Response Deprivation and Instrumental ... - Semantic Scholar
Response
Deprivation
and
Instrumental
in the Controlled-Amount JAMES ALLISON
Performance
Paradigm
AND WILLIAM
TIMBERLAKE
The response deprivation analysis. previously applied to a paradigm frequently used in free operant experiments, was applied successfully to another paradigm frequently used in discrete trials experiments. Each trial ended when the rat performed I lick at an empty tube (E). and either 10 or 100 licks at a second tube containing either saccharin (S) or water (W). Baseline trials were begun by exposing both tubes. Contingency trials required I instrumental E-lick for access to the second tube. Rate of instrumental responding. relative to baseline. was directly proportional to the value required if the subject were to perform the contingent licks at their baseline rates. The results also confirmed the predicted functional relations between the absolute rate of instrumental responding and the number of contingent S-licks. the number of contingent W-licks, and hours of water deprivation. Critical implications for the concept of instrumental reinforcement were discussed.
This paper shows how two disparate paradigms for investigating instrumental responding can be unified by viewing instrumental responding as a means of tracking the baseline rate of the contingent response. One of these, the controlled-time paradigm, is frequently used in free operant experiments. The experimenter terminates each trial (synonymous with “session”) when a particular time has elapsed, all trials have the same duration, and the amount of responding is left to the subject. Discrete trials experiments frequently use the second type. the controlled-amount paradigm. Here the experimenter terminates each trial when a particular amount of responding has elapsed, all trials allow the same amount of responding, and the duration of the trial is left to the subject. Each of these paradigms provides a rate measure of responding for the trial as a whole, broadly defined as amount of responding/set. Response rate generally refers to frequency/set, but in the present analysis, amount may refer to any unit that can be used to specify a response This research was supported by Grant GB-36209 from the National Science Foundation We thank Frank Restle and James Dinsmoor for their comments on an early version of the manuscript, and Lance Trexler and Lee Hoyman for their assistance in the laboratory. Requests for reprints should be sent to James Allison, Department of Psychology. Indiana University, Bloomington. IN 47401. I’2 Copyright All
rights
#) uf
1975 reproduction
hy
Academic in
Presr, any
form
Inc. reserved.
Printed
in the
United
Statea
DEPRIVATION
AND
PERFORMANCE
123
requirement. In the controlled-time paradigm, the subject varies response rate from one trial to the next by varying the amount of responding. Rate is directly related to amount. In the controlled-amount paradigm, the subject varies response rate from one trial to the next by varying the latency of the response. Rate is inversely related to latency. It has been hypothesized that instrumental performance in the controlled-time paradigm depends on response deprivation (Timberlake & Allison, 1974). A contingency is said to deprive the subject of the contingent response when the following is true: If the subject were to perform the instrumental response at only its baseline rate, the subject would necessarily perform the contingent response at less than its baseline IXtC Evidence from the controlled-time paradigm suggests that response deprivation is a necessary and sufficient antecedent for instrumental performance. That is, the subject will perform an instrumental response above its baseline rate if, and only if, the contingency deprives the subject of the contingent response (Allison & Timberlake, 1974; Eisenberger, Karpman & Trattner, 1967; Premack, 1965; Timberlake & Allison, 1974). The next few paragraphs extend previous theoretical work by showing how response deprivation can be identified as an antecedent condition in the controlled-amount paradigm. In this type of experiment, response deprivation depends partly on the temporal patterning of the two behaviors during the baseline trial and partly on the terms enforced during the contingency trial. To demonstrate these dependencies, it is convenient to consider an experiment in which each trial is ended when the rat has made 1 lick at an empty tube (1 E-lick), and 10 licks at a saccharin tube (10 S-licks). The baseline trial is begun by presenting both tubes. Amount of responding is controlled by removing the empty tube when I E-lick occurs, and the saccharin tube when the 10th S-lick occurs. Figure I presents a hypothetical baseline pattern in which the rat happens to complete the 10 S-licks before the 1 E-lick (Pattern 1, Fig. 1). In a contingency trial, the subject must complete the instrumental response before it begins the contingent response. For example. if the experimenter designates 1 E-lick as the instrumental response, the contingency trial is begun by presenting the empty tube alone. The saccharin tube does not appear until 1 E-lick has occurred. At that time the empt:y tube is removed, and the saccharin tube is presented, remaining available until the 10th S-lick occurs. In the context of Pattern I, the contingency just described would deprive the subject of the contingent response. This statement can be verified by noting that if the rat were to complete the instrumental E-lick at the baseline latency shown in Pattern I, it would necessarily make
174
,ALLISON
AND
‘I’IMBERLAKE
6
SEC
FIG. the empty
first,
I. Six 10th.
hypothetical and
baseline 100th
lick
patterns at a saccharin
(S I. S IO, and tube:
E signifies
S 100 the
signify latency
the of
latencies
of
licking
an
tube).
each af the 10 S-licks later than it did in baseline. so each contingent S-lick would occur at a rate lower than its baseline rate. The rat could perform each S-lick at its baseline rate only by completing the E-lick earlier than it did during the baseline trial. If response deprivation is a sufficient antecedent condition for instrumental performance. as appears to be true in controlled-time experiments. this contingency should decrease the latency of the E-lick relative to the baseline latency. On the other hand, if the rat typically behaved in the baseline trial as shown in Pattern 7 (see Fig. I). this very same contingency would not deprive the rat of the contingent response. Were the rat to perform the instrumental E-lick at the baseline latency shown in Pattern 2, each of the 10 contingent S-licks could be performed at its baseline latency. If response deprivation is a necessary antecedent condition for instrumental performance, as appears to be true in controlled-time experiments, this contingency should not decrease the latency of the E-lick relative to the baseline latency. Pattern 3 (see Fig. 1) would also satisfy the response deprivation condition. If the subject performed the instrumental E-lick at the baseline latency, it would necessarily perform the first contingent S-lick later than the baseline latency. However, one or more of the remaining S-licks could occur at their baseline latencies. The general case for the controlled-amount paradigm follows immedi-
DEPRIVATION
AND
PERFORMANCE
125
ately from the discussion of Patterns 1,1, and 3. A contingency deprives the subject of the contingent response if, and only if, in the baseline trial the subject begins the contingent response before completing the instrumental response. Figure 1 includes three additional baseline patterns. of the same general type as Pattern 1, which might conceivably occur if the experimenter allowed 100 S-licks rather than 10. These patterns were selected for comparison with Pattern I because of their theoretical implications for functional relations between the absolute rate of instrumental responding and magnitude of reward. Recall that the present analysis views instrumental responding as a means of tracking the baseline rate of the contingent response. It follows that Pattern 6 is the only one of the three loo-lick patterns that would require a shorter latency of the instrumental E-lick, hence a higher rate of instrumental responding than the IO-lick Pattern 1. Pattern 5 would require a lower rate of instrumental responding than Pattern I. Pattern 4 would require the same rate as Pattern 1. In ~controlled-time experiments which employ fixed-ratio schedules. the rate of instrumental responding appears to decrease as magnitude of reward increases (Collier. 1977). The present analysis explains this functional relationship by noting that if each food reward is relatively large, the subject can track the baseline rate of eating by performing relatively few instrumental lever presses (see Timberlake & Allison, 1974). In controlled-amount experiments, the opposite functional relationship is generally obtained. That is, the rate of instrumental responding increases as magnitude of reward increases (see Hall, 1966, pp. 18 I - 183). The present analysis explains this type of finding by assuming that baseline trials would reveal behavioral patterns like Patterns I and 6, rather than Patterns 1 and 4, or 1 and 5. The crucial distinguishing feature of Patterns 1 and 6 is that the subject allowed 100 S-licks begins licking saccharin sooner than the subject allowed only IO S-licks. If such a result were obtained in baseline, then the analysis implies that the subject rewarded with 100 contingent S-licks would perform an instrumental E-lick sooner than the subject rewarded with IO S-licks. The rate of the instrumental E-lick would therefore increase with magnitude of reward. Oulr theoretical analysis of the controlled-amount paradigm leads to several conclusions which can now be summarized in general form. A contingency should produce a latency of instrumental responding less than the baseline latency if, in the baseline trial, the subject initiates the contingent response before completing the instrumental response (response deprivation). The absolute latency of instrumental responding produced by such a contingency should be an increasing function of the
126
ALLISON
AND
TIMBERLAKE
baseline latency of initiating the contingent response. If the contingency does not satisfy the response deprivation condition, then the latency of instrumental responding produced by the contingency should not differ from the baseline latency. EXPERIMENT
1
This explanation of the assumed functional relations between instrumental responding and magnitude of reward was tested by performing the necessary baseline measurements. Such measurements have not previously been reported in the literature. Next, the assumed functional relations were checked by shifting each subject from the baseline condition to the contingency condition. This second step was considered essential for two reasons. First. the controlled-amount literature does contain some exceptions to the rule that rate of instrumental responding increases with magnitude of reward. Second. previous investigations have employed responses, response requirements. and procedures which are not exactly parallel to those used here. Methocls Subjects. The subjects were 23 male albino rats obtained from Harlan Industries (Cumberland, Indiana). approximately 80 days old at the beginning of the experiment. Appurcrtus. The test enclosure was a box approximately 25 cm square and 18.5 cm high. Three walls, the ceiling, and the grid floor were metal: the fourth wall was a hinged Plexiglas door. Two l.S-cm holes in the rear wall were centered 3.5 cm above the floor and were 8 cm apart. The left hole gave access to an empty metaf drinking tube. and the right hole to an identical tube filled with .4% saccharin solution. The tip of each tube was located in the center of the access hole. and recessed about 7 mm behind the inside surface of the wall; the rat could easily contact the tip by extending the tongue through the access hole. Each hole could be opened or closed independently by means of a thin metal flag interposed between the tip of the tube and the outer surface of the wall. The two flags were raised (opening a hole) and lowered (closing a hole) by electric motors. Licks were monitored by drinkometer circuits grounded to the grid floor. Experimental events were controlled by solid-state circuitry and a film programmer located in an adjoining room. Electromechanical counters used as . I -set clocks recorded the latency of the E-lick, the first S-lick, and the last S-lick. Procechre. Subjects were housed two to a cage. Prior to any experimental treatment, half of the cages were assigned randomly to the 100~lick condition. and half to the IO-lick condition. One subject died
DEPRIVATION
AND
PERFORMANCE
127
early in the experiment, leaving 10 subjects in the IO-lick group, and 1 1 in the lOO-lick group. The experiment was divided into four successive phases: familiarization, precontingency baseline, contingency, and postcontingency baseline. D.uing the first part of the familiarization phase the subjects had continuous access to food and water in the home cage. On the first day, each subject, together with its cage mate, spent 5 min in the apparatus with the holes closed. This procedure was repeated on the second day with both holes open, and the rats stayed in the apparatus until some licks at each tube were observed. Subjects were run alone on all remaining trials. Each subject received five familiarization trials, one per day, under the baseline condition to which it had been assigned. Each of these trials ended when the rat made the prescribed number of licks (1 E-lick, and 10 or 100 S-licks), or when 1 hr elapsed. Subjects that failed to lick in two or more of these trials (three in the IO-lick group, and five in the 100~lick group) received 5-7 additional trials under acute water deprivation ranging from 16 to 48 hr until they made the prescribed number of licks within 1 hr. In the last six days of this phase the rats were adapted to a I-hr watering schedule which was maintained throughout the rest of the experiment. For the remainder of the experiment the rats were run at 23-23 hr of water deprivation, one trial per day, 7 days per wk. Prior to each trial, the experimenter placed the rat in the apparatus, closed the acrylic door, and entered the adjoining room where the control system was located. The trial was started approximately 30 set after placement. Baseline trials were started by opening both holes simultaneously. The left hole stayed open until the rat made one lick at the empty tube, and the right hole stayed open until the rat made the prescribed number of licks at the saccharin tube. Contingency trials differed in that (a) they were started by opening the left hole alone, and (b) one lick at the empty tube closed the left hole and simultaneously opened the right hole. Each subject received 18 precontingency baseline trials, 27 contingency trials, and 12 postcontingency baseline trials. Results and Discussion Latency data were reduced by calculating trial-block medians for each subject. Because some of the resulting distributions were markedly skewed, statistical inference depended largely on nonparametric Wilcoxon, Mann-Whitney, and Friedman tests (Siegel, 1956), and group medians were used as measures of central tendency. Although theory predicted the direction of several effects, all directional tests reported here are conservative I-tailed tests. The rejection region was .05. The latency of the first S-lick. the last S-lick, and the E-lick are plotted
128
ALLISON
AND
TIMBERLAKE
30 EASE
CONTINGENCY
BASE
c
Oc.. S-13
14-m
IS-21
43-4S
W-S?
TRIAL
FIG. 2. Latency of the E-lick (open circles) and the first and last S-licks (closed circles) for the 1O-lick group.
BASE
I r
: fn
CONTINGENCY
0
E-LICK
.
S-LICK
IO-
TRIAL
3. Latency of the E-lick (open circles) and the first and last S-licks (closed circles) for the 100~lick group. FIG.
DEPRIVATION
AND
PERFORMANCE
129
in Fig. 2 for the IO-lick group and in Fig. 3 for the 100~lick group. Inspection of trial-by-trial plots indicated that these measures were reasonably stable over the last 10 trials of the precontingency baseline phase, so the data from this phase are presented in the form of two 5-trial blocks. The remaining data are plotted in 3-trial blocks. Baseline “recovery” was evaluated by comparing the last precontingency block with the last postcontingency block (Trials 14- 18 vs Trials 5.5-57 in Figs. 2 and 3). The IO-lick group returned to the baseline latency of the E-lick. as did the loo-lick group. However, the latency of both the first and last S-licks showed a significant decrease in both the IO-lick group and the loo-lick group. Given the experimental design used here, it is not possible to identify the source of these baseline shifts in saccharin licking. The possible sources include the contingency training which intervened between the two baseline blocks, age, and the amount of prior experience with the general experimental environment, with saccharin licking, and with the I-hr watering schedule. In the present context, the more important question is whether the two sets of baseline data lead to the same conclusions in terms of the response deprivation analysis. Assuming that thirsty rats perform instrumentally for saccharin, the response deprivation analysis requires that the first S-lick would occur significantly sooner than the E-lick under baseline conditions. Both sets of baseline data showed that both groups satisfied this response deprivation condition. In the last precontingency block the condition was satisfied by all of the lo-lick subjects and 9/ 11 100~lick subjects. In the last postcontingency block the condition was satisfied by 9/10 of the IO-lick subjects and 9/l 1 of the loo-lick subjects. Assuming that the absolute rate of instrumental responding increases with magnitude of reward in the controlled-amount paradigm, the response deprivation analysis predicted that the loo-lick group would perform the first S-lick sooner than the 1O-lick group under baseline conditions. Both sets of baseline data supported this prediction. The difference was significant in both the last precontingency block and the last postcontingency block. In summary both sets of baseline data showed that each gronp satisfied the response deprivation condition, performing the first S-lick before the E-lick, and that the lOO-lick group performed the first S-lick before the IO-lick group. Given these baseline results, the analysis makes two straightforward predictions for the contingency phase. First, both groups should perform the E-lick sooner than they did in baseline. Second, the loo-lick group should perform the E-lick before the IO-lick group; the absolute rate of instrumental responding should be an increasing function of the magnitude of reward.
130
ALLISON
AND
TIMBERLAKE
The results supported both of these predictions. In the first block ot contingency trials the groups were indistinguishable. but in the remaining blocks the IOO-lick group performed the E-lick sooner than the IO-lick group. Statistical analysis of the final block of contingency trials showed that the difference was reliable. Comparisons between this bloch and the final baseline blocks showed that 70/31 subjects performed the E-lick sooner than they did in precontingency baseline, and 16/I! I sooner than in postcontingency baseline (7/ IO in the I O-lick group. and 9/ I I in the 1OO-lick group). The results also showed that the effects of the contingency can not be understood by supposing that the E-lick was nothing more than a direct substitute (Allison & Timberlake, 1974) for the S-licks. This substitution hypothesis implies that at the very outset of contingency training the latency of the E-lick would reproduce the baseline latency of the first S-lick. Because the groups differed significantly in the baseline latency of the first S-lick, the substitution hypothesis would predict a corresponding difference in the latency of the E-lick at the outset of contingency training. Contrary to this prediction, there was no significant difference in the first block of contingency trials. Our measurement of the baseline latency of the first S-lick may raise a question that would not ordinarily arise in experiments on instrumental performance. The question is whether the asymptotic latency of the instrumental E-lick was short enough to qualify as instrumental performance, in view of another response whose latency was generally shorter. This other response was the first S-lick as measured under baseline conditions. The question probably would not arise had the instrumental E-lick reached an asymptote well below the baseline latency of the first S-lick. One answer is that convention defines instrumental performance in terms of a baseline measure of the instrumental response, and not in terms of a baseline measure of the contingent response. The latter measure has no obvious rationale as a criterion for instrumental performance, and we know of no precedent for its use in that capacity. It should also be mentioned that the response deprivation analysis offers some reason to doubt that the subject will complete the instrumental requirement much before the baseline latency of initiating the contingent response. The theoretical function of the instrumental response is to permit the subject to begin the contingent response at a time which is near, but not before, the baseline latency of initiation. The response deprivation analysis thus implies an a priori limit on asymptotic instrumental performance (Timberlake & Allison. 1974). The preceding analyses focus on adjustments in the latency of the E-lick as a means of tracking the baseline rate of saccharin licking.
DEPRIVATION
“t-
w” VJ
AND
131
PERFORMANCE
.A A
100
V’ t
e--
-* l
ot --I~
i-l
1 1 j ,
19-21
FIG.
IO-lick
4. Time between the instrumental group (broken line) and the IOO-lick
43-45
E-lick group
and the (continuous
first
contingent line).
S-lick
for
the
Saccharin licking itself revealed a similar but independent adjustment. Upon performing the E-lick, the rat then performed the first S-lick progressively faster as training continued. Figure 4 shows this adjustment by plotting the time between the instrumental E-lick and the first contingent S-lick across the nine blocks of contingency trials. All 31 subjects showed a decrease in this measure between the first block and the last. Had the rats not made this adjustment, a higher asymptotic rate of instrumental responding would have been required in tracking the baseline rate of the contingent response. As Fig. 4 suggests, the 100~lick group made this adjustment more quickly than the IO-lick group. The groups did not differ significantly in the first block of contingency trials or the last five, but did differ significantly in the second block, the third. and the fourth. There was one additional adjustment in saccharin licking which might have emerged in contingency training. Suppose that on a particular trial the rat performed the first contingent S-lick later than the baseline latency. The consequence would be a tracking error. However, the rat could then hold the overall size of the error to a minimum by performing the remaining S-licks earlier than usual. This possibility was examined by several analyses which focused on the time between the first S-lick and the last, but there was no conclusive evidence that this type of adjustment occurred.
133
AI.I.ISON
AND
TIMBERLAKE
above can be understood as adaptlbe I‘he ad,lustments dibcu~ed response\ which the subject performs in tracking a specific quantity. the baseline rate of the contingent response. Note in particular that there is no need to attribute reinforcing properties to the contingent response in explaining the fact that the contingency decreased the latency of the E-lick relative to baseline. That is, it is not necessary to assume that contingent S-licks automatically strengthened the tendency to perform the instrumental E-lick. and thus decreased the latency of the E-lick. The decrease is admittedly compatible with the concept of reinforcement, but this should not be allowed to obscure the fact that the results can be explained in some detail without a reinforcement concept. It follows that previous results of this type, generally explained in terms of a hypothetical strengthening of the instrumental response. might be explained empirically, in terms of response deprivation. had the experimenter performed the necessary baseline measurements. As we have just pointed out. the observed variations in the latency of the E-lick between baseline and contingency can be explained without assuming that contingent S-licks reinforced the instrumental E-lick. Yet. most reinforcement theorists would insist that for the thirsty rat. saccharin-flavored water ought to have reinforcing properties. It was therefore considered important to examine sequential features of the behavior for evidence of such properties. An ES response pattern is one in which the E-lick was followed by the S-licks (see Pattern 2. Fig. I ). Another pattern that could have occurred in this experiment is SE (see Pattern I, Fig. I). The one remaining pattern is SES (one or more S-licks, followed by the E-lick, followed in turn by one or more S-licks: see Pattern 3. Fig. I). If S-licks automatically strengthen a preceding E-lick. then any occurrence of the ES pattern should have increased the probability of the E-lick. It follows that if an ES pattern occurred on a particular trial. the ES pattern should have tended to recur on the next trial. The data did not support this expectation. The transition matrices in Table I show the relative frequency with which each pattern was repeated or shifted on the next trial during the precontingency baseline phase ( I8 trials). and the postcontingency baseline phase (I 3 trials). All sub.jects in a group were pooled in computing these matrices. Table I reveals a pronounced tendency to shift from ES to some other pattern, contrary to the supposition that the E-lick was reinforced by consequent S-licks. Every ES row shows that the relative frequency of repeating ES was less than .5. The only pattern that tended to recur was the SE pattern. Every SE row shows that the relative frequency of repeating SE was greater than .S. Patterns displayed by individual subjects confirmed these impressions. A sub.ject was classified as a repeater of a particular pattern if it repeated
DEPRIVATION
AND
133
PERFORMANCE
TABLE 1 TRANSITION MATRICES Precontingency
Postcontingency ~~-___-~~~
Trial II + I pattern
Trial ,I pattern
ES
SE
SES
IO-lick
ES SE SES
.44 .29 .37
.47 .62 .44
.08 .08 .I9
loo-lick
ES SE SES
.26 .30 .30
.62 .68 .25 ~~~~
ES SE SES
.36 .30 .33
.54 .66 .33
Group
_~-Combined
Relative frequency on Trial n ~~ ~~~--
.~~. -
Trial )I + 1 pattern
Relative frequency on Trial II
ES
SE
.35 .56 .09 ~~-
.30 .25 .oo
.70 .73 1.00
.I1 .02 .45 ~..~~
.28 .61 .I1
.30 .I8 .oo
.67 .82 1.00
.04 .oo .oo
22 .77 .Ol
.I0 .05 .33
.31 .59 .I0
.30 .22 .oo
.68 .78 1.00
.02 .Ol .oo
.25 .74 .Ol
.-~
SES .OO .Ol .oo ~~~~ ~__
_~-.
.27 .72 .Ol ~~ ~_
the pattern more frequently than it shifted to some other pattern on the next trial. All 2 I subjects had an opportunity to repeat ES and SE during the precontingency baseline phase. Only six of these were ES repeaters. but 17 were SE repeaters. In the postcontingency baseline phase, 70 subjects had an opportunity to repeat ES, but only two of these were ES repeaters. All 2 I had an opportunity to repeat SE, and 18 were SE repeaters. Perhaps the individual ES trial did reinforce the E-lick, but the effect was too small to be detected. This objection would not apply to the first postcontingency baseline trial. This was a trial which immediately followed 27 consecutive ES trials in which S-licks followed the E-lick in close temporal contiguity (see Fig. 4). If these 27 trials had any cumulative reinforcement effect upon the E-lick, it should have been manifest during the first postcontingency trial. Specifically, a relatively large proportion of subjects should have repeated the ES pattern during this trial. The data did not support this expectation. Figure 5 plots the relative frequency of the ES pattern across all phases of the experiment in 3-trial blocks, and includes a separate plot for the first postcontingency trial. As Fig. 5 indicates, contingency training had no effect on the relative frequency of the ES pattern. Among the IO-lick group, this measure showed no significant variation across the IO blocks of baseline trials plotted in Fig. 5. The same result was obtained among the IOO-lick group. Between the last block of precontingency trials and the first post-
ALLISON
BASE
AND
TIMBERLAKE
BASE
CONTINGENCY
.8
.6
FIG. 5. Mean and the IOO-lick
relative frequency of the ES pattern group (open circles).
for the IO-lick
group
(closed
circles)
contingency trial, only 4/ 10 subjects in the IO-lick group showed an increase in the relative frequency of the ES pattern. Three decreased, and three showed no change. Only 31 I I subjects in the IOO-lick group showed an increase. Six decreased, and two showed no change. In summary, the results presented in Table I and Fig. 5 offer no support for the notion that the E-lick was reinforced by consequent S-licks. EXPERIMENT
2
As mentioned before, in the controlled-amount paradigm instrumental performance generally increases with magnitude of reward. Here we report an exception, examine its implications for the response deprivation analysis, and extend the analysis to the functional relationship between thirst and instrumental performance. Experiment 7 was conducted in essentially the same way as Expt I, with three noteworthy changes. First, water (W) was used instead of saccharin solution. Second, on alternate days the subjects were tested shortly before or shortly after their regular watering in the home cage. Third, baselines were not measured until after the contingency phase. Subjects and apparatus. The subjects were eight male albino rats obtained from Harlan Industries, approximately 90 days old at the beginning of the experiment. The apparatus was the same as that used in Expt
DEPRIVATION
AND
PERFORMANCE
135
I, except that the right-hand tube contained water rather than saccharin solution. The measures recorded were the latency of the E-lick, the first W-lick, and the last W-lick. Procedure. For 10 days prior to the first contingency trial, the subjects were adapted to a 30-min watering schedule which was maintained throughout the experiment; food was always available in the home cage. On each of the two days before the first contingency trial, the subjects were familiarized with the experimental environment by placing them in the apparatus for 5 min with neither tube accessible. Contingency and baseline trials were conducted as in Expt 1; the responses were 1 E-lick and IO or 100 W-licks. Half of the subjects were assigned randomly to the IO-lick group and half to the loo-lick group. On any particular day, the subject received its daily trial during the 60-min interval before or after watering in the home cage. Beforewatering trials alternated with after-watering trials, and the two possible trial sequences were counterbalanced within each group. Each subject received 54 contingency trials and 12 postcontingency baseline trials. Since it was anticipated that the subjects might take an excessively long time to perform the prescribed number of licks during the trials conducted after watering, a 300-set cutoff was used on each trial. Specifically, a clock began timing the 300-set cutoff interval at the beginning of the trial. Upon the occurrence of the E-lick and the first W-lick, the clock was reset to zero and then continued to run. The trial ended when the subject made the prescribed number of licks or when the cutoff timer registered 300 sec. On cutoff trials, missing latencies were recorded as 300 set, the latency of the E-lick plus 300 set, or the latency of the first W-lick plus 300 set, depending on which event marked the beginning of the cutoff interval. Results
and Discussion
The’oretical analyses were based on Table 2, which presents the latency of the first W-lick, the last W-lick, and the E-lick for the baseline phase. In the present context, the response deprivation condition is satisfied if the latency of the first W-lick is significantly less than that of the E-lick under baseline conditions. The group satisfied this criterion both when tested before watering and when tested after watering. The prediction for the contingency phase is that the group should perform the E-lick sooner than in baseline, both before and after watering. Additional predictions were obtained by analyzing the baseline latency of the first W-lick. When tested before watering, the subjects performed the first W-lick significantly sooner than when tested after watering. This effect is consistent with results reported by Belles (1962); when rats were given free access to food and water after going without
136
ALLISON
AND
TIMBERLAKE
TABLE BASFI
INP
2
LA-I~NCI~S.
EXPT
?
Response
Group
Test
IO
time
First
Laht
W-lick
W-lick
Before After
100
2.7
4.4
8.8
(7.2)
s3.s
80.2
83.9
(56.8)
34.3
(14.4)
Before After
” Parenthetical
entries
are
latencies
E-lick”
2.‘)
22.5
43.0
60.3
obtained
in the
last
16.5.15 (48.
two
blocks
I)
of contingency
trials.
water for various intervals, the latency of the drinking response decreased as prior water deprivation increased. The number of W-licks had no significant effect on the latency of the first W-lick-unlike Expt I, in which the latency of the first S-lick was inversely related to the number of S-licks. There was no significant interaction between time of testing and number of W-licks. These results concerning the baseline latency of the first W-lick have 140 -
c
AFTER
s
10 60 100
u
BEFORE 20 :
OL‘ I-6
FIG. 100.lick circles),
6. Latency group Expt
Z-30 TRIAL
of the
(continuous 2.
instrumental lines),
E-lick before
watering
for
the
IO-lick
(closed
group circles)
(broken and
after
lines) watering
and
the (open
DEPRIVATION
AND
137
PERFORMANCE
three clear implications for the contingency phase. First, subjects rewarded with 10 W-licks should perform the instrumental E-lick at approximately the same latency as those rewarded with 100 W-licks (see Patterns I and 4, Fig. I)-unlike Expt 1, in which the latency of the instrumental E-lick was inversely related to the number of S-licks. Second, the subjects should perform the instrumental E-lick significantly sooner when tested before watering than when tested after watering. Third, the effect -just mentioned should not depend on magnitude of reward. The results of the contingency phase confirmed all of these implications. Figure 6 shows the latency of the instrumental E-lick in 3-trial blocks, with magnitude of reward and time of testing as parameters. Statistical analyses of the last two blocks combined (see the parenthetical entries in Table 2) showed that the subjects performed the E-lick significantly sooner than in baseline, both before and after watering. The latency of the instrumental E-lick did not vary significantly with magnitude of reward, and the subjects performed the instrumental E-lick significantly sooner when tested before watering than when tested after watering. There was no significant interaction. It was noted in Expt I that contingent S-licks had no discernible effect on the tendency to perform the E-lick first. The relative frequency of the ES response pattern was relatively low, both before and after the 27 consecutive ES trials of contingency training. The corresponding comparison was not available in Expt 2, because baseline trials were not conducted before the 54 contingency trials. However, it should be noted that during the first baseline trial only one subject performed the E-lick first, and the other seven performed the W-licks first. Similar results were obtained across the remaining baseline trials. Out of a total of 96 opport.unities (eight subjects, each receiving I2 baseline trials), the E-lick occurred first only 13 times, and one subject was responsible for 10 of these occasions. Among the remaining opportunities W-licks occurred first 77 times. and six times the trial timed out before any licking occurred. These results provide little if any indication that the E-lick was reinforced by consequent W-licks. Amount
of Change
Shown
by Individual
Subjects
Baseline patterns of the sort that appear in Fig. 1 show how much change a contingency will require, if the subject is to begin the contingent response at the baseline latency. If the contingency deprives the subject of the contingent response, then the contingency requires a decrease in the latency of the instrumental response relative to baseline. In the context of Expt 1, the decrease required was approximately the baseline latency of the E-lick, minus the baseline latency of the first
138
ALLISON
AND
TIMBERLAKE
S-lick (see Pattern 1, and Patterns 3 through 6).’ On the other hand, if the contingency does not deprive the subject of the contingent response. then the contingency requires no change relative to the baseline latency (see Pattern 2). A stringent test of the analysis is whether individual subjects showed changes in the latency of the E-lick which were directly related to the amount of change required for the subject to begin the contingent response at the baseline latency.? This aspect of the analysis was tested by pooling the 21 subjects of Expt I, and the eight of Expt 3. for a total of 29 independent observations. The data base for Expt I consisted of the final block of precontingency baseline trials, and the final contingency block. The eight observations from Expt 3 had the same data base as Table 3: before-watering data were averaged with after-watering data to obtain eight independent observations. The results clearly supported the analysis. Figure 7 plots the decrease shown by each of the 29 subjects against the decrease required if the subject were to begin the contingent response at its baseline latency. As Figure 7 suggests, the decrease obtained can be described as an increasing linear function of the decrease required. The regression line which appears in Fig. 7 was obtained by the least-squares method (Walker & Lev, 1969); its equation is Y = .96(X) - .13. The slope constant. .96, had a standard error of .062. Similar results were obtained using the final block of postcontingency baseline trials and the final contingency block as the data base for Expt 1 (see Fig. 8). A decrease in the latency of the instrumental response is generally attributed to reinforcement. However, the response deprivation analysis has two critical implications for this use of the reinforcement concept which are clearly illustrated by the linear regression equation. Results predicted by the response deprivation analysis fall into two major classes. In one of these classes the concept of instrumental reinforcement is untenable, as the results in this class provide no empirical support for the concept. Illustrative results are obtained by setting X = 0 in ’ This measure is a slight underestimate of the decrease required, because it omits a reaction-time parameter-the least possible interval between the completion of the instrumental response, and the beginning of the contingent response. The predictions derived here are independent of this parameter. so it is ignored for the time being. 2 This measure, the amount of change required, corresponds precisely to a measure which we have developed for controlled-time experiments (Timberlake & Allison. 1974). Both measures express the amount of response deprivation produced by a contingency schedule. In controlled-time experiments, contingencies generally produce an increase in the rate of the instrumental response. relative to baseline. which is directly related to the amount of response deprivation (Timberlake & Allison. 1974). The present analysis implies that the same relationship will hold in controlled-amount experiments.
DEPRIVATION
0
30
AND
60 DECREASE
60
139
PERFORMANCE
.
EXP.
0
EXP. 2
120 REWIRED
IS0
I
160
210
(SEC1
FIG. 7. Decrease in the latency of the E-lick, relative to baseline, as a function of the decrease required in Expt I (closed circles) and Expt 2 (open circles). The data base for Expt I consisted of the final block of precontingency baseline trials, and the final contingency block. The eight observations from Expt 2 had the same data base as Table 2; before-watering data were averaged with after-watering data to obtain eight independent observations.
the regression equation and evaluating Y. If the contingency did not deprive the subject of the contingent response (X = O), the subject would incur no response deprivation if it performed the instrumental response at the baseline latency. The contingency required no decrease in the latency of the instrumental response relative to baseline, and no decrease occurred. Illustrative results that belong to the second class are obtained by setting X > 0 in the regression equation and evaluating Y. If the contingency did deprive the subject of the contingent response (X > O), the subject would incur some response deprivation if it performed the instrumental response at the baseline latency. The contingency required a decrease, and a decrease was obtained. Results which fall into this second class can be explained in terms of reinforcement, but here the concept is not necessary. The reason is that both classes are predicted by an alternative concept, response deprivation. The conclusion is plain. In experiments which support the response
140
ALLISON
AND
TIMBERLAKE
.
EXP. I
c. EXP. 2
FIG. decrease
8. Decrease required
in the in
Expt
latency I (closed
Expt I consisted of the final block gency block. The eight observations before-watering data were averaged
of the
E-lick,
circles)
relative
and
Expt
to baseline, 2 (open
circles).
as a function The
data
of the base
for
of postcontingency baseline trials, and the final continfrom Expt 2 had the same data base as Table 2: with after-watering data to obtain eight independent
observations.
deprivution unnecessary
analysis, the concept or untenable. GENERAL
of
instrumentul
reinforcement
is
DISCUSSION
By viewing instrumental performance as a means of tracking the baseline rate of the contingent response, it is possible to explain a wide variety of functional relations observed in controlled-time experiments (see Timberlake & Allison, 1974, for a review). The present work shows that the same analysis can be applied successfully to controlled-amount experiments, and thereby unifies two familiar paradigms that have traditionally received separate theoretical treatments. The principal supportive evidence reported here can be summarized in the form of two explanatory propositions. One refers to the rate of the instrumental response during the contingency trial, relative to the baseline rate. Specifically, the contingency produces an increase in the rate of the instrumental response which is directly proportional to the increase required if the subject is to perform the contingent response at its baseline rate. The second applies to absolute rates produced by contingencies that deprive the subject of the contingent response. Specifically.
DEPRIVATION
AND
PERFORMANCE
141
the functional relation between the rate of the instrumental response and a particular independent variable parallels the relation between this variable and the baseline rate of initiating the contingent response. The present experiments identified three variables which supported this second proposition: hours of water deprivation, number of contingent W-licks, and number of contingent S-licks. It should be apparent that these propositions do not predict the subject’s behavior under baseline conditions. That is a problem for theories of thirst or incentive motivation. The intent of these propositions is to predict instrumental performance by comparing baseline behavior with the behavioral requirements of the contingency. In some cases the baseline behavior may be concordant with accepted principles of learning and motivation, as when the thirsty rat licks water sooner than does the less thirsty rat, licks a saccharin tube sooner than an empty tube, or licks a saccharin tube sooner when the incentive is 100 licks than when the incentive is only 10 licks. Discordant cases may demand further inquiry, as when 100 water licks appear to provide no more incentive motivation than 10 water licks. Since random assignment procedures can easily go awry when the number of subjects is small, this discordant finding from Expt 2 (four subjects per group) could have resulted from biased sampling. Demonstrable instances are fairly common. In one deliberate example, a group of rats rewarded with 10 food pellets in the goal box of a runway ran significantly faster than another group rewarded with only one pellet, but a preselected subgroup of the onepellet subjects ran just as fast as a preselected subgroup of the IO-pellet subjects (Allison, 1964; Atkinson, 1964, p. 285). A major implication of the response deprivation analysis is that results of this sort are best understood by removing the contingency, and observing the subjects’ behavior under baseline conditions. Unlike reinforcement theory, the present approach requires no postulation of a hypothetical process which is supposed to strengthen the instrumental response. It offers predictive explanations, rather than post hoc ‘explanations, for the variability shown by putative reinforcers in their effects upon rate of instrumental responding relative to baseline. Two clear examples of this variability have now been documented in both the controlled-amount paradigm, and the controlled-time paradigm (Allison & Timber-lake, 1974; Premack, 1965; Timberlake & Allison, 1974). The two examples are saccharin-licking and water-licking, which “reinforce” the instrumental response if the contingency deprives the subject of the contingent response, but somehow fail to “reinforce” the same instrumental response if the contingency does not deprive the sub,ject of the contingent response. We should temper our conclusions by noting that the supporting experiments from the controlled-amount para-
142
ALLISON
AND
TIMBERLAKE
digm used an instrumental response (licking an empty tube) which was similar topographically to the contingent response (licking a saccharin tube or a water tube). It remains to be seen whether the same results will be obtained for more disparate responses. However, it should also be noted that the supporting experiments from the controlled-time paradigm include some which used wheel running as the instrumental response (Premack, 1965; Timberlake & Allison. 1974). We suggest that it is time to consider carefully whether the concept of instrumental reinforcement is useful or even supportable in the prediction of instrumental performance, and that the alternative concept of response deprivation has many important advantages. REFERENCES ALLISON, J. Strength of preference for food. magnitude instrumental conditioning. Jo~rnul f$ Compurutir~e 1964.57,
of food reward. and
Physiological
and performance
in
Psychology.
2 17-213.
ALLISON, J., & TIMBERLAKE, W. Instrumental and contingent saccharin-licking in rats: Response deprivation and reinforcement. Learning and Motivation, 1974,5, 23 l-247. ATKINSON, J. W. An introduction to motivation. New York: Van Nostrand, 1964. BOLLES, R. C. The readiness to eat and drink: The effect of deprivation conditions. Journul of Compurutive und Physiological Psychology, 1962. 55, 230-234. COLLIER, G. Reinforcement magnitude in free feeding. Paper presented at the annual meeting of the Psychonomic Society, 1972. EISENBERGER, R., KARPMAN, M.. & TRATTNER, J. What is the necessary and sufficient condition for reinforcement in the contingency situation? Journul of Experimenful Psychology, 1967, 74, 342-350. HALL, J. F. The psyci~olog~ of learning. Philadelphia: Lippincott, 1966. PREMACK. D. Reinforcement theory. In D. Levine (Ed.), Nehrosku symposium on motivution. Lincoln: University of Nebraska Press. 1965. SIEGEL, S. Nonpurumetric stutistics ,fbr the hehuviorul scienc~es. New York: McGrawHill. 1956. TIMBERLAKE. W., & ALLISON, J. Response deprivation: An empirical approach to instrumental performance. Psychological Review.. 1974. 81, I46- 164. WALKER, H. M., & LEV, J. Elementury stu/isticcr/ methods. New York: Holt. Rinehart, & Winston, 1969. Received Revised
February 25. 1974 July 30, 1974
Deprivation
and
Instrumental
in the Controlled-Amount JAMES ALLISON
Performance
Paradigm
AND WILLIAM
TIMBERLAKE
The response deprivation analysis. previously applied to a paradigm frequently used in free operant experiments, was applied successfully to another paradigm frequently used in discrete trials experiments. Each trial ended when the rat performed I lick at an empty tube (E). and either 10 or 100 licks at a second tube containing either saccharin (S) or water (W). Baseline trials were begun by exposing both tubes. Contingency trials required I instrumental E-lick for access to the second tube. Rate of instrumental responding. relative to baseline. was directly proportional to the value required if the subject were to perform the contingent licks at their baseline rates. The results also confirmed the predicted functional relations between the absolute rate of instrumental responding and the number of contingent S-licks. the number of contingent W-licks, and hours of water deprivation. Critical implications for the concept of instrumental reinforcement were discussed.
This paper shows how two disparate paradigms for investigating instrumental responding can be unified by viewing instrumental responding as a means of tracking the baseline rate of the contingent response. One of these, the controlled-time paradigm, is frequently used in free operant experiments. The experimenter terminates each trial (synonymous with “session”) when a particular time has elapsed, all trials have the same duration, and the amount of responding is left to the subject. Discrete trials experiments frequently use the second type. the controlled-amount paradigm. Here the experimenter terminates each trial when a particular amount of responding has elapsed, all trials allow the same amount of responding, and the duration of the trial is left to the subject. Each of these paradigms provides a rate measure of responding for the trial as a whole, broadly defined as amount of responding/set. Response rate generally refers to frequency/set, but in the present analysis, amount may refer to any unit that can be used to specify a response This research was supported by Grant GB-36209 from the National Science Foundation We thank Frank Restle and James Dinsmoor for their comments on an early version of the manuscript, and Lance Trexler and Lee Hoyman for their assistance in the laboratory. Requests for reprints should be sent to James Allison, Department of Psychology. Indiana University, Bloomington. IN 47401. I’2 Copyright All
rights
#) uf
1975 reproduction
hy
Academic in
Presr, any
form
Inc. reserved.
Printed
in the
United
Statea
DEPRIVATION
AND
PERFORMANCE
123
requirement. In the controlled-time paradigm, the subject varies response rate from one trial to the next by varying the amount of responding. Rate is directly related to amount. In the controlled-amount paradigm, the subject varies response rate from one trial to the next by varying the latency of the response. Rate is inversely related to latency. It has been hypothesized that instrumental performance in the controlled-time paradigm depends on response deprivation (Timberlake & Allison, 1974). A contingency is said to deprive the subject of the contingent response when the following is true: If the subject were to perform the instrumental response at only its baseline rate, the subject would necessarily perform the contingent response at less than its baseline IXtC Evidence from the controlled-time paradigm suggests that response deprivation is a necessary and sufficient antecedent for instrumental performance. That is, the subject will perform an instrumental response above its baseline rate if, and only if, the contingency deprives the subject of the contingent response (Allison & Timberlake, 1974; Eisenberger, Karpman & Trattner, 1967; Premack, 1965; Timberlake & Allison, 1974). The next few paragraphs extend previous theoretical work by showing how response deprivation can be identified as an antecedent condition in the controlled-amount paradigm. In this type of experiment, response deprivation depends partly on the temporal patterning of the two behaviors during the baseline trial and partly on the terms enforced during the contingency trial. To demonstrate these dependencies, it is convenient to consider an experiment in which each trial is ended when the rat has made 1 lick at an empty tube (1 E-lick), and 10 licks at a saccharin tube (10 S-licks). The baseline trial is begun by presenting both tubes. Amount of responding is controlled by removing the empty tube when I E-lick occurs, and the saccharin tube when the 10th S-lick occurs. Figure I presents a hypothetical baseline pattern in which the rat happens to complete the 10 S-licks before the 1 E-lick (Pattern 1, Fig. 1). In a contingency trial, the subject must complete the instrumental response before it begins the contingent response. For example. if the experimenter designates 1 E-lick as the instrumental response, the contingency trial is begun by presenting the empty tube alone. The saccharin tube does not appear until 1 E-lick has occurred. At that time the empt:y tube is removed, and the saccharin tube is presented, remaining available until the 10th S-lick occurs. In the context of Pattern I, the contingency just described would deprive the subject of the contingent response. This statement can be verified by noting that if the rat were to complete the instrumental E-lick at the baseline latency shown in Pattern I, it would necessarily make
174
,ALLISON
AND
‘I’IMBERLAKE
6
SEC
FIG. the empty
first,
I. Six 10th.
hypothetical and
baseline 100th
lick
patterns at a saccharin
(S I. S IO, and tube:
E signifies
S 100 the
signify latency
the of
latencies
of
licking
an
tube).
each af the 10 S-licks later than it did in baseline. so each contingent S-lick would occur at a rate lower than its baseline rate. The rat could perform each S-lick at its baseline rate only by completing the E-lick earlier than it did during the baseline trial. If response deprivation is a sufficient antecedent condition for instrumental performance. as appears to be true in controlled-time experiments. this contingency should decrease the latency of the E-lick relative to the baseline latency. On the other hand, if the rat typically behaved in the baseline trial as shown in Pattern 7 (see Fig. I). this very same contingency would not deprive the rat of the contingent response. Were the rat to perform the instrumental E-lick at the baseline latency shown in Pattern 2, each of the 10 contingent S-licks could be performed at its baseline latency. If response deprivation is a necessary antecedent condition for instrumental performance, as appears to be true in controlled-time experiments, this contingency should not decrease the latency of the E-lick relative to the baseline latency. Pattern 3 (see Fig. 1) would also satisfy the response deprivation condition. If the subject performed the instrumental E-lick at the baseline latency, it would necessarily perform the first contingent S-lick later than the baseline latency. However, one or more of the remaining S-licks could occur at their baseline latencies. The general case for the controlled-amount paradigm follows immedi-
DEPRIVATION
AND
PERFORMANCE
125
ately from the discussion of Patterns 1,1, and 3. A contingency deprives the subject of the contingent response if, and only if, in the baseline trial the subject begins the contingent response before completing the instrumental response. Figure 1 includes three additional baseline patterns. of the same general type as Pattern 1, which might conceivably occur if the experimenter allowed 100 S-licks rather than 10. These patterns were selected for comparison with Pattern I because of their theoretical implications for functional relations between the absolute rate of instrumental responding and magnitude of reward. Recall that the present analysis views instrumental responding as a means of tracking the baseline rate of the contingent response. It follows that Pattern 6 is the only one of the three loo-lick patterns that would require a shorter latency of the instrumental E-lick, hence a higher rate of instrumental responding than the IO-lick Pattern 1. Pattern 5 would require a lower rate of instrumental responding than Pattern I. Pattern 4 would require the same rate as Pattern 1. In ~controlled-time experiments which employ fixed-ratio schedules. the rate of instrumental responding appears to decrease as magnitude of reward increases (Collier. 1977). The present analysis explains this functional relationship by noting that if each food reward is relatively large, the subject can track the baseline rate of eating by performing relatively few instrumental lever presses (see Timberlake & Allison, 1974). In controlled-amount experiments, the opposite functional relationship is generally obtained. That is, the rate of instrumental responding increases as magnitude of reward increases (see Hall, 1966, pp. 18 I - 183). The present analysis explains this type of finding by assuming that baseline trials would reveal behavioral patterns like Patterns I and 6, rather than Patterns 1 and 4, or 1 and 5. The crucial distinguishing feature of Patterns 1 and 6 is that the subject allowed 100 S-licks begins licking saccharin sooner than the subject allowed only IO S-licks. If such a result were obtained in baseline, then the analysis implies that the subject rewarded with 100 contingent S-licks would perform an instrumental E-lick sooner than the subject rewarded with IO S-licks. The rate of the instrumental E-lick would therefore increase with magnitude of reward. Oulr theoretical analysis of the controlled-amount paradigm leads to several conclusions which can now be summarized in general form. A contingency should produce a latency of instrumental responding less than the baseline latency if, in the baseline trial, the subject initiates the contingent response before completing the instrumental response (response deprivation). The absolute latency of instrumental responding produced by such a contingency should be an increasing function of the
126
ALLISON
AND
TIMBERLAKE
baseline latency of initiating the contingent response. If the contingency does not satisfy the response deprivation condition, then the latency of instrumental responding produced by the contingency should not differ from the baseline latency. EXPERIMENT
1
This explanation of the assumed functional relations between instrumental responding and magnitude of reward was tested by performing the necessary baseline measurements. Such measurements have not previously been reported in the literature. Next, the assumed functional relations were checked by shifting each subject from the baseline condition to the contingency condition. This second step was considered essential for two reasons. First. the controlled-amount literature does contain some exceptions to the rule that rate of instrumental responding increases with magnitude of reward. Second. previous investigations have employed responses, response requirements. and procedures which are not exactly parallel to those used here. Methocls Subjects. The subjects were 23 male albino rats obtained from Harlan Industries (Cumberland, Indiana). approximately 80 days old at the beginning of the experiment. Appurcrtus. The test enclosure was a box approximately 25 cm square and 18.5 cm high. Three walls, the ceiling, and the grid floor were metal: the fourth wall was a hinged Plexiglas door. Two l.S-cm holes in the rear wall were centered 3.5 cm above the floor and were 8 cm apart. The left hole gave access to an empty metaf drinking tube. and the right hole to an identical tube filled with .4% saccharin solution. The tip of each tube was located in the center of the access hole. and recessed about 7 mm behind the inside surface of the wall; the rat could easily contact the tip by extending the tongue through the access hole. Each hole could be opened or closed independently by means of a thin metal flag interposed between the tip of the tube and the outer surface of the wall. The two flags were raised (opening a hole) and lowered (closing a hole) by electric motors. Licks were monitored by drinkometer circuits grounded to the grid floor. Experimental events were controlled by solid-state circuitry and a film programmer located in an adjoining room. Electromechanical counters used as . I -set clocks recorded the latency of the E-lick, the first S-lick, and the last S-lick. Procechre. Subjects were housed two to a cage. Prior to any experimental treatment, half of the cages were assigned randomly to the 100~lick condition. and half to the IO-lick condition. One subject died
DEPRIVATION
AND
PERFORMANCE
127
early in the experiment, leaving 10 subjects in the IO-lick group, and 1 1 in the lOO-lick group. The experiment was divided into four successive phases: familiarization, precontingency baseline, contingency, and postcontingency baseline. D.uing the first part of the familiarization phase the subjects had continuous access to food and water in the home cage. On the first day, each subject, together with its cage mate, spent 5 min in the apparatus with the holes closed. This procedure was repeated on the second day with both holes open, and the rats stayed in the apparatus until some licks at each tube were observed. Subjects were run alone on all remaining trials. Each subject received five familiarization trials, one per day, under the baseline condition to which it had been assigned. Each of these trials ended when the rat made the prescribed number of licks (1 E-lick, and 10 or 100 S-licks), or when 1 hr elapsed. Subjects that failed to lick in two or more of these trials (three in the IO-lick group, and five in the 100~lick group) received 5-7 additional trials under acute water deprivation ranging from 16 to 48 hr until they made the prescribed number of licks within 1 hr. In the last six days of this phase the rats were adapted to a I-hr watering schedule which was maintained throughout the rest of the experiment. For the remainder of the experiment the rats were run at 23-23 hr of water deprivation, one trial per day, 7 days per wk. Prior to each trial, the experimenter placed the rat in the apparatus, closed the acrylic door, and entered the adjoining room where the control system was located. The trial was started approximately 30 set after placement. Baseline trials were started by opening both holes simultaneously. The left hole stayed open until the rat made one lick at the empty tube, and the right hole stayed open until the rat made the prescribed number of licks at the saccharin tube. Contingency trials differed in that (a) they were started by opening the left hole alone, and (b) one lick at the empty tube closed the left hole and simultaneously opened the right hole. Each subject received 18 precontingency baseline trials, 27 contingency trials, and 12 postcontingency baseline trials. Results and Discussion Latency data were reduced by calculating trial-block medians for each subject. Because some of the resulting distributions were markedly skewed, statistical inference depended largely on nonparametric Wilcoxon, Mann-Whitney, and Friedman tests (Siegel, 1956), and group medians were used as measures of central tendency. Although theory predicted the direction of several effects, all directional tests reported here are conservative I-tailed tests. The rejection region was .05. The latency of the first S-lick. the last S-lick, and the E-lick are plotted
128
ALLISON
AND
TIMBERLAKE
30 EASE
CONTINGENCY
BASE
c
Oc.. S-13
14-m
IS-21
43-4S
W-S?
TRIAL
FIG. 2. Latency of the E-lick (open circles) and the first and last S-licks (closed circles) for the 1O-lick group.
BASE
I r
: fn
CONTINGENCY
0
E-LICK
.
S-LICK
IO-
TRIAL
3. Latency of the E-lick (open circles) and the first and last S-licks (closed circles) for the 100~lick group. FIG.
DEPRIVATION
AND
PERFORMANCE
129
in Fig. 2 for the IO-lick group and in Fig. 3 for the 100~lick group. Inspection of trial-by-trial plots indicated that these measures were reasonably stable over the last 10 trials of the precontingency baseline phase, so the data from this phase are presented in the form of two 5-trial blocks. The remaining data are plotted in 3-trial blocks. Baseline “recovery” was evaluated by comparing the last precontingency block with the last postcontingency block (Trials 14- 18 vs Trials 5.5-57 in Figs. 2 and 3). The IO-lick group returned to the baseline latency of the E-lick. as did the loo-lick group. However, the latency of both the first and last S-licks showed a significant decrease in both the IO-lick group and the loo-lick group. Given the experimental design used here, it is not possible to identify the source of these baseline shifts in saccharin licking. The possible sources include the contingency training which intervened between the two baseline blocks, age, and the amount of prior experience with the general experimental environment, with saccharin licking, and with the I-hr watering schedule. In the present context, the more important question is whether the two sets of baseline data lead to the same conclusions in terms of the response deprivation analysis. Assuming that thirsty rats perform instrumentally for saccharin, the response deprivation analysis requires that the first S-lick would occur significantly sooner than the E-lick under baseline conditions. Both sets of baseline data showed that both groups satisfied this response deprivation condition. In the last precontingency block the condition was satisfied by all of the lo-lick subjects and 9/ 11 100~lick subjects. In the last postcontingency block the condition was satisfied by 9/10 of the IO-lick subjects and 9/l 1 of the loo-lick subjects. Assuming that the absolute rate of instrumental responding increases with magnitude of reward in the controlled-amount paradigm, the response deprivation analysis predicted that the loo-lick group would perform the first S-lick sooner than the 1O-lick group under baseline conditions. Both sets of baseline data supported this prediction. The difference was significant in both the last precontingency block and the last postcontingency block. In summary both sets of baseline data showed that each gronp satisfied the response deprivation condition, performing the first S-lick before the E-lick, and that the lOO-lick group performed the first S-lick before the IO-lick group. Given these baseline results, the analysis makes two straightforward predictions for the contingency phase. First, both groups should perform the E-lick sooner than they did in baseline. Second, the loo-lick group should perform the E-lick before the IO-lick group; the absolute rate of instrumental responding should be an increasing function of the magnitude of reward.
130
ALLISON
AND
TIMBERLAKE
The results supported both of these predictions. In the first block ot contingency trials the groups were indistinguishable. but in the remaining blocks the IOO-lick group performed the E-lick sooner than the IO-lick group. Statistical analysis of the final block of contingency trials showed that the difference was reliable. Comparisons between this bloch and the final baseline blocks showed that 70/31 subjects performed the E-lick sooner than they did in precontingency baseline, and 16/I! I sooner than in postcontingency baseline (7/ IO in the I O-lick group. and 9/ I I in the 1OO-lick group). The results also showed that the effects of the contingency can not be understood by supposing that the E-lick was nothing more than a direct substitute (Allison & Timberlake, 1974) for the S-licks. This substitution hypothesis implies that at the very outset of contingency training the latency of the E-lick would reproduce the baseline latency of the first S-lick. Because the groups differed significantly in the baseline latency of the first S-lick, the substitution hypothesis would predict a corresponding difference in the latency of the E-lick at the outset of contingency training. Contrary to this prediction, there was no significant difference in the first block of contingency trials. Our measurement of the baseline latency of the first S-lick may raise a question that would not ordinarily arise in experiments on instrumental performance. The question is whether the asymptotic latency of the instrumental E-lick was short enough to qualify as instrumental performance, in view of another response whose latency was generally shorter. This other response was the first S-lick as measured under baseline conditions. The question probably would not arise had the instrumental E-lick reached an asymptote well below the baseline latency of the first S-lick. One answer is that convention defines instrumental performance in terms of a baseline measure of the instrumental response, and not in terms of a baseline measure of the contingent response. The latter measure has no obvious rationale as a criterion for instrumental performance, and we know of no precedent for its use in that capacity. It should also be mentioned that the response deprivation analysis offers some reason to doubt that the subject will complete the instrumental requirement much before the baseline latency of initiating the contingent response. The theoretical function of the instrumental response is to permit the subject to begin the contingent response at a time which is near, but not before, the baseline latency of initiation. The response deprivation analysis thus implies an a priori limit on asymptotic instrumental performance (Timberlake & Allison. 1974). The preceding analyses focus on adjustments in the latency of the E-lick as a means of tracking the baseline rate of saccharin licking.
DEPRIVATION
“t-
w” VJ
AND
131
PERFORMANCE
.A A
100
V’ t
e--
-* l
ot --I~
i-l
1 1 j ,
19-21
FIG.
IO-lick
4. Time between the instrumental group (broken line) and the IOO-lick
43-45
E-lick group
and the (continuous
first
contingent line).
S-lick
for
the
Saccharin licking itself revealed a similar but independent adjustment. Upon performing the E-lick, the rat then performed the first S-lick progressively faster as training continued. Figure 4 shows this adjustment by plotting the time between the instrumental E-lick and the first contingent S-lick across the nine blocks of contingency trials. All 31 subjects showed a decrease in this measure between the first block and the last. Had the rats not made this adjustment, a higher asymptotic rate of instrumental responding would have been required in tracking the baseline rate of the contingent response. As Fig. 4 suggests, the 100~lick group made this adjustment more quickly than the IO-lick group. The groups did not differ significantly in the first block of contingency trials or the last five, but did differ significantly in the second block, the third. and the fourth. There was one additional adjustment in saccharin licking which might have emerged in contingency training. Suppose that on a particular trial the rat performed the first contingent S-lick later than the baseline latency. The consequence would be a tracking error. However, the rat could then hold the overall size of the error to a minimum by performing the remaining S-licks earlier than usual. This possibility was examined by several analyses which focused on the time between the first S-lick and the last, but there was no conclusive evidence that this type of adjustment occurred.
133
AI.I.ISON
AND
TIMBERLAKE
above can be understood as adaptlbe I‘he ad,lustments dibcu~ed response\ which the subject performs in tracking a specific quantity. the baseline rate of the contingent response. Note in particular that there is no need to attribute reinforcing properties to the contingent response in explaining the fact that the contingency decreased the latency of the E-lick relative to baseline. That is, it is not necessary to assume that contingent S-licks automatically strengthened the tendency to perform the instrumental E-lick. and thus decreased the latency of the E-lick. The decrease is admittedly compatible with the concept of reinforcement, but this should not be allowed to obscure the fact that the results can be explained in some detail without a reinforcement concept. It follows that previous results of this type, generally explained in terms of a hypothetical strengthening of the instrumental response. might be explained empirically, in terms of response deprivation. had the experimenter performed the necessary baseline measurements. As we have just pointed out. the observed variations in the latency of the E-lick between baseline and contingency can be explained without assuming that contingent S-licks reinforced the instrumental E-lick. Yet. most reinforcement theorists would insist that for the thirsty rat. saccharin-flavored water ought to have reinforcing properties. It was therefore considered important to examine sequential features of the behavior for evidence of such properties. An ES response pattern is one in which the E-lick was followed by the S-licks (see Pattern 2. Fig. I ). Another pattern that could have occurred in this experiment is SE (see Pattern I, Fig. I). The one remaining pattern is SES (one or more S-licks, followed by the E-lick, followed in turn by one or more S-licks: see Pattern 3. Fig. I). If S-licks automatically strengthen a preceding E-lick. then any occurrence of the ES pattern should have increased the probability of the E-lick. It follows that if an ES pattern occurred on a particular trial. the ES pattern should have tended to recur on the next trial. The data did not support this expectation. The transition matrices in Table I show the relative frequency with which each pattern was repeated or shifted on the next trial during the precontingency baseline phase ( I8 trials). and the postcontingency baseline phase (I 3 trials). All sub.jects in a group were pooled in computing these matrices. Table I reveals a pronounced tendency to shift from ES to some other pattern, contrary to the supposition that the E-lick was reinforced by consequent S-licks. Every ES row shows that the relative frequency of repeating ES was less than .5. The only pattern that tended to recur was the SE pattern. Every SE row shows that the relative frequency of repeating SE was greater than .S. Patterns displayed by individual subjects confirmed these impressions. A sub.ject was classified as a repeater of a particular pattern if it repeated
DEPRIVATION
AND
133
PERFORMANCE
TABLE 1 TRANSITION MATRICES Precontingency
Postcontingency ~~-___-~~~
Trial II + I pattern
Trial ,I pattern
ES
SE
SES
IO-lick
ES SE SES
.44 .29 .37
.47 .62 .44
.08 .08 .I9
loo-lick
ES SE SES
.26 .30 .30
.62 .68 .25 ~~~~
ES SE SES
.36 .30 .33
.54 .66 .33
Group
_~-Combined
Relative frequency on Trial n ~~ ~~~--
.~~. -
Trial )I + 1 pattern
Relative frequency on Trial II
ES
SE
.35 .56 .09 ~~-
.30 .25 .oo
.70 .73 1.00
.I1 .02 .45 ~..~~
.28 .61 .I1
.30 .I8 .oo
.67 .82 1.00
.04 .oo .oo
22 .77 .Ol
.I0 .05 .33
.31 .59 .I0
.30 .22 .oo
.68 .78 1.00
.02 .Ol .oo
.25 .74 .Ol
.-~
SES .OO .Ol .oo ~~~~ ~__
_~-.
.27 .72 .Ol ~~ ~_
the pattern more frequently than it shifted to some other pattern on the next trial. All 2 I subjects had an opportunity to repeat ES and SE during the precontingency baseline phase. Only six of these were ES repeaters. but 17 were SE repeaters. In the postcontingency baseline phase, 70 subjects had an opportunity to repeat ES, but only two of these were ES repeaters. All 2 I had an opportunity to repeat SE, and 18 were SE repeaters. Perhaps the individual ES trial did reinforce the E-lick, but the effect was too small to be detected. This objection would not apply to the first postcontingency baseline trial. This was a trial which immediately followed 27 consecutive ES trials in which S-licks followed the E-lick in close temporal contiguity (see Fig. 4). If these 27 trials had any cumulative reinforcement effect upon the E-lick, it should have been manifest during the first postcontingency trial. Specifically, a relatively large proportion of subjects should have repeated the ES pattern during this trial. The data did not support this expectation. Figure 5 plots the relative frequency of the ES pattern across all phases of the experiment in 3-trial blocks, and includes a separate plot for the first postcontingency trial. As Fig. 5 indicates, contingency training had no effect on the relative frequency of the ES pattern. Among the IO-lick group, this measure showed no significant variation across the IO blocks of baseline trials plotted in Fig. 5. The same result was obtained among the IOO-lick group. Between the last block of precontingency trials and the first post-
ALLISON
BASE
AND
TIMBERLAKE
BASE
CONTINGENCY
.8
.6
FIG. 5. Mean and the IOO-lick
relative frequency of the ES pattern group (open circles).
for the IO-lick
group
(closed
circles)
contingency trial, only 4/ 10 subjects in the IO-lick group showed an increase in the relative frequency of the ES pattern. Three decreased, and three showed no change. Only 31 I I subjects in the IOO-lick group showed an increase. Six decreased, and two showed no change. In summary, the results presented in Table I and Fig. 5 offer no support for the notion that the E-lick was reinforced by consequent S-licks. EXPERIMENT
2
As mentioned before, in the controlled-amount paradigm instrumental performance generally increases with magnitude of reward. Here we report an exception, examine its implications for the response deprivation analysis, and extend the analysis to the functional relationship between thirst and instrumental performance. Experiment 7 was conducted in essentially the same way as Expt I, with three noteworthy changes. First, water (W) was used instead of saccharin solution. Second, on alternate days the subjects were tested shortly before or shortly after their regular watering in the home cage. Third, baselines were not measured until after the contingency phase. Subjects and apparatus. The subjects were eight male albino rats obtained from Harlan Industries, approximately 90 days old at the beginning of the experiment. The apparatus was the same as that used in Expt
DEPRIVATION
AND
PERFORMANCE
135
I, except that the right-hand tube contained water rather than saccharin solution. The measures recorded were the latency of the E-lick, the first W-lick, and the last W-lick. Procedure. For 10 days prior to the first contingency trial, the subjects were adapted to a 30-min watering schedule which was maintained throughout the experiment; food was always available in the home cage. On each of the two days before the first contingency trial, the subjects were familiarized with the experimental environment by placing them in the apparatus for 5 min with neither tube accessible. Contingency and baseline trials were conducted as in Expt 1; the responses were 1 E-lick and IO or 100 W-licks. Half of the subjects were assigned randomly to the IO-lick group and half to the loo-lick group. On any particular day, the subject received its daily trial during the 60-min interval before or after watering in the home cage. Beforewatering trials alternated with after-watering trials, and the two possible trial sequences were counterbalanced within each group. Each subject received 54 contingency trials and 12 postcontingency baseline trials. Since it was anticipated that the subjects might take an excessively long time to perform the prescribed number of licks during the trials conducted after watering, a 300-set cutoff was used on each trial. Specifically, a clock began timing the 300-set cutoff interval at the beginning of the trial. Upon the occurrence of the E-lick and the first W-lick, the clock was reset to zero and then continued to run. The trial ended when the subject made the prescribed number of licks or when the cutoff timer registered 300 sec. On cutoff trials, missing latencies were recorded as 300 set, the latency of the E-lick plus 300 set, or the latency of the first W-lick plus 300 set, depending on which event marked the beginning of the cutoff interval. Results
and Discussion
The’oretical analyses were based on Table 2, which presents the latency of the first W-lick, the last W-lick, and the E-lick for the baseline phase. In the present context, the response deprivation condition is satisfied if the latency of the first W-lick is significantly less than that of the E-lick under baseline conditions. The group satisfied this criterion both when tested before watering and when tested after watering. The prediction for the contingency phase is that the group should perform the E-lick sooner than in baseline, both before and after watering. Additional predictions were obtained by analyzing the baseline latency of the first W-lick. When tested before watering, the subjects performed the first W-lick significantly sooner than when tested after watering. This effect is consistent with results reported by Belles (1962); when rats were given free access to food and water after going without
136
ALLISON
AND
TIMBERLAKE
TABLE BASFI
INP
2
LA-I~NCI~S.
EXPT
?
Response
Group
Test
IO
time
First
Laht
W-lick
W-lick
Before After
100
2.7
4.4
8.8
(7.2)
s3.s
80.2
83.9
(56.8)
34.3
(14.4)
Before After
” Parenthetical
entries
are
latencies
E-lick”
2.‘)
22.5
43.0
60.3
obtained
in the
last
16.5.15 (48.
two
blocks
I)
of contingency
trials.
water for various intervals, the latency of the drinking response decreased as prior water deprivation increased. The number of W-licks had no significant effect on the latency of the first W-lick-unlike Expt I, in which the latency of the first S-lick was inversely related to the number of S-licks. There was no significant interaction between time of testing and number of W-licks. These results concerning the baseline latency of the first W-lick have 140 -
c
AFTER
s
10 60 100
u
BEFORE 20 :
OL‘ I-6
FIG. 100.lick circles),
6. Latency group Expt
Z-30 TRIAL
of the
(continuous 2.
instrumental lines),
E-lick before
watering
for
the
IO-lick
(closed
group circles)
(broken and
after
lines) watering
and
the (open
DEPRIVATION
AND
137
PERFORMANCE
three clear implications for the contingency phase. First, subjects rewarded with 10 W-licks should perform the instrumental E-lick at approximately the same latency as those rewarded with 100 W-licks (see Patterns I and 4, Fig. I)-unlike Expt 1, in which the latency of the instrumental E-lick was inversely related to the number of S-licks. Second, the subjects should perform the instrumental E-lick significantly sooner when tested before watering than when tested after watering. Third, the effect -just mentioned should not depend on magnitude of reward. The results of the contingency phase confirmed all of these implications. Figure 6 shows the latency of the instrumental E-lick in 3-trial blocks, with magnitude of reward and time of testing as parameters. Statistical analyses of the last two blocks combined (see the parenthetical entries in Table 2) showed that the subjects performed the E-lick significantly sooner than in baseline, both before and after watering. The latency of the instrumental E-lick did not vary significantly with magnitude of reward, and the subjects performed the instrumental E-lick significantly sooner when tested before watering than when tested after watering. There was no significant interaction. It was noted in Expt I that contingent S-licks had no discernible effect on the tendency to perform the E-lick first. The relative frequency of the ES response pattern was relatively low, both before and after the 27 consecutive ES trials of contingency training. The corresponding comparison was not available in Expt 2, because baseline trials were not conducted before the 54 contingency trials. However, it should be noted that during the first baseline trial only one subject performed the E-lick first, and the other seven performed the W-licks first. Similar results were obtained across the remaining baseline trials. Out of a total of 96 opport.unities (eight subjects, each receiving I2 baseline trials), the E-lick occurred first only 13 times, and one subject was responsible for 10 of these occasions. Among the remaining opportunities W-licks occurred first 77 times. and six times the trial timed out before any licking occurred. These results provide little if any indication that the E-lick was reinforced by consequent W-licks. Amount
of Change
Shown
by Individual
Subjects
Baseline patterns of the sort that appear in Fig. 1 show how much change a contingency will require, if the subject is to begin the contingent response at the baseline latency. If the contingency deprives the subject of the contingent response, then the contingency requires a decrease in the latency of the instrumental response relative to baseline. In the context of Expt 1, the decrease required was approximately the baseline latency of the E-lick, minus the baseline latency of the first
138
ALLISON
AND
TIMBERLAKE
S-lick (see Pattern 1, and Patterns 3 through 6).’ On the other hand, if the contingency does not deprive the subject of the contingent response. then the contingency requires no change relative to the baseline latency (see Pattern 2). A stringent test of the analysis is whether individual subjects showed changes in the latency of the E-lick which were directly related to the amount of change required for the subject to begin the contingent response at the baseline latency.? This aspect of the analysis was tested by pooling the 21 subjects of Expt I, and the eight of Expt 3. for a total of 29 independent observations. The data base for Expt I consisted of the final block of precontingency baseline trials, and the final contingency block. The eight observations from Expt 3 had the same data base as Table 3: before-watering data were averaged with after-watering data to obtain eight independent observations. The results clearly supported the analysis. Figure 7 plots the decrease shown by each of the 29 subjects against the decrease required if the subject were to begin the contingent response at its baseline latency. As Figure 7 suggests, the decrease obtained can be described as an increasing linear function of the decrease required. The regression line which appears in Fig. 7 was obtained by the least-squares method (Walker & Lev, 1969); its equation is Y = .96(X) - .13. The slope constant. .96, had a standard error of .062. Similar results were obtained using the final block of postcontingency baseline trials and the final contingency block as the data base for Expt 1 (see Fig. 8). A decrease in the latency of the instrumental response is generally attributed to reinforcement. However, the response deprivation analysis has two critical implications for this use of the reinforcement concept which are clearly illustrated by the linear regression equation. Results predicted by the response deprivation analysis fall into two major classes. In one of these classes the concept of instrumental reinforcement is untenable, as the results in this class provide no empirical support for the concept. Illustrative results are obtained by setting X = 0 in ’ This measure is a slight underestimate of the decrease required, because it omits a reaction-time parameter-the least possible interval between the completion of the instrumental response, and the beginning of the contingent response. The predictions derived here are independent of this parameter. so it is ignored for the time being. 2 This measure, the amount of change required, corresponds precisely to a measure which we have developed for controlled-time experiments (Timberlake & Allison. 1974). Both measures express the amount of response deprivation produced by a contingency schedule. In controlled-time experiments, contingencies generally produce an increase in the rate of the instrumental response. relative to baseline. which is directly related to the amount of response deprivation (Timberlake & Allison. 1974). The present analysis implies that the same relationship will hold in controlled-amount experiments.
DEPRIVATION
0
30
AND
60 DECREASE
60
139
PERFORMANCE
.
EXP.
0
EXP. 2
120 REWIRED
IS0
I
160
210
(SEC1
FIG. 7. Decrease in the latency of the E-lick, relative to baseline, as a function of the decrease required in Expt I (closed circles) and Expt 2 (open circles). The data base for Expt I consisted of the final block of precontingency baseline trials, and the final contingency block. The eight observations from Expt 2 had the same data base as Table 2; before-watering data were averaged with after-watering data to obtain eight independent observations.
the regression equation and evaluating Y. If the contingency did not deprive the subject of the contingent response (X = O), the subject would incur no response deprivation if it performed the instrumental response at the baseline latency. The contingency required no decrease in the latency of the instrumental response relative to baseline, and no decrease occurred. Illustrative results that belong to the second class are obtained by setting X > 0 in the regression equation and evaluating Y. If the contingency did deprive the subject of the contingent response (X > O), the subject would incur some response deprivation if it performed the instrumental response at the baseline latency. The contingency required a decrease, and a decrease was obtained. Results which fall into this second class can be explained in terms of reinforcement, but here the concept is not necessary. The reason is that both classes are predicted by an alternative concept, response deprivation. The conclusion is plain. In experiments which support the response
140
ALLISON
AND
TIMBERLAKE
.
EXP. I
c. EXP. 2
FIG. decrease
8. Decrease required
in the in
Expt
latency I (closed
Expt I consisted of the final block gency block. The eight observations before-watering data were averaged
of the
E-lick,
circles)
relative
and
Expt
to baseline, 2 (open
circles).
as a function The
data
of the base
for
of postcontingency baseline trials, and the final continfrom Expt 2 had the same data base as Table 2: with after-watering data to obtain eight independent
observations.
deprivution unnecessary
analysis, the concept or untenable. GENERAL
of
instrumentul
reinforcement
is
DISCUSSION
By viewing instrumental performance as a means of tracking the baseline rate of the contingent response, it is possible to explain a wide variety of functional relations observed in controlled-time experiments (see Timberlake & Allison, 1974, for a review). The present work shows that the same analysis can be applied successfully to controlled-amount experiments, and thereby unifies two familiar paradigms that have traditionally received separate theoretical treatments. The principal supportive evidence reported here can be summarized in the form of two explanatory propositions. One refers to the rate of the instrumental response during the contingency trial, relative to the baseline rate. Specifically, the contingency produces an increase in the rate of the instrumental response which is directly proportional to the increase required if the subject is to perform the contingent response at its baseline rate. The second applies to absolute rates produced by contingencies that deprive the subject of the contingent response. Specifically.
DEPRIVATION
AND
PERFORMANCE
141
the functional relation between the rate of the instrumental response and a particular independent variable parallels the relation between this variable and the baseline rate of initiating the contingent response. The present experiments identified three variables which supported this second proposition: hours of water deprivation, number of contingent W-licks, and number of contingent S-licks. It should be apparent that these propositions do not predict the subject’s behavior under baseline conditions. That is a problem for theories of thirst or incentive motivation. The intent of these propositions is to predict instrumental performance by comparing baseline behavior with the behavioral requirements of the contingency. In some cases the baseline behavior may be concordant with accepted principles of learning and motivation, as when the thirsty rat licks water sooner than does the less thirsty rat, licks a saccharin tube sooner than an empty tube, or licks a saccharin tube sooner when the incentive is 100 licks than when the incentive is only 10 licks. Discordant cases may demand further inquiry, as when 100 water licks appear to provide no more incentive motivation than 10 water licks. Since random assignment procedures can easily go awry when the number of subjects is small, this discordant finding from Expt 2 (four subjects per group) could have resulted from biased sampling. Demonstrable instances are fairly common. In one deliberate example, a group of rats rewarded with 10 food pellets in the goal box of a runway ran significantly faster than another group rewarded with only one pellet, but a preselected subgroup of the onepellet subjects ran just as fast as a preselected subgroup of the IO-pellet subjects (Allison, 1964; Atkinson, 1964, p. 285). A major implication of the response deprivation analysis is that results of this sort are best understood by removing the contingency, and observing the subjects’ behavior under baseline conditions. Unlike reinforcement theory, the present approach requires no postulation of a hypothetical process which is supposed to strengthen the instrumental response. It offers predictive explanations, rather than post hoc ‘explanations, for the variability shown by putative reinforcers in their effects upon rate of instrumental responding relative to baseline. Two clear examples of this variability have now been documented in both the controlled-amount paradigm, and the controlled-time paradigm (Allison & Timber-lake, 1974; Premack, 1965; Timberlake & Allison, 1974). The two examples are saccharin-licking and water-licking, which “reinforce” the instrumental response if the contingency deprives the subject of the contingent response, but somehow fail to “reinforce” the same instrumental response if the contingency does not deprive the sub,ject of the contingent response. We should temper our conclusions by noting that the supporting experiments from the controlled-amount para-
142
ALLISON
AND
TIMBERLAKE
digm used an instrumental response (licking an empty tube) which was similar topographically to the contingent response (licking a saccharin tube or a water tube). It remains to be seen whether the same results will be obtained for more disparate responses. However, it should also be noted that the supporting experiments from the controlled-time paradigm include some which used wheel running as the instrumental response (Premack, 1965; Timberlake & Allison. 1974). We suggest that it is time to consider carefully whether the concept of instrumental reinforcement is useful or even supportable in the prediction of instrumental performance, and that the alternative concept of response deprivation has many important advantages. REFERENCES ALLISON, J. Strength of preference for food. magnitude instrumental conditioning. Jo~rnul f$ Compurutir~e 1964.57,
of food reward. and
Physiological
and performance
in
Psychology.
2 17-213.
ALLISON, J., & TIMBERLAKE, W. Instrumental and contingent saccharin-licking in rats: Response deprivation and reinforcement. Learning and Motivation, 1974,5, 23 l-247. ATKINSON, J. W. An introduction to motivation. New York: Van Nostrand, 1964. BOLLES, R. C. The readiness to eat and drink: The effect of deprivation conditions. Journul of Compurutive und Physiological Psychology, 1962. 55, 230-234. COLLIER, G. Reinforcement magnitude in free feeding. Paper presented at the annual meeting of the Psychonomic Society, 1972. EISENBERGER, R., KARPMAN, M.. & TRATTNER, J. What is the necessary and sufficient condition for reinforcement in the contingency situation? Journul of Experimenful Psychology, 1967, 74, 342-350. HALL, J. F. The psyci~olog~ of learning. Philadelphia: Lippincott, 1966. PREMACK. D. Reinforcement theory. In D. Levine (Ed.), Nehrosku symposium on motivution. Lincoln: University of Nebraska Press. 1965. SIEGEL, S. Nonpurumetric stutistics ,fbr the hehuviorul scienc~es. New York: McGrawHill. 1956. TIMBERLAKE. W., & ALLISON, J. Response deprivation: An empirical approach to instrumental performance. Psychological Review.. 1974. 81, I46- 164. WALKER, H. M., & LEV, J. Elementury stu/isticcr/ methods. New York: Holt. Rinehart, & Winston, 1969. Received Revised
February 25. 1974 July 30, 1974
