higher levels of responding to the CS.
Thirty experimentally naive pigeons (15 White King, 15 White Carneaux) obtained from the Palmetto Pigeon Plant, Sumter, South Carolina, were used. The birds were housed individually and exposed to a 12-hr light/dark cycle. All birds were allowed free access to grit and water in their home cages and were maintained at 75% of their free feeding weights.
This experiment employed five Lehigh Valley 3-key experimental chambers (Model SEC-002) identical to those used in Experiment 1. The control of inputs and outputs from these chambers was performed by MED-Associates MED-PC[R] system software employing a data resolution of 10 ms. The stimuli used for the CS and cover cues were a white "X" on a black surround and a blue keylight, respectively.
As in Experiment 1, duration of US (food) presentation was 1 s, timed from the moment at which the bird's head interrupted the magazine photo-cell beam. In contrast to Experiment 1, a 10-s limited hold was programmed so that if a bird did not eat within 10 s of any US presentation, the hopper was lowered and the session proceeded. The limited hold was never encountered during cover-cue pretraining, and rarely encountered (.01% of trials) under the signaled random procedure.
Group assignment and counterbalancing of daily session times. All birds were exposed to the same magazine training protocol as in Experiment 1, and were then randomly assigned to six groups. The absolute and relative durations of cues were manipulated across groups. For four groups, CS and cover cue durations were equal at 4, 12, or 36 s--CS4-Cov4, CS12-Cov12, CS12-Cov12D, and CS36-Cov36 groups. For the other two groups, CS duration was 12 s and cover duration was 4 or 36 s--CS12-Cov4, and CS12-Cov36 groups, respectively.
Counterbalancing between groups and daily session times could not be performed on a daily basis owing to the unequal number of groups (6) and chambers (5). For this reason, all birds in a group were run at the same time of day. The time of day, however, changed in a systematic fashion across sessions on a 12-day cycle. Running times for all groups were ranked 1 (first group to run that day) through 6 (last group to run that day) for the first day in a cycle. On Days 2, 4, 6, 8, 10, and 12 of a cycle, Ranks 1 and 2, 3 and 4, and 5 and 6 were reversed from the running rank of the previous day. On Days 1, 3, 5, 7, 9, and 11 of a cycle, Ranks 2 and 3 and Ranks 4 and 5 were reversed from their previous day's ranks. This manner of changing running times balanced the exposure of groups to running times over a 12-session cycle.
Cover-cue pretraining. After the groups were formed, all birds were exposed to a cover-cue pretraining procedure. Birds in each group were exposed to the same cover-cue duration for which they were scheduled during the signaled random phase of the experiment. All cover-cue presentations were immediately followed by presentation of the US, and there were 20 trials per session. The mean ITI duration was 10 times the value of the cover-cue duration, [t.sub.c], and the range was [t.sub.c] to 19[t.sub.c]. Nineteen ITI values were evenly distributed within the range, with the addition of an ITI value equal to 10[t.sub.c]. The resulting mean ITI values yielded equivalent ratios of ITI to trial duration, as well as cycle to trial duration, across groups. All birds were exposed to this procedure for 12 sessions, with 20 trials per session.
Signaled random procedure. Following cover-cue pretraining, all birds were exposed to the signaled random procedure. Each session was comprised of 250 time-base periods of a duration equal to that of the CS for a given group. There were 20 CS periods and 230 no-CS periods. The US was presented at the end of 4 of the CS periods and 46 of the no-CS periods so that that p(US|CS) = p(US|[begin strikethrough]CS[end strikethrough]) = .20. Each of the US presentations during a no-CS period was preceded by the cover cue; that is, p(US|cover CS) = 1.0. The order of events within a session was determined by a random selection-without-replacement procedure. The experiment was conducted over 84 sessions.
For the equal-stimulus-duration groups (CS4-Cov4, CS12-Cov12, CS12-Cov12D, and CS36-Cov36), there were 184 stimulus-free periods, each equal in duration to the time-base duration (see Table 1 for total durations of the CS, cover stimulus, and stimulus-free periods). For the CS12-Cov4 group, which was exposed to a cover cue duration (4 s) shorter than the time base (12 s), periods containing the cover cue commenced with an 8-s stimulus-free interval, followed by the 4-s CS. For this group, the total stimulus-free time included 184 stimulus-free periods, 12 s in duration, plus 46 8-s intervals corresponding to time left in periods containing a cover stimulus. For the CS12-Cov36 group, which was exposed to a cover-cue duration longer than the 12-s time base, some of the periods that were stimulus free for the other 12-s CS-duration groups were filled with the cover stimulus. For each of the 46 cover-cue presentations, three 12-s time-base periods were filled with the stimulus. Because session duration was held constant among all of the 12-s CS-duration groups, there were correspondingly fewer stimulus-free periods for the CS12-Cov36 group than for the other 12-s CS-duration groups (see Table 1).
For all groups but CS12-Cov12D, one trial could begin immediately after another had finished, and the order of events within a session was randomly determined. A constraint was placed on the distribution of events for the CS12-Cov12D group to partially control for inter-US interval differences between the CS12-Cov12 and CS12-Cov36 groups that could influence responding (Jenkins, Barnes, & Barrera, 1981). For the CS12-Cov12D group, a 24-s trial-free period was imposed after each CS presentation. This trial distribution constraint maintained the same CS-cover trial intervals as in the CS12-Cov36 group (24 s interval + 12 s cover cue = 36 s) and made the US-US intervals between groups more similar.
All birds acquired stimulus-directed responding during the 12 sessions of cover-cue pretraining. Mixed-design analyses of variance (ANOVA) were applied to the two measures of performance--proportion of trials with a peck and rate of responding--using sessions and groups as factors. Although response tendency appeared to be weaker under longer CS-US intervals for both measures, analyses indicated a significant effect of sessions, F(11, 48) = 3.80 and 2.99, for the proportion and rate measures, respectively, p < .05, but not of groups, F(5, 24) = 0.50 and 1.25, p > .05, and no session x group interaction, F(55, 216) = 1.34 and 1.31, p > .05. For the CS4-Cov4, CS4-Cov12, CS12-Cov12, CS12-Cov12D, CS12-Cov36, and CS36-Cov36 groups, respectively, proportions of trials with a response in the final session were .51, .60, .73, .75, .50, and .29, and rates of responding were .57, .31, .63, .22, .12, and .04 responses per s.
CS-Directed Responding in the Signaled Random Procedure
Owing to significant skewness in the data for CS and cover-cue responding under the signaled random procedure, statistical analyses used the Kruskal-Wallis H test.
[FIGURE 2 OMITTED]
Figure 2 shows responding during the CS across the seven 12-session blocks of this phase. The top and bottom panels show the mean proportion of trials with a peck and mean pecks per second, respectively. As training progressed, the level of responding diverged between groups with equal CS and cover-cue durations (open symbols) versus those with unequal CS and cover-cue durations (filled symbols). At the end of training, the proportion of trials with a peck and response rate were higher for each equal duration group in comparison to the corresponding group matched for either CS or cover-cue duration (statistical analyses are reported below). In addition, level of responding for all groups was higher than that observed in Experiment 1, suggesting a possible effect of cover-cue pretraining in Experiment 2.
Although there was a significant effect of group for proportion of trials with a response and rate of responding averaged across all seven blocks, H(5) = 17.39 and 16.89, respectively, p < .05, Figure 2 indicates that performance within several groups differed from the beginning to the end of training. A comparison of the mean level of responding in Blocks 1 and 2 versus the mean level in Blocks 6 and 7 indicated that proportion of trials with a response changed in opposite directions for the CS12-Cov12D and CS12-Cov36 groups, ts(4) = 4.24; p < .05; Wilcoxon Signed Ranks test. For rate of responding, the CS4-Cov4 and CS12-Cov12D groups showed an increase, while the CS12-Cov36 group showed a decrease from Blocks 1 and 2 to Blocks 6 and 7, ts(4) = 4.24. Accordingly, the planned between-group comparisons were conducted for performance at the end of training. These analyses are summarized for the mean of Sessions 61-84 (Blocks 6 and 7) in Figure 3. The top and bottom panels show the mean proportion of trials with a peck and the mean number of pecks per second, respectively. Open and filled bars represent equal stimulus-duration and unequal stimulus-duration groups, respectively. For both measures, level of responding was higher for all equal stimulus-duration groups in comparison to the unequal stimulus-duration comparison groups. A Kruskal-Wallis H test applied to the mean of Blocks 6 and 7 revealed a significant overall effect of group, H(5) = 19.31 and 19.68 for proportion of trials with a response and rate of responding, respectively, p < .05. Results of pairwise comparisons between individual groups confirm this apparent stimulus-duration-equality effect for both measures of performance. As shown in Tables 2 and 3, when groups were matched for CS duration, level of responding for the equal-duration groups (CS12-Cov12 and CS12-Cov12D) was significantly higher in comparison with the two unequal duration groups (CS12-Cov4 and CS12-Cov36) for all eight comparisons. In addition, comparisons of groups matched for cover-stimulus duration (CS4-Cov4 versus CS12-Cov4; CS36-Cov36 versus CS12-Cov36) also indicate a stimulus-duration-equality effect (3 of 4 comparisons are significant).
Figure 3 also indicates an effect of absolute stimulus duration for the equal-duration conditions. Level of responding decreased with increasing stimulus duration, though this effect was significant in only 5 of 10 comparisons (Tables 2 and 3).
[FIGURE 3 OMITTED]
Cover-Cue-Directed Responding in the Signaled Random Procedure
Figure 4 shows responding during the cover cue across the seven 12-session blocks under the signaled random procedure, in the same format as Figure 2. In contrast to performance during the CS, between-group differences early in this phase appeared to diminish with continued training. Analyses of group means across the 84 sessions revealed a significant group effect for rate of responding, H(5) = 14.66, p < .05, but not for proportion of trials with a response, H(5) = 10.86, p = .054. Differences in level of cover-cue-directed responding between the mean of Blocks 1 and 2 and the mean of Blocks 6 and 7 were significant only for response rate for the CS12-Cov4 group, t(4) = 4.24; p < .05.
Figure 5 shows responding during the cover cue for the last 24 sessions in the same format as Figure 3. There was a significant effect of group for mean rate of responding, H(5) = 13.77, p < .05, but not for mean proportion of trials with a response, H(5) = 9.18, p > .05. The results of individual pairwise comparisons among the groups are shown in Tables 4 and 5. All significant differences involve the two groups exposed to the 36-s cover stimulus (CS12-Cov36 and CS36-Cov36 groups), which supported very low levels of responding.
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
The present experiment provides strong support for the stimulus-duration-equality interpretation of the cover-stimulus effect. Further, it is the first to show the effect when the cover stimulus was both shorter and longer than the CS in the same experiment. It also extends the equality effect to CS and cover-stimulus durations ranging from 4 to 36 s and indicates that, contrary to a possibility raised by Williams (1994), the cover stimulus effect can be obtained with cover-stimulus durations of less than 10 s. Although not explicitly designed for this purpose, the present study also provides some evidence regarding the possibility that effectiveness of a cover-stimulus of a given duration may vary inversely with interreinforcer interval (Williams, 1994). For the two pairs of groups within which cover stimulus duration was held constant (CS4-Cov4 and CS12-Cov4; CS36-Cov36 and CS12-Cov36), cover stimulus effectiveness was not predicted by cover duration relative to interreinforcer interval. For the 4-s cover-duration groups, the cover-stimulus effect was greater in the context of the shorter interreinforcer interval (CS4-Cov4); for the 36-s cover groups, the opposite was the case--the cover-stimulus effect was stronger for the CS36-Cov36 group.
By lengthening the training procedure to 84 sessions, Experiment 2 demonstrated that the stimulus-duration-equality effect emerges relatively slowly as performance levels increase for the equal-stimulus-duration groups and decrease for the unequal-stimulus-duration groups. The gradual emergence of effects of relative stimulus duration observed in the present experiment and in Experiment 1 may account for the apparent absence of a duration similarity effect in Experiment 11 of Jenkins et al. (1981) that employed only 10 sessions of training. It also complicates interpretation of data from the signaled random procedure in general. As can be seen in Figure 2, differing conclusions may be drawn as a function of the number of sessions evaluated.
The difference in response tendency among the groups with equal CS and cover durations (CS4-Cov4, CS12-Cov12, CS12-Cov12D, and CS36-Cov36 groups) deserves comment. Because the temporal parameters used for each of these groups yield identical ratios of intertrial interval to trial duration and of interreinforcer interval (cycle) to trial duration for both the CS and cover stimuli, Scalar Expectancy Theory (SET) predicts comparable levels of acquired associative strength across these groups (Gibbon & Balsam, 1981). Nonetheless, the level of responding in the presence of both the CS and cover stimulus varied inversely with stimulus duration, and some significant differences were observed in terminal performance (Tables 2-5). These findings are consistent with those reported by Gibbon et al. (1977) who found that, although acquisition speed of autoshaped responding was well predicted by SET, terminal levels of responding varied inversely with trial duration.
The pattern of results for the cover stimulus relative to CS-directed responding differed from those of Experiment 1 in which the independent variable appeared to have opposite effects on level of cover versus CS responding (Figure 1, proportion of trials with a peck, only). In Experiment 2, cover and CS responding appeared to vary directly (Figures 2 and 4, both measures). The latter results are consistent with competition and comparator theories, according to which group differences in CS-directed responding would be supported by differential effectiveness of the cover cue in preventing context conditioning. The inconsistency between Experiments 1 and 2 may be attributable to the effects of cover-cue pretraining in Experiment 2. Cover-cue pretraining established levels of responding to the cover cue that did not change significantly for most groups during the course of the signaled random phase, and may represent a primacy effect. The fact that differences among groups in responding to the cover cue during the latter phase were restricted to conditions with the longest CS and cover-cue durations (Tables 4 & 5) suggests that responding to the cover cue may have been sensitive to the effects of trial duration or US presentation rate (see General Discussion) in both phases of the study for this cue. Thus, owing to well established response levels in pretraining, responding to the cover cue during the signaled random phase may have been relatively intransigent despite the contingencies of reinforcement prevailing during that phase. Although US presentation rate during the signaled random phase may have influenced responding to the CS as well, the prevailing contingencies of reinforcement sufficed to generate group differences in acquisition during the course of training as was also the case in previous studies employing cover-cue pretraining (Durlach, 1983).
Data generated by the signaled random procedure have been reasonably well accommodated by competition (e.g., Durlach, 1983; Goddard & Jenkins, 1987, Experiment 1) and comparator models (e.g., Cooper et al., 1990; Gunther & Miller, 2000) of associative learning. According to competition models, the CS competes with background stimuli (context) for the associative value supported by the US. When the US is presented both in the presence and absence of the CS in the unsignaled procedure, the unpaired US presentations allow the context to accrue strength relative to the CS. Under that procedure, the context effectively blocks acquisition of responding to the CS. Under the signaled random procedure, where US presentations are always preceded by either the cover cue or the CS, the context accrues little value and the CS gradually increases in associative strength. In contrast to competition models, comparator models argue that all stimuli that are contiguous with US presentation acquire associative strength independently. However, behavioral expression of the associative strength acquired by a given stimulus is controlled by its strength relative to that of the other stimuli present during conditioning (comparator stimuli). Thus, under the unsignaled random procedure, both the CS and the context acquire associative strength, but the CS does not control responding because its strength is not comparatively higher than the strength of the background stimuli (context). Under the signaled random procedure, both the CS and context acquire value as they do under the unsignaled random procedure, but the effective associative strength of the context is weakened owing to the presence of the cover cue whenever the US is presented. In Cooper et al. (1990), for example, the cover cue weakens the context by deleting the US with which it is paired from the calculation of background strength.
Neither the competition nor comparator models readily accommodate the stimulus-duration-equality effect. Under competition models, for instance the Rescorla-Wagner model, level of responding to the CS is predicted to vary with the associative strength of the cover cue which, in turn, varies inversely with cue duration. Therefore, in contrast to data from signaled random procedure experiments, competition models predict higher levels of responding when the cover cue is shorter than the CS in comparison to when both cues share the same duration. The deletion-comparator model of Cooper et al. (1990) similarly predicts that responding to the CS will be higher when the cover stimulus is shorter than, rather than equal to, the CS duration. The comparator model of Miller and Matzel (1988) predicts that high levels of responding will always develop under the signaled random procedure when the cover cue is associated with the US.
The difficulty of the Rescorla-Wagner model in accurately predicting the stimulus-duration-equality effect can be alleviated by applying a common-elements generalization analysis (Rescorla, 1976). Application of the Rescorla-Wagner (1972) model to the study of generalization suggests that cues control generalized responding to the extent that they contain common elements. Rescorla argued that when two cues, a high-frequency tone and a low-frequency tone for instance, are both reinforced, their common element (the auditory element, in this case) is also reinforced. The procedure may be described as an AX+/BX+ paradigm, where X refers to the common auditory elements shared by both A and B. The conceptualization of generalization through common elements suggests that cues that generalize to one another may also show additive or competitive effects.
Rescorla's (1976) common-element approach to the Rescorla-Wagner model can be applied to the signaled random procedure. Under the equal stimulus conditions, the CS (A), cover (B), and temporal (X) cues participate in an AX+/BX+ conditioning paradigm. The common element, X, is reinforced more frequently than components A or B because both AX and BX are followed by the US. The unequal stimulus conditions can be described as an AX+/BY+ procedure. In the absence of common temporal elements, cues A and B should acquire the same level of associative value as their corresponding temporal components, X and Y respectively, assuming equal saliences across elements. The result is that the element, X, accrues more associative value in the equal stimulus condition than in the unequal condition.
Computer simulations of the signaled random procedures of Experiments 1 and 2 were conducted using the common-elements approach to the Rescorla-Wagner model. For each element, total value accruing to that element on a given trial, [DELTA]V, is found by:
[DELTA]V = [alpha][beta]([lambda] - [V.sub.tot]) (1)
The equation indicates that [DELTA]V is a function of the learning rate parameters of the CS ([alpha]) and of the US ([beta]), multiplied by the difference between the total amount of value which can accrue to the CS ([lambda]) and the amount of value already acquired by it ([V.sub.tot]). The value of [V.sub.tot] is equal to the sum of the associative values of all stimuli (e.g., stimuli A, B ... n) present during a trial and can be represented by:
[V.sub.tot] = [V.sub.A] + [V.sub.B] +...+ [V.sub.n] (2)
In the present simulation, [V.sub.tot] for CS trials (AX) is composed of the associative value of the visual component (A) of the CS, plus the temporal component ([X.sub.A]) of the CS, plus background value; [V.sub.tot] for cover stimulus trials (BX) is composed of the associative value of the visual component (B) of the cover stimulus, plus the temporal component ([X.sub.B]) of the cover stimulus, plus background value.
The learning rate parameter values selected for these simulations were those used by Rescorla and Wagner (1972). The value of [alpha], for both the temporal and nontemporal CS components, was set to .50; the [alpha]-value for the background was equal to .10. The learning rate parameter value for the US, [beta], was .10 and the [beta]-value during nonreinforced trials was .05. Estimations of V from this model were calculated with a [lambda] value equal to 100. Simulations calculated changes in value by using a time base period equal to the shortest stimulus (CS or cover cue) duration for a given experimental condition. Estimations of V for the nontemporal and temporal elements were made within each simulation. The simulations were conducted by using the same session and trial structure to which birds were exposed in a given experiment. Mean estimates of V were obtained for each component from three independent simulations.
Results of the simulations are shown in Table 6. As shown by the column labeled AX, total associative value for the CS is nearly equivalent for the equal and unequal stimulus-duration conditions of each experimental phase. Therefore the stimulus-duration-equality effect is not attributable to differences in total CS strength in the equal versus unequal stimulus-duration conditions. Similarly, the stimulus-duration-equality effect cannot be attributed to associative strength acquired by the nontemporal element of the CS (A). This can be seen by comparing the predicted values of A for the equal and unequal conditions of Experiment 1 and the equal and unequal pairs of groups with identical CS or cover-stimulus durations in Experiment 2 (e.g., CS4-Cov4 versus CS12-Cov4; CS12-Cov12 versus CS12-Cov36). For each comparison, the value of A was lower for the equal-duration versus the unequal-duration conditions. Only the predicted associative values of the temporal element ([X.sub.A]) were consistent with the stimulus-duration-equality effect. For all comparisons, the value of [X.sub.A] was higher for the equal versus the unequal stimulus-duration condition. Therefore, according to the model, the stimulus-duration-equality effect is attributable to differentially higher associative strength acquired by the shared temporal element under the equal-stimulus-duration conditions.
The simulation does not anticipate differences in rate of acquisition or in postacquisition response strength among the CS4-Cov4, CS12-Cov12, and CS36-Cov36 groups; however, the data indicate an inverse relation between response tendency and CS duration among these groups (Figures 4 and 5, Tables 2-5). As previously mentioned, Scalar Expectancy Theory also fails to predict this effect (Gibbon & Balsam, 1981). As level of responding in Pavlovian preparations has been shown to vary with amount or rate of food delivery (Gibbon et al., 1977; Mustaca, Gabelli, Papini, & Balsam, 1991; Ploog, 2001; but see Gibbon, Farrell, Locurto, Duncan, & Terrace, 1980), it is plausible that the effect of increased CS duration is mediated by the lengthening of programmed cycle duration and concomitant decrease in absolute reinforcer rate across groups. The effects of trial spacing may reflect arousal or activation phenomena in conditioning preparations (Killeen, Hanson, & Osborne, 1978; Tomie, Silberman, Williams, & Pohorecky, 2002). Accordingly, the present findings are consistent with separable roles of relative trial spacing as supporting associatively based performance, and absolute trial spacing as affecting performance via arousal mechanisms.
In order to test the roles of associative value and reinforcer rate in an account of performance in Experiment 2, each of the dependent measures (proportion, rate) for individual pigeons (N = 30) was subjected to a multiple regression analysis with associative value, V, and reinforcer rate as predictors. The analysis was based on data from the last two blocks of the training phase (Figures 3 & 5). For responding to the CS, the model yielded significant positive beta weights for each predictor with proportion (multiple [R.sup.2]=.57) and with rate (multiple [R.sup.2]=.72) measures. For responding to the cover stimulus, all beta weights were positive, but none was significant. For Experiment 1, reinforcer rate was equivalent in the two conditions; for data from the last two blocks, the correlation of proportion with V was significant for the CS and cover cue, r = .54 and .55, respectively, N = 15, p < .05, but rate did not correlate with V for either stimulus. The correlational analysis supports the common elements model for responding to the CS in both studies. Although responding to the cover cue conformed to the model in Experiment 1, there were no significant correlations with V for that cue in Experiment 2. It is possible that substantial pretraining with the cover cue in Experiment 2 could have exerted a strong primacy effect upon responding to that cue (see e.g., Hemmes, Brown, Jakubow, & Cabeza de Vaca, 1997).
Interpretation of the stimulus-duration-equality effect in terms of control by the common temporal element represents a departure from conventional applications of the Rescorla-Wagner model, in that behavior control is attributed to the associative value of a single element of the CS compound, as opposed to the sum of values for all available elements. The present interpretation is also novel in assigning associative value to a temporal element; nonetheless, the stimulus properties of CS duration have been amply demonstrated (Blaisdell, Denniston, & Miller, 1998; Brown et al., 1993; Cole, Barnet, & Miller, 1995; Ohyama, Gibbon, Deich, & Balsam, 1999). A complication in this analysis is the fact that temporal information is not available until the end of a conditioning trial. This implies that temporal cues alone cannot account for the differences in the anticipatory responding observed between conditions in the present study. However, it may be supposed that learned associations between the visual element of the CS and CS duration mediated the effect of temporal cues upon level of conditioned responding. Within-compound associations have been well documented in Pavlovian conditioning (e.g., Heth, 1985; Rescorla & Durlach, 1981). In the present study acquired associations between the visual cue and duration permit associative value accruing to duration cues to be signaled by the visual component of the cue on each trial. An implication of this account is that the role of the visual component is primarily one of signaling the prevailing temporal cue, rather than serving as an excitor.
While the common-elements model was able to account for the present results, its predictive power is vulnerable to violation of its assumptions, including the assignment of values to the free parameters and the assumption that perceived CS duration is equal to the trial duration. However, the model fares better than plausible empirical accounts that appeal to simpler explanations. An examination of differences among conditions in the intertrial interval duration, or rate of US presentation during the CS, during the cover cue, or the difference between those values (i.e., contrast effects) does not yield an account of the observed differences in performance. It appears that a mechanism that focuses on the role of CS duration as a cue may have more explanatory power than one that appeals to the role of CS duration as a determinant of excitatory strength. In any case, whether a common-elements account will prove to have merit, the present results join those of other recent lines of research indicating that a complete theoretical account of Pavlovian conditioning will need to incorporate the role of temporal factors.
Table 1 Duration and Frequency of Events in Experiments 1 and 2* Group CS time Cover time Stimulus-free time Session time Experiment 1 CS15-Cov15 15X18=270 15X51=765 15X51=765 1800 CS15-Cov5 15X18=270 5X51=255 (15X51)+(10X51)=1275 1800 Experiment 2 CS4-Cov4 4X20=80 4X46=184 4X184=736 1000 CS12-Cov4 12X20=240 4X46=184 (12X184)+(46X8)=257 6 3000 CS12-Cov12 12X20=240 12X46=552 12X184=2208 3000 CS12-Cov12D 12X20=240 12X46=552 12X184=2208 3000 CS12-Cov36 12X20=240 36X46=1656 (12X184)-(24X46)=1104 3000 CS36-Cov36 36X20=720 36X46=1656 36X184=6624 9000 *In cells with a product, the first and second entries are duration and number of the event, respectively. All time values are in seconds. Table 2 Analysis of Mean Proportion of CS Trials with a Response, Experiment 2, Sessions 61-84, Values of H (df = 1) for Each Pairwise Comparison Among All Groups Groups Group CS12- CS12- CS12- CS12- CS36- Cov4 Cov12 Cov12D Cov36 Cov36 CS4-Cov4 6.81** .27 2.45 6.81** 5.77* CS12-Cov4 -- 5.77* 3.93* .04 2.45 CS12-Cov12 -- -- .53 6.81** 3.93* CS12-Cov12D -- -- -- 5.77* .53 CS12-Cov36 -- -- -- -- 4.81* * p < .05. ** p < .01. Table 3 Analysis of Mean Rate of Responding During CS Trials, Experiment 2, Sessions 61-84, Values of H (df = 1) for Each Pairwise Comparison Among All Groups Groups Group CS12- CS12- CS12- CS12- CS36- Cov4 Cov12 Cov12D Cov36 Cov36 CS4-Cov4 6.81** 3.15 6.81** 6.81** 6.81** CS12-Cov4 -- 5.77* 3.93* .04 1.84 CS12-Cov12 -- -- .53 4.81* 4.81* CS12-Cov12D -- -- -- 3.93* 1.84 CS12-Cov36 -- -- -- -- 1.32 * p < .05. ** p < .01. Table 4 Analysis of Mean Proportion of Cover Trials with a Response, Experiment 2, Sessions 61-84, Values of H (df = 1) for Each Pairwise Comparison Among All Groups Groups Group CS12- CS12- CS12- CS12- CS36- Cov4 Cov12 Cov12D Cov36 Cov36 CS4-Cov4 1.84 .09 .09 4.81* 2.45 CS12-Cov4 -- 1.84 .09 1.84 .88 CS12-Cov12 -- -- 1.32 5.77* 3.93* CS12-Cov12D -- -- -- 1.32 .54 CS12-Cov36 -- -- -- -- .53 * p < .05. Table 5 Analysis of Mean Rate of Responding During Cover Trials, Experiment 2, Sessions 61-84, Values of H (df = 1) for Each Pairwise Comparison Among All Groups Groups Group CS12- CS12- CS12- CS12- CS36- Cov4 Cov12 Cov12D Cov36 Cov36 CS4-Cov4 1.84 .53 1.32 6.81** 4.84* CS12-Cov4 -- .53 .01 4.81* 3.17 CS12-Cov12 -- -- .53 6.81** 3.96* CS12-Cov12D -- -- -- 2.80 1.85 CS12-Cov36 -- -- -- -- .09 * p < .05. ** p < .01. Table 6 Values of Nontemporal CS Element, A, Nontemporal Cover Element, B, and Temporal Element, X, for Experiments 1 and 2 as Predicted by Rescorla-Wagner (1972) Model Stimulus A B [X.sub.A] [X.sub.B] AX BX Group Experiment 1 CS15-Cov5 16.3 50.0 16.3 50.0 32.6 100.0 CS15-Cov15 4.6 25.8 30.4 30.4 35.0 56.2 Experiment 2, Signaled Random Procedure CS4-Cov4 -9.1 32.3 23.2 23.2 14.1 55.5 CS12-Cov4 6.9 50.0 6.9 50.0 13.8 100.0 CS12-Cov12 -9.1 32.3 23.2 23.2 14.1 55.5 CS12-Cov12D -8.7 31.1 22.4 22.4 13.7 53.5 CS12-Cov36 6.6 12.7 6.6 12.7 13.2 25.4 CS36-Cov36 -9.1 32.3 23.2 23.2 14.1 55.5 Note. Column headings A, B refer to estimated associative strength of the nontemporal elements of the CS and cover cues respectively; [X.sub.A], [X.sub.B] refer to the estimated strength of the temporal elements associated with cues A, B; and AX and BX refer to the combined strength for the CS and cover cues, respectively.
BLAISDELL, A. P., DENNISTON, J. C., & MILLER, R. R. (1998). Temporal encoding as a determinant of overshadowing. Journal of Experimental Psychology: Animal Behavior Processes, 24, 72-83.
BROWN, B. L., HEMMES, N. S., CABEZA DE VACA, S., & PAGANO, C. (1993). Sign- and goal-tracking during delay and trace autoshaping in pigeons. Animal Learning and Behavior, 21, 360-368.
COLE, R. P., BARNET, R. C., & MILLER, R. R. (1995). Temporal encoding in trace conditioning. Animal Learning and Behavior, 23, 144-153.
COOPER, L. D., ARONSON, L., BALSAM, P. D., & GIBBON, J. (1990). Duration of signals for intertrial reinforcement and nonreinforcement in random control procedures. Journal of Experimental Psychology: Animal Behavior Processes, 16, 14-26.
DURLACH, P. J. (1982). Pavlovian learning and performance when a CS and US are uncorrelated. In M. Commons, R. J. Herrnstein, & A. R. Wagner (Eds.), Quantitative analyses of behavior, Volume III: Acquisition (pp. 173-193). Cambridge, MA: Ballinger.
DURLACH, P. J. (1983). Effect of signaling intertrial unconditioned stimuli in autoshaping. Journal of Experimental Psychology: Animal Behavior Processes, 9, 374-389.
GIBBON, J., BALDOCK, M. D., LOCURTO, C., GOLD, L., & TERRACE, H. S. (1977). Trial and intertrial durations in autoshaping. Journal of Experimental Psychology: Animal Behavior Processes, 3, 264-284.
GIBBON, J., & BALSAM, P. (1981). Spreading association in time. In C. M. Locurto, H. S. Terrace, & J. Gibbon (Eds.), Autoshaping and conditioning theory (pp. 219-253). New York: Academic Press.
GIBBON, J., FARRELL, L., LOCURTO, C. M., DUNCAN, H., & TERRACE, H. S. (1980). Partial reinforcement in autoshaping with pigeons. Animal Learning and Behavior, 8, 45-59.
GODDARD, M. J., & JENKINS, H. M. (1987). Effect of signaling extra unconditioned stimuli on autoshaping. Animal Learning & Behavior, 15, 40-46.
GUNTHER, L. M., & MILLER, R. R. (2000). Prevention of the degraded-contingency effect by signalling training trials. Quarterly Journal of Experimental Psychology, 53B, 97-119.
HEMMES, N. S., BROWN, B. L., JAKUBOW, J. J., & CABEZA DE VACA, S. (1997). Determinants of response recovery in extinction following response elimination. Learning and Motivation, 28, 542-557.
HETH, C. D. (1985). Within-compound associations of taste and temperature. Learning and Motivation, 16, 413-422.
INMAN, R. L. (1974). Use of a t-statistic as an approximation of the exact distribution of the Wilcoxon Signed Ranks Test statistic. Communications in Statistics, 3, 795-806.
JENKINS, H. M., BARNES, R. A., & BARRERA, F. J. (1981). Why autoshaping depends on trial spacing. In C. M. Locurto, H. S. Terrace, & J. Gibbon (Eds.), Autoshaping and conditioning theory (pp. 255-284). New York: Academic Press.
KILLEEN, P. R., HANSON, S. J., OSBORNE, S. R. (1978). Arousal: Its genesis and manifestation as response rate. Psychological Review, 85, 571-581.
LATTAL, K. M. (1999). Trial and intertrial durations in Pavlovian conditioning: Issues of learning and performance. Journal of Experimental Psychology: Animal Behavior Processes, 25, 433-450.
MILLER, R. R., & MATZEL, L. D. (1988). The comparator hypothesis: A response rule for the expression of associations. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 22, pp. 51-92). San Diego, CA: Academic Press.
MUSTACA, A. E., GABELLI, F., PAPINI, M. R., & BALSAM, P. D. (1991). The effects of varying the interreinforcement interval on appetitive contextual conditioning. Animal Learning & Behavior, 19, 125-138.
OHYAMA, T., GIBBON, J., DEICH, J. D., & BALSAM, P. D. (1999). Temporal control during maintenance and extinction of conditioned keypecking in ring doves. Animal Learning and Behavior, 27, 89-98.
RAMSEY, P. P., & RAMSEY, P. H. (1990). Simple tests of nonnormality in small samples. Journal of Quality Technology, 22, 299-309.
PLOOG, B. O. (2001). Net amount of food affects autoshaped response rate, response latency, and gape amplitude in pigeons. Learning & Motivation, 32, 383-400.
RESCORLA, R. A. (1976). Stimulus generalization: Some predictions from a model of Pavlovian conditioning. Journal of Experimental Psychology: Animal Behavior Processes, 2, 88-96.
RESCORLA, R. A., & DURLACH, P. J. (1981). Within-event learning in Pavlovian conditioning. In N. E. Spear & R. R. Miller (Eds.), Information processing in animals: Memory mechanisms (pp. 81-111). Hillsdale, NJ: Erlbaum.
RESCORLA, R. A., & WAGNER, A. R. (1972). In A. Black & W. F. Prokasy (Eds.), Classical Conditioning II: Current research and theory (pp. 64-99). New York: Appleton-Century-Crofts.
TOMIE, A., SILBERMAN, Y., WILLIAMS, K., & POHORECKY, L. A. (2002). Pavlovian autoshaping procedures increase plasma corticosterone levels in rats. Pharmacology, Biochemistry, and Behavior, 72, 507-513.
WILLIAMS, B. A. (1994). Contingency theory and the effect of the duration of signals for noncontingent reinforcement. Psychonomic Bulletin and Review, 1, 111-114.
JAMES J. JAKUBOW, BRUCE L. BROWN, and NANCY S. HEMMES
The Graduate Center of the City University of New York and Queens College of the City University of New York
This research was supported in part by PCS-CUNY Grant #669568 to N. S. Hemmes and B. L. Brown. James J. Jakubow is now affiliated with Florida Atlantic University.
Correspondence should be addressed to Nancy Hemmes, Department of Psychology, Queens College-CUNY, Flushing NY 11367. (E-mail: email@example.com).
|Printer friendly Cite/link Email Feedback|
|Author:||Jakubow, James J.; Brown, Bruce L.; Hemmes, Nancy S.|
|Publication:||The Psychological Record|
|Date:||Mar 22, 2004|
|Previous Article:||Probability of equivalence formation: familiar stimuli and training sequence.|
|Next Article:||Does the temporal placement of food-pellet reinforcement alter induction when rats respond on a three-component multiple schedule?|