Title: Free-operant avoidance of time-out from response-independent food presentation by pigeons pecking a key
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00102817/00001
 Material Information
Title: Free-operant avoidance of time-out from response-independent food presentation by pigeons pecking a key
Physical Description: Book
Language: English
Creator: Galbicka, Gregory, 1956-
Copyright Date: 1981
 Record Information
Bibliographic ID: UF00102817
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: oclc - 07968343
ltuf - ABS2186

Full Text







To all nineteenth-century scientists, living or dead

--magnetism notwithstanding--


The verbal behavior which follows is, in a very real sense,

not my own. Rather, it is the culmination of interactions

with a verbal community which, particularly during the past

seven years, has progressively shaped that verbal behavior.

I would hope that this influence is clear enough in the behav-

ior itself, such that others will call it "their own." Should

the contingencies responsible for this behavior have been less

than exact, however, I would like to formalize some of them,

and in that way express my gratitude.

First, the the members of the supervisory committee:

To Marc N. Branch, for his guidance and support

during the past four years, which, while substantial,

was never totally complete, lest my behavior become

rule-governed and echoic.

To Edward F. Malagodi, for providing an integra-

tive intraverbal repertoire found in no other single

person, as well as sound advice.

To Henry S. Pennypacker, most responsible for

the development of a minimal behavioral repertoire

from which extensions were later formed, for not pro-

viding autoclitics until they were manded, thus strength-

ening my own.

To Robert J. Waldbillig, for providing additional

"inside information." May we indeed by more closely al-

igned in the future.

And to John A. Cornell, whose central tendency was

always to understand this vast amount of verbiage.

Next, to my parents, Joseph A. and Dorothy M. Galbicka, and

other members of my family, as well as to Janet Siwy, for supporting

and nuturing my conviction in a natural science of behavior more

strongly than they may imagine.

To the associated members of the Main Branch Lab, for tech-

nical as well as intellectual support.

To the New Wave Behaviorists, for providing a verbal commun-

ity which did not severely punish novel verbal statements at odds

with the prevailing contingencies, while heavily reinforcing oc-

casional ones that were not.

To the other faculty and students (graduate and undergraduate)

associated with the Experimental Analysis of Behavior at Fort Skinner,

for accepting and, I am afraid, occasionally reinforcing certain

behavioral eccentricities.

To Janie Partin, who independently acquired an intraverbal

repertoire comparable to my own, for occasionally telling me what

I should be saying.

And finally, to John T. McArthur and Ray A. Preston, without

whom I would never have "ventured so far from home," intellectually

or physically.

To all these the Tall Kid says "Thanks," hoping that my behavior in

the future will be such that they will accept the recognition they deserve.












* 1

Deletion Procedures .
Delay Procedures .
Stimulus Modification Procedures

Contiguous vs. Consequent Control
Two-Factor Theories
One-Factor Theories


Method .
Results .
Discussion .

Method .
Results .
Discussion .

Method .
Results .
Discussion .




. 4
. 4
. 11
S 18

S . 23
S 25
. 33

S 41

. 50
. 51
. 56
S 67

. 77
. 81
S 88

S 93
S. 107





. .



1 List of conditions and summary measures for
each subject during Experiment I . . . . . 55

2 List of conditions and summary measures for
each subject during Experiment II . . . . . 82

3 List of conditions and summary measures for
each subject during Experiment III . . .. . .. 95



1 Response rates during time-in for each subject
as a function of the RT interval . . . . . . .

2 Numbers of time-outs delivered to each subject
as a function of the RT interval . . . . . . .

3 Representative cumulative records for each subject
under the different RT intervals . . . . . . .

4 Relative frequency and conditional probability
distributions for IRTs in fifths of the RT interval . .

5 Cumulative records depicting transistions in
key pecking of P-6441 between different RT intervals . .

6 Daily session response rates for each subject
under the delay. and yoked-VT contingencies . . . .

7 Cumulative records from selected sessions under the
delay-contingency and under the corresponding yoked-VT sessii

8 Response rates on the delay-key during time-in for
P-7820 and P-6441 as a function of the RT interval . . .

9 Representative cumulative records for P-7820 and P-6441
under the RT intervals of the two-key procedure . . .

10 Numbers of time-outs delivered to P-7820 and P-6441
under the two-key procedure as a function of the RT interval

11 Relative frequency and conditional probability distributions
of IRTs in fifths of the RT interval under the two-key
procedure for P-7820 and P-6441 . . . . . . .

on 87






Abstract of Dissertation Presented to the Graduate Council of the
University of Florida in Partial Fulfillment of the Requirements
for the Degree of Doctor of Philosophy



Gregory Galbicka

August, 1981

Chairman: Marc N. Branch
Major Department: Psychology

The present experiments examined the behavior of food-deprived

pigeons pecking a small translucent disc ("key"). Food was occasion-

ally presented independently of responding, except during signalled

"time-out" periods, during which food was never presented. Key peck-

ing during "time-in" postponed the next time-out according to a free-

operant avoidance paradigm. Successive time-outs followed one an-

other at 5-sec intervals (i.e., the time-out--time-out interval=5

sec) unless a response occurred during time-in, in which case the

next time-out occurred x sec after the last response, where x was

the length of the response--time-out interval.

During Experiment I, stimuli correlated with time-in and time-

out were projected on the key. Lengthening the response--time-out

interval while maintaining a constant time-out--time-out interval



progressively decreased response rates during time-in for all sub-


During Experiieant II, the importance of the delay contingency

in maintaining the key pecking observed was examined by presenting

time-outs response-independently at variable intervals matched to

ones obtained under a proceeding free-operant avoidance condition.

Response rates for all subjects decreased when the delay contingency

was suspended in this manner.

The independent contributions of responding maintained by time-

out-postponement and responding elicited by the time-in and time-out

stimuli were examined with a two-key procedure during Experiment III.

Responses to a continuously illuminated "delay" key during time-in

postponed time-out, signalled by stimuli projected on a separate

"signal" key. Response rates on the delay key during time-in for

two subjects decreased as the response--time-out interval was length-

ened. Responding on the signal key was unsystematically related to

the response--time-out interval, and generally occurred at very low

rates. The third subject responded on the delay and signal keys at

comparable rates, and response rates on the delay key during time-

in were unaffected by changes in the response--time-out interval.

Thus, free-operant time-out-postponement may control key peck-

ing in the relative absence of elicited pecking, but elicited re-

sponding may contribute to the behavior observed.


The analysis of operant behavior (i.e., behavior modified by

its consequences) may be partitioned into four broad classes, de-

pending on whether increases or decreases in some characteristic

property of a consequent stimulus are programmed as response con-

sequences, and the effect of such consequences on the subsequent

frequency of said response. (Morse and Kelleher (1977) present

a classification scheme similar to the one here but emphasize con-

sequent stimulus presentation or termination, respectively, rather

than increases or decreases in some aspect of the stimulus.) In-

creases in the frequency of a response via response-contingent in-

creases in some characteristic (e.g., frequency, magnitude, dura-

tion, etc.) of a consequent stimulus are generally termed instances

of "positive reinforcement." Increases in response frequency pro-

duced in this manner are distinguished from increases produced via

response-contingent decreases in consequent aversiveve") stimula-

tion, which are termed "negative reinforcement." The common use

of the term "reinforcement" highlights the increases in response

rate observed with both manipulations, regardless of whether re-

sponses increase ("positive") or decrease ("negative") some aspect

of consequent stimulation. The analysis of these two types of con-

trol has developed simultaneously and for the most part independently.

However, a number of parallels exist in both the development of pro-

cedures and interpretations (see Hineline (1977) for a recent treat-

ment of these s-imilarities). The remaining two classes of analysis

are those in which response frequency decreases following a history

of response-contingent stimulation ("punishment"), and may be simi-

larly classed as "positive" or "negative." (The present paper is

concerned primarily with reinforcement operations and in particular

negative reinforcement, and as such punishment will subsequently be

discussed only occasionally.)

What follows is a review of the study of negative reinforcement,

first with respect to the procedures that have been employed and

second in terms of the controlling variables suggested as necessary

for the acquisition and maintenance of behavior under this type of

aversive control. The majority of the data to be reviewed involve

the behavior of rats, and in particular rats depressing a small,

generally rectangular piece of metal protruding perpendicularly from

one wall of an experimental chamber (i.e., "bar" or "lever" press-

ing). This subject/response combination has so frequently been

used in the analysis of negative reinforcement contingencies that

exceptions need be, and have been, noted. The research conducted

in the present studies represents one such exception and involves

the modification of a frequently used procedure in the analysis of

negative reinforcement. Under this modified procedure, food-deprived

pigeons pecked a small translucent disc ("key"), and each peck post-

poned for a fixed period the occurrence of a signalled period of

time-out from response-independent food presentation.

One final note is in order before proceeding. The response

most often studied under negative reinforcement contingencies has

not always been the lever press. Prior to the mid-1950's, a num-

ber of investigators (e.g., Mowrer & Lamoreaux, 1942) examined the

effects of negative reinforcement contingencies on running in a

"shuttle-box." This apparatus consists of a rectangular enclosure

partitioned along the major axis into "sides" and some means of de-

termining which side the subject occupies at different points in

time. The response defined by such an apparatus is movement to the

opposite side. Results from these experiments have, not imperti-

nently, largely been excluded from the review. Running in a shut-

tle-box is affected by negative reinforcement contingencies in ways

comparable to effects noted when bar pressing is the measured re-

sponse. Since studies involving the latter are generally more con-

temporary, they have taken precedence over the former in the pre-

sent review. Where important, however, either for historical rea-

sons or in discussing response topography per se, results from ex-

periments involving "shuttling" have been included.


The procedures most often employed in the study of the main-

tenance of behavior by aversive stimuli may be broadly classified

in terms of whether a response is experimentally programmed to

delete, delay, or otherwise modify some characteristic property of

the aversive stimulus. While the functional effects of these pro-

cedures may not be so clearly discernible, classification in terms

of the experimentally programmed effects of responses provides a

convenient, if arbitrary, means of distinction.

Deletion Procedures

Aversive events may be presented continuously or intermittently.

When a response terminates a continuously present aversive event

for some period of time, an escape paradigm is defined. The ear-

liest studies involving such procedures were conducted by invesi-

gators interested in respondent conditioning (e.g., Bechterev, 1913;

Pavlov, 1927). Bechterev (1913), using electric shock delivered

through a panel on which a dog's foot rested, reliably observed

flexion following the onset of shock. Since in many cases flexion

resulted in the termination of shock, it is difficult to discern

whether the response was "elicited" by the presentation of the


"unconditional stimulus" or was maintained as an effective escape

response through response-contingent termination of shock.

Under the procedure described above, a single response resulted

in termination of the aversive stimulus. Dinsmoor and his colleagues

(e.g., Dinsmoor, 1968; Dinsmoor & Winograd, 1958) developed a pro-

cedure under which responses terminated shock according to a vari-

able-interval (VI) schedule. Under this procedure, in the presence

of continuous electric shock, the first response after some average

interval of time produced a fixed period of shock-free time. They

reported that rates of bar pressing in rats were directly related

to the intensity of shock delivered (Dinsmoor & Winograd, 1958) and

that responding maintained by escape from continuous shock was simi-

lar to responding maintained by a comparable VI schedule of dele-

tion of frequent, irregularly-spaced brief shocks (Dinsmoor, 1968).

(Whether this latter procedure should be termed "escape" is debat-

able, since shock was not continuously present. The distinction

may be of little use, however, if the behavior maintained is simi-

lar. Rather, it may be more fruitful, as Hineline (1977) suggested,

to consider both procedures as shock-deletion procedures which vary

only with respect to the background density of aversive stimulation.)

Deletion procedures involving more intermittent shock include

both free-operant procedures (where the opportunity to respond is

continuously present) and discrete-trials procedures (where response

opportunity is restricted). Sidman (1966) maintained bar pressing

in rats under a procedure he termed "fixed-cycle avoidance" where

a brief shock was scheduled to occur every t sec. A single response

anytime during the inter-shock-interval cancelled the delivery of

shock at the end of that cycle. DeVilliers (1972, 1974) modified

this procedure such that shocks were scheduled to occur at variable

rather than fixed intervals ("VI shock deletion"). Rats' bar press-

ing rates under this latter procedure were linearly related to the

number of shocks deleted (deVilliers, 1974). When an independent

VI shock deletion schedule was programmed simultaneously for each

of two responses, relative response rates and relative rates of shock

deletion were also linearly related (Logue and deVilliers, 1978).

This latter relationship is similar to those obtained under analo-

gous procedures involving positive reinforcement (See deVilliers

(1977) for a recent review of this literature). Additionally, de-

Villiers (1974) observed both positive and negative behavioral con-

trast (cf. Reynolds, 1961; Schwartz & Gamzu, 1977) with rats lever

pressing under a multiple schedule (in which two or more schedules,

or "componenents," alternate in succession, each in the presence

of a unique exteroceptive stimulus) with VI shock deletion schedules

in the two components. He also noted that these contrast effects

increased with decreases in component duration, an effect similar

to that obtained under schedules of positive reinforcement (e.g.,

Shimp & Wheatley, 1971; Todorov, 1972).

Herrnstein and Hineline (1966) developed a procedure under which

shocks could be delivered at random intervals according to one of

two constant-probability distributions of inter-shock-intervals.

One distribution, the "post-shock" distribution, was effective if a

response had not occurred since the last delivered shock. A single

response deleted shocks scheduled by this distribution, and the shock

delivery was subsequently controlled by the second ("post-response")

distribution until a shock was delivered according to that distribu-

tion. Following this shock delivery, control reverted back to the

post-shock distribution. Unlike the previously described shock-

deletion procedures, not all shocks could be deleted by responding;

responding deleted only those shocks scheduled according to the post-

shock distribution. Responding was maintained under this procedure

whenever the frequency of shock programmed by the post-response dis-

tribution.was less than the frequency scheduled by the post-shock

distribution, even with shock frequencies as close as six and nine

per min, respectively, scheduled by the post-response and post-shock


A classification system originally developed by Schoenfeld and

Cole (1972) to describe schedules of positive reinforcement can be

extended to describe certain other free-operant negative reinforce-

ment procedures. The system involves continuously repeating cycles

T sec in duration. These cycles are further divided into subcycles

denoted t and t The first response during t deletes shock

scheduled to occur at the end of T. All other responses are ineffec-

tive. Hurwitz and Millenson (1961) programmed cycles of constant
duration comprised of a t and following t period, and manipulated

the relative amount of time occupied by tD in T. With increases in

the relative duration of t response rates first increased then de-

creased in a %mnner similar to relations obtained by Hearst (1960)

with comparable manipulations under at: analogous schedule of response-

contingent food presentation. Sidman (1962a) obtained results simi-

lar to Hurwitz and Millenson's when tD occurred at the end of each

cycle; however, when tD was shifted towards the middle of T (such

that tA periods occurred both prior to and after tD), responding

ceased. Kadden, Schoenfeld and Snapper (1974) independently manipu-

lated the probabilities of shock given a response during tD and given

no response, and found that bar pressing of rats was maintained when-

ever the former was less than the latter. They also noted that re-

sponse rates were positively accelerated across T when t occupied a

large portion of T, provided the probability of shock given a response

in tD was greater than 0.0 but less than that which suppressed respond-


Responding under the free-operant procedures described thus far

deletes only the scheduled occurrence of brief, intermittent aversive

stimulus presentations. Responding may also be maintained when it

terminates a stimulus correlated with aversive stimulation, as well

as deleting the aversive stimulus itself. Hake and Campbell (1972)

produced positively accelerated patterns of lever pressing in squirrel

monkeys responding under a fixed-interval schedule of stimulus-shock

complex termination. In the presence of a unique exteroceptive

stimulus, brief shocks were delivered every 30 sec, and the first

response after 3 min terminated for 3 min the stimulus correlated

with shock delivery and interrupted the periodic delivery of shock.

Byrd (1977), also using squirrel monkeys, maintained patterns of

lever pressing characterized by a pause followed by an abrupt tran-

sition to a high rate under a procedure similar to Hake and Cambell's

except that a fixed number of responses (fixed-ratio) was required

for termination, and Kelleher and Morse (1964) maintained schedule

appropriate responding in squirrel monkeys when fixed-interval and

fixed-ratio schedules of stimulus-shock complex termination alter-

nated in the presence of distinctive stimuli.

Krasnegor, Brady and Findley (1971) examined various pairs of

ratio requirements under a slightly different stimulus-shock complex

termination procedure. Sessions consisted of repeating 90 sec cycles.

The first 30 sec of each cycle was signalled by the illumination of

a blue light (SI), unless the subjects (rhesus monkeys) emitted a

fixed number of lever presses. Completion of the ratio requirement

terminated S1 and produced a "blackout" for the remainder of the

90 sec cycle. If the ratio requirement was not met within the first

30 sec, the blue light was replaced by a green one (S2) for 30 sec,

again unless a fixed number of responses (not necessarily the same

as that required in the presence of S1) was emitted within 30 sec.

Completion of the ratio requirement in S2 likewise produced a black-

out for the remainder of that cycle. If neither requirement was met,

S2 was followed immediately by illumination of a red light for 3 sec,
during which 3 brief shocks were delivered at one-sec intervals, then

by a 27-sec blackout. Subjects almost invariably completed one of

the two ratio requirements and, hence, few shocks were delivered.

When the ratio requirements in the presence of S1 and S2 were identical,


responding was slightly "biased" towards S2 (i.e., slightly more ratios

were completed during S2 than during S-I). When the requirement in

one component was varied while maintaining the requirement in the

other component constant at 30, the frequency with which ratios were

completed during the varied component was inversely related to the

ratio value but was directly related to this value in the constant


Studies of discrete-trials deletion procedures are less numerous

than the above described free-operant ones. Hineline and Herrnstein

(1970) used a discrete-trials procedure they termed "fixed-cycle

deletion" to maintain bar pressing in rats. Twenty-sec trials ended

with presentation of a brief shock unless a single response occurred

during the trial, in which case the lever immediately retracted for

the remainderof the trial and shock presentation at the end of that

trial was deleted. Neffinger and Gibbon (1975) and Flye and Gibbon

(1979) extended Gibbon, Berryman and Thompson's (1974) tA~tion of

"contingency spaces" to the analysis of aversive control by negative

reinforcement with "partial avoidance" procedures. Like Hineline

and Herrnstein's procedure, brief shocks were scheduled to occur at

the end of each 20-sec trial unless a response occurred. As in

fixed-cycle deletion procedures, responses were followed by retrac-

tion of the lever or produced some other exteroceptive stimulus change.

However, unlike fixed-cycle deletion procedures, where the probability

of shock given a response is always 0.00 and 1.00 given no response,

partial avoidance procedures allow the probability of shock delivery

given a response to be varied independently of the probability of shock

given no response. Bar pressing in rats typically was maintained whenever

the probability of shock given a response was less than that given no re-

sponse. (A few subjects responded when the two probabilities were equal,

however this responding was characteristically short-latency responses,

and may have been either shock-elicited or -induced.)

Delay Procedures

Unlike the deletion procedures, delay (or postponement) procedures

do not cancel discrete occurrences of aversive events, but rather postpone

them for some period of time. (Again, the distinction is between the experi-

mentally programmed effects of a response, not possible functional ones.)

As with deletion procedures, delay procedures vary with respect to such para-

meters as continuous vs. intermittent aversive stimulation, fixed vs. variable

delays, free-operant vs. discrete-trials procedures, etc.

Probably the most frequently used delay procedure is "free-operant

avoidance" (Sidman, 1953). Typically, in the absence of responding, brief

inescapable shocks occuratfixed intervals, specified by the "shock-shock" (SS)

interval. Responses delay the next shock for some fixed period of time, termed

the response-shock (RS) interval. Delays do not cumulate; that is, shock al-

ways occurs x sec (specified by the RS interval) after the most recent response.

Sidman studied bar pressing of rats which resulted in shock-postponement un-

der this procedure. However, a variety of species, responses and aversive

events have been used with surprising generality of effects. Rats bar press

under control of noise-postponement (Knutson & Bailey, 1974), postponement

of time-out from response-independent food presentation (D'Andrea,

1971) or food presentations (Smith & Clark, 1972), and have "shuttled"

(e.g., Harman & Hineline, in Hineline, 1977; Libby & Church, 1974)

or run in a wheel (Weisman & Litner, 1969) when such responses de-

layed presentation of brief electric shocks. Dogs have acquired

panel press (e.g., Rescorla, 1969) or shuttle (Rescorla, 1968) re-

sponses under shock-postponement procedures. Key pecking (e.g.,

Ferrari, Todorov & Graeff, 1973; Todorov, Ferrari & de Souza, 1974)

or treadle pressing (e.g., Dinsmoor & Sears, 1973; Foree & LoLordo,

1970; Jowaisas, 1977; Rilling & Budnick, 1975; Smith & Keller, 1970)

of pigeons have been comparably controlled under similar shock-

postponement procedures, and key pecking which results in postpone-

ment of time-out from response-independent food presentation (Thomas,

1965a, 1965b) has also been examined. Numerous studies have reported

free-operant avoidance in rhesus and squirrel monkeys (including Clark

and Smith's (1977) of food-postponement), and humans have turned

levers when such responses postponed time-out from response-independent

money presentations (e.g., Baron & Galazio, 1976; Baron & Kaufman,


Reliable acquisition of free-operant avoidance depends on a rela-

tively short SS interval with respect to the RS interval (Leaf, 1965;

Sidman, 1962; but, cf. Clark & Hull, 1966) although the absolute values

required for acquisition may vary depending on the response employed

(see, for example, Harman & Hineline, in Hineline, 1977). Responding

is also (possibly even more) reliably engendered when the basic proce-

dure is modified to provide variable SS (Bolles & Popp, 1964), RS

(Sidman & Boren, 1957a) or SS and RS (Hineline, 1977) intervals.

Patterns of responding engendered under free-operant avoidance pro-

cedures consist of fairly constant overall rates of responding,

with transient increases in response rate ("bursts") immediately

after delivery of the aversive stimulus (e.g., Boren, 1961; Ellen &

Wilson, 1964). These increases may be due to responding elicited

by aversive stimulus presentation (e.g., see Hutchinson, 1977) or

maintained by adventitious escape contingencies. Boren (1961) pro-

vided suggestive evidence for the latter notion with a free-operant

avoidance paradigm involving two levers, one effective only during

RS and one only during SS intervals, respectively. With an SS

interval=0 (a delay analog of escape), bursting occurred predomi-

nantly during RS intervals, but occurred more frequently on the

SS-interval lever, the one associated with shock termination.

Although overall rates of responding are generally fairly con-

stant under free-operant avoidance procedures, the distribution of

times between successive responses, or interresponse-times (IRTs),

has on occasion been reported to be non-random. Specifically, the

conditional probability of an IRT (i.e., the probability of an IRT

of length t sec given at least that amount of time has elapsed since

the preceding response) has been reported to increase with increas-

ingly long IRTs for rats shuttling (e.g., Libby & Church, 1974) or

bar pressing (Sidman, 1966) and pigeons key pecking (Jowaisas, 1977)

under free-operant shock-postponement procedures. This type of tem-

poral control need not develop, however, prior to or concurrent with

the development of stable performance in terms of overall response

rate, and may appear in some subjects only after extended exposure

under the procedure (see, for example, Sidman, 1966, particularly

pp. 464. ff.).

Overall rates of responding maintained by free-operant avoidance

procedures depend on the values of both the SS and RS intervals. Re-

sponse rate generally increases rapidly to a maximum as the RS inter-

val is increased to a value equal to or slightly less than the SS

interval, then declines exponentially with longer RS-interval values

(e.g., Clark & Hull, 1966; Klein & Rilling, 1972; Sidman 1953; Thomas,

1965a; Todorov, Ferrari & de Souza, 1974). Verhave (1959) obtained

similar functions, with the exception of higher absolute response

rates than those reported by Sidman (1953), when the number of re-

sponses required to initiate a new RS interval was raised from the

traditional single response to eight responses. Additionally, re-

sponse rates are directly related to both the intensity (Boren &

Sidman, 1959; Klein & Rilling, 1972; Leander, 1973) and duration

(Leander, 1973) of shock presented.

The standard free-operant avoidance procedure may be modified in

a number of ways other than the ones cited above. One modification,

termed "discriminated free-operant avoidance" (e.g., Sidman, 1955,

1957; Sidman & Boren, 1957b, 1957c), involves the interpolation of

exteroceptive stimuli differentially correlated with the delivery of

shock. As these procedures played an important role in the develop-

ment of theoretical accounts of aversive control, detailed discussion

will be reserved until later. Other variations of the basic proce-

dure have been examined. Boren and Sidman (1957a) modified the stan-

dard free-operant avoidance paradigm to examine the effects of deliver-

ing only a percentage of shocks actually scheduled. Response rates

were generally unaffected across a wide range of values until the

probability dropped to approximately 0.05. These results are compar-

able to those obtained by Neffinger and Gibbon (1975) and Flye and

Gibbon (1979) with their discrete-trials partial avoidance deletion

procedure,where responding was maintained at substantial levels when
as many as 95% of shocks scheduled to occur on trials with a response

were cancelled.

Sidman (1962b) and Field and Boren (1963) modified the paradigm

to produce an "adjusting avoidance" procedure. Under the standard

free-operant avoidance procedure, shock is postponed to occur x sec

after the last response, where x is the value of the RS interval.

Under adjusting avoidance schedules, each response postpones shock

for x sec, and the delay cumulates across successive responses to

some maximum value. Thus, with an RS interval=15 sec, two responses

spaced 8 sec apart would result in shock presentation being post-

poned until 23 sec after the first response under standard free-

operant avoidance, but until 30 sec after the first response under

an adjusting schedule. Responding maintained under adjusting avoid-

ance schedules usually results in shock being greatly postponed, and

the degree to which it is postponed is decreased with the addition

of stimuli correlated with the time to shock presentation (Field &

Boren, 1963). Sidman (1966) pointed out another interesting dif-

ference between standard free-operant and adjusting avoidance pro-

cedures by noting that adjusting schedules do not provide differen-

tial consequences for spaced-responding. Under free-operant avoid-

ance "efficient" responding (measured in terms of the number of

responses per shock presentation) should be characterized by fairly

widely-spaced responses since a response occurring soon after a

response does not provide as much additional delay as one occurring

relatively later. Indeed, as mentioned above, responding under free-

operant avoidance procedures is typically characterized by an in-

crease in the conditional probability of a response with increasing

post-response time. Adjusting avoidance procedures do not provide

differential delay of aversive stimulation following spaced responses,

since each response adds a specified delay regardless of where it

occurs in the RS interval. As such, one might not expect to see the

increase in response probability with increasing post-response time.

Sidman (1962b) reported exactly that; conditional probabilities of

interresponse-times remained relatively constant or decreased at

longer interresponse-times under such a procedure.

The last free-operant delay procedure to be discussed is that

described by Sidman (1966) termed "fixed-interval avoidance" (more

for the patterns of behavior produced, it seems, than for a specifi-

cation of fixed periods of time between successive shock presentations).

Sidman modified the deletion procedures which specify t and t

periods during which responding is ineffective or deletes shock

scheduled to occur at the end of tD, respectively. Responses during

t_ under fixed-interval avoidance do not cancel the presentation of

shock but rather prolongs the tt period (and, thus, delay shock) for

some fixed period of time. Patterns of responding engendered under

this procedure were typically positively accelerated, more so than

those reported by Kadden, Schoenfeld and Snapper (1974) under stan-

dard deletion procedures, possibly because each cycle under fixed-

interval avoidance begins with shock presentation, which could act

as a discriminative stimulus signalling the beginning of t thus

controlling a decreased rate of responding early in each cycle.

Hineline (1970) developed a discrete-trials procedure in which

one shock was delivered during every 20 sec trial. Each trial began

with insertion of a retractable lever into the chamber and, if no re-

sponse occurred within 8 sec, a brief shock was delivered on the

eighth sec. Two sec later the lever was retracted for 10 sec, and

then reinserted to begin the next trial. If a response occurred prior

to the delivery of shock, the lever immediately retracted, and shock

delivery was delayed until 18 sec into the trial. Responding was

reliably maintained under this procedure. Benodict (1975) modified

this procedure to differentiate latencies to a response. With dif-

ferent groups of rats, the delay to shock was either directly or in-

versely related to response latency. Thus, for the "long-latency-

long-delay" group, each sec in the latency to a response from trial

onset added a sec to the delay to shock achieved by a response, while

for the "short-latency-long-delay" group each additional sec in the

response latency subtracted a sec from the delay. Response latencies

for the first group were generally longer than those for the second.

Gardner and Lewis (1976, 1977) and Lewis, Gardner and Hutton

(1976) developed a discrete-trials procedure to incorporate multiple

delayed shock presentations. In the absence of responding, brieF

shocks were delivered (in the presence of a distinctive stimulus) at

random intervals averaging 30 sec. A single response in the presence

of this stimulus initiated a 3-min alternate condition correlated

with a second stimulus, during which six shocks were delivered one

sec apart, beginning either 10, 88 or 165 sec (under different exper-

imental phases) after the response. (Further responses during the

alternate condition had no scheduled consequences.) The percentage

of time rats spent in the alternate condition was directly related

to the delay to the first shock following a bar press which initiated

the alternate condition, and substantial responding was maintained

when a response delayed the presentation of as many as 12 similarly-

spaced shocks for 105 sec. Similar results were also obtained with

pigeons key pecking under a similar shock-delay procedure (Gardner &

Lewis, 1977).

Stimulus Modification Procedures

The third and final class of aversive control paradigm includes

those procedures which provide response-dependent modulation of some

physical property of the aversive stimulus. While all deletion pro-

cedures may be considered to modify either the duration and/or the

intensity of aversive stimulation, only those procedures which do not

concomitantly decrease the frequency or otherwise change the temporal

distribution of aversive events in time will be considered in this


Weiss and Laties (1959, 1963) modified the traditional escape

paradigm by programming response-contingent intensity-reduction,

rather than termination, of a continuously present shock which in-

creased in intensity every sec without a response. This procedure

has become known as "shock-titration" or "fractional escape."

Response rates under such procedures are inversely related to the

time between successive increments in shock intensity in the absence

of responding, and decreases in the value of this parameter, as well

as increasing the number of responses required for a decrement in

shock intensity, results in an increase in the median intensity of

shock delivered (Wiess & Laties, 1959). Powell and Peck (1969)

reported that acquisition of lever pressing by rats was more reliable

under a procedure where each lever press reduced for 20 sec the inten-

sity of shocks delivered every five sec than with a standard free-

operant avoidance procedure with an SS interval=5 sec and an RS

interval=20 sec. Bersh and Alloy (1978) maintained lever pressing

under IRT
less than,-t sec to be effective) of shock-intensity reduction.

Following an appropriate IRT, the intensity of shocks delivered at

random intervals averaging 6 sec was reduced from 1.6- to 0.75-mA

for a 15-sec period. Response rates in the presence of occasional

low-intensity shock were inversely related for all subjects to the

value of the required IRT. In a subsequent experiment (Bersh &

Alloy, 1980), shock intensity was held constant at 1.6-mA and IRT-

contingent shock-duration reduction (from 1.0- to 0.3-sec) was pro-

granmed for one group of subjects while shock durations of matched

duration were presented independent of responding to a second group.

Responding was well maintained in the former while subjects in the

latter group responded rarely if at all.

Lewis, Gardner and Lopatto (1980) developed a procedure, simi-

lar to the delay procedure of Gardner and Lewis (1976) previously

described, to examine negative reinforcement via reduction in shock

duration. Shocks were presented every 30 sec. Shock duration was

2.0 sec unless a response occurred, which resulted in a 3-min change

in stimulus conditions ("alternate condition") during which shock

duration was reduced. Bar pressing by rats was acquired when shock

duration in the alternate condition was 0.1 sec. When shock dura-

tion was subsequently varied systematically across experimental phases,

the percent of session-time rats spent in the alternate condition was

inversely related to the duration of shock presented during it. Addi-

tionally, responding was maintained in experienced subjects and

acquired in naive ones when the procedure was modified such that all

but the first shock delivered during the alternate condition were

shorter (0.1-sec) than the 2.0 sec shocks delivered in the absence

of responding.

The final procedures to be considered under this heading do

not produce direct response-contingent modulation of aversive stimuli,

but rather produce stimuli correlated with imminent aversive stimula-

tion. Responses under such procedures result in the termination of

a stimulus condition during which brief shocks are presented at

randomly-spaced intervals ("unsignalled condition"), and onset of a

second stimulus condition ("signalled condition") which remains in

effect for some fixed period of time. In the presence of the signal-

led stimulus condition, shocks are delivered as before, but are brief-

ly (e.g., for 5 sec) preceded by the onset of an additional stimulus

(e.g., a tone) which terminates with shock delivery. Thus, during

the signalled shock component, shock is never delivered in the presence

of the component-correlated stimulus alone (a "safety" signal), but

rather only in the presence of the component-correlated stimulus and

the pre-shock stimulus (a "warning" stimulus). Responding which

results in the presentation of safety stimuli increases with increases

in shock intensity (Harsh & Badia, 1975) and is maintained when sig-

nalled shocks are of greater intensity or longer duration (Badia,

Culbertson & Harsh, 1973) or occur more frequently (Badia, Coker &

Harsh, 1973) than unsignalled shocks.

Under these procedures, the probability of shock delivery in

the presence of the safety stimuli (i.e., Pr(S/Ss)) is 0.00 and 1.00

in the presence of the warning stimulus (S ). Badia, Harsh, Coker

and Abbot (1976) examined bar pressing in rats under conditions inter-

mediate to these. They manipulated the Pr(S/S ) by only occasionally

presenting shock at the end of the warning stimulus (while maintain-

ing the Pr(S/Ss)=0.00 and found that responding was little affected.

As the Pr(S/S ) was decreased, subjects responded at rates similar


to those obtained when every warning stimulus was followed by shock.

In contrast, increasing the Pr(S/Ss), by occasionally delivering

shock without the pre-shock stimulus, while holding the Pr(S/S )

constant, produced systematic decreases in response rates.


Continguous vs. Consequent Control

The systematic analysis of behavior controlled by aversive

stimuli most likely began with the defensive conditioning studies

of Pavlov (1927) and Bechterev (1913). Pavlov studied salivation

elicited by injection of acidic solutions into the mouth, while

Bechterev concentrated on leg flexion elicited by presentation of

shock to the feet of dogs. Both found that responding could be

elicited by presentation of a stimulus correlated with aversive

stimulation, and suggested that this responding was acquired simply

due to the correlation 6f conditioned and unconditioned stimuli.

It did not take long, however, for others (e.g., Schlosberg, 1934;

Zener, 1937) to suggest that the consequences of the acquired re-

sponse may be of importance in the maintenance of responding. The

first experimental test of this notion (Schlosberg, 1934) was,

surprisingly, anything but convincing. Rats for which a tail flick

prevented delivery of electric shock responded at levels as low as

others that were shocked response-independently. Other investiga-

tors, however, later found clear differences in responding as a

function of whether or not the response could terminate or delete

an aversive event (e.g., Brogden, Lipman & Culler, 1938; Hunter,

1935). Brogden, Lipman and Culler (1938), for example, found that

presentation of shock following a 2-sec tone engendered wheel

running in guinea pigs confined to the wheel, regardless of

whether shock was presented after every tone presentation or

only those during which a response did not occur. This response

was maintained, however, only in subjects exposed to response-

contingent shock deletion. When running had no scheduled con-

sequences, it diminished. Thus, it appeared that the conse-

quences of at least some responses did importantly determine

whether a response was maintained; response-contingent termina-

tion ("escape") or postponement ("avoidance") of aversive stimu-

lation maintained responding more readily than did situations void

of such consequences.

While these results answered a simple question, they raised

many more. In particular, which of the many possible dimensions of

a consequence is/are necessary and sufficient for the maintenance

of behavior through negative reinforcement? Situations involving

escape responses were easily handled by Thorndike's (1914) law of

effect, since the termination of an aversive event was undoubtedly

a "satisfying state of affairs" and should thus lead to an increase

in the subsequent probability of that response. Avoidance responses,

however, were somewhat enigmatic in that the consequence of a suc-

cessful avoidance response was the non-occurrence of an event, or

the occurrence of a non-event, both of which were difficult to recon-

cile with the predominant learning theories of the time (e.g.,

Guthrie, 1935; Hull, 1943; Skinner, 1953), all of which to some

degree or other emphasized stimulus-response (or response-stimulus)

contiguity in the modification of behavior. The avoidance response

clearly lacks this property, and as such alternate behavioral mech-

anisms seemed imperative.

Two-Factor Theories

A procedural detail, remnant from the classical conditioning

heritage of aversive control procedures, provided the first and

most long-lived account of avoidance behavior. Since early studies

not reviewed here had emphasized stimulus-stimulus correlations, it

was common to present an exteroceptive stimulus prior to the onset

of the aversive stimulus. For example, in the Brogden et al. (1938)

study, a tone was presented 2 sec prior to the delivery of shock to

guinea pigs confined in a running wheel. Rotation of the wheel

greater than one inch terminated the tone and precluded shock deli-

very, if the response occurred prior to shock onset. (Responses

during shock had no scheduled consequences.) Thus, each response

during the tone did produce an immediate environmental consequence,

termination of the tone, and also deleted the upcoming scheduled

shock. This led to the development of a number of "two-factor

theories" (e.g., Mowrer and Lamoreaux, 1942; Schoenfeld, 1950;

Sidman, 1953). Although differing to some degree, all suggested

that what appeared to be "avoidance" responses were actually the

result of interactions between more "fundamental" (contiguous) pro-

cesses. The earliest and most influential propogator of two-factor

theory was 0. H. Mowrer. Mowrer proposed that the "avoidance"

response was initially learned through its termination of the

aversive stimulus. Borrowing from Freud (1936), he further con-

tended that stimuli paired with aversive stimulation came to eli-

cit (presumably through a respondent-conditioning process)

"anxiety" or "Fear". Responses which terminate such stimuli are

subsequently maintained by the immediate anxiety- or fear-

reduction they provide. Thus, what appear to be avoidance re-

sponses are actually responses maintained by escape from acquired

motivational states elicited by stimuli paired with aversive stimu-

lation (i.e., "anxiety" or "fear"). Schoenfeld (1950) rejected

the postulation of acquired motivation states in deference to a

more descriptive analysis. He suggested that paired stimuli

(again through respondent conditioning) acquired aversive charac-

teristics, and it was escape from these "conditioned" aversive

stimuli that was responsible for the maintenance of responses in

the absence of primary aversive stimulation.

The free-operant avoidance paradigm developed by Sidman (1953)

provided the first major attack on two-factor theories of the type

above (even though he suggested one of his own). Since according

to two-factor theories responding is maintained by the termination

of exteroceptive stimuli correlated with aversive stimulus presen-

tation, a procedure that does not provide such stimuli should not

maintain responding. Obviously, such a procedure does maintain

responding, and thus apparently refutes the importance of condi-

tioned motivational states or aversive stimulus termination.

Sidman suggested instead that the responding observed was

determined by the interaction of two "gradients of punishment".

One, the "distribution-of-punishment" gradient, referred to the

increased probability that "non-avoidance" responses would be

followed by an aversive event as the RS interval was decreased.

The distribution-of-punishment gradient was offered to account

for increases in response rate observed with decreases in the

length of the RS interval to values equal to or slightly less

than the SS interval. That rates eventually reached a maximum

suggested to Sidman the action of a "delay-of-punishment" gradi-

ent, which, at even shorter RS intervals, resulted in increasing

suppression of the avoidance response itself. Although plausible,

this interpretation never received prolonged attention, most pro-

bably due to the inability to measure directly changes in "non-

avoidance" behavior.

Anger (1963) took a somewhat different view of free-operant

avoidance and concluded that it signalled the death of two-factor

theory prematurely. Since aversive stimuli, when they occur under

free-operant avoidance procedures, almost always follow a response

by a fixed amount of time, the passage of time without a response

is differentially correlated with delivery of the aversive stimulus.

In effect, 'long time since the last response" becomes a conditioned

aversive stimulus which can only be terminated by responding. Es-

cape from such "conditioned aversive temporal stimuli" may thus

serve as the basis for negative reinforcement responsible for the

maintenance of responding under free-operant avoidance procedures.

Patterns of responding maintained by such procedures, characterized

in rats by increases in the probability of a response with increas-

ing post-response time (e.g., Libby & Church, 1974; Sidman, 1966),

provide circumstantial support for such a notion.

As mentioned earlier, one of the major procedural modifications

of the free-operant avoidance paradigm involves the interpolation of

exteroceptive stimuli which signal impending aversive stimulation.

These "discriminated free-operant avoidance" procedures have pro-

vided evidence which, if not directly counter to two-factor theories,

necessitate assumptions which basically make the theory untestable.

Sidman (1955) trained cats and rats to bar press under a free-

operant avoidance procedure with RS and SS intervals equal to 20 sec.

Once stable responding was engendered under this procedure, he added

a light 5 sec prior to the presentation of each shock (a "pre-shock"

or "warning" stimulus). Responses postponed shock as before (i.e.,

RS=SS=20 sec) and also postponed the light if they occurred prior

to light onset (RL=SL=15 sec) or terminated the light in its pre-

sence. He found that the majority (60-75%) of responses occurred in

the presence of the light, and the probability of a response in-

creased slightly with increasing time since light onset. Similar

effects have subsequently been reported with rhesus monkeys respond-

ing under a comparable procedure (Hyman, 1969). Ulrich, Holz and

Azrin (1964) exposed rats to a similar procedure (i.e., RS=20 sec,

RL=15 sec) with the exception that the SS interval equalled 5 sec

and the SL interval 0 sec (i.e., the light was continuously present

following shock until a response occurred). They found an even

greater preponderance of responses occurred following the onset

of the warning stimulus. It is not clear from a two-factor account

why responding which delayed onset of the warning stimulus (a "con-

ditioned aversive stimulus") was not maintained. Possibly the

light was a "weaker" aversive stimulus, capable of maintaining

responding by response-contingent termination but not delay of

the warning stimulus. It is also somewhat surprising that respond-

ing was more probable later in the light period. Responses which

immediately terminate the light presumably provide a greater amount

of negative reinforcement and should, therefore, be more probable.

Two-factor theories might point to heightened temporal stimulus

control to account for the observed increase in response probabil-

ity with increasing time since light onset.

Two additional experiments (Sidman, 1957; Sidman & Boren,

1957b) offer further difficulties for two-factor theories. In

both studies, the value of the RS interval depended on the presence

or absence of the pre-shock stimulus. For example, with a 5-sec

pre-shock light, a response in the absence of the light postponed

each shock for 20 sec (i.e., RS=20 sec), concomitantly postponing

light onset for 15 sec (i.e., RL=15 sec). Should the subject pause

longer than 15 sec, and hence the pre-shock stimulus be presented,

these contingencies no longer held. Rather, the light remained on

until a shock was delivered (i.e., RL=0 sec), and the RS interval,

in the presence of the light, was reduced to the duration of the

pre-shock stimulus (i.e., RS=5 sec). Sidman and Boren (1957b) found

that with RS and RL intervals equal to 15 and 20 sec respectively in

absence of a 5-sec pre-shock light, and 0 and 5 sec respectively in

its presence, rats typically bar pressed only in the absence of the

light. Once the light period was initiated, the subjects generally

"waited out" the interval, received a shock, and then resumed re-

sponding. Sidman (1957) parametrically manipulated the RS interval

while maintaining a constant RL interval, and subsequently examined

a range of RL intervals in the presence of a constant RS interval.

Responding under both these procedures can be summarized as occurring

in the presence of the stimulus correlated with the less stringent

response requirement (i.e., that which provided the greater delay to

shock per response). These results are analogous tothose obtained by

Krasnegor, Brady and Findley (1971) under their sequential deletion

procedure, and argue against a simple two-factor account, since re-

sponding varied systematically with changes in the delay achieved by

a response to both the light and to shock, even though light pre-

sentation was consistently paired with shock under all conditions.

In a second experiment, Sidman and Boren (1975c) compared two

procedures. After training rats to bar press under a free-operant

avoidance procedure with RS=SS=20 sec, they programmed RL and RS

intervals equal to 16 and 20 sec, respectively, in the absence of

the light, but always presented shock 4 sec after light onset (i.e.,

responses during the light had no experimentally arranged conse-

quences). Under this procedure, responding in the presence of the

light decreased to very low levels. Response rates prior to light

onset, after an initial transitory increase, returned to levels

comparable to those obtained under initial free-operant avoidance

training with no warning stimulus. In contrast, when responding in

the presence of the light terminated it and postponed shock (simi-

lar to Sidman (1955)), responding occurred predominantly in the

presence of the light. Both procedures, then, allowed for the esta-

blishment of the light as a conditioned aversive stimulus through

respondent conditioning, since shock only occurred in the presence

of the light, yet the former maintained a much higher rate of re-

sponding in the absence than in the presence of the warning stimulus.

Data obtained under procedures other than those just described

also argue against a simple two-factor account of avoidance. For

example, Bolles, Stokes and Younger (1966) compared two response

topographies (running in a wheel and shuttling) under procedures

which allowed for all possible combinations of response-contingent

avoidance ("A"), shock termination ("T "), and/or warning stimulus

termination ("T "). That is, with different groups of rats, re-

sponses resulted either in nothing or in one of the following set

of possible consequences: A, Ts, Tw, ATs, ATw, TsTw, or ATsTw.

Termination of the warning stimulus only (i.e., shock still occurred)

did not maintain more shuttling than that obtained when shuttling

had no effect, and both responses occurred more reliably when re-

sponding avoided shock, independent of stimulus termination.

Kamin, Briner and Black (1963) used a slightly different tactic

to argue against the "anxiety-provoking" aspects of warning stimuli.

They trained subjects to respond under a signalled shock deletion

schedule, and, later each day, exposed the, to a schedule of response-

contingent food presentation. Occasionally, they interpolated brief

presentations of the pre-shock stimulus during the schedule of food

presentation, and found that the degree to which behavior maintained

by the presentation of food was suppressed by this stimulus was

inversely related to the amount of responding maintained in the pre-

sence of that stimulus under the shock-deletion procedure. Since

suppression of appetitive behavior by stimuli paired with the presen-

tation of aversive events is typically viewed as a measure of the

amount of "anxiety" produced by such stimuli, these data suggest that

responding under the shock-deletion schedule was maintained better

when the pre-shock stimuli evoked less, not greater amounts of


One-Factor Theories

Results such as the ones mentioned above have led to a general

decreased acceptance of two-factor theories in favor of other inter-

pretations (but, see Dinsmoor, 1977). Herrnstein (1969) offered

a "one-factor" theory based on the notion of shock frequency reduc-

tion. He argued that "the reinforcement for avoidance behavior is

a reduction in time of aversive stimulations . ." (p.67). Herrn-

stein offers as evidence for this position data obtained under the

previously described deletion procedure developed by Herrnstein

and Hineline (1966). Under this procedure, shocks were scheduled

according to two random distributions of inter-shock-intervals, and

responding determined which distribution delivered the next shock.

Herrnstein and Hineline found that responding was maintained when

it decreased the frequency of shock delivery by as little as three

shocks/min under that obtained in the absence of responding. It is

important to note that responding does not achieve a specified

fixed delay to shock under this procedure, although the average

delay may be longer with response-contingent decreases in overall

shock frequency, provided the subject does not make multiple re-

sponses between shock presentations. Since responding is main-

tained in the absence of any exteroceptive stimuli correlated with

shock delivery, and in the absence of a programmed temporal con-

tingency between responding and shock presentation, Herrnstein

(1969) argued for the preeminence of shock frequency reduction

(SFR) as the basis of negative reinforcement. That Logue and


deVilliers (1974) found relative response rates to be systematically

related to the relative SFR achieved by one of two responses and

not to the relative number of shocks actually delivered provides

additional support for this notion.

Hineline (1970),however, pointed out that all procedures em-

ployed to that time had confounded at least two of three possible

controlling variables: 1) escape from conditioned aversive stimuli

(exteroceptive, temporal or organismic), 2) SFR and 3) the (average)

delay to the next aversive event. Thus, he argued thatstatements

suggesting the controlling variable were most likely premature.

He developed a discrete-trials procedure (Hineline, 1970) in an

attempt to separate these variables. As previously discussed, the

procedure consisted of 20-sec trials during which a single shock

was delivered. In the absence of a response, a retractable lever

was inserted into the chamber for 10 sec, a brief shock was deli-

vered at the 8th sec, and then the lever was withdrawn for 10 sec.

A response prior to 8 sec resulted in immediate retraction of the

lever, and delayed shock presentation until the 18th sec of the

trial. Thus, responding produced no overall SFR, since one shock

was presented during every trial regardless of a response. The

stimulus paired with shock delivery (i.e., "lever in" vs. "lever

out") depended on whether a response occurred, since shocks were

presented while the lever was inserted on trials without a response

and while retracted if a response occurred prior to shock delivery.

Thus, two-factor accounts might predict an oscillation in responding,

initially escaping "lever in" by responding, then escaping "lever

out" by not responding. Indeed, consistent responding would reli-

ably produce rather than terminate the conditioned aversive stimu-

lus, since shocks would always be presented when the lever was re-

tracted, the immediate consequence of every lever press. Hineline

found that responding was reliably and consistently maintained under

this procedure. Consequently, he argued for the importance of delay

to aversive stimulation as a controlling variable in the maintenance

of responding by negative reinforcement, an argument which might be

termed the "procrastination hypothesis": responding will be main-

tained when it achieves a delay to the next aversive event longer

than that in the absence of a response, independent of any reduc-

tion in the overall frequency of aversive stimulation. The dif-

ferentiation of response latencies as a function of the amount of

delay they achieve (Benodict, 1976) further argues for delay as an

important controlling variable.

Herrnstein (1969) took exception to this view, arguing that

the results obtained could be viewed in terms of "shock-frequency

reduction under the control of discriminative stimuli" (p. 69).

That is, the presentation of the lever signals the opportunity

to decrease the frequency of shock in the presence of the lever

to zero. He cites the results of a second experiment by Hineline

(1970) to support this interpretation. During this experiment,

the paradigm just mentioned was modified such that responses still

retracted the lever but resulted in shock delivery exactly 10 sec


after the response. Thus, the trial length and frequency of shock

delivery were inversely and directly related, respectively, to the

latency to a response. Responding was not maintained under this

procedure, which Hsrrnstein suggested resulted from equal frequen-

cies of shock (one per 10 sec) in the presence or absence of the

lever. Hineline, however, argued that delay was sufficient to main-

tain responding in the absence of any overall SFR (Experiment 1)

but not in the presence of an increase in the overall frequency

of shock (Experiment 2).

Lambert, Bersh, Hineline and Smith (1973) provided additional

evidence against SFR as sufficient for the maintenance of responding

by negative reinforcement. They used a procedure modeled after

Hineline (1970), with 60-sec trials and lever presentation restric-

ted to the first 10 sec. In the absence of a response, five shocks

were delivered during each trial at one-sec intervals beginning at

the 11th sec. A lever press cancelled the delivery of these shocks,

but resulted in the immediate presentation of a single shock. Thus,

responding produced a decrease in shock frequency but also resulted

in a decrease in the delay to the first shock. Responding was not

maintained under this procedure, which the authors suggested as an

indictment of the one-factor theory based solely on SFR. Although

suggestive, this procedure differs from all those previously dis-

cussed in that responding results in the immediate presentation

of an aversive event, and as such may provide for alternative inter-

pretations. Specifically, responding may not have been maintained

due to inadequate negative reinforcement based on shock-delay but

rather may have been actively suppressed by punishment. (Although

some (e.g., Oinsmnoor, 1954, 1977; Skinner, 1953) have argued that

avoidance and punishment may not be independent processes, that

contention has been challenged (e.g., Galbicka & Branch, 1981;

Rachlin & Herrnstein, 1969) for reasons which, while germane to

the present discussion, will be deferred to keep the scope of the

review manageable.) Response-contingent aversive stimulation, of

course, is nothing more than one limiting condition with regard to

Sidman's (1953) notion of "delay-of-punishment" gradients (i.e.,

delay=0 sec). Thus, it is conceivable that all delay procedures

can be analyzed in terms not of the negative reinforcement they

provide via increases in the delay to aversive stimulation, but

rather in terms of the decreased response suppression produced by

delayed punishment. Such an analysis, as mentioned earlier, is

handicapped by the inability to measure punished "non-avoidance"

behavior, except when the delay equals zero (i.e., when aversive

stimulation is response-contingent) as it was in the Lambert et al.

study. Hence, although conceptually plausible, such a notion is

very difficult (if not impossible) to verify experimentally.

Alternative conceptualizations in terms of the relation between

a specified response and aversive stimulation (e.g., response-

contingent delay), while possibly incomplete, allow experimental


A systematic examination of the effects of delayed aversive

stimulation in the absence of any SFR conducted by Gardner and

Lewis (1974) provided more telling evidence for the importance of

this factor in the maintenance of responding under negative rein-

forcement paradigms. Under their procedure, responding produced

a 3-min change in stimulus conditions and interrupted the delivery

of shock otherwise scheduled to occur at random intervals averag-

ing 30 sec. Under the "alternate condition" initiated by a response,

the same number of shocks which would, on the average, have been

presented had a response not occurred (i.e., 6) were delivered.

Thus, no overall SFR could occur. During the alternate condition

shocks were delivered at one-sec intervals, beginning after a spe-

cified delay from the response which initiated the condition. That

delay was, across experimental phases, either 10, 88 or 165 sec.

The percent of time rats spent in the alternate condition (and,

thus, the rate of responding in its absence) was directly related

to the delay to the first shock achieved by a response. Addition-

ally, substantial responding was maintained when as many as twelve

shocks (representing a 100% shock-frequency increase) were delivered

during the alternate condition and the delay to the first shock

was 105 sec. This systematic effect of delay on response rate

argues strongly for delay as an important determinant of the main-

tenance of responding by negative reinforcement, independent of

(and in the face of an increase in) the frequency of shock deli-


In a subsequent experiment, Lewis, Gardner and Hutton (1977)

suggested that delay to the first aversive event may not be

necessary in and of itself, provided that subsequent delays occur.

They modified their previous procedure such that all but one (or

two) of the shocks delivered during the alternate condition were

delayed. The nondelayed shock(s) were delivered at exactly the

same time they would have been delivered had a response not

occurred. Responding was maintained in all subjects when one non-

delayed shock was presented, and in two of six subjects (both of

which had previously been exposed to the "one nondelayed shock"

condition) when the first two shocks delivered were not delayed

by a response. Similar results were obtained with comparable

manipulations on pigeons' key pecking (Gardner & Lewis, 1977).

The authors argued that these results implicate delays to the

second and, to a less degree, third shock in a series of shocks

as controlling behavior. Shull, Spear and Bryson (1981) recently

reported that delays to the second (and to a lesser degree third)

food presentations were systematically related to response rate

under a procedure analogous to that just described but not invol-

ving delayed presentations of food. Although these data indicate

that delays are capable of controlling responding, an account in

terms of SFR, where the frequency of aversive stimulation is cal-

culated using non-arithmetic averages, cannot be discounted at

the present time. Shull et al. chose delay values carefully so

as to preclude totally frequency differences, regardless of the


averaging method used. Whether similar parameters of delayed aver-

sive stimulation would systematically control responding remains

to be determined.

Only changes in the distribution of aversive events in time

have to this point been considered as possible controlling vari-

ables. It is clear from the data obtained under procedures which

provide response-contingent reduction in physical characteristics

of the aversive stimulus that such reductions can maintain respon-

ding in the absence of concomitant changes in the frequency of such

events. Although some of these studies may be reconciled with the

previous discussion by allocating shock frequencies to two separate

classes dependent on some characteristic of the aversive stimulus

(e.g., responding increases the delay to the next "intense" or

"long" shock), procedures such as Weiss and Laties' (1959, 1963)

shock-titration schedules, where the characteristic property of

the aversive stimulus varies across a semi-continuous range, make

such analyses difficult. Furthermore, demonstration of lawful

interactions between physical properties of aversive events, for

example the similar rates of responding maintained under free-

operant avoidance procedures as a function of the product of the

durations and intensities of shocks presented (Leander, 1973),

suggest that accounts which emphasize factors other than changes

in the temporal distribution of (classes of) aversive events may

extend the precision possible in the prediction and control over

behavior by negative reinforcement contingencies.


Under free-operant avoidance paradigms (Sidman, 1953), re-

sponses postpone the occurrence of an aversive stimulus for x sec.

Otherwise these events occur every t sec. These two temporal para-

meters are typically termed the response-shock (RS) and shock-

shock (SS) intervals, respectively, underscoring a bias towards the

use of electric shock as an aversive stimulus. Investigators in-

terested in the control of behavior by response-contingent delay

or deletion of aversive events (i.e., by negative reinforcement),

apparently swayed by the logistical and behavioral superiority of

electric shock (see Azrin & Holz, 1966), have selected it almost

to the exclusion of other aversive events.

This bias generally may be of little concern, unless one is

interested in studying negative reinforcement using pigeons as

subjects. Pigeons have generally been considered prime subjects

in the analysis of positive reinforcement contingencies (e.g.,

Ferster, 1953), and the analysis of operant behavior (i.e., beha-

vior modified by its consequences) under such contingencies owes

a great deal to them (e.g., Ferster & Skinner, 1957). In contrast,

they have been used infrequently in the analysis of behavior con-

trolled by negative reinforcement contingencies, and until very

recently, their absence was highly conspicuous. One reason for


this apparent omission may be purely logistical. Electric shock

is most often delivered through the feet of subjects standing on

an electrified floor. Pigeons, however, do not conduct electricity

readily through their feet, unless special procedures, such as ap-

plication of graphite (Ferster & Skinner, 1957) and extremely high

electrical currents are used. Implanted electrodes (e.g., Azrin,

1959) may circumvent these problems at the expense of daily moni-

toring and cleaning, as well as the increased risk of infection.

Thus, many invesitgators may simply be responding to these incon-

veniences and opting for more (at least structurally if not behav-

iorally) "cooperative" subjects, such as rats and monkeys. How-

ever, recent suggestions (e.g., Bolles, 1970, 1973) that the "laws"

of negative reinforcement may lack "interpecies generality" (Sid-

man, 1960) have spurred a renewed interest in the demonstration

of pigeon behavior acquired and maintained by negative reinforce-

ment contingencies.

A second bias, interacting with the one previously mentioned,

appears to have increased further the difficulty of providing dem-

onstrations of negatively-reinforced behavior in pigeons. Speci-

fically, it appears that the acquisition of key pecking, the most

widely measured behavior of pigeons, may be retarded by the use of

electric shock as an aversive stimulus. Shock elicits neck-con-

traction in pigeons (Smith, Gustavson & Gregor, 1972), a response

incompatible with the targeted response, the key peck. Thus, com-

petition between elicited and operant responses may impede the ac-

quisition of the latter. Indeed, it appears that a number of

other responses of pigeons, such as head lifting (Hoffman & Fles-

cher, 1959), shuttling (e.g., Baum, 1973; MacPhail, 1968) and treadle

pressing (e.g., Dinsmoor & Sears, 1973. Foree & LoLordo, 1970; Klein

& Rilling, 1972, 1974; Rilling & Budnick, 1975; Smith & Keller,

1970) may be acquired more readily than key pecking under negative

reinforcement contingencies. While studies demonstrating the con-

trol of responses other than key pecking by negative reinforcement

indicate that the behavior of pigeons is not completely "immune"

to negative reinforcement, they do so at the expense of direct com-

parison to the extant literature on key pecking under other sources

of control.

Sidman (1960) noted that differences between species and re-

ponse classes may more directly reflect important differences in

the training procedures used than in the inherent characteristics

of the classes. This appears to be at least partially the case

with key pecking under the control of shock-reduction. A number

of investigators have now demonstrated that key pecking can be

shaped by response-contingent termination of trains of increasingly-

intense brief shocks (e.g., Alves de Morares & Todorov, 1977;

Ferrari & Todorov, 1980; Ferrari, Todorov & Graeff, 1973; Hineline

& Rachlin, 1969a, 1969b; Rachlin & Hineline, 1967; Todorov, Ferrari

& de Souza, 1974) or that control can be transferred from positive

to negative reinforcement contingencies (e.g., Foree & LoLordo,

1974; Gardner & Lewis, 1977; Lewis, Lewin, Stoyack & Muehleisen,

1974; but cf. Schwartz & Coulter, 1971). However, the necessary

and sufficient parameters responsible for the acquisition of key

pecking have not been delineated fully, and the shaping procedures

may be somewhat arduous, requiring a larqe investment of experi-

menter time or a degree of initial prctraining involving positive

reinforcement. Given these difficulties, it is doubtful that many

investigators will be willing to use pigeons as subjects when

studying negative reinforcement.

The problems encountered when attempting to study negative

reinforcement contingencies with pigeons do not stem primarily

from either the use of the key peck as the measured response or

electric shock as the aversive stimulus, but rather from the in-

teractions inherent in their combined use. Hence, two possible

experimental tactics are available. One involves measuring re-

sponses other than key pecking under negative reinforcement pro-

cedures based on electric shock-termination, -postponement, -in-

tensity-reduction, etc. Such procedures, as discussed above, have,

by and large been successful. Unfortunately, the results obtained

may not be readily generalizable to the vast literature involving

key pecking maintained by contingencies other than negative rein-

forcement, in that responses other than key pecking may not be

as free as key pecking to vary across some measured dimension.

For example, it is doubtful that a pigeon can press a treadle or

lift its head as frequently for extended periods of time as it

can peck a key. While this assumption may with further analysis

prove incorrect, the fact remains that key pecking under sources


of control other than negative reinforcement has been and continues

to be more frequently analyzed than any other behavior of pigeons.

Given the availability of this reference source, it would seem to

be a disservice to relinquish it on the basis of methodological

problems alone. The other available tactic, developing procedures

involving key-peck-contingent reductions in some physical charac-

teristic of an aversive stimulus other than electric shock, may

be useful in this regard.

The prime candidate for an alternative aversive stimulus is

"time-out," a signalled period during which positive reinforce-

ment is never experimentally programmed. Time-out has been used

successfully as an aversive stimulus with rats (D'Andrea, 1971;

McMillan, 1967), monkeys (e.g., Ferster, 1958), pigeons (e.g.,

Ferster, 1958; Thomas, 1965a, 1965b) and humans (Baron & Galazio,

1976; Baron & Kaufman, 1966). The presentation of time-out does

not require any special subject preparation or apparatus, and is

thus logistically simpler to program than electric shock presen-


The use of time-out with pigeons is additionally alluring,

in that it may be used both to engender and maintain key pecking.

Specifically, a number of investigators have reported that food-

deprived pigeons will peck at a stimulus differentially correlated

with the presentation of food (see Schwartz & Gamzu (1977) for a

recent review of this literature). Since the presentation of

time-out requires that it alternate with another stimulus

situation during which reinforcement is available ("time-in"),

one can "autoshape" (Brown & Jenkins, 1968) key pecking by locating

the time-in stimulus behind the response key. Thus, key pecking

may be reliably acquired with little or no preliminary training

(other than training the subject to eat from the food magazine).

Although ai aid in the acquisition of key pecking, the pre-

sence of stimulus-reinforcer contingencies requires that their

contribution to key pecking subsequently-maintained by negative

reinforcement contingencies be assessed. (As an aside, the term

"stimulus-reinforcer" should in the present context be translated

"stimulus-food presentation," as it is most commonly used by

others with schedules of response-independent or -dependent food

presentation. It is not meant to describe differential relations

between the presentation of particular stimuli and the occurrence

of negative reinforcement.) Specifically, intermittent presen-

tations of food or shock may elicit or otherwise "induce" be-

haviors (see, for example, Hutchinson (1977) and Staddon (1977)

for recent reviews of these effects) which are or are not topo-

graphically similar to the operant response under investigation.

To the extent that responses evoked by non-operant sources of con-

trol are topographically similar to the operant response, the

separate contributions of operant and non-operant contingencies

may "additively" interact in producing the overall frequency of

the observed response. Conversely, if the operant and non-operant

response classes are incompatible, the contingencies responsible

for these response classes will produce a degree of "competition"

between the two classes, since the occurrence of a member of one

class will (momentarily) preclude the occurrence of a member of

the other class. For example, key pecking directed towards stim-

uli differentially correlated with food delivery would be indis-

tinguishable (on the basis of frequency, at least) from operant

key pecks when these stimuli are projected on the operandum used

to measure the latter. Thus, changes in the frequency of elicited

key pecking resulting indirectly from manipulations of the operant

contingency (i.e., resulting from changes in the frequency of, for

example, time-out presentations and not from changes in the direct

consequences of responding) might obscure functional relations

between the operant contingency and the frequency of operant key


Interactions such as the ones described above between op-

erant and non-operant response classes may be examined in a num-

ber of ways. One method involves response-independent presenta-

tion of stimulus events in precisely the same manner as under a

previous condition during which the operant contingency was in

force. Differences between responding maintained in the presence

oF the operant contingency and that maintained in its absence but

with "yoked" distributions of stimulus presentations allow evalu-

ation of the importance of the operant contingency in maintaining

the behavior observed under the former (see, e.g., Coulson, Coul-

son & Gardner, 1970; Smith, 1973). They do not, however, allow

for direct measurement of the independent contributions of operant

and non-operant contingencies to behavior under such procedures.

That is, responding maintained in the absence of a specific op-

erant contingency (i.e., under the yoked procedure) may be pri-

marily non-operant or may reflect the occurrence of spurious cor-

relations between responses and stimulus events, leading to the

"superstitious" (see Skinner, 1948) maintenance of operant behav-

ior, or some combination thereof.

A more direct method of assessment involves "topographical

tagging" (e.g., Catania, 1971, 1973; Keller, 1974) of different

response classes, such that responses maintained by one set of

contingencies are primarily directed towards one operandum, while

responses under an alternate source of control occur on a sepa-

rate operandum. With respect to the two sources of control over

key pecking discussed above, such a "tagging" operation involves

removing stimuli differentially correlated with the delivery of

food from the key on which responses result in the presentation

of reinforcement. Since any stimuli subsequently projected on

this key are no longer differentially correlated with reinforce-

ment, they should elicit no pecking, and responses on this key

should primarily be under the control of operant contingencies.

Elicited pecking should be directed towards the re-located

reinforcement-correlated stimuli. By projecting these stimuli

on a key other than that used to define operant responses, elic-

ited responses may not only be separated from operant ones, but

additionally may independently be measured.

The present set of studies examined key pecking maintained

under free-operant avoidance of time-out from response-indepen-

dent food presentation. Experiment I involved manipulation of

the response--titme-out (RS) interval (the amount of time each

key peck postponed the next time-out) while maintaining a con-

stant time-out--time-out interval (which specified the time be-

tween successive time-outs in the absence of a response). Ex-

periment II examined the effect of removing the postponement

contingency while presenting comparable frequencies and distri-

butions of time-out response-independently. Finally, the in-

dependent contributions of the postponement and stimulus-rein-

forcer contingencies to the behavior observed were examined dur-

ing Experiment III.


Rates of responding under free-operant avoidance procedures

generally increase as the delay achieved by a response increases

to a value equal to or slightly less than the time between suc-

cessive aversive events in the absence of responding, and then

decrease exponentially as the response-contingent delay is length-

ened further (e.g., Clark & Hull, 1966; Klein & Rilling, 1972;

Sidman, 1953; Thomas, 1965a; Todorov, Ferrari & de Souza, 1974).

Responding is generally maintained at a fairly constant rate

with occasional transient increases.("bursts") following the

presentation of an aversive event (Boren, 1961; Ellen & Wilson,

1964). Although the overall response rate is fairly constant,

the times between successive responses, or interresponse-times

(IRTs), often are distributed non-randomly. The conditional

probability of an IRT=t sec (i.e., the probability of a response

t sec after the preceding one given that t or more sec have e-

lapsed) has been reported to increase as a function of IRT length

with rats (e.g., Libby & Church, 1974; Sidman, 1966) and pigeons

(Jowaisas, 1977). This type of temporal control need not always

develop prior to or concurrent with the development of stable

performance (in terms of overall rates of responding) under free-

operant avoidance paradigms (Sidman, 1966, particularly pp. 464 ff).


The present experiment extends Thomas's (1965a) analysis of

key pecking maintained under free-operant avoidance of time-out

from response-independent food presentation. In Thomas's study,

food was briefly presented response-independently at random intervals

averaging l(or 3) min, except during 5 min time-outs that occurred

every 10 sec in the absence of a key peck. Each key peck post-

poned the next time-out (under different experimental phases)

for either 20, 30, 60, or 120 sec. The present study examined

key pecking under a similar time-out-postponement procedure, but

differed with respect to parameters of time-out duration and the

length of response contingent time-out delay. Responses postponed

the occurrence of the next 20-sec time-out from response-indepen-

dent food presentation for some fixed period of time (either 5,

10, 20, or 40 sec). Otherwise, time-outs were presented every

5 sec. Except during time-out, food was briefly presented at

random intervals averaging 30 sec. In addition to these para-

metric differences between the two studies, the present study

provides a more extensive analysis than Thomas's by including

response rates during both time-in and time-out, number of time-

outs delivered and IRT distributions.



Three adult male, White Carneaux pigeons (Columba livia),

previously used in an undergraduate laboratory course, were

initially reduced to 80% of their free-feeding weights (475, 400

and 415 g for P-7820, P-9275 and P-6441, respectively). They were

subsequently maintained on a 23 h deprivation regimen, with post-

session feeding of mixed grain restricted so as to maintain the

specified level of body weight. Each subject was individually housed

in a colony room and had continuous access to water and health grit

between experimental sessions.


Experimental sessions were conducted in a pigeon condition-

ing unit similar to that described by Ferster and Skinner (1957),

measuring 30 cm wide by 31 cm long by 31 cm deep. An aluminum

sheet served as the front wall of the chamber, all other walls were

painted flat black. The front wall contained three standard re-

sponse keys (R. Gerbrands, Co.) located 22 cm above the grid floor.

Only the center key, located 15.5 cm from either side edge of the

front wall, was operative. A static force of 0.15 N applied to

the key briefly activated a "feedback" relay mounted behind the

front wall and was recorded as a response. The key could be trans-

illuminated with light from either of two 1.1-W, 28-V dc lamps lo-

cated behind it and which were covered with either a white or red

cap, respectively. A plexiglas extension mounted on the key pro-

truded 0.15 cm beyond the front wall. (Such extensions have oc-

casionally been reported to aid in the acquisition of key pecking

under autoshaping procedures (e.g., Rachlin, 1969).) The other

two keys did not extend beyond the frontwall and remained dark

throughout the course of the experiment. Directly beneath the cen-

ter key was a 6 cm by 5 cm aperture through which mixed grain could

be presented, dependent on the operation of a solenoid-driven rain

feeder. A 1.1-W, 28-V dc lamp located directly above and behind

this aperture was lit during grain presentations. General illumi-

nation during experimental sessions was provided by two 1.1-4, 28-V

dc houselights located in either upper corner of the front wall and

mounted behind reflectors which prevented direct downward illumi-

nation. The chamber was located in a room where white masking noise

was continuously present. Noise from a ventilation fan mounted on

the chamber ceiling also helped mask extraneous sounds. A PDP-8/f

minicomputer, located in an adjacent room and operating under the

SKED (Snapper, Stephens & Lee, 1974) or SuperSKED (Snapper & Inglis,

1978) software systems, programmed stimuli and recorded data. Also,

cumulative response records of each session were generated by a

Gerbrands (Model C-3) Cumulative Recorder located in the same room

as the computer.


Daily, one-hour sessions consisted of two possible stimulus

situations with respect to keylight color; continuous transillumi-

nation with either white or red light. In the presence of the white

light ("time-in"), grain was presented for 2 sec at random intervals

averaging 30 sec (determined by a probability generator set equal

to 0.03 and sampled every sec), whereas in the presence of a red

keylight grain was never presented ("time-out"). Grain was also

never presented within 3 sec of time-out termination but could oc-

cur immediately prior to a time-out or immediately after a response.

Food presentation was not delayed after responses in an attempt to

maintain the frequency of food presentation constant in the pre-

sence of large changes in response rate. Responses to the white

key produced a "click" from the feedback relay and postponed the

next time-out for the amount of time specified by the RT interval.

In the absence of responding, time-outs followed one another at

5-sec intervals (i.e., TT interval=5 sec). Neither the TT or RT

interval timed during food presentations. Termination of each

20-sec time-out was signalled by a brief (0.25 sec) darkening of

the houselights in addition to reillumination of the key with white

light. The RT interval length was manipulated across phases in

successive order from 20 to 10 to 40 to 5 sec, while maintaining

the TT interval equal to 5 sec. Each condition remained in effect

until response rates in time-in and numbers of time-outs presented

per session showed minimal variablility and no apparent trends for

20 consecutive daily sessions as determined by visual examination

of graphic displays of these measures. The number of sessions

each subject was exposed to the different RT interval values is

presented in Table 1.

Since all subjects had histories involving magazine training

and key pecking, little preliminary training was involved. One

subject (P-7820) was exposed to a variety of preliminary "pilot"

conditions, during which a number of variables (e.g., session


(sec) (ORDER) ____________
Ti e -in T i'e-t

P-782n 5 72 43.62 0.71 41.7
(4) (36.00-53.10) (0.24-1.5 ) (33-53)
10 96 20.82 0.33 46.3
(2) (16.50-25.32) (0.10-0.78) (35-65)
20 171 12.18 0.24 25.0
(1) (10.50-15.36) (0.00-0.86) (9-39)
40 71 7.80 0.10 13.0
(3) (6.42-11.16) (0.24-1.54) (5-32)

P-9275 5 42 21.72 0.27 105.2
(4) (17.34-26.10) (0.14-0.49) (95-114)
10 100 17.58 0.19 41.0
(2) (15.06-19.86) (0.00-0.44) (30-72)
20 74 8.94 0.14 31.5
(1) (7.14-10.44) (0.00-0.44) (24-47)
40 30 5.28 0.10 27.5
(3) (4.20-6.35) (0.00-0.25) (17-35)

P-6441 5 69 39.16 2.53 25.0
(4) (32.70-42.60) (1.29-3.95) (13-42)
10 45 20.83 1. 2 11.3
(2) ( P ;. 0-26.6 } ( .17--.71) (5-1s)
20 I 12. 0' .7; .
(1) (I0.44-Ii.]05 -. ". ;) (7-33)
30 : 9.50 r.r :. .
(:) (5.,0-13.? ) 7i .: -:. (C-C B)

"Vai L..:; shc,'., are tx.' ,- ^; ,'- ;: rv' 'r:[ :-, '.: : y''., : :r rr
ed.h '*.i in it-rx 'l ,r.::s r-: _.*. '.', : s--..r';ed u.'i::. th'. :rrL :- --- r:. .. ., .- -

length, stimulus conditions, feeder cycle length, etc.) were un-

systematically manipulated. The others were placed directly un-

der the first experimental condition, with two slight exceptions.

First, the feeder cycle was initially set at 4 sec. This value

remained in effect until visual observation showed that each sub-

ject ate readily following the presentation of the magazine, at

which point the cycle duration was reduced and subsequently main-

tained at 2 sec. Second, the probability of food presentation

each sec was adjusted so that approximately one-half of the ini-

tial time-in periods contained a food presentation. This value

was maintained until the first key peck, after which it was re-

duced to the value used subsequently throughout the experiment

(i.e., 0.03).


Both subjects initially placed directly under the experi-

mental procedure pecked the response key within 15 min of the be-

ginning of the first session. For all subjects, pecking occurred

first during time-in, and always occurred more frequently during

time-in than during time-out. Informal visual observations sug-

gested that pecks during time-out generally occurred during the

first few sec of a time-out, and often were initiated prior to

time-out presentation.

Table 1 presents the means and ranges of summary measures

obtained across the last 20 sessions of each condition for each

subject. Response rates during time-in, graphically depicted in

Fig. 1, were generally linearly decreasing functions of RT inter-

val length when plotted on double-log arithmetic axes. (The cor-

respondence between the mean and median response rates observed

during the last 20 sessions under each RT interval suggested that

daily session variability in this measure was generally normally

distributed.) The exceptions to this generalization were the data

from P-9275, whose response rate deviated from the linear function

at the shortest RT interval. In contrast, response rates during

time-out were not reliably different under the different condi-

tions. Although the mean time-out response rate in some cases

showed systematic decreases with increases in RT interval length,

the ranges of values at all RT intervals show considerable over-


The relationship between RT interval and the number of time-

outs delivered per session (see Fig. 2) was less consistent across

subjects than that between RT interval and time-in response rate.

Although there was a tendency for time-out presentations to de-

crease with longer RT intervals, these decreases were not syste-

matically related to RT interval length. It is of interest to

note the large differences in response rate under different RT in-

tervals in the presence of similar frequencies of time-out presen-

tation. For example, compare response and time-out rates for P-7820

under RT interval=5 and 10 sec and for P-6441 under RT interval=5

and 20 sec.

Figure 1. Response rates during time-in for each subject
as a function of the RT interval. Points, horizontal lines
and vertical bars represent means, medians and ranges of values
observed, respectively, during the last 20 days exposure to
each RT interval. Points have been slightly displaced to in-
crease clarity. Note the logarithmic axes.

o P-7820
a P-9275
e P-6441




3- E A;~ 't ~i e~!










Figure 2. Numbers of time-outs delivered to each subject
as a function of the RT interval. Figure characteristics are
the same as those for Fig. 1.


O -)

( 10- '
H-- 5
a P-7820
E P-9275
m P-6441
1 I 2
5 10 20 40

Cumulative response records obtained under the different RT

interval values are presented for each subject in Fig. 3. Response

rates were fairly constantwithin sessions,with occasional bursts

following time-out (some examples of these have been denoted in

the figure by arrows). Conspicuously absent from these records

are any large "warm-up" effects (i.e., decreased response and in-

creased aversive stimulus rates early in the session; see, e.g.,

Hineline, 1978a, 1978b; Sidman, 1966) characteristically shown

by rats, but as yet unreported in pigeons, under comparable shock-

postponement procedures. Although a careful analysis of such ef-

fects was not conducted, informal observation suggested that re-

sponse rates were usually higher early in the session than late

(e.g., see records for P-7820 at RT interval=5 and 10 sec, P-9275

at 5 and 20 sec, and P-6441 at 5 sec).

Patterns of responding within the RT interval are presented

in Fig. 4. (Due to a program malfunction data are available for

only some conditions.) Shown are the relative frequencies and

conditional probabilities (i.e., IRTs/Op) of IRTs in each fifth

of the RT interval averaged across the last 5 sessions under each

condition. All subjects showed a high relative frequency of IRTs

in the first class interval ("bin"), usually with monotonically

decreasing frequencies thereafter. Three of the eight condi-

tional probability distributions, however, show clear increases

in the conditional probability of an IRT at values longer than the

second bin. Subject P-7820 produced U-shaped distributions of

Figure 3. Representative cumulative records for each subject under the different
RT intervals. In each record, time reads from left to right. The upper (response)
pen stepped vertically with each response during time-in, reset to baseline after 550
responses, and was deflected during time-out (and the motor stopped). Defections of
the lower (event) pen denote food presentations. Records were selected from sessions
that most closely approximated both the median response rate and number of time-outs
delivered at each RT interval.

/ P-7820

R/1 /







R-T=20" .

R-T 40"





r~r'r-.r.r in ---e -- r. ,- --

30 min

"' '"~ "~-"-"""""'~""" -"~-' 'T~~'


Figure 4. Relative frequency (filled circles) and con-
ditional probability (open circles) distributions for IRTs in
fifths of the RT interval. Points and vertical bars represent
means and ranges, respectively, of values observed during the
last 5 sessions at each RT interval. (Due to an apparatus mal-
function, data are only available for some RT intervals.) Ranges
of points without vertical bars are contained within the point.

R-T=5 sec


R-T=20sec R-T=40 sac



m1 7

(I-sec bins)


1 2 3 4 5
12 45
(2-sec bins)

(4-sec bins)


1 2 345
(8-sec bins)

















IRTs/Op at all RT intervals, although the differences in condi-

tional probabilities between IRT classes were small at RT interval=

10 sec. Interresponse-time distributions of P-9275 showed little

temporal control over responding by RT interval at any RT inter-

val length, whereas distributions for P-6441 showed temporal con-

trol over IRTs at RT interval=5 sec but not 20 sec.

The development of steady-state performance occasionally re-

quired a great deal of exposure to the different RT interval values,

due more to short-term oscillations in response rate than to grad-

ual transitions between experimental phases. Indeed, initial tran-

sitions were generally very rapid. Data depicting such transi-

tions for P-6441, a representative subject, are presented in Fig. 5

in the form of cumulative response records. On the left are cumu-

lative records obtained during the last day of exposure to one RT

interval, and on the right those obtained under the first day of

exposure to the subsequent RT interval. Comparison of records on

the right with those immediately below it on the left (the first

and last day of exposure to a specific RT interval, respectively)

reveals that response rates approximating those obtained after

extended exposure were generally obtained by the last third of the

first session under the new RT interval value.


Key pecking was reliably and quickly engendered in all sub-

jects under the present procedure. It should be noted that the

Figure 5. Cumulative records depicting transistions in
key pecking of P-6441 between different RT intervals. Records
on the left show responding during the last session at a specific
RT interval value, those on the right during the first session
of the subsequent value. The first and last day of exposure to a
particular RT interval is shown, respectively, ina record on the
right and the one immediately beneath it on the left. Recording
characteristics are the same as those for Fig. 3.


S- 41 ,i- T=20,sec Rm- R-T=1O sPc

/ /

R-T-O sec R-T=40 sac

Tr ner a'i 7 ve T ws a i. Is w sfs s i n ii/ifa-u r rvn e man rs / TfmiiaTi uTreai ear irsse assas111iT'r
R-T 40 sec i-R-T=5 se

R-T=5 secj --R-T=20 sec

"TT-T T ~riri.30t w m i rf lr r ry 'p' a r .. s i-. i m a s f f r i l m' s, r
30 mrin

procedure used to establish key pecking in two of the subjects

(P-9275 and P-6441) was merely a modification of the procedure

used by Brown and Jenkins (1968) to "autoshape" key pecking. Un-

der their procedure, one stimulus was reliably followed after 8 sec

by the presentation of food, while food was never presented in the

presence of a second stimulus. Under the present procedure, the

time-in and time-out stimuli, respectively, defined a similar

differential relation with respect to food delivery (see, e.g.,

Gamzu & Williams, 1971, 1973), with the exception that food pre-

sentation could occur anytime during the presence of the former

stimulus, not just following it.

The linear relation obtained in the present study between

time-in response rate and RT interval (when plotted on double-

log axes) was comparable to ones reported by Thomas (1965a), who

used a similar procedure, and by others using analagous shock-

postponement procedures (e.g., Clark & Hull, 1966; Klein & Ril-

ling, 1972; Sidman, 1953; Todorov, Ferrari & de Souza, 1974),

suggesting that responding was maintained in similar ways.

Whether response rates would have decreased at even shorter RT

intervals, as it does in other species (e.g., Sidman, 1953) can-

not be readily determined, since RT intervals shorter than the

TT interval were not examined. However, P-9275's deviation from

the linear function towards a lower response rate at RT interval=

5 sec may be indicative of that kind of process.

Although the functions obtained were similar, the absolute

response rates observed in the present study were much greater

than those obtained by Thomas (1965a). Response rates observed by

Thomas ranged frci approximately 0.4-3.6 R/min, but from approxi-

mately 5.0-40.0 R/min under the present procedure. Given the num-

ber of procedural differences between the two studies, it is im-

possible to interpret these differences. However, differences in

time-out duration and/or rate of food presentation, while producing

different absolute response rates, did not substantially change

the form of the functions obtained here relating response rate

during time-in to RT interval from that observed by Thomas. This

suggests that the "behavioral process" represented by this func-

tion may be largely independent of variables other than the length

of response-contingent delay of time-out.

One procedural feature seemed to enhance performance greatly

during initial pilot studies. During pilot experiments, as in

Thomas's study, the time-in and time-out stimuli were signalled

originally only by colors of the keylight, with the houselight

remaining continuously illuminated. Under this pilot procedure,

presentation of the time-out stimulus was followed typically by

the subject turning away from the stimulus and engaging in other

behavior (e.g., pecking the floor, preening, etc.). As such, the

subject would typically be facing away from the response key when

time-out ended. Often, the subsequent time-in period would e-

lapse and another time-out ensue. A brief darkening of the house-

light at time-out termination was therefore employed in an attempt

to provide a discriminative stimulus which would be effective

regardless of the subject's orientation with respect to the re-

sponse key. The addition of this stimulus resulted in rapid ter-

mination of other ongoing behavior when time-in began and immedi-

ate approach towards the response key, thus enhancing performance.

The patterns of responding maintained under the present pro-

cedure were also highly comparable to those reported for rats bar

pressing under comparable shock-postponement procedures (e.g.,

Boren, 1961; Ellen & Wilson, 1964) with the exception of the ap-

parent absence of a large warm-up effect (e.g., Hineline, 1978a,

1978b; Sidman, 1966). If further analysis substantiates the lack

of warm-up, it may lend credence to Hineline's (1978a) sugges-

tion that such effects reflect habituation of non-operant behaviors

evoked by aversive stimulation. It is possible that elicited key

pecking directed at the stimulus signalling food delivery inter-

acted with negatively-reinforced key pecking in ways functionally

similar to the interactions suggested by Hineline. However,

rather than "competing" with operant behavior as Hineline sug-

gests it does under free-operant shock-postponement procedures,

elicited behavior under the present procedure may have combined

additively with negatively-reinforced key pecking since the food-

correlated stimulus was located on the response key. Hence, re-

sponse rates early in the session might be expected to be increased,

rather than decreased early in the session, due to this transient

additive influence. There are presently no data describing within-

session changes in elicited pecking to either support or refute

this interpretation.

The inconsistent development of temporal control over re-

sponding by RT interval length (as evidenced by increasing con-

ditional IRT probabilities with increases in IRT length) across

subjects and RT intervals cannot at the present time be inter-

preted with confidence. Sidman (1966) suggested that such con-

trol may occasionally develop slowly, and that stable rates of

responding may be maintained without concomitant evidence of tem-

poral control by RT interval length. Although each RT interval

was in effect for a comparatively large number of sessions during

the present study, the number necessary for this type of control

to develop (if it does reliably develop) is at the present time

unknown. It is clear from the present results that such control

did not develop rapidly. However, since RT intervals were changed

during the present study without regard to the distribution of

IRTs, it is possible that extended exposure to each value would

have resulted in increased temporal control over responding. It

may be of interest in this regard that P-7820 (who most consis-

tently evidenced some degree of temporal control over responding)

was generally exposed to each RT interval longer than the other


In any event, it is clear that the vast majority of IRTs

fell within the range specified by the first bin in each dis-

tribution. It has been suggested that certain topographical

characteristics of key pecking may result in the transduction

of extremely short (less than 0.8 sec) IRTs which do not appear

to be subject to control by positive reinforcement contingencies

(e.g., Blough, 1966; Shimp, 1973). Whether this is the case here

is difficult to ascertain given the relatively large class inter-

vals used in the IRT analysis.

Although the present results suggest comparability between

free-operant avoidance of time-out and avoidance of other aver-

sive events, other interpretations are possible. First, the

absence of a delay between a key peck and subsequent presenta-

tion of food may have led to the "superstitious" maintenance of

key pecking. Although this claim cannot be refuted directly, it

is difficult to see how decreases in the RT interval could pro-

duce increases in the frequency of accidental correlations be-

tween responses and food presentations necessary to account for

the systematic increases in response rate observed at progres-

sively shorter RT intervals. It might, in fact, be easier to

argue the converse. That is, the frequency of spurious cor-

relations might be expected to increase with increases in the

amount of time-in time (and, hence, at longer RT intervals).

However, it is possible that response rate increases initially

produced by decreases in RT interval length may subsequently in-

crease the probability of close contiguity or frequency of dif-

ferential accidental correlations between responses and food

presentations. These correlations may then further increase

response rate, thus tending to overestimate the increase pro-

duced by the postponement contingency alone. However, such

effects depend on the initial modification of behavior by the

postponement contingency, and cannot, therefore, suggest that

the avoidance contingencies were of little importance.

A second alternative interpretation of the present results

involves the role of elicited key pecking under procedures in-

volvina differential stimulus-reinforcer contingencies. As pre-

viously discussed, food-deprived pigeons will peck at stimuli

differentially correlated with the delivery of food. Addition-

ally, the rate of pecking appears to increase as the duration of

a signal differentially correlated with food presentation de-

creases relative to the duration of stimuli signalling either

the absence of food (e.g., Baldock, 1974; Terrace, Gibbon, Far-

rell & Baldock, 1975) or a decreased frequency of food presen-

tation (Spealman, 1976). Applied to the present situation, the

relative duration of the time-in stimulus (signalling imminent

food presentation) decreased with increases in the number of time-

outs (signalling the absence of food) presented. Since the num-

ber of time-outs delivered tended to increase with decreases in

the RT interval (and thus the relative "time-in" time decreased),

the increased rates of responding at shorter RT intervals might

be ascribed solely to increases in the rate of elicited key peck-

ing. Although the large differences noted in response rates un-

der the different RT interval values in the presence of similar

rates of time-out presentation (see Figs. 1 and 2, particularly

data for P-7820 at RT interval=5 and 10 sec and for P-6441 at

RT interval=5 and 20 sec) argue against such an interpretation,

it might be argued that, while frequencies of time-out under the

different conditions were comparable, the distributions of inter-

time-out-intervals (times between successive time-out presenta-

tions) may not have been, thus producing differences in the rate

of elicited pecking. Data obtained in the second and third ex-

periments suggest, however, that an interpretation based solely

on elicited key pecking cannot account for the performance gen-

erated under this procedure.


Interpretation of the results of Experiment I solely in terms

of elicited key pecking would suggest that the response-contingent

postponement of time-out was of little importance in the mainte-

nance of different rates of key pecking except in providing a

changing number of time-out presentations. If this interpretation

is correct, then presentation of similar distributions of time-

out independently of responding should produce little change in the

rates of key pecking maintained. Conversely, interpretations

based on the negative reinforcement provided by response-contingent

delay of aversive stimulation would suggest that removal of such a

contingency (i.e., programming "extinction") would result in

decreases in response rate.

A number of extinction procedures have been used in the analy-

sis of behavior maintained under free-operant avoidance procedures.

The first consists simply of no longer presenting the aversive

stimulus, and generally results in fairly rapid reductions in

response rate (e.g., Boren & Sidman, 1957b; Schnidman, 1958).

Although the effects of this manipulation are similar to extinc-

tion of positively reinforced behavior (where responding no longer

results in the presentation of reinforcement), it has been argued


that such a manipulation decreases responding through behavioral

mechanisms other than extinction qua extinction. Davenport, Coger

and Spector (1970), for example, have argued that the removal of

shock reduces response rates either by removing the "motivation"

that usually increases the effectiveness of aversive-stimulus delay

as a reinforcer (much as pre-session feeding decreases responding

maintained by response-contingent food presentation), or by rein-

forcing all responses equally, since all responses "delay" aver-

sive stimulation for equal amounts of time. They suggest that a

more "proper" extinction procedure would involve elimination of

any response-contingent delay to the next aversive event, not

elimination of the event itself, since the former, not the latter,

presumably constitutes reinforcement under free-operant avoidance

paradigms. Thus, for example, they would argue that, after train-

ing under a free-operant shock-postponement procedure with an RS

interval=15 sec and an SS interval=5 sec, extinction would involve

presentation of shock every 5 sec, independent of responding. This

procedure does generally decrease response rates (e.g., Davenport,

Coger & Spector, 1970; Davenport & Olson, 1968) but also produces

a more or less drastic change in the prevailing stimulus conditions,

depending on the level of responding maintained originally under

the postponement procedure. That is, for an animal reliably

responding within the RS interval (and thus reliably delaying

shock), the sudden presentation of shock ever 5 sec serves as a

highly discriminable change in the prevailing contingencies. Coulson,

Coulson and Gardner (1970), recognizing this fact, have suggested

a third procedure for the extinction of negatively reinforced beha-

vior, involving the suspension of any response-contingent delay

while maintaining comparable frequencies and distributions of

response-independent aversive-event delivery. This procedure is

also effective in reducing response rates, however, responding

generally decreases more slowly and does not cease altogether

(Coulson, Coulson & Gardner, 1970; Smith, 1973).

This last procedure is most relevant to the question of the

role of elicited key pecking under the time-out delay procedure

of Experiment I. To reiterate, key pecking maintained under the

time-out-delay procedure should not be greatly affected by response-

independent presentation of similar temporal distributions of time-

outs if such responding is solely elicited by occasional presenta-

tion of the time-in stimulus. If, however, key pecking under this

procedure depends on response-contingent delay of time-out, remov-

al of the delay contingency, even in the presence of continued oc-

casional presentation of time-out, should reduce response rates.

The present experiment examined responding maintained under such

a procedure and provided data on the recoverability of the per-

formance engendered during Experiment I.


Subjects and Apparatus

The subjects and apparatus used were the same as in Experi-

ment I.


Directly following exposure to RT interval=5 sec under Experi-

ment I, each subject was returned to a previously examined RT inter-

val; P-7820 to RT interval=10 sec, P-9275 to 40 sec and P-6441 to

20 sec. Exposure to a particular RT interval value was not ran-

domly determined. Subject P-6441 finished the initial series first

and subsequently was returned to the first value studied (i.e.,

RT interval=20 sec). In an attempt to minimize differences in the

number of time-outs presented to each subject, P-9275 was subse-

quently exposed to RT interval=40 sec. Subject P-7820 was exposed

to RT interval=10 sec in order to obtain data, albeit between sub-

jects, at each of the three longer RT intervals. Responding was

allowed to stabilize at these values according to the stability

criterion outlined in Experiment I, and the sequential inter-time-

out-intervals during each session were recorded. Subsequently,

the delay contingency was suspended, and time-outs were occasion-

ally presented at variable times independently of responding, with

distributions "yoked" to those of the previous phase by program-

ming inter-time-out-intervals equivalent to those obtained during

the 20th preceding session (when the delay contingency was in

effect). That is, the distribution of sequential inter-time-out-

intervals from each of the last twenty sessions under the time-out

delay contingency was programmed in the same ordinal position for

a single session. Since time-out presentations occurred at varying

times response-independently and were yoked to presentations under

the delay contingency, this condition was termed a "yoked variable-

time" (yoked-VT) schedule of time-out presentation. All other para-

meters of this procedure were equivalent to those in Experiment I.

The yoked-VT schedule remained in effect for 20 sessions, at which

point the delay contingency was reinstituted at the RT interval

value prevailing prior to implementation of the yoked-VT schedule.


Absolute response rates and numbers of time-outs presented

under the initial reexposure to the RT interval and the subsequent

yoked-VT schedule are presented for each subject in Table 2. For

each subject, the rate of responding during reexposure to the RT

interval was very similar to that obtained during exposure to that

value in Experiment I. The correspondence between the number of

time-outs delivered during the first and second exposures was also

good, although differences may be noted (e.g., data for P-7820).

Response rates maintained under the postponement procedure

and the subsequent yoked-VT procedure are shown, as a percent of

the mean response rate under the postponement procedure, in Fig. 6.

(Absolute response rates and numbers of time-outs presented during

each session shown in Fig. 6 have been included in the Appendix.)








P-7820 RT'interval=10 sec,


P-6441 RT interval=20 sec


P-9275 RT interval=40 sec




















aV, lues are the means of the
each procedure. Ranges are

last 20 (delay conditions) or 5 (yoked-VT)
shown in parentheses.

sessions under






-I---- -------------

Figure 6. Daily session response rates for each subject
under the delay and yoked-VT contingencies. Data were taken
from the last 20 sessions under the delay contingency (to the
left of the vertical line) for from the 20 sessions the yoked-VT
contingency was in effect (to the right). Response rates are
expressed as a percentage of the mean rate during the last 20
sessions under the delay contingency.

R-T=O sec



R-T=20 sec

Yoked VT TO


Yoked VT:TO

10 20 30 40





Removing the response-contingent delay decreased rates of respond-

ing during time-in by at least 50% in every subject within the 20

sessions the yoked-VT schedule was in effect. Neither the degree

of relative suppression nor the rate with which responding was sup-

pressed appeared to be systematically related to the RT interval

value under the postponement procedure. Absolute response rates

under the yoked-VT procedure were, by contrast, inversely related

to the RT interval (after 20 sessions exposure to this procedure).

Intrasession patterns of responding maintained in each subject

under the delay and the yoked-VT procedures are depicted in Fig. 7.

Cumulative response records presented were taken from the last

twenty sessions under the delay procedure or from the twenty days

the yoked-VT schedule was in effect. Three pairs of records are

shown for each subject, taken from the 1st, 11th and 20th session.

The pairs of records are from sessions during which the delay con-

tingency was in effect (labelled "R"), and from the sessions under

the yoked-VT procedure (labelled "Y") which were matched with re-

spect to time-out presentation to the former. (The records have

been overlayed to highlight the correspondence between inter-time-

out-intervals under the two procedures.) While response rates un-

der the yoked-VT procedure generally progressively declined with

extended exposure to this contingency, no systematic differences

were observed in the patterns of responding. The subjects exposed

to the two longer RT intervals prior to the yoked phase (i.e.,

P-6441 and P-9275) both paused for long periods during the middle

Figure 7. Cumulative records from selected sessions under the delay contingency
and under the corresponding Yoked-VT session. Records were taken from the 1st, ilth or
20th session of the last 20 sessions the delay contingency was in effect and of the Yoked-VT
procedure. Records from sessions under the delay contingency (labeled "R" in the ficiure)
have been overlapped with records from corresponding sessions under the yoked-VT procedure
(labeled "Y" ). Event pen tracings have been removed fom each record, otherwise recording
characteristics are the same as those for Fig. 3. (Occassional extended deflections of the
response pen were the result of an infrequent mechanical failure which did not stop the
motor during time-out.)

20th Session

~--- 2-

(R-T-40) R

30 M
30 Min



ist Session

Ilth Session

of the 20th session under the yoked-VT procedure, and responded

earlier and later in the session at rates reduced below those un-

der the delay procedure. Such intrasession variations in response

rate were not observed with P-7820, who, although responding at

rates lower than those under the delay contingency, still responded

at substantial rates after 20 days under the yoked-VT procedure.

Reinstatement of the postponement contingency increased re-

sponse rates to values comparable to or above those observed prior

to the yoked-VT procedure. The length of this transition varied

between subjects, with P-7820, P-6441 and P-9275 responding at

rates equal to or above the mean rate under the postponement pro-

cedure after 10, 25 or 1 session, respectively.


The results of Experiment II indicated that 1) response rates

(and to a lesser degree time-out rates) obtained under reexposure

to an RT interval value were highly comparable to those obtained

under initial exposure to that value during Experiment I, and 2)

the delay contingency was essential in maintaining the rates of

responding observed under free-operant avoidance procedures when

time-out is used as an aversive stimulus. The decreases in response

rate observed when distributions of response-independent time-outs

were exactly matched to those of the immediately preceding response-

contingent delay procedure argue against the notion that key pecking

observed under the delay procedure was governed solely by stimulus-

reinforcer contingencies.

Although the present results indicate that the delay contin-

gency was crucial in determining the rates of responding engen-

dered, they do not rule out the possibility that elicited key peck-

ing directed at the time-in stimulus occurred and thus contributed

to the overall rate of responding. Recall that the absolute rate

of responding under the yoked-VT procedure was inversely related

(after 20 sessions exposure) to the value of the preceding RT inter-

val value. At least two interpretations of this effect are possible.

First, it might be suggested that shorter delays occur between

responses and subsequent response-independent time-outs as the aver-

age inter-time-out-interval decreases. This may increase the like-

lihood that "superstitious" negatively-reinforced responding will

occur at higher rates when the yoked distribution is taken from a

preceding condition involving a shorter RT interval (and hence,

possibly shorter inter-time-out-intervals). Alternatively, it

might be argued that decreasing the RT interval increases the num-

ber of time-outs delivered (or the probability of short inter-time-

out-intervals) and thus engenders more elicited pecking by decreas-

ing the relative amount of "time-in" time (cf., Terrace, Gibbon,

Farrell & Baldock, 1975). Both the "superstitious maintenance"

and the elicitationn" notions rest on the assumption that the num-

ber of time-outs delivered increases with decreases in the RT inter-

val. Although an attempt was made to minimize differences between

subjects in the number of time-outs delivered during the yoked-VT

condition, some between-subject differences way be noted (see

Table 2).

These two interpretations cannot be evaluated independently

under the procedures used thus far; either, both or neither could

be correct. Assuming the elicitation notion to be true, it is

possible that the slopes of the function relating response rate

during time-in to RT interval obtained during Experiment I are

greater than they would be in the absence of elicited behavior

directed at the stimuli on the response key (given the added

assumption that shorter RT intervals elicit more signal-directed

pecking). It is entirely possible that key pecking is controlled

by response-contingent delay of time-out, but not by the length

of the consequent delay. Experiment III provided evidence sug-

gesting, however, that this is not the case.


In food-deprived pigeons, pecking directed at some stimulus

typically occurs only when that stimulus signals a higher fre-

quency or probability of food presentation than that signalled in

the absence of that stimulus (e.g., Brown & Jenkins, 1968; Gamzu &

Williams, 1973; Keller, 1974). Additionally, the probability and/

or rate of elicited pecking is inversely related to both the rela-

tive and absolute duration of the positively-correlated stimulus

(e.g., Baldock, 1974; Spealman, 1978; Terrace, Gibbon, Farrell &

Baldock, 1975). Hence, a continuously present stimulus (i.e.,

one of long duration which is non-differentially correlated with

the presentation of food) should be less effective in eliciting

pecking than shorter, differentially-correlated stimuli.

Keller (1974) developed a procedure under which key pecks to

one operandum (the "food" key) occasionally resulted in the pre-

sentation of food while responses to a second key had no scheduled

consequences. Presented on this latter key (the "signal" key)

were stimuli correlated with the availability of reinforcement

contingent on responses to the food key which was continuously il-

luminated with a single stimulus. Keller argued that responses

to the signal key were predominantly elicited key pecks, since


University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - Version 2.9.7 - mvs