You are on page 1of 13

Psychological Bulletin Copyright 1988 by the American Psychological Association, Inc.

1988. VolT 103, No. 1,44-56 0033-2909/88/J00.75

Behavioral Momentum and the Partial Reinforcement Effect

John A. Nevin
University of New Hampshire

Free-operant behavior is more resistant to change when the rate of reinforcement is high than when
it is low. The usual partial reinforcement extinction effect, demonstrating greater resistance to extinc-
tion after intermittent than after continuous reinforcement, seems to contradict this generalization.
However, most free-operant extinction data are reported as response totals, which confound the
initial levels of responding and the rate at which responding decreases over the course of extinction.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

A reanalysis shows that after extended training, the slope of the extinction curve is shallower after
This document is copyrighted by the American Psychological Association or one of its allied publishers.

continuous reinforcement than after intermittent reinforcement, suggesting greater rather than less
resistance to change. These results, which hold for both independent-groups and within-subject com-
parisons, support the general finding that resistance to change of free-operant behavior is a positive
function of the rate of reinforcement. This generalization does not, however, hold for discrete-trial
performance. I discuss some consequences of these analyses for applications of behavioral research
results.

A series of experiments conducted over the past dozen years Generality and Consistency
supports a simple generalization: The greater the rate of rein-
forcement for a free-operant response, the more it resists change In addition to pigeons, rats and monkeys have served as sub-
(Nevin, 1974, 1979; Nevin, Mandell, & Atak, 1983; Nevin, jects in experiments of this sort, and many other procedures
Mandell, & Yarensky, 1981). A standard experiment uses pi- for assessing resistance to change have been used: introducing
geons as subjects, deprived to 80% of their free-feeding weights. concurrent reinforcement for an alternative response, either sig-
After brief training to eat from the grain magazine in a conven- naled (Nevin et al., 1981) or unsignaled (Pliskoff, Shull, & Gol-
tional pigeon chamber and then to peck the key, with food after lub, 1968); varying hours of deprivation (Carlton, 1961)orbody
each peck, the birds are exposed to a multiple schedule of rein- weight (Herrnstein & Loveland, 1974); presenting stimuli sig-
forcement in which two distinctive lights on the key signal two naling unavoidable electric shock (Blackman, 1968; Lyon,
different schedule components. For example, the key might be 1963); punishing the reinforced response (Bouzas, 1978); in-
lighted red for 1 min, during which pecks would be reinforced creasing response effort (Elsmore, 1971); and simply tracking
at variable intervals averaging 1 min (60 reinforcers per hr), and response rates through circadian changes in activity (Elsmore
then green for the next 1 min, during which pecks would be &Hursh, 1982).
reinforced at variable intervals averaging 3 min (20 reinforcers Whatever the assessment procedure, it must be applied
per hr). (Such a schedule would be denoted mult VI 1-min, VI equally to the two components: For example, if punishment is
3-min.) Training would continue until response rates in both used to assess resistance to change, it must be arranged so that
components became stable—perhaps 40 daily 1-hr sessions. experienced shock rates are the same in both components, even
Then the resistance to change of these asymptotic response when response rates differ. Prefeeding and deprivation proce-
rates would be assessed by introducing a new variable uni- dures are especially useful in this respect, because motivation
formly with respect to both components, such as free food dur- levels and internal stimulation are the same regardless of which
ing dark-key periods between schedule components (e.g., component is in effect. Also, all these procedures leave the base-
Nevin, 1974) or prefeeding in the home cage (e.g., Eckerman, line schedule of reinforcement unchanged. Under these condi-
1968). Both procedures have a common result: The rate of re- tions, the results are uniform and clear: Relative to their base-
sponding changes less, relative to its baseline training level, in line levels, the rate of responding in the component with the
the component with the higher rate of reinforcement (red, VI higher rate of reinforcement is less affected than the rate of re-
1-min in this case). sponding in the component with the lower rate of reinforce-
ment.
Normally, baseline response rate is higher in the component
with the higher rate of reinforcement. However, resistance to
[ am indebted to L. Slade for assistance with literature searches and change is independent of baseline response rates. For example,
data analyses, to E. Fickett for help with the conduct of the study re-
Nevin (1974, Experiment 5) established low response rates in a
ported in this article, to W. Baum and V. Benassi for many stimulating
VI 1-min component and high rates in a VI 3-min component
discussions, and to two reviewers for helpful comments on an earlier
version of this article.
by means of tandem interresponse-time requirements. These
Correspondence concerning this article should be addressed to John rate differences are opposite to those usually obtained with
A. Nevin, Department of Psychology, University of New Hampshire, standard VI schedules. Nevertheless, resistance to change rela-
Durham, New Hampshire 03824. tive to baseline remained greater in the VI 1-min component.

44
BEHAVIORAL MOMENTUM AND THE PARTIAL REINFORCEMENT EFFECT 45

Path, Fields, Malott, and Grossen (1983) repeated Nevin's Conversely, the data reviewed above suggest that rate of rein-
(1974) method of using tandem interresponse-time require- forcement (but not response rate) determines resistance to
ments to produce different response rates but arranged identi- change relative to baseline.
cal reinforcement rates in both components and found no The rate of reinforcement obtained by a free-operant re-
difference in resistance to change relative to baseline levels. sponse in a signaled schedule component may affect resistance
Conversely, if the training schedules establish similar response to change either because the response is reinforced more often
rates despite differences in reinforcement rates, resistance to (an operant, response-reinforcer relation) or because the stimu-
change remains greater in the component with the higher rate lus signaling that component enjoys a more favorable correla-
of reinforcement (e.g., Blackman, 1968;Nevinetal., 1983). tion with reinforcement (a Pavlovian, stimulus-reinforcer rela-
The findings of research on free-operant resistance to change tion). To evaluate these alternatives, Nevin (1984) maintained
are consistent with the findings of research on schedule prefer- key pecking with a common VI 2-min schedule in three initial
ence. A standard procedure, known as concurrent chain sched- multiple-schedule components that were followed by noncon-
ules, provides intermittent access to one signaled schedule for tingent transitions to different signaled reinforcement rates.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

responses on one of two simultaneously available choice keys One was followed by a higher rate of reinforcement, the second
This document is copyrighted by the American Psychological Association or one of its allied publishers.

and intermittent access to a second schedule, signaled by a by the same rate, and the third by nonreinforcement. Respond-
different stimulus, for responses on the other key. In this situa- ing was more resistant to prefeeding or extinction in the VI 2-
tion, subjects uniformly prefer the schedule with the higher ob- min component that preceded the higher rate of reinforcement
tained rate of reinforcement, regardless of differences in re- than in the VI 2-min component that preceded nonreinforce-
sponse-reinforcer contingencies arranged by the schedules and ment. Thus, differential stimulus-reinforcer relations in a stim-
consequent differences in response rate (e.g., Autor, 1969; ulus sequence affected resistance to change in initial compo-
Herrnstein, 1964; see Nevin, 1979, for review of data relevant nents in which operant, response-reinforcer relations were
to resistance to change). The magnitude of the preference varies equated.
with a number of procedural parameters, but the direction of Nevin et al. (1987) explored a more complex procedure that
preference is clear. Thus, the generality of the ordinal relation compared response-contingent and noncontingent transitions
between resistance to change and reinforcement rate across di- to components with a higher rate of food reinforcement or with
verse assessment procedures fits nicely with the ordering of nonreinforcement. They found that contingent transitions to a
schedule preferences. higher rate of reinforcement maintained higher response rates
in the initial component than did noncontingent transitions,
demonstrating that an upward transition in the rate of food re-
Measurement Considerations
inforcement was itself a reinforcer. However, relative resistance
The measure of resistance to change is (B0 — BX)/B0, where to extinction was unaffected by this operant contingency when
B0 is baseline response rate, x is the value of the new variable the Pavlovian, stimulus-reinforcer relations were the same.
used to assess resistance to change, and Bx is the response rate Nevin et al. (1987) also repeated Nevin's (1984) finding that
at that value. The smaller the value of this measure, the greater resistance to extinction was greater in the initial component
is resistance to change. When this measure is plotted as a func- that preceded an increase in rate of reinforcement than in the
tion of AT, the slope of the function characterizes resistance to initial component that preceded nonreinforcement when both
change: The shallower the slope, the greater is resistance to transitions were noncontingent. Nevin et al. (1987) therefore
change. suggested that maintained response rate is determined primar-
In many of Nevin and his associates' analyses, response rates ily by operant, response-reinforcer contingencies, whereas re-
relative to baseline are plotted on semilogarithmic axes, for sistance to change is determined primarily by Pavlovian, stimu-
three reasons. First, a lot of assessment operations produce low lus-reinforcer relations.
response rates, so that floor effects could obscure slope differ- Although this distinction may apply with special force to
ences; this problem is avoided by the use of logarithms, which within-subject, multiple-schedule research with differential
range to minus infinity. Second, the semilogarithmic functions stimulus-reinforcer relations arranged within sessions, it may
are often roughly linear, facilitating slope comparisons. Third, also apply to independent-groups research designs in which one
the slope of a semilogarithmic function is invariant with respect group obtains reinforcement at a high rate and thus experiences
to multiplication or division, so that slopes are unaffected by a strong environment-reinforcement correlation, whereas an-
changes in baseline values. (See Nevin, Smith, & Roberts, 1987, other obtains reinforcement at a low rate and thus experiences a
for further discussion of measurement issues.) relatively weak environment-reinforcement correlation. With
The arguments above suggest that rate of responding main- respect to the relation between rate of reinforcement and resis-
tained under constant baseline conditions and the slope of the tance to change, free-operant studies with independent groups
function characterizing the resistance to change of a free oper- are in qualitative agreement with multiple-schedule studies (see
ant are independent both at the level of measurement and func- below). However, the effects of response-reinforcer and envi-
tionally, in experimental data. Different processes may deter- ronment-reinforcer relations have not been separated in inde-
mine these distinct aspects of behavior. For example, ratio pendent-groups research designs.
schedules routinely generate higher response rates than do in-
terval schedules, even when the obtained rates of reinforcement
A Mechanical Metaphor
are the same. Thus, response rates are clearly determined by The independence of maintained response rate and resis-
the schedule contingencies as well as by rates of reinforcement. tance to change, coupled with the suggestion that these aspects
46 JOHN A. NEV1N

of behavior may depend on separate processes, led Nevin et al. havior modification may be seen as maximizing the momentum
(1983) to an analogy derived from Newtonian mechanics. A of desirable behavior and minimizing the momentum of unde-
moving body has both velocity and mass, and under constant sirable behavior.
conditions, it continues in motion at a constant velocity. When Given the necessity of measuring relative behavioral mass,
it is acted on by an external force, its velocity changes in propor- one must also measure behavioral momentum in relative terms.
tion to the force and in inverse proportion to its mass. If baseline Thus, if mi/m2 gives the mass ratio of two performances and
response rate under constant conditions is considered analo- Bt/B2 gives their velocity ratio, their momentum ratio is m\B\l
gous to the initial velocity and a new variable, such as prefeed- m2B2. Many studies of maintained response rate in multiple
ing, is taken as the external force, the change in response rate is schedules have shown that
inversely proportional to the behavioral analogue of mass.
Nevin et al. (1983) proposed a method for quantifying behav- Bi/B2 = (4)
ioral mass, by analogy to the fundamental Newtonian relation
where r\ and r^ are the rates of reinforcement obtained by Bt
and B2, respectively. The parameter a reflects sensitivity of re-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(1)
This document is copyrighted by the American Psychological Association or one of its allied publishers.

sponse-rate ratios to reinforcement-rate ratios, and c reflects


where v stands for velocity, F for force, and m for mass. inherent biases attributable to differences in the reinforcement
Rewritten in the behavioral notation above, contingencies that produce different response rates when the
reinforcement rates are equal. (See McSweeney, Farmer, Dou-
(Ba- a = x/m, (2) gan, & Whipple, 1986, for a thorough quantitative review of the
where (B0 - BX)/B0 is the change in response rate relative to literature.)
baseline (analogous to Au) and x is the value of the variable used Nevin et al. (1981) and Nevin et al. (1983) reported data con-
to assess resistance to change (analogous to F). Because this sistent with the conjecture that mass ratios are similarly related
term is dimensionless, behavioral mass (m) must be expressed to relative reinforcement rates:
in the same units as the assessment variable (x). But the units
of x vary from one experiment to another, according to whether m\lmi = (H/^)*, (5)
prefeeding, alternative reinforcement, punishment, or some where the notation is as above except that b reflects the sensitiv-
other procedure is used to assess resistance to change, thus pre- ity of mass ratios to reinforcement-rate ratios. Momentum ra-
cluding the possibility of a transsituational, absolute scale of tios are then given by the product of Equations 4 and 5:
behavioral mass. This problem may be resolved by evaluating
+b
the mass of one performance relative to another by finding a (6)
pair of values of the new variable, xt and x2, that produce equal
changes in response rate relative to baseline for the two perfor- If this conjecture is correct, relative momentum depends on rel-
mances. Then ative rate of reinforcement and on whatever operant or Pavlov-
ian factors determine the values of the exponents a and b and
the bias multiplier c. Many quantitative, parametric studies will
or be required for evaluating Equation 6 and assessing the deter-
(3) miners of its parameters.

That is, the ratio of the values of the assessment variable that Behavioral Momentum and Resistance to Extinction
give equal changes in the two performances, relative to their
baselines, is equal to the ratio of the behavioral masses of those This may or may not prove to be a fruitful approach to the
performances. Nevin et al. (198 3) showed that when mass ratios study of learned behavior, but before going much further, the
were measured in this way, the ratios were invariant with re- approach must deal with one apparent counterinstance to the
spect to the magnitude of the new variable and were internally generalization that more frequent reinforcement leads to
consistent across three sets of reinforcement rates, thus meeting greater resistance to change, namely, the well-known partial re-
the basic requirements of ratio-scale measurement. inforcement extinction effect (PRE). As every student of intro-
In classical physics, momentum is the product of mass and ductory psychology knows (or should know?), responding is
velocity. Likewise, behavioral momentum may be construed as more persistent during extinction after training with intermit-
the product of resistance to change (mass) and baseline re- tent reinforcement than after training with continuous rein-
sponse rate (velocity). Thus construed, behavioral momentum forcement (CRF). Clearly, any intermittent schedule arranges a
is closely identified with the general notion of learning. A per- rate of reinforcement lower than that provided by CRF. Thus,
formance is said to be well learned if it is executed frequently a serious problem arises.
and reliably during training and if it persists with little or no Resistance to extinction may in fact be a counterinstance to
impairment under new circumstances (high velocity and large the general relation between resistance to change and rate of
mass). The notion of learning would not be invoked, however, reinforcement—a serious counterinstance because of the theo-
if performance ceased altogether as soon as training ended (zero retical and practical significance of persistence in the face of
mass), whatever its response rate during training, or if the per- nonreinforcement. There may, however, be no conflict. Extinc-
formance never occurred (zero velocity). Thus, the effectiveness tion may give discrepant results because the schedules of rein-
of a training program may be characterized by its effects on forcement are no longer in effect during the assessment proce-
behavioral momentum. For practical purposes, the goal of be- dure, unlike all other assessment procedures reviewed above.
BEHAVIORAL MOMENTUM AND THE PARTIAL REINFORCEMENT EFFECT 47

Alternatively, the relation between reinforcement rate and resis- deprivation from 23 hr to 1 hr. Response rates maintained by
tance to change (behavioral mass) may be nonmonotonic, turn- CRF decreased substantially less than did those maintained by
ing down again as rate of reinforcement becomes very high and RR schedules requiring an average of 10,20,30, or 40 responses
approaches CRF. Another possibility is that the consistent out- per reinforcer. Thus, CRF appears to maintain responding that
comes of resistance-to-change research depend on within-sub- is more resistant to changes in deprivation than intermittent
ject comparisons after extended training in multiple-schedule reinforcement schedules, and the result is general across within-
procedures. Independent-group designs of the sort most com- subject, multiple-schedule procedures and group designs.
monly used in the study of the PRE may give different results. Another method for assessing resistance to change is to pro-
Conversely, the PRE may depend on the use of independent- vide alternative reinforcement on a separate key or lever. With
groups designs and relatively brief training. Finally, the PRE pigeons and intermittent reinforcement schedules, alternative
itself may need to be reexamined. reinforcement produces results that are entirely consistent with
those for changes in food deprivation (Nevin et al., 1981), and
Is Extinction a Counterinstance? the same appears to be true for rats and for schedules including
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

CRF. Lea and Roper (1977) arranged for rats to obtain food
This document is copyrighted by the American Psychological Association or one of its allied publishers.

In several experiments, researchers have demonstrated that according to CRF or various fixed-ratio (FR) schedules, ranging
after multiple-schedule training with two intermittent sched- from FR 1 (i.e., CRF) to FR 16 in one compartment of a two-
ules, extinction proceeds relatively less rapidly in the presence compartment chamber over successive conditions. With each
of the signal previously correlated with the higher rate of rein- schedule, a lever in the second compartment sometimes made
forcement. Gollub and Urban (1958) trained pigeons on mult food available according to an FR 8 schedule and sometimes
VI 3-min, VI 9-min schedules with unusually long compo- did not. When food was available, responding in the first com-
nents— 15 and 45 min, respectively—so that the number of re- partment decreased least when its schedule was CRF and most
inforcers was the same, on average, in both components. They when it was FR 16. Thus, both deprivation change and alterna-
then discontinued food reinforcement and alternated the com- tive reinforcement procedures suggest that CRF maintains per-
ponent stimuli every 12.5 min until responding decreased to formances that are more resistant to change than are those
near-zero levels. For all four birds, key pecking persisted longer maintained by intermittent reinforcement schedules.
in the presence of the stimulus correlated with the VI 3-min
schedule. Nevin (1974, Experiment 2) repeated these results
Is the Relation Between Resistance to Change and
with pigeons trained on mult VI 2-min, VI 6-min schedules
Reinforcement Rate Unique to Within-Subject Designs?
with 30-s components in both training and testing; Nevin et al.
(1983) repeated them, again with pigeons, on mult VI 28-s, VI At least two studies suggest that data obtained with indepen-
86-s; mult VI 86-s, VI 360-s; and mult VI 28-s, VI 360-s, with dent groups may be functionally similar to those obtained
1-min components in both training and testing. Cohen (1986) within subjects. I described the first, by P. Jenkins (1978), above
recently confirmed these results with rats trained on mult fixed- in conjunction with Carlton's (1961) within-subject results. In
interval (FI) 30-s, FT 120-s schedules and on mult random-inter- the second, Church and Raymond (1967) trained separate
val (RI) 30-s, RI 120-s schedules, so the result is general across groups of rats on VI 12-s or VI 5-min schedules and then intro-
species. Thus, resistance to extinction following multiple- duced a VI 2-min schedule of punishing electric shocks. There
schedule training with intermittent reinforcement in both com- were substantially greater reductions, relative to baseline, in the
ponents appears to conform to the general pattern that resis- VI 5-min group than in the VI 12-s group. More recently, Bou-
tance to change is positively related to reinforcement rate. zas (1978) trained pigeons on mult VI 1-min, VI 4-min sched-
ules and then introduced a VI 30-s schedule of punishing
shocks in both components. He observed larger relative de-
Is the Relation Nonmonotonic?
creases in response rate in the VI 4-min component. At least
The extinction studies cited above used relatively infrequent with respect to deprivation and punishment, then, resistance to
reinforcement, even in the component with the higher rate of change appears to be ordered similarly in independent-groups
reinforcement, and it may be that still more frequent reinforce- and within-subject, multiple-schedule research designs.
ment would lead to a reduction in resistance to change. Several
studies, with rats as subjects but with different procedures, sug- Is the Standard PRE Observed Within
gest that this is not the case. Carlton (1961) trained two rats on
Subjects on Multiple Schedules?
mult CRF VI 2-min schedules and two rats on mult CRF FI
2-min schedules of food reinforcement. He then varied food Several experimenters have addressed this question, but the
deprivation from 0 to 65 hr. There was virtually no change in answer is not clear. In the earliest study, Hearst (1961, Experi-
CRF responding by any of the rats, whereas there were large, ment 2) trained four pigeons on a four-component multiple
systematic variations in response rates maintained by the 2- schedule with nonreinforcement, CRF, variable-ratio (VR) 4,
min interval schedules for all rats. and VR 10 schedules correlated with distinctive key colors.
Carlton's (1961) result was confirmed by P. Jenkins (1978), Each component remained in effect until five reinforcers had
who used independent groups of rats trained on CRF or differ- been obtained, except the nonreinforcement component,
ent random-ratio (RR) schedules. (I consider this study in more which terminated after 2 min. After 17 training sessions, with
detail below). After training, extinguishing, and reconditioning 5 exposures to each component per session, Hearst conducted
lever pressing to stable levels, P. Jenkins (1978) reduced food 3 sessions of extinction, with each component appearing 75
48 JOHN A. NEVIN

times in all, for 30 s each time. Responses in the presence of trate on the free-operant extinction literature in this review.
each color were summed for the 3 sessions and plotted against Discussion of discrete-trial research is reserved for a later sec-
the training schedule value to demonstrate a clear PRE: The tion of this article.
number of responses in extinction increased monotonically
with the size of the ratio schedule in training. Evidently, the Slopes of Extinction Curves
usual PRE can be demonstrated in a within-subject, multiple-
schedule procedure much like that used by Nevin (e.g., 1974; The most common measure of the free-operant PRE is total
see also Nevin et al., 1983;Nevinetal., 1981) to evaluate resis- number of responses in some fixed period of extinction or to
tance to change. some extinction criterion. Such measures fail to distinguish the
This conclusion is complicated considerably by the work of initial level of responding (velocity in the momentum meta-
Pavlik and his associates. Pavlik, Carlton, and Manto (1965) phor) from the rate of decrease of response rate (which is in-
replicated Hearst's (1961) result with rats in a two-lever, two- versely related to behavioral mass). Two performances with
component multiple schedule in which components changed af- different initial levels may extinguish at the same rate (indicat-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ter a predetermined number of responses or reinforcers. The ing that they have the same mass) and yet differ markedly in
This document is copyrighted by the American Psychological Association or one of its allied publishers.

interpretation of their results is complicated, however, by the total number of responses, and two performances that start at
occurrence of numerous responses on the previously incorrect different initial levels but extinguish at rates that are inversely
lever during extinction, suggesting the breakdown of stimulus related to their levels may be similar in total number of re-
control. More important, the results appear to be quite different sponses. This fairly obvious point was made some time ago by
when schedule components are arranged on a single lever and Clark (1959), who provided a quantitative account of extinction
alternate regularly in time, regardless of behavior, as in the ma- that is directly relevant to the analysis of resistance to change.
jority of research on resistance to change cited above. Pavlik Clark's (1959) account was based on Skinner's (1938) sugges-
and Carlton (1965) trained rats on mult CRF VR 3 schedules tion that momentary response rate in extinction (r in Clark's,
with 30-s components for four 20-min sessions and then con- 1959, notation) was proportional to the number of responses
ducted a single session of extinction. They observed more re- remaining in the "reflex reserve." If R is the total number of
sponses in the presence of the signal for the former CRF sched- responses emitted up to time / and Cis total responses to extinc-
ule than for the VR 3 schedule and termed the result a "reverse tion (i.e., the size of the reserve),
PRE." They also trained two control groups on either CRF or
VR 3 for four 10-min sessions and observed the standard PRE R = C(l - <?-*')• (7)
during extinction when the latter two groups were compared. This implies that momentary response rate (the first derivative
Pavlik, Carlton, Lehr, and Hendrickson (1967) replicated these of total responses as a function of time) is
results with more extensive training and testing, and Adams,
Nemeth, and Pavlik (1982) replicated again with female (as op- r = dRIdt = kCe'to /o
posed to male) rats and sucrose (as opposed to food) reinforce-
or, in logarithm form,
ment. Thus, the reverse PRE may be reasonably general, for
individual rats at least.
log r = log kC - kt, (8)
Although most of the data reported by Pavlik and his associ-
ates (Pavlik & Carlton, 1965; Pavlik etal., 1967; Pavlik, Carlton, where k, the rate constant of the exponential decay process, is
& Manto, 1965) support the generalization that resistance to given by the slope of the function relating response rate to time
change is directly related to reinforcement rate, they are based in semilogarithmic coordinates and is independent of C. This is
on a limited range of schedules. Moreover, VR 3 can in fact the form in which many resistance-to-change results have been
result in a fairly high rate of reinforcement if the subjects re- analyzed; t (time in extinction) is the variable used to assess
spond rapidly. Hearst's (1961) data, based on a somewhat wider resistance to change. Therefore, the shallower the slope of the
range of schedule values, do not support the generalization, and extinction function, the greater is behavioral mass (i.e., resis-
neither do the results of the many studies that compare extinc- tance to change), regardless of the total number of responses.
tion after CRF and intermittent reinforcement with indepen- In the present reanalysis, I compare the slopes of functions that
dent groups. For assessment methods other than extinction, describe the decrease in response rate across successive sessions
there is no conflict between within-subject and independent- of extinction after training with CRF or intermittent reinforce-
groups research. Thus, the PRE as usually reported remains ment.
inconsistent with the effects of other assessment methods on I located five studies in which subjects were trained on CRF
resistance to change. and one or more intermittent schedules with equal numbers of
reinforcers and in which extinction data were reported over the
course of three or more sessions so that slopes could be esti-
Reanalysis of the PRE
mated. I believe that all published studies meeting these criteria
To resolve the discrepancy between extinction and other as- are included in the reanalyses.
sessment methods when CRF is compared with intermittent 1. Boren (1961) studied response rate and resistance to ex-
reinforcement, I examined a number of extinction studies per- tinction as functions of the FR schedule used in training. After
formed during the past 40 years and attempted to isolate the initial lever-press training, six groups of rats were trained for 10
critical variables. Because the resistance-to-change literature sessions each on CRF, FR 3, FR 6, FR 10, FR 15, or FR 21,
cited above used free-operant methods exclusively, I concen- with 50 reinforcers per session. They were then exposed to five
BEHAVIORAL MOMENTUM AND THE PARTIAL REINFORCEMENT EFFECT 49

1-hr extinction sessions on consecutive days. Average response 1.00


rates for Training Sessions 9 and 10 were reported and were
taken as baseline; average response rates in each extinction ses- .50
sion were used to estimate resistance to change over the course
of extinction. Complete tables of data were given in Boren's .20
(1953) doctoral dissertation.
2. P. Jenkins (1978) studied response rate, resistance to ex- .10
tinction, and resistance to deprivation change (see above) as
functions of the RR schedule used during training. He followed .05
Boren's (1953, 1961) general method closely, using six groups
of rats trained for seven sessions of 50 reinforcers each on CRF,
RR 5, RR 10, RR 20, RR 30, and RR 40 schedules. He then
.02
conducted five 50-min extinction sessions on consecutive days.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1.00
Average response rates for Training Sessions 5, 6, and 7 were
This document is copyrighted by the American Psychological Association or one of its allied publishers.

reported and were taken as baseline, and average response rates


.50
in each extinction session were used to estimate resistance to
change over the course of extinction. Complete tables of data
were given in P. Jenkins's (1977) master's thesis. .20
3. W. O. Jenkins, McFann, and Clayton (1950) studied resis-
Wilson (1954) Borer. (I960
tance to extinction after training with CRF or VI 1-min sched- .10 • CRF
O • CRF
ules in three ways: six 1-hr sessions on consecutive days, six 1- o FI .33 (4.7) • FR 15 (58)
.05 a FI I (70) i FR 21 (.78)
hr sessions separated by 2 hr, or one continuous 8-hr session. b £ FI 2 (11.2)
o. [500]
The subjects were pigeons, and extinction was preceded by five [240
sessions of training, with 40 reinforcers per session. Baseline .02
data were not reported, but complete tables of extinction data
were given in the article. For comparability with other analyses, Successive Sessions of Extinction
only the data for six daily 1-hr extinction sessions are treated
Figure 1. Six sets of extinction curves for free-operant behavior are pre-
here. sented as proportions of response rate in the first session of extinction,
4. W. O. Jenkins and Rigby (1950) studied resistance to ex- plotted on semilogarithmic axes to permit comparison of their slopes.
tinction in independent groups of rats after training with CRF, Numbers of reinforcers during training are shown in brackets for each
FI1 -min, and FI 2-min schedules. Three 1 -hr sessions of extinc- set. If ratio schedules were used, I estimated the average time between
tion followed 90 reinforcers on these schedules. A fourth group reinforcers from baseline data; it is shown in parentheses (in minutes).
was trained on CRF for 2,400 reinforcers, which required as Across studies, the slopes of the extinction curves after continuous rein-
much total training time as the 90 FI 2-min reinforcers. Base- forcement (CRF) become shallower than they do after the various inter-
mittent schedules as the number of reinforcers in training increases.
line data were not reported, but complete tables of extinction
(VI = variable interval; FI = fixed interval; RR = random ratio; FR =
data were provided in the article.
fixed ratio.)
5. Wilson (1954b) studied response rate and resistance to
extinction as functions of the FI schedule used during training.
He also followed Boren's (1953,1961) general method, training
independent groups of rats on CRF, FI 10-s, FI 20-s, FI 1-min, procedure. Portions of the data are shown in Figure 1; for rough
FI 2-min, FI 4-min, and FI 8-min schedules. One set of groups comparability across studies, only those groups that experi-
was trained for 15 reinforcers, and a second set of groups was enced average interreinforcement intervals between 0.33 and 2
trained for 240 reinforcers before five 50-min extinction ses- min are shown, together with the CRF data. The number of
sions were given on consecutive days. No baseline data were reinforcers given during training is shown in brackets. Here it
reported for the 15-reinforcer groups; therefore, changes rela- appears that as the number of reinforcers increases, the slopes
tive to baseline (average response rates for the last 35 reinforc- of the extinction functions shift relative to each other. After 15
ers) were analyzed for the 240-reinforcer groups only. In his dis- or 90 reinforcers, the slope of the CRF extinction function is
sertation, however, Wilson (1954a) tabulated average response generally steeper than for the intermittent schedules; at 200 or
rates for all groups during extinction. 240 reinforcers, the slopes become similar; and at 350 reinforc-
Wilson (1954b) conducted a second experiment, in which ers, the CRF function may be slightly shallower. After 500 rein-
four additional groups of rats were trained for 50, 85, 120, or forcers, there is no doubt about the result: Responding in ex-
500 reinforcers on an FI 2-min schedule before extinction test- tinction after CRF is decidedly more persistent than after inter-
ing. Thus, including the 15- and 240-reinforcer groups from the mittent reinforcement, relative to its initial level.
first experiment, a function relating extinction responding to Confirming data for independent groups come from the stud-
number of intermittently scheduled reinforcers is also available. ies by Pavlik and his associates (Adams et al., 1982; Pavlik et
For each of the five studies described above, response rates in al., 1967) cited above. Adams et al. (1982) trained their subjects
each session of extinction were expressed as proportions of the for 100 min, and Pavlik et al. (1967) trained their subjects for
response rate in the initial extinction session in order to evalu- 220 min before beginning extinction. Both groups of authors
ate the rate of change of response rate within the extinction reported a conventional between-groups PRE when they exam-
50 JOHN A. KEVIN

Mowrer 8 Adams et al. Pavlik et al.


Jones (1945) (1982) (1967)

in
c
o .5
Q. c
en o
(B tn
u>
<u
[l40rft] [230 mlnj
c
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

o
This document is copyrighted by the American Psychological Association or one of its allied publishers.

CRF -B -• C R F - B
•CRF-B o—o V R 3 - B -o VR3 - B
S. .5 o FR3- B •-• CRF-W • CRF - W
o 0-0 V R 3 - W o VR3-W

I 2 3 1 2 3 1 2
Sessions of Extinction
Figure 2. Further data, in the form of Figure 1, showing that the slopes of extinction curves after continuous
reinforcement (CRF) become shallower than they do after intermittent reinforcement with extended train-
ing. The studies by Adams, Nemeth, & Pavlik (1982) and Pavlik, Carlton, Lehr, & Hendrickson (1967)
included subjects trained with both schedules, to permit within-subject (W) as well as between-groups (B)
comparisons, (rft = total reinforcers before extinction; 100 and 220 min = total training time before extinc-
tion; FR =fixedratio; VR - variable ratio.)

ined simple response totals; however, if the data are reexpressed examined.) Although the data are less complete than one might
as proportions of response rate in the first session of extinction, wish, they suggest that the rate of extinction is also slower after
responding appears more persistent after CRF than after VR 3, CRF than after intermittent reinforcement in within-subject
and the difference increases with duration of training. comparisons, at least after 220 min of training.
The Pavlik et al. (1967) data are summarized in Figure 2,
together with those of an early systematic study of the PRE for
Initial Decrements Relative to Baseline
free-operant behavior by Mowrer and Jones (1945). Mowrer
and Jones trained independent groups of rats on CRF, FR 2, The slopes of the extinction curves presented above do not, of
FR 3, and FR 4 for 140 reinforcers and then conducted three course, give the whole story: They say nothing about the initial
20-min sessions of extinction on successive days, an extinction decrement in extinction relative to baseline response rates. The
procedure comparable to that used by Pavlik and his associates studies by Boren (1961), P. Jenkins (1978), and Wilson (1954b)
(Adams et al., 1982; Pavlik et al., 1967). As the figure shows, provided the baseline data that are required for the analysis.
FR 3 performance extinguished more slowly than CRF perfor- However, all three studies reported response rates without cor-
mance after 140 reinforcers (the data are tabled in Mowrer and rection for eating time. Thus, momentary baseline response
Jones's article), but the reverse is true for the data of Adams et rates are likely to be underestimated, and the ratios of extinc-
al. (1982) after ten 10-min sessions of training (which probably tion response rates to baseline response rates are likely to be
resulted in over 500 reinforcers for the CRF group) and for the overestimated for schedules with frequent reinforcement.
data of Pavlik et al. (1967) after twenty-two 10-min training Response rates for the first session of extinction are shown as
sessions (which probably resulted in well over 1,000 reinforcers proportions of baseline response rates for all three studies in
for the CRF group). Thus, it appears that after extensive train- Figure 3. Two specifications of the training schedule were used:
ing, the rate of extinction after CRF is slower than after inter- (a) reinforcers per response, which was experimentally con-
mittent reinforcement for independent groups. This result is, trolled in the Boren (1961) and P. Jenkins (1978) studies but
of course, consistent with the generalization that resistance to had to be estimated from Wilson's (1954b) baseline data, and
change—construed as the inverse of the slope of the function (b) reinforcers per hour, which was experimentally controlled in
relating response rate to the value of some assessment opera- the Wilson (1954b) study but had to be estimated from Boren's
tion—is an increasing function of the rate of reinforcement. (1961) and P. Jenkins's (1978) baseline data.
Figure 2 also contains within-subject comparisons from Ad- The upper panel of Figure 3 shows that response rate in the
ams et al. (1982) and Pavlik et al. (1967), in the form of response first session of extinction was a generally decreasing function of
rates relative to those in the first session of extinction. (Recall reinforcers per response; that is, the richer the training sched-
that these data showed a reverse PRE when response totals were ule, the greater the relative reduction at the onset of extinction.
BEHAVIORAL MOMENTUM AND THE PARTIAL REINFORCEMENT EFFECT 51

part of the stimulus situation in which responding is reinforced;


c (b) with respect to these aspects of the stimulus situation, the
o difference between CRF and extinction is greater than between
intermittent reinforcement and extinction because nonrein-
forcement and its consequences have been experienced during
training; (c) in the study of stimulus generalization, larger re-
X
o sponse decrements are routinely observed with larger changes
in stimulus values; and (d) therefore, one should find a larger
decrement in extinction after CRF than after intermittent rein-
c forcement. Figure 3 confirms this line of argument and, more-
o
over, suggests that the relevant stimulus dimension is the rate of
8.8 reinforcement obtained during training.
o
w The schedule in effect immediately before the transition from
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

training to extinction may not be the crucial determiner of gen-


This document is copyrighted by the American Psychological Association or one of its allied publishers.

eralization decrement. H. M. Jenkins (1962) and Theios (1962)


•= o.
£ "01 .10 i.o interpolated periods of CRF between partial reinforcement and
c Reinforcers per response in baseline extinction in discrete-trial procedures and found that the usual
PRE was only slightly attenuated (Theios) or even, perhaps, en-
Wil*on(l954) oFI «CRF hanced (H. M. Jenkins, 1962) as a result. Evidently, the novelty
Bar«n(l96l) aFR "CRF rather than the abruptness of nonreinforcement determines the
a> Jenkins (1978) ARR »CRF initial decrement. In any case, there must inevitably be differ-
c ential exposure to nonreinforcement across groups in any be-
— .6
«
tween-groups study of the PRE, and the stimulus consequences
U) of this differential exposure to nonreinforcement complicate
D
XI the measurement of resistance to change.

Confounding Effects of Generalization Decrement

Nevin (1974,1979; see discussion above) has emphasized the


2 need to apply a single, uniform operation to the performances
o
a. being compared for resistance to change. Otherwise, the assess-
o
ment is confounded: A given result could arise either because
the performances differ in persistence or because the assessment
10 100 1000 operation was differentially applied. The extinction procedure
Reinforcers per hour in baseline evidently includes two factors: the termination of reinforce-
ment and the change in the stimulus situation. The former is
Figure 3. Response rates during the first session of extinction, expressed
uniform with respect to all performances undergoing extinc-
as proportions of the final response rates maintained during training,
are related to either reinforcers per response (upper panel) or reinforcers tion, but the latter is not, being greater for CRF or other sched-
per hour (lower panel) experienced during training. The decreasing ules that arrange high reinforcement rates than for low rein-
slopes of all functions may be understood in relation to the generaliza- forcement rates.
tion-decrement hypothesis of extinction, according to which the reduc- In the terms of the momentum metaphor, it is as if two bodies
tion in responding depends on the change in the stimulus situation from with different masses and moving at different velocities were
training. (Fl = fixed interval; FR = fixed ratio; RR = random ratio; subjected to forces with two components. The heavier, slower
CRF = continuous reinforcement.) body represents the mass and velocity established by CRF, and
the lighter, faster body represents the mass and velocity estab-
lished by intermittent reinforcement. One component of the
All functions would be steeper if baseline response rates were opposing force is like friction (analogous to the termination of
corrected for eating time. The Wilson (1954b) data agree with reinforcement): It is applied equally to both, continuously
the Boren (1961) and P. Jenkins (1978) data at rich schedule throughout their deceleration. The other component is like the
values but are consistently higher at leaner schedule values. The application of a brake (analogous to generalization decrement):
lower panel shows the same trend with respect to reinforcers per It is applied differentially to the two bodies—more strongly to
hour, but now there is substantial agreement across the three the heavier and slower—with diminishing force as they slow
data sets over the full range of their overlap. down. When the braking forces terminate, only friction is oper-
These results exemplify the generalization-decrement hy- ative, and the heavier body will decelerate less rapidly than the
pothesis, which has been advanced repeatedly to explain many lighter body, regardless of their velocities when the brakes are
aspects of the PRE and related extinction phenomena. (For re- removed. Therefore, the effect of the differential braking com-
view and discussion, see Mackintosh, 1974; Staddon, 1975.) ponent must be removed if the relative masses of the two bodies
Briefly, the hypothesis asserts that (a) the delivery of reinforcers are to be estimated accurately.
and the consequences of nonreinforcement are themselves a This is what the examination of changes in response rate rela-
52 JOHN A. NEVIN

live to the initial session of extinction was intended to accom- resistance to extinction after training on FT 2-min schedules,
plish. Unfortunately, the generalization-decrement component over the range from 15 to 500 reinforcers. Examination of this
of the difference between training and extinction lasts for an data set (not shown) reveals a systematic increase in the number
unknown period of time. For want of a better approach, I have of responses made during 5 extinction sessions but little if any
used response rate in the first whole session of extinction, rela- evidence of a consistent change in the slope of the extinction
tive to baseline, to characterize the generalization-decrement curves. Other relevant data come from a study by Lentz and
effect and then used that first session as a baseline for assessing Cohen (1980), who trained pigeons on a VI f-min schedule and
the subsequent rate of extinction. Although this procedure does gave free food in the experimental chamber at various stages of
not alter the slope of the extinction curve, which is invariant training. The decrement in response rate with free food, relative
(in semilogarithmic coordinates) with respect to the choice of to the immediately preceding baseline rate, suggests a gradual
baseline, it is admittedly arbitrary. Each successive daily session increase in resistance to change as training progresses, at least
of extinction involves removing the subject from its home cage, up to 48 sessions and perhaps beyond, with 60 reinforcers per
placing it in the experimental chamber, and turning on the session. Again, no asymptote is well-defined. However, for prac-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

houselight or other signal that a session has begun. During tical purposes, the data suggest that the difference in resistance
This document is copyrighted by the American Psychological Association or one of its allied publishers.

training, this stimulus complex was followed sooner (if CRF) to change between continuously and intermittently reinforced
or later (if an intermittent schedule) by reinforcers. The daily responding may become maximal after approximately 1,000 to
reinstatement of these beginning-of-session cues may make gen- 3,000 reinforcers.
eralization from training to extinction a potent factor for some
time. Accordingly, an unbiased assessment of resistance to ex-
A Confirming Experiment
tinction must somehow eliminate the differential stimulus con-
sequences of the onset of an extinction session after CRF or Nevin (1985) examined extinction after CRF and after inter-
intermittent reinforcement. An experimental effort to do just mittent reinforcement in a within-subject design with extended
this is reported below. training, in a procedure that attempted to equalize the decre-
mental effects of the transition to nonreinforcement by provid-
ing repeated extinction after an explicit signal. In a standard
Effects of Number of Reinforcers
two-key pigeon chamber, three pigeons were exposed to daily
Before further discussion of the generalization-decrement sessions that always began with reinforcement and always ended
problem, it is necessary to consider how resistance to extinction with extinction. On odd-numbered days, the left, red key was
is affected by the length of training, which appears as an impor- lighted, and a 3-s food reinforcer was given after every key peck.
tant parameter across the various studies considered in Figures After 50 reinforcers, the key and chamber lights went dark for
1 and 2. Several studies permit direct examination of resistance 10 s and were then relighted for 64 min of nonreinforcement.
to extinction as a function of number of reinforcers. Wilson The same procedure was used on even-numbered days except
(1954b) reported data for resistance to change after CRF train- that the right, green key was lighted and 50 food reinforcers
ing for 15 or 240 reinforcers, and Boren (1961), who used the were given according to a VI 30-s schedule before the 10-s
same apparatus as Wilson (1954b) and followed virtually identi- blackout and the 64-min extinction period. Thus, the birds ex-
cal procedures, provided data for 500 reinforcers. W. O. Jenkins perienced daily repeated extinction sessions after both CRF and
and Rigby (1950) provided data for 90 and 2,400 reinforcers; intermittent reinforcement, arranged on keys that differed in
D'Amato, Schiff, and Jagoda (1962) provided data for 200,400, color and location, with the transitions to extinction signaled
800, and 1,600 reinforcers; and Furomoto (1971; see also 1966) by blackout in both cases. After 8 initial sessions—4 left red
provided data for conditions ranging from 3 to 10,000 rein- CRF and 4 right green VI 30 s—consisting of 50 reinforcers
forcers. each but without the 64-min extinction period, this procedure
Examination of the Wilson (1954b) and Boren (1961) data in remained in effect for 125 sessions. As a result, the subjects re-
Figure 1 suggests that resistance to extinction of continuously ceived over 3,000 reinforcers on each schedule before the end
reinforced responding increases from 15 to 240 to 500 reinforc- of training.
ers. The extinction curves of W. O. Jenkins and Rigby (1950)— It is well-known that repeated alternations of training and
not shown—are shallower after 2,400 reinforcers than after 90 extinction reduce total responses during extinction, but other
reinforcers. D'Amato et al. (1962) and Furomoto (1966) ana- aspects of extinction performance may also change under these
lyzed their data in the manner suggested by Clark (1959). D'A- conditions. Clark (1964) subjected pigeons to repeated alter-
mato et al. (1962) found little effect of number of reinforcers on nations of VI 1-min reinforcement and extinction and found
the slope of the extinction curve, except that the slope for 1,600 that the initial extinction response rate decreased and the slope
reinforcers was shallowest, whereas Furomoto (1966) found of the extinction curve became increasingly negative over 20
that her extinction curves became shallower up to 1,000 rein- such repetitions. Clark and Taylor (1960) performed a similar
forcers. In general, continuously reinforced responding appears experiment with rats on CRF schedules and found similar
to become more resistant to extinction as training continues, at trends over 12 such repetitions. Thus, both the level and the
least up to 1,000 reinforcers, but no asymptote is well-defined slope of the extinction curves may be expected to change over
by the extant data. the course of repeated training and extinction in the present
Unfortunately, there are many fewer studies of resistance to study.
change of intermittently reinforced responding in which num- The results, broken down into successive 16-min segments of
ber of reinforcers is varied. Wilson (1954b) provided data on extinction, pooled over blocks of sessions, and averaged for the
BEHAVIORAL MOMENTUM AND THE PARTIAL REINFORCEMENT EFFECT 53

Days 75-125 tion curves within the 64-min extinction period early in train-
ing, consistent with Figures 1 and 2. A substantial difference
was evident during the final 50 sessions, with CRF leading to
• CRF
more persistent (albeit lower level) responding. For both CRF
o VI 30sec
and VI 30-s conditions, the final slopes were steeper than early
in training, exactly as in the data of Clark (1964) and Clark and
Taylor (1960).
Although these differences held for all three birds, the post-
CRF extinction rates were so low during the final 50 sessions
that a few sporadic, random bursts of responses could have al-
tered the picture considerably. Response bursts occurring
within the first 16-min segment of extinction may be inter-
preted as "spontaneous recovery" and may affect the results.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

For example, if the VI 30-s schedule generates greater spontane-


This document is copyrighted by the American Psychological Association or one of its allied publishers.

ous recovery than does CRF during the initial segment of ex-
tinction, relative response rates in subsequent segments will be
artifactually lower following VI 30-s reinforcement than follow-
ing CRF. However, there are no independent data bearing on
this possibility, and in any case, the extinction curves differed
in slope as well as level. Thus, the results are consistent with
other within-subject and between-groups data showing that af-
ter extended training, CRF engenders shallower extinction
curves than does intermittent reinforcement. In summary, it
appears that slower extinction after CRF than after intermittent
reinforcement is the normative asymptotic result.

Comparison With Findings of Discrete-Trial Research

In this review, I have concentrated on free-operant behavior,


for which the natural dependent variable is the rate at which an
unconstrained response is repeated during an extended period
3 4 of time, for example, an experimental session or a signaled
schedule component. The principal reason for this emphasis is
Successive Quarters of Extinction that the quantitative study of behavioral momentum has used
Figure 4. The upper panels show the rates of responding throughout free-operant methods and has used changes in response rate as
repeated 64-min sessions of extinction, following 50 reinforcers on ei- its dependent measure. The overwhelming majority of research
ther continuous reinforcement (CRF) or variable interval (VI) 30-s on the PRE, however, has used discrete-trial methods, most
schedules and a 10-s time-out. The lower panels show the averaged pro- commonly with rats in runways, in which the natural depen-
portions of response rates during the first 16 min of extinction. Overall
dent variable is the within-trial speed of a highly constrained
levels of responding are lower after many repeated alternations of rein-
response that is extended in space and time. The data of runway
forcement and extinction (upper panels), and the slopes of the extinc-
studies tell a different story from those of free-operant research.
tion curves are steeper (lower panels). Also, with extended training, the
slope of the extinction curve following CRF is shallower than it is follow- With respect to the PRE during extinction, between-groups
ing VI 30-s reinforcement. (CRF = continuous reinforcement; VI = research has demonstrated repeatedly that running persists at
variable interval.) high speeds for many more trials after training on a partial rein-
forcement schedule than after CRF, even when running speed is
expressed as a proportion of speed during the initial extinction
3 subjects, are presented in Figure 4. The upper left panel shows trials. (See, e.g., Weinstock, 1954, Figure 2, where this result is
that there were substantially more responses throughout extinc- evident by inspection.) This ordinal outcome is quite robust
tion after VI 30-s reinforcement than after CRF during the early with respect to such parameters as percentage of reinforced tri-
sessions, and the upper right panel shows that the levels had als, length of training, magnitude of the reinforcer, and trial
fallen sharply by the final 50 sessions of the procedure. On aver- spacing. Moreover, it is not confined to rats in runways with
age, there were more key pecks following VI than following CRF running speed as the measure but appears also with pigeons
even after extended training, but there were several sessions of pecking a briefly lighted key with probability of response as the
both kinds in which no pecks at all occurred. measure (e.g., H. M. Jenkins, 1962) and in many other prepara-
For each subject, pooled response rates in the second, third, tions as well (see Mackintosh, 1974, for review). Within-subject
and fourth 16-min segments of extinction were expressed as analyses have sometimes obtained a reverse PRE (e.g., Pavlik,
proportions of the rate during the first segment and then aver- Carlton, & Hughes, 1965; Pavlik & Collier, 1977), but other re-
aged. The results, presented in the lower panels of Figure 4, search has failed to replicate the reverse PRE within subjects
show that there was little difference in the slopes of the extinc- (e.g., Amsel, Rashotte, & MacKinnon, 1966; Rashotte & Am-
54 JOHN A. NEVIN

sel, 1968). It is not at all clear how these discrete-trial results can ment be used to select and strengthen qualitative variants of the
be reconciled with those for free-operant extinction research class, as in skill training? In brief, is it the rate of performance,
reviewed above. the quality, or both that count? If a high rate is desirable, with
The differences do not stop with resistance to extinction. For or without differential reinforcement with respect to quality, by
example, running speed is also more resistant to satiation implication the performance must be treated as a free operant,
(Haas, Shessel, Willner, & Rescorla, 1970) and to punishment and the results of free-operant research should be relevant.
(Brown & Wagner, 1964) when it is maintained by partial rein- Third, with respect to persistence in everyday life, is it desir-
forcement than when it is maintained by CRF, contrary to the able that the total number of repetitions of the newly estab-
free-operant results cited above. Also, across trials, running lished behavior be large, or is the long-term maintenance of a
speed decreases more rapidly in the goal section of the runway stable rate, even at a low level, more important? To the extent
than in the start section, during both extinction (e.g., Wagner, that data from the animal laboratory are applicable, it ought to
1961) and satiation (Rescorla, 1977). If running is viewed as a be possible to achieve both, for example, by the use of a short
chain, this result is exactly contrary to findings with free-oper- ratio schedule that generates high rates as a result of the ratio
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ant chain schedules, in which response rate is more resistant to contingency and substantial resistance to change as a result of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

change at the end of the chain than at the beginning (see Nevin the relatively high rate of reinforcement obtained. Indeed, this
et al., 1981, for data and review). The functional differences is the sort of procedure that should maximize behavioral mo-
between discrete-trial and free-operant results appear to extend mentum—the product of mass and velocity—in the Newtonian
well beyond the PRE. metaphor described above.
There is one area of clear agreement, however. In discrete- Fourth, it is by no means clear that the only relevant data
trial as well as free-operant choice procedures, subjects uni- are those taken from research on resistance to extinction. No
formly prefer stimuli correlated with CRF over stimuli corre- responsible therapist would establish a class of behavior that
lated with partial or intermittent reinforcement (e.g., D'Amato, goes unreinforced in everyday life; indeed, that would be a
Lachman, & Kivy, 1958; vom Saal, 1972). This preference is definition of undesirable behavior. Accordingly, it is important
opposite in direction to the relative persistence of discrete-trial to consider resistance to challenges other than nonreinforce-
performance and thus raises some important questions of inter- ment, such as competition from alternative activities and their
nal consistency. For example, these choice results suggest that reinforcers, or suppression by occasional punishers that may
partially reinforced behavior should be less resistant than con- result from the uncontrolled conditions of everyday life. More-
tinuously reinforced behavior to competition from a reinforced over, reductions in newly acquired behavior due to generaliza-
alternative, which is inconsistent with their relative persistence tion decrement are likelier to arise from differences between the
during extinction, satiation, or punishment. A full discussion therapeutic setting and everyday life than from stimulus differ-
of this inconsistency and of the functional differences between ences between reinforcement and extinction. Given all these
free-operant and discrete-trial persistence would be both considerations, it may be quite inappropriate, in many cases, to
lengthy and speculative and is beyond the scope of this article. base a therapeutic program on the PRE as studied in discrete-
trial alleyway research. Nation and Woods (1980) argued that
Some Consequences for Behavior therapists should consider these discrete-trial results in devising
treatment strategies, and so they should; but therapists should
Modification and Therapy
also give full consideration to the principles that have emerged
In applied and therapeutic settings, one common goal of be- from research on free-operant resistance to change and use
havior modification or therapy is to define a class of desirable them as appropriate when they seem applicable to the nature of
behavior, make its occurrence more probable within the thera- the behavior in question and to the desired outcomes of therapy.
peutic setting, and ensure its effective persistence in everyday
life. In this section, I give a brief and general overview of the Conclusion
implications of the data reviewed above for this goal; more de-
tailed consideration of specific therapeutic problems and meth- The PRE for free-operant behavior, as usually reported, is
ods is needed but is beyond the scope of this article. thoroughly confounded. The usual measure of resistance to ex-
The first question pertains to the class of behavior being es- tinction confounds the level of responding and the slope of the
tablished. If it is operant—that is, affected by its conse- extinction curve, and the extinction procedure itself confounds
quences—is it best construed as free operant, initiated in the decremental effects of stimulus change with the effects of nonre-
presence of appropriate stimuli but otherwise unconstrained? inforcement. When subjects are given extended training before
Or is it blocked or enabled by external factors, in a fashion anal- extinction, when the rate of decrease of responding is measured
ogous to discrete-trial procedures? In view of the apparent separately during the course of extinction, and when stimulus-
functional differences between free-operant and discrete-trial change effects are controlled, the rate of decrease of responding
persistence results reviewed above, this question may be impor- appears to be lower following training with CRF than it is fol-
tant for the treatment strategy. lowing training with intermittent reinforcement. Thus, in free-
Second, once the desired behavior is established, whether by operant research, there is no discrepancy between the effects of
shaping, modeling, or instruction, what reinforcement contin- extinction and the many other procedures demonstrating that
gencies should be used? Should the rate of occurrence of re- resistance to change is a positive function of the rate of rein-
peated instances of the response class be maximized by the use forcement. Accordingly, theoretical analyses of behavioral mo-
of a ratio schedule, for example, or should differential reinforce- mentum can proceed with reasonable confidence in the consis-
BEHAVIORAL MOMENTUM AND THE PARTIAL REINFORCEMENT EFFECT 55

tency of the data. Unfortunately, the results of discrete-trial rate, latency, and resistance to change. Journal of the Experimental
studies are not, in general, consistent with the free-operant Analysis of Behavior, 39, 267-274.
findings, suggesting that therapeutic applications of laboratory Furomoto, L. (1966). Extinction in the pigeon after continuous rein-
research must be considered with great care. forcement: Effects of number of reinforced responses. Unpublished
doctoral dissertation, Harvard University, Cambridge, MA.
Furomoto, L. (1971). Extinction in the pigeon after continuous rein-
References forcement: Effects of number of reinforced responses. Psychological
Reports, 28, 331-338.
Adams, J. E.Nemeth, R. V., &Pavlik, W. B. (1982). Between-and with-
Gollub, L. R., & Urban, J. T. (1958). The accentuation of a rate differ-
in-subjects PRE with sucrose incentives. Bulletin of the Psychonomic
ence during extinction. Journal of the Experimental Analysis of Be-
Society, 20,261-262.
havior, 1, 365-369.
Amsel, A., Rashotte, M. E., & MacKinnon, J. R. (1966). Partial rein-
Haas, R B., Shessel, F. M., Willner, H. S., & Rescorla, R. A. (1970).
forcement effects within subjects and between subjects. Psychological
The effect of satiation following partial reinforcement. Psychonomic
Monographs, 80(20, Whole No. 628).
Science, IS, 296-297.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Autor, S. (1969). The strength of conditioned reinforcers as a function


Hearst, E. (1961). Resistance-to-extinction functions in the single or-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

of frequency and probability of reinforcement. In D. P. Hendry (Ed.),


ganism. Journal of the Experimental Analysis of Behavior, 4, 133-
Conditioned reinforcement (pp. 127-162). Homewood, IL: Dorsey.
144.
Blackman, D. E. (1968). Response rate, reinforcement frequency, and
Herrnstein, R. J. (1964). Secondary reinforcement and rate of primary
conditioned suppression. Journal of the Experimental Analysis ofBe-
reinforcement. Journal of the Experimental Analysis of Behavior, 7,
havior, //, 503-516.
27-36.
Boren, J. J. (1953). Response rate and resistance to extinction as func-
Herrnstein, R. J., & Loveland, D. H. (1974). Hunger and contrast in a
tions of the fixed ratio. Unpublished doctoral dissertation, Columbia
University, New York. multiple schedule. Journal of the Experimental Analysis of Behavior,
Boren, J. J. (1961). Resistance to extinction as a function of the fixed 24, 107-116.
ratio. Journal of Experimental Psychology, 61, 304-308. Jenkins, H. M. (1962). Resistance to extinction when partial reinforce-
Bouzas, A. (1978). The relative law of effect: Effects of shock intensity ment is followed by regular reinforcement. Journal of Experimental
on response strength in multiple schedules. Journal of the Experimen- Psychology, 64, 441-450.
tal Analysis of Behavior, 30, 307-314. Jenkins, P. (1977). Resistance to extinction, and satiation following
Brown, R. X, & Wagner, A. R. (1964). Resistance to punishment and training on random ratio schedules. Unpublished master's thesis,
extinction following training with shock or nonreinforcement. Jour- University of New Hampshire, Durham.
nal of Experimental Psychology, 68, 503-507. Jenkins, P. (1978). Resistance to extinction and satiation following
Carlton, P. L. (1961). The interacting effects of deprivation and rein- training on random-ratio schedules of reinforcement. Psychological
forcement schedule. Journal of the Experimental Analysis ofBehav- Record, 25,471^178.
ior, 4, 379-381. Jenkins, W. O., McFann, H, & Clayton, F. L. (1950). A methodological
Church, R. M., & Raymond, G. A. (1967). Influence of the schedule of study of extinction following aperiodic and continuous reinforce-
positive reinforcement on punished behavior. Journal ofComparative ment. Journal of Comparative and Physiological Psychology, 43, 155-
and Physiological Psychology, 63, 329-332. 167.
Clark, F. C. (1959). Some quantitative properties of operant extinction Jenkins, W, O., & Rigby, M. K. (1950). Partial (periodic) versus continu-
fate.. Psychological Reports, 5, 131-139. ous reinforcement in resistance to extinction. Journal ofComparative
Clark, F. C. (1964). Effects of repeated VI reinforcement and extinction and Physiological Psychology, 43, 30-40.
upon operant behavior. Psychological Reports, 15, 943-955. Lea, S. E. G., & Roper, T. J. (1977). Demand for food on fixed-ratio
Clark, F. C., & Taylor, B. W. (1960). Effects of repeated extinction of schedules as a function of the quality of concurrently available food.
an operant on characteristics of extinction. Psychological Reports, 6, Journal of the Experimental Analysis of Behavior, 27, 371-380.
226. Lentz, B. E., & Cohen, S. L. (1980). The effect of prior training on the
Cohen, S. L. (1986). A pharmacological examination of the resistance- contra-freeloading phenomenon. Bulletin of the Psychonomic Soci-
to-change hypothesis of response strength. Journal of the Experimen- ety, 15, 48-50.
tal Analysis of Behavior, 46, 363-379. Lyon, D. O. (1963). Frequency of reinforcement as a parameter of con-
D'Amato, M. R., Lachman, R., & Kivy, P. (1958). Secondary reinforce- ditioned suppression. Journal of the Experimental Analysis of Behav-
ment as affected by reward schedule and the testing situation. Journal
ior, 6, 95-98.
of Comparative and Physiological Psychology, 51,734-741.
Mackintosh, N. J. (1974). The psychology of animal learning. New
D'Amato, M. R., Schiff, D., & Jagoda, H. (1962). Resistance to extinc-
York: Academic Press.
tion after varying amounts of discriminative or nondiscriminative in-
McSweeney, F. K., Farmer, V. A., Dougan, J. D., & Whipple, J. E.
strumental training. Journal of Experimental Psychology, 64, 526-
(1986). The generalized matching law as a description of multiple-
532.
schedule responding. Journal of the Experimental Analysis ofBehav-
Eckerman, C. (1968, August). Deprivation conditions and chained
ior,45, 83-101.
schedule and multiple schedule performance. Paper presented at the
Mowrer, O. H., & Jones, H. (1945). Habit strength as a function of the
convention of the American Psychological Association, San Fran-
pattern of reinforcement. Journal of Experimental Psychology, 35,
cisco, CA.
Elsmore, T. F. (1971). Effects of response effort on discrimination per- 293-311.
formance. Psychological Record, 21, 17-24. Nation, J. R., & Woods, D. J. (1980). Persistence: The role of partial
Elsmore, T. F., & Hursh, S. (1982). Orcadian rhythms in operant behav- reinforcement in psychotherapy. Journal of Experimental Psychol-
ior of animals under laboratory conditions. In F. M. Brown & R. C. ogy: General, 109, 175-207.
Graeber (Eds.), Rhythmic aspects of behavior (pp. 273-310). Hills- Nevin, J. A. (1974). Response strength in multiple schedules. Journal
dale, NJ: Erlbaum. of the Experimental Analysis of Behavior. 21, 389-408.
Path, S. J., Fields, L., Malott, M. K., & Grossen, D. (1983). Response Nevin, J. A. (1979). Reinforcement schedules and response strength. In
56 JOHN A. NEV1N

M. D. Zeiler & P. Harzem (Eds.), Reinforcement and the organization PRF and CRF training in different runways, at different times of day,
of behaviour (pp. 117-158). Chichester, England: Wiley. by different experimenters. Psychonomic Science, 11, 315-316.
Nevin, J. A. (1984). Pavlovian determiners of behavioral momentum. Rescorla, R. A. (1977). Pavlovian second-order conditioning: Some im-
Animal Learning and Behavior, 12,363-370. plications for instrumental behavior. In H. Davis & H. M. B. Hurwitz
Nevin, J. A. (1985, May). Reinforcement, extinction, and behavioral (Eds.), Operant-Pavlovian interactions (pp. 133-164). Hillsdale, NJ:
momentum. Invited address at the meetings of the Association for Erlbaum.
Behavior Analysis, Columbus, OH. Skinner, B. F. (1938). The behavior of organisms. New York: Appleton-
Nevin, J. A., Mandell, C, & Atak, J. R. (1983). The analysis of behav- Century-Crofts.
ioral momentum. Journal of the Experimental Analysis of Behavior, Staddon, J. E. R. (1975). Learning as adaptation. In W. K. Estes (Ed.),
39,49-59. Handbook of learning and cognitive processes. Vol. 2: Conditioning
Nevin, 3. A., Mandell, C, & Yarensky, P. (1981). Response rate and and behavior theory (pp. 37-98). Hillsdale, NJ: Erlbaum.
resistance to change in chained schedules. Journal of Experimental Theios, J. (1962). The partial reinforcement effect sustained through
Psychology: Animal Behavior Processes, 7, 278-294. blocks of continuous reinforcement. Journal of Experimental Psy-
Nevin, J. A., Smith, L. D., & Roberts, J. (1987). Does contingent rein- chology, 64, 1-6.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

forcement strengthen operant behavior? Journal of the Experimental vom Saal, W. (1972). Choice between stimuli previously presented sepa-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Analysis of Behavior. 48, 17-33. rately. Learning and Motivation, 3, 209-222.


Pavlik, W. B., & Carlton, P. L. (1965). A reversed partial-reinforcement Wagner, A. R. (1961). Effects of amount and percentage of reinforce-
effect. Journal of Experimental Psychology, 70, 417-423. ment and number of acquisition trials on conditioning and extinc-
Pavlik, W. B., Carlton, P. L., & Hughes, R. A. (1965). Partial reinforce- tion. Journal of Experimental Psychology, 62,234-242.
ment effects in a runway: Between- and within-Ss. Psychonomic Sci- Weinstock, S. (1954). Resistance to extinction of a running response
ence, 3, 203-204. following partial reinforcement under widely spaced trials. Journal
Pavlik, W. B., Carlton, P. L., Lehr, R., & Hendrickson, C. (1967). Sup- of Comparative and Physiological Psychology, 47,318-322.
plementary reports: A reversed PRE. Journal of Experimental Psy- Wilson, M. P. (1954a). Periodic reinforcement interval and number of
chology, 75, 274-276. periodic reinforcements as parameters of response strength. Unpub-
Pavlik, W. B., Carlton, P. L., & Manto, P. G. (1965). A further study of lished doctoral dissertation (includes data tables not in Wilson,
the partial reinforcement effect within-subjects. Psychonomic Sci- 1954b), Columbia University, New York.
ence, 3, 533-534. Wilson, M. P. (1954b). Periodic reinforcement interval and number of
Pavlik, W. B., & Collier, A. C. (1977). Magnitude and schedule of rein- periodic reinforcements as parameters of response strength. Journal
forcement in rats' resistance to extinction—within subjects. Ameri- of Comparative and Physiological Psychology, 47, 51-56.
can Journal of Psychology, 90, 195-205.
Pliskoff, S. S., Shull, R. L., & Gollub, L. R. (1968). The relation between
response rates and reinforcement rates in a multiple schedule. Jour- Received August 11, 1986
nal of the Experimental Analysis of Behavior, 11, 271-284. Revision received April 9,1987
Rashotte, M. E., & Amsel, A. (1968). The generalized PRE: Within-S Accepted May 5, 1987 •

Call for Nominations for Editor of the Journal of Abnormal Psychology

The Publications and Communications Board has opened nominations for the editorship of the
Journal of Abnormal Psychology for the years 1990-1995. Don Fowles is the incumbent editor.
Candidates must be members of APA and should be available to start receiving manuscripts in
early 1989 to prepare for issues published in 1990. Please note that the P&C Board encourages
more participation by women and ethnic minority men and women in the publication process
and would particularly welcome such nominees. To nominate candidates, prepare a statement
of one page or less in support of each candidate. Submit nominations no later than March 1,
1988, to

Amado Padilla
Department of Psychology
University of California-Los Angeles
Los Angeles, California 90024

Other members of the search committee are Michael J. Goldstein and Anthony J. Marsella.

You might also like