Professional Documents
Culture Documents
Comportamwnto Operacional Contigencia Não Verbal
Comportamwnto Operacional Contigencia Não Verbal
FRODE SVARTDAL
I thank Magne A. Flaten for his comments on a previous version of this article. Requests
for reprints should be sent to Frode Svartdal, Department of Psychology, University of
Tromso, 9OtKl Tromso, Norway. E-mail address: frodes@mack.uit.no.
406
0023-9690191 $3.00
Copyright Q 1991 by Academic Press, Inc.
All rights of reproduction in any form reserved.
MODULATION OF RULE-GOVERNED BEHAVIOR 407
Zettle, Rosenfarb, & Korn, 1986; Hayes, Brownstein, Haas, & Greenway,
1986; Lippman & Meyer, 1967).
Reduced sensitivity to nonverbal contingencies should be expected in
situations where verbal stimuli easily interfere with arranged contingen-
cies. But when verbal control is minimized (Davey, 1989; Matthews,
Shimoff, Catania, & Sagvolden, 1977; Reber, 1976; Shimoff, Catania, &
Matthews, 1981; Wearden, 1988) or when behavior is inaccessible or too
complex to allow verbal control (e.g., Hefferline, Keenan, & Harford,
1958; Neuringer, 1986), nonverbal control may be facilitated. Further-
more, since verbal control of behavior requires attentional effort (e.g.,
Schneider & Fisk, 1983), verbal guidance of a behavioral property is likely
to prevent other concurrent behavioral properties from being salient to
the actor.
When describing their own behavior, actors generally prefer higher
levels of action identification (i.e., why the action is performed or what
its effects and implications are) to lower levels of action identification
(i.e., details of an action; how an action is done) (Vallacher & Wegner,
1987). If an action is guided by a higher level of identification, attention
to lower level attributes of that action is reduced (Vallacher & Wegner,
1987; Wegner, Vallacher, Macomber, Wood, & Arps, 1984). It may be
assumed, therefore, that low-level properties of behavior-i.e., subtle
response dimensions related to topography, temporal properties, and
force, the so-called modality attributes of action (e.g., Rescher, 1970)-
are likely to be nonsalient to actors when higher-level properties are
verbally controlled. The control of low-level properties of behavior in
such situations must therefore involve nonverbal mechanisms, of which
operant contingency control may be an example. Since a significant pro-
portion of human behavioral variability is related to modality attributes
of behavior, e.g., behavioral style, paralinguistic behavior, and emotional
modulation of action (Rosenfeld, 1982; Scherer, 1982; Scherer & Ekman,
1982), an enhanced understanding of this control is of interest.
The purpose of the present study was to investigate mechanisms that
may be effective in generating low-level behavior changes by arranged
nonverbal operant contingencies when the role of verbal control of that
behavior is minimized. The setup was essentially similar to that used by
Svartdal (1989), but with some important extensions. In that experiment,
nonverbal operant contingencies modified a nonsalient and low-level at-
tribute of responding (i.e., speed of pressing) when rules guided a higher-
level attribute of that behavior (i.e., the number of presses emitted).
Subjects were instructed to count brief auditory stimuli (“clicks”) in stim-
ulus presentation periods and to indicate the number counted by pressing
a key the corresponding number of times in subsequent response periods.
“Correct” answers resulted in “feedbacks.” Unknown to the subjects,
those feedbacks (i.e., reinforcers) depended on the speed of pressing
408 FRODE SVARTDAL
METHOD
Subjects
Twenty-four undergraduates participated in the experiment. Each was
paid 100 NOK (equivalent to $14). The subjects were randomly assigned
to different experimental groups prior to arrival.
Procedure
The experimental session consisted of 130 trials, each with a 6.1-s stim-
ulus-presentation period and a subsequent 6.4-s response period. Each
trial was separated by a pause of 1 s. The first 30 trials were “blind,”
during which the subjects were instructed to count and respond as ac-
curately as possible with no “feedback.” Speed of pressing during baseline
Trials 1l-30 permitted individual calibration of baseline speed. Following
the baseline trials, all subjects had 100 experimental trials. Transition
between baseline and experimental trials was signalled by orange lamps
for 2 s and 3 x l-s lOOO-Hz tones.
The number of clicks generated during each stimulus-presentation pe-
riod was randomly determined and varied from trial to trial (mean =
10.85; range 4-17). Since the clicks were generated by a software random
number generator starting from the same seed at the outset of the ex-
perimental session, the sequence of random clicks was identical for all
subjects. Because the clicks were of short duration and occasionally oc-
410 FRODE SVARTDAL
curred at brief intervals (min. = 170 ms), counting was sometimes quite
difficult. No information about the actual number of clicks presented was
provided to the subjects.
The computer recorded values from an S-bit analog-to-digital converter
with intervals of 85 ms (i.e., 75 samples per response period), corre-
sponding to the analog output from the tachometer. Pressing typically
stabilized by 1.5 s into the response period. Because pressing also would
end early in some response periods due to few presented clicks, speed
data were selected from the 1.7-s interval, starting 1.7 s into each response
period. The mean speed of pressing (i.e., averaged interresponse times
(IRTs) in the selected time window) in baseline Trials 11-30 was used
for individual identification of the reinforcement criterion, which was
automatically calculated by the computer following Trial 30.
Feedback stimuli depended on the averaged IRTs computed from the
analog values recorded in the selected time window in the response period.
Modification of the selected behavioral unit by the arranged feedback
stimuli would be required in order for the behavioral unit to be called
an operant and the feedback stimuli reinforcers (e.g., Catania, 1973). The
IRT durations were updated by the tachometer based on the rate of
pressing during the response period. The tachometer was reset if no
presses occurred within a l-s interval following a response. The integrator
time constant was permanently set at 500 ms.
Three experimental conditions were arranged. In the fast condition,
reinforcers were presented at the end of the 6.4-s response period if the
mean speed of pressing during the selected time window was above the
level identified during the subject’s baseline trials. If reinforcers were
obtained on at least five successive trials, the reinforcement requirement
was increased, corresponding to a rate change of 4.8 responses/min for
the following trials. In the slow condition, reinforcers were presented
according to the same regime, except that the speed had to be below the
baseline criterion and that the reinforcement criterion was decreased after
any five consecutively reinforced trials. In the random condition, rein-
forcers occurred independently of speed and independently of counting
at a rate of about 50% of the experimental trials.
Instructions and verbal control. Before the experiment started, each
subject was briefly interviewed to assess whether she/he had any knowl-
edge of the experiment. This examination was kept very general to prevent
any communication of information to the subject about the nature of the
experiment. The subject was then seated in front of the console and
received the following instructions:
This is an experiment on how people count random events. When this lamp [ex-
perimenter indicates] is red. you will hear a number of clicks here [experimenter
indicates the loudspeaker]. You must count the number of clicks. Then a brief
MODULATION OF RULE-GOVERNED BEHAVIOR 411
pause follows. After the pause this lamp [experimenter indicates] will turn green.
Your task is now-while the green lamp is on-to press this key [experimenter
indicates] the same number of times as the number of clicks you just heard. After
a while the green lamp will be turned off, and following a short break the sequence
will be repeated.
The number of clicks will vary randomly during the experiment. Your task is
simply to count the number of clicks you hear when the red lamp is on, and then
press the corresponding number of times when the green lamp is on.
The clicks are quite faint and may at times occur with short intervals. This may
sometimes make counting rather difficult. Therefore your answer will most probably
not always be correct.
In the first part of the experiment your answers will be “blind”; you will not
get any message as to whether the answer was correct or not. Just try to answer
as accurately as possible. After a while you will hear a signal [later demonstrated]
meaning that from now on, you will get feedback if your answer was correct. If
your answer was correct, you will get a message [later demonstrated]; if you don’t
answer correctly, you will get no message.
Here [experimenter indicates] you can monitor the number of correct answers
you have achieved. You may obtain a maximum of 60 correct answers. Very few
people can accomplish this. Try as well as you can to obtain as many correct
answers as possible.
Do you have any questions?
Following the instructions, any questions were answered and a short dem-
onstration of the procedure was given. Then the experiment was started.
After the completion of the experiment, each subject was checked for
insight into the experimental manipulation by use of a questionnaire and
an interview. This check involved three separate criteria. First, if the
subject complied with the instructions, counting would by necessity have
been her/his primary preoccupation during the experimental session, mak-
ing it unlikely that other explicit strategies had been applied in the sit-
uation. Second, each subject was asked about strategies applied, both
with respect to counting and to responding. Finally, the subject was ex-
plicitly queried whether any strategy related to speed-that is, faster or
slower pressing, or rhythmicity-had been used. Based on these criteria,
a subject was classified as an “insight” subject if speed in any way was
reported or implied as relevant to responding.
After the completion of the session, the subject was given a general
description of the experiment. This explanation was kept very brief and
uninformative, as the subject was promised a more comprehensive ex-
planation later. Finally, the subject was asked not to tell others about
the nature of the experimental task.
Analysis of data. Speed data were sampled from the selected 1.7-s time
window (i.e., Samples 21-40) of every response period. Because the
general speed levels differed among subjects, individual data were trans-
formed by dividing each observed speed value by that value -t mean
baseline value, giving the formula d, = d,/(d, + baseline). Thus, a
412 FRODE SVARTDAL
TABLE 1
Number of Criterion Changes, Last Trial vs Baseline, and Mean Speed Changes,
Transformed Values
1s - 4* ,521
2s - 2+ .488
3s ,498
4s (T Ig:) (.349)
5s - 6’ .43s
6S - 2* ,429
7s - 1* ,525
23s - 1* ,478
9F + 4* ,531
10 F + 3* 525
11 F + 3* 510
12 F + 7* 543
13 F + 4* 521
14 F + 3* 503
15 F + lo* ,584
16 F + 5’ 609
17 R - ,511
18 R - .522
19 R - ,503
20 R - ,531
21 R - ,507
22 R - ,554
23 R - ,478
24 R - 571
Note. Trials 91-130 relative to baseline; non-adapting subjects in italics; insight subject
in parentheses. S = Slow; F = Fast; R = Random.
A FAST
l luNoavl
0.45 - I-l_LAm~-l_l v SLOW
816212 3 4 5 6 7 6 9 10
FIG. 1. Mean group speed changes relative to baseline (Fast, Slow, and Random groups).
Data are averaged over blocks of 10 trials, and one nonadapting subject in the Slow group
(Subject 7) is removed from the data.
Early trials
J , , , , , , , , , , , iz
7 6 9 10 11 12 19 14 15 16 I?
cncklevel
Late trials
'I 6 9 10 11 12 13 14 15 16 17
clldcbal
FIG. 2. Mean group speed changes on different click levels/trial, first two experimental
trials/click level (a), and last four experimental trials/click level (b). See text for explanation.
416 FRODE SVARTDAL
the mean of the two first experimental trials on each click level (mean
Trial Number 47; click levels 16 and 17 were excluded in this figure, since
click level 16 had only one trial (Trial Number 70) and click level 17 had
only four trials (Trial Numbers 57, 60, 69, and 74). Contrast analyses of
the final speed data showed that the fast and slow, and the slow and
random groups were significantly different (F(1, 20) = 10.56, p < .005;
F(1, 20) = 5.29, p < .05, respectively). No significant difference appeared
in the comparison of the random and fast groups (F(1, 20) = .97).
To sum up, the speed data showed that (1) the arranged speed con-
tingencies generated adaptive behavior in 13 of 15 subjects, (2) the random
contingency produced a mean speed of responding that was located be-
tween the fast and slow groups, but was somewhat biased in the direction
of faster pressing, and (3) adaptation to the arranged contingencies was
not affected by the number of clicks/trial when the slow and fast groups
were compared on late trials.
Bias in Counting
The observed speed changes could alternatively be attributed to biases
in counting as a function of the arranged speed contingencies, i.e., to a
general increase (fast group) or decrease (slow group) in the number of
presses emitted. Bias in counting was measured by subtracting the number
of presses/trial from the number of clicks/trial. Thus, a positive figure
indicates too few presses; a negative indicates too many presses. The
mean of these differences on the last 40 experimental trials relative to
baseline Trials 11-30 was used as an index of counting bias.
The fast and slow groups differed marginally in the direction of the
bias prediction (fast group mean - .45; slow group mean = .32), but the
group difference was nonsignificant (t(13) = .84). The random contin-
gency generated no bias in counting (mean difference = - .04).
The slow and fast groups differed too little to account for the differences
in speed of responding. If it is assumed that the subjects on average
responded for 4 s during the response periods, the mean speed change
in the fast group (.541 or about + 39 responses/min) would imply a mean
counting bias of - 2.6 presses per trial. Conversely, the mean speed change
in the slow group (.482 or about - 12 responses/min, one nonadapting
subject removed from the data) implies a mean bias of + .78 per trial.
Both figures are far from the actual biases observed.
The mean absolute deviances in counting (i.e., number of clicks/trial-
responses/trial) in experimental Trials 91-130 was 1.43 in the slow group
and .08 in the fast group (t(13) = 1.40, p < .20). That in the random
group was .55. Thus, all groups tended to emit somewhat fewer presses
than were required, and this tendency was marginally stronger in the slow
group than in the fast group.
In sum, even though subjects in the slow group slightly reduced their
MODULATION OF RULE-GOVERNED BEHAVIOR 417
counts and subjects in the fast group slightly increased their counts relative
to baseline levels, biases in counting could not explain the observed dif-
ferences in speed of responding between those groups.
Response-Reinforcer Asynchronies
The delay hypothesis predicts that the speed differences between the
slow and fast groups will increase with increasing levels of clicks/trial,
i.e., with reduced response-reinforcer delays. Figure 2b indicates a dif-
ference between the slow and fast groups in the predicted direction. This
result was partially supported by a multivariate ANOVA, which showed
an interaction between groups (fast and slow) and click levels that ap-
proached significance (F(l0, 4) = 5.09, p < .lO).
It may be the case that reliable conditioning had already been estab-
lished in late trials and that differential effects of long versus short delays
would be more apparent in middle trials. Speed data from middle trials
(experimental Trials 10-59) did not, however, differ as a function of few
versus many clicks/trial between the fast and slow groups; but the overall
fast versus slow group speed difference was not as marked as in late trials
(F(1, 20) = 4.18, p < .lO). The present data therefore only render partial
support for the delay of reinforcement hypothesis.
DISCUSSION
The present study substantiates the idea that low-level and nonsalient
attributes of rule-governed behavior may be modulated by nonverbal
operant contingencies independently of the subject’s awareness of the
critical contingency. Thus, Svartdal’s (1989) main finding was confirmed.
Modification of the selected IRTs in the fast and slow groups by arranged
consequences demonstrates that this behavior can be called an operant
and the feedback stimuli reinforcers.
The experiment investigated factors that may generate such changes.
Bias in counting was discounted as a determining mechanism, even though
a marginal trend appeared in the data in the direction of support for the
bias hypothesis. The hypothesis that speed of pressing was altered through
a reinforcement process was found to be more probable, since reliable
correlations between the speed contingencies and speed of pressing were
observed. This hypothesis was further supported by the gradual rather
than abrupt behavior changes over trials and by the fact that behavioral
changes in the predicted directions occurred without the subjects’ ver-
balization of the contingency.
Several behavior-controlling mechanisms that do not relate to learning
per se may interact with the associative mechanisms assumed to generate
the systematic speed changes observed in this experiment. Counting biases
may be one such mechanism, but support for this mechanism was not
found. Subjects in the random group showed a gradual increase in speed
418 FRODE SVARTDAL
of pressing in the first part of the experimental trials, even though this
group was not subjected to a fast speed contingency. This effect may be
attributed to practice effects or to accidental reinforcement. It is inter-
esting to note, however, that this “superstitious” responding seemed to
deteriorate over time, even if a high speed level of pressing would have
resulted in a high rate of reinforcers throughout the experiment. This may
indicate that behavior, even if apparently adaptive, is not necessarily
stereotypic (cf. Schwartz, 1982; Page & Neuringer, 1985).
The weak effect of varying delays of reinforcement is perhaps not
surprising. Delays varied from 0 to 5 s, which may be within the optimal
range required for conditioning (e.g., Killeen, 1981; Mackintosh, 1974).
Reinforcer presentations “informed” the subjects that the required re-
sponse was “correct,” but this information was essentially noninformative
in relation to the instructed task. The relative difficulty of the counting
task probably prevented the subjects from suspecting that they had been
misinformed in cases where they observed discrepancies between the clicks
counted and the presses emitted. The interesting point in this context is,
however, that the arranged speed contingencies affected behavior in a
reliable manner. Obviously, the subjects were able to detect a signal
embedded in noise (Killeen, 1981) despite varying response-reinforcer
asynchronies, and they apparently did so in a manner that did not involve
conscious processing of response or contingency requirements.
The complexity of an arranged operant contingency must be assumed
to affect the probability that a subject is able to verbalize that contingency.
The nature of the response to be conditioned is a critical part of the
contingency, and it may be assumed that salient responses are easier to
detect and describe than are nonsalient ones. Therefore, the verbalization
of a response-reinforcer contingency may be easy if the consequences
follow a discrete and easily detectable response; when the response is
continuously variable and difficult to describe, however, the response-
reinforcer contingency may be much more difficult to verbalize. In the
present context, the response was nonsalient in two senses. First, the
response was made nonsalient by making another behavior attribute (i.e.,
instructed counting) salient. Second, the response was defined as an an-
alog, modulating attribute of behavior. Both factors should increase the
probability that the response and the response-reinforcer relation will be
obscured, and therefore reduce the probability that they will be con-
sciously processed by the subject.
This analysis has implications for the question of awareness in condi-
tioning. It is often believed that awareness about the arranged contingency
is crucial for classical as well as operant conditioning (Brewer, 1974;
Dawson & Schell, 1987; Wearden, 1988). Dawson and Schell (1987) con-
clude that “classical autonomic conditioning critically involves conscious
cognitive processes” (p. 34), and Wearden (1988) claims that “operant
MODULATION OF RULE-GOVERNED BEHAVIOR 419
REFERENCES
Baron, A., & Galizio, M. (1983). Instructional control of human operant behavior. Psy-
chological Record, 33, 495-520.
Bentall, R. P., & Lowe, C. F. (1987). The role of verbal behavior in human learning: III.
Instructional effects in children. Journal of the Experimental Analysis of Behavior, 47,
177-190.
Bentall, R. P., Lowe, C. F., & Beasty, A. (1985). The role of verbal behavior in human
learning: II. Developmental differences. Journal of the Experimental Analysis of Be-
havior, 43, 165-181.
Berry, D. C., & Broadbent, D. E. (1988). Interactive tasks and the implicit-explicit dis-
tinction. British Journal of Psychology, 79, 251-272.
Brewer, K. S. (1974). There is no convincing evidence for operant or classical conditioning
in adult humans. In W. B. Weimer & D. S. Palermo (Eds.), Cognition and the symbolic
process. Hillsdale, NJ: Erlbaum.
Catania, A. C. (1973). The concept of the operant in the analysis of behavior. Behaviorism,
1, 103-116.
Davey, G. (1989). Ecological learning theory. London: Routledge.
Dawson, M. E., & Schell, A. M. (1987). Human autonomic and skeletal classical condi-
tioning: The role of conscious cognitive factors. In G. C. L. Davey (Ed.), Cognitive
processeses and Pavlovian conditioning. Chichester: Wiley.
Galizio, M. (1979). Contingency-shaped and rule-governed behavior: Instructional control
of human loss avoidance. Journal of the Experimental Analysis of Behavior, 31, 53-
70.
Harzem, P., Lowe, C. F., & Bagshaw, M. (1978). Verbal control in human operant behavior.
The Psychological Record, 28, 405-423.
Hayes, S. C., Brownstein, A. J., Haas, J. R., & Greenway, D. E. (1986). Instructions,
multiple schedules, and extinction: Distinguishing rule-governed from schedule-con-
trolled behavior. Journal of the Experimental Analysis of Behavior, 46, 137-147.
Hayes, S. C., Brownstein, A. J., Zettle, R. D., Rosenfarb, I., & Korn, Z. (1986). Rule-
governed behavior and sensitivity to changing consequences of responding. Journal of
the Experimental Analysis of Behavior, 45, 237-256.
Hefferline, R. F., Keenan, B., & Harford, R. A. (1958). Escape and avoidance conditioning
in human subjects without their observation of the response. Science, 130, 1338-1339.
Killeen, P. R. (1981). Learning as causal inference. In M. L. Commons & J. A. Nevin
(Eds.), Quantitative analyses of behavior: Discriminative properties of reinforcement
schedules. New York: Pergamon Press.
420 FRODE SVARTDAL