You are on page 1of 15

The Behavior Analyst 2012, 35, 213–227 No.

2 (Fall)

Operant Variability: A Conceptual Analysis


Lourenço de Souza Barba
Universidade de São Paulo, Brazil
Some researchers claim that variability is an operant dimension of behavior. The present paper
reviews the concept of operant behavior and emphasizes that differentiation is the behavioral
process that demonstrates an operant relation. Differentiation is conceived as change in the overlap
between two probability distributions: the distribution of reinforcement probability as a function of
some response property (S distribution) and the probability distribution of the response property
itself (R distribution). This concept implies that the differentiation process can be measured only if
S distribution and R distribution are both established on the same response property. To determine
whether the differentially reinforced behavioral variability fits the proposed concept of operant
behavior, I examine the main procedures (lag n and threshold procedures) and the main dependent
variable (U value) employed in the studies of operant variability. Because lag n and threshold
procedures establish their S distributions on properties distinct from U value, differentiation
cannot be measured over the change in U value. I conclude that studies of operant variability
have failed to provide a direct demonstration that variability is an operant dimension of behavior.
Hence, studies in which measures of variability provide a basis to measure differentiation can
better support the claim that variability is an operant dimension of behavior.
Key words: operant variability, lag n procedure, threshold procedure, U value

Neuringer (2002, 2003, 2004, 2009) view of operant variability might


claims that variability is an operant contribute to the enrichment of this
dimension of behavior. Like other area of study. Therefore the present
dimensions of behavior (force, topog- paper provides a conceptual analysis
raphy, duration, location), variability of Neuringer’s claim in light of Catan-
is, according to Neuringer, a dimen- ia’s (1973, 1998) concept of operant
sion that may control consequences behavior. To perform this analysis, the
of behavior and may be controlled by present work (a) reviews Catania’s
them. Many studies have shown that concept of operant behavior, empha-
some measures of behavioral variabil- sizing that differentiation is the behav-
ity are, in fact, precisely controlled by ioral process that demonstrates an
contingencies of reinforcement, and operant relation; (b) examines the
they have given rise to behavioral procedures and measures of variability
technologies that have been success- often adopted in studies of operant
fully applied in a variety of fields, such variability; (c) shows that the most
as autism, depression, skills training, common measure of variability does
problem solving, and creativity (Neur- not enable one to measure differenti-
inger, 2002). The present article does ation process in these studies; and (d)
not contest that, as a practical matter, claims that procedures in which mea-
variability in behavior can be modi- sures of variability allow measurement
fied by contingencies of reinforce- of differentiation can offer better
ment. However, despite the successful evidence of operant variability.
application of technologies derived
from variability studies, an analytic OPERANT BEHAVIOR
According to Skinner (1969,
I thank Patrick Diamond for reviewing the
style of this paper. I also thank Armando pp. 105–132) an operant relation is
Machado for his helpful comments on its demonstrated when some responses
content. Finally, I thank Maria de Lourdes produce environmental changes and
Passos for her helpful comments on the these changes modify the rate at
content and style of this article.
Correspondence concerning this article should which the responses are emitted.
be addressed to the author at lourenbarba@ Therefore, his definition of operant
yahoo.com.br. behavior involves a relation between

213
214 LOURENÇO DE SOUZA BARBA

some responses and their conse- When an experimenter arranges


quences (a contingency relation) and contingencies that differentially rein-
the effects of this contingency rela- force some response, he or she adopts
tion on the rate of responding. Con- a reinforcement criterion based on
sequences that increase the probability some response property. This rein-
of responding are called reinforcers. forcement criterion defines a descrip-
Skinner pointed out that reinforce- tive response class and provides the
ment is always contingent on some basis for differential reinforcement,
properties of responses. The delivery that is, the reinforcement of some
of a food pellet that reinforces lever responses but not others (Catania,
presses is contingent on certain prop- 1998). According to Catania (1973), a
erties of the responses. To close the descriptive response class may also
circuit, a lever press must have some be defined in terms of a distribution
minimal force, and must occur at some of conditional probabilities of rein-
range of angles and location in space. forcement. This distribution (a stim-
Hence, reinforcement is contingent on ulus distribution or S distribution, as
responses that meet those criteria. I will refer to it henceforth) function-
‘‘Thus a set of contingencies defines ally relates probabilities of reinforce-
an operant,’’ according to Skinner ment to the values some response
(1969, p. 131). Each of these contin- property may assume. If the experi-
gencies is established on particular menter selects the response prop-
properties of responses. Responses erty force and requires a minimum
with the properties required by the amount of force (x) for a reinforcer
contingencies, and whose frequency is to be presented, he or she establishes
affected by the reinforcement, consti- an S distribution that is shown in
tute an operant response class. Figure 1.
Skinner (1969) also noted that Before differential reinforcement
‘‘Contingencies cannot always be takes place, the property force exhib-
detected on a given occasion. Al- its a particular probability distribu-
though a response is reinforced, we tion in the organism’s repertoire.
cannot be sure what property satisfied Responses with a force higher than
the contingencies and hence defines x may be rare when differential
the operant’’ (p. 131). It should be reinforcement begins, but differential
noted, however, that the specific reinforcement can engender an in-
properties that satisfy contingencies crease in the frequency of responses
and the properties that come to occur with a force higher than x. Thus, the
more often may not be the same ones. differential reinforcement of respons-
Reinforcement can affect the frequen- es with a force higher than x can
cy of responses that do not produce it. change the probability distribution of
This spread of the effects of reinforce- the property force in the organism’s
ment to responses that do not pro- repertoire. Hence, it is possible to
duce it is called induction (Catania, distinguish two probability distribu-
1998). Catania (1973, 1998) proposed tions: one that prevails before differ-
a concept of operant behavior that ential reinforcement goes into effect
distinguishes these two response class- (an initial, or baseline, distribution of
es: the class of responses that produce some response property, hereafter the
some consequences (descriptive re- Ri distribution) and another distri-
sponse class) and the class of respons- bution that prevails after differential
es whose frequency is affected by the reinforcement had an effect on the
contingency relation (functional re- baseline distribution (a final distribu-
sponse class). Catania (1973, 1998) tion of the response property, the Rf
claimed that the concept of the distribution). According to Catania’s
operant emerges from the relation (1973) proposal, the probabilities that
between these two response classes. define the Ri and Rf distributions are
OPERANT VARIABILITY 215

occurs when ‘‘the distribution of


emitted responses [comes] to conform
closely to the boundaries of the class
of reinforced responses’’ (Catania,
1998, p. 117). In other words, a dif-
ferentiation process occurs when the
Rf distribution more closely con-
forms to the S distribution than does
the Ri distribution. Thus, the differ-
entiation process implies that the
Figure 1. S distribution that is established
overlap between the S distribution
when the experimenter selects the response and the Rf distribution is higher than
property of force and requires a minimum the overlap between the S distribu-
amount of force (x) to reinforce the response. tion and the Ri distribution. Catania
If the force exerted by response is lower than (1973) proposed the standard Pear-
x, the reinforcement probability is 0. If the
force is equal to or higher than x, the son product-moment correlation co-
reinforcement probability is 1. efficient (r) as a measure of the degree
of overlap between an S and an R
distribution. Because a differentiation
process entails an increase in the
derived from the relative frequencies magnitude of the correlation, the var-
at which the values the property as- iation of r constitutes the measure of
sumes are observed to occur. Hence, if this process. According to Catania
responses with a force higher than x (1973), ‘‘An operant relation is dem-
initially occur at a low relative fre- onstrated when the distribution of
quency in the organism’s repertoire stimulus probabilities along some di-
(Ri), it may be said, alternatively, that mension of responding increases the
such responses occur with a low prob- correlation between that distribution
ability. A continuous property like and the distribution of response prob-
force may assume noninteger values. abilities along the same dimension’’
In such a case, the range of values the (p. 106). Thus, Catania’s (1973, 1998)
property may assume can be divided concept of the operant also implies
into equal intervals. The experimenter that differentiation is the process that
can therefore measure the relative demonstrates an operant relation.
frequency at which responses in each Therefore, the definition of an
interval occur. These relative frequen- operant proposed by Catania (1973)
cies represent the probabilities that involves three distributions that rep-
define the Ri and Rf distributions, resent probabilities ‘‘as a function of
according to Catania. some property of the response’’
The change in the probability (p. 105). These distributions (all with
distribution of the property force is functional properties) are the follow-
an example of what Skinner called ing: an Ri distribution, an S distribu-
‘‘quantitative differentiation’’ (1938, tion, and an Rf distribution. Figure 2
p. 338). Skinner’s conceptual frame- shows a hypothetical case in which
work puts response differentiation at the overlap between an S distribution
the center of operant conditioning, and an Rf distribution is higher than
given that the latter process necessar- the overlap between the S distribution
ily involves the former (Galbicka, and an Ri distribution. The change in
1988). It could be said, moreover, overlap shows the usual effect of the
that ‘‘response differentiation is op- differential reinforcement of respons-
erant conditioning, and vice versa’’ es with force higher than x.
(Galbicka, 1988, p. 343). A differen- It should be noted that a change
tiation process, which is the usual in correlation requires at least two
effect of differential reinforcement, values (an initial correlation and a
216 LOURENÇO DE SOUZA BARBA

Figure 2. A hypothetical case in which the overlap between an S distribution and a final R
distribution is higher than the overlap between the S distribution and an initial R distribution.
The range of values that the property force may assume was divided into 10 class intervals to
establish R distributions. Initially, responses with force higher than x occur with low probability
(before the differential reinforcement of responses with force higher than x). The final R
distribution represents a performance in which responses with force higher than x occur with
high probability (after the differential reinforcement of responses with force higher than x).

final correlation), and these correla- response properties if these properties


tions should only be calculated for S were correlated with force. If higher
and R distributions that are estab- forces required longer durations,
lished on the same property. Thus, for example, the procedure would
even if differential reinforcement also change the Ri distribution of the
changes R distributions of several property duration. However, only the
response properties, only the change change in Ri distribution established
in the R distribution of that property on the property force would allow
on which the S distribution was one to measure the differentiation
established enables one to measure process, because the S distribution
the differentiation process. was established on that property.
Correlated properties. In the exam- When an experimenter differentially
ple above, the S distribution was reinforces responses with forces high-
established on the property force er than x, he or she manipulates an
and the differentiation process would aspect of the organism’s environment.
have to be measured by a change in Manipulation of a single environmen-
the Ri distribution established on tal variable can change several re-
that property. However, the differen- sponse properties. Response differen-
tial reinforcement of responses with a tiation is the behavioral process that
force higher than x could also affect links the manipulation (differential
the probability distribution of other reinforcement) to the behavioral
OPERANT VARIABILITY 217

change that must be considered in different only if it fell within a


order to show operant conditioning. temporal class interval different from
If variability is an operant dimen- that of the immediately preceding
sion of behavior, as Neuringer (2002, IRT (with temporal class intervals
2003, 2009) claims, it should be defined by the experimenters). Thus,
possible to differentially reinforce var- Schoenfeld et al. used a lag n
iable behavior and to obtain a differ- procedure with n 5 1. The subjects
entiation process as a result. That is, in this study exhibited a pattern of
if Neuringer’s supposition holds, it alternating IRTs belonging to two
should be possible to establish an S class intervals. The rats that dis-
distribution on the property variability played this pattern earned all avail-
and to measure the differentiation able reinforcers with the lowest pos-
process by modifying an Ri distribu- sible level of variability.
tion established on this same proper- Schwartz (1982) also tried to dif-
ty. In fact, the differentiation process ferentially reinforce variable behavior
could only be measured by changes in with a lag n procedure, but he con-
distributions from Ri to Rf estab- ditioned key pecking in pigeons
lished on the property variability. The distributed across two keys, a left
analysis of methods employed in stud- key (L) and a right key (R). Rein-
ies of operant variability will show, forcement was contingent on exactly
however, that the effects of differen- four pecks on the left key and exactly
tial reinforcement of variable behavior four pecks on the right key, in any
have often been measured through order. In addition, the contingency
changes in response properties on required pigeons to vary response
which the S distribution was not sequences. The variability require-
established. The present analysis focus- ment was met only if the pattern of
es on procedures and dependent vari- the current sequence differed from the
ables that have been the most common immediately preceding n patterns. In
in the studies of operant variability, an initial series of eight-peck sequenc-
specifically, the lag n, threshold proce- es like LLLLRRRR, LLRRLLRR,
dures, percentage of reinforcement, LLRRLLRR, LLLLRRRR, LRLR-
and U value as dependent variables. LRLR (emitted in that order, with n
5 1), only the third sequence would
PROCEDURES USED fail to be reinforced because it re-
IN STUDIES OF plicated the preceding sequence.
OPERANT VARIABILITY Schwartz obtained low and decreas-
ing variability during lag n sessions.
The Lag n Procedure In addition, the variability contin-
A lag n contingency reinforces gency involved some intermittency
responses that differ from each of of reinforcement, because sequences
the previous n responses (differences that did not meet the variability
being defined in terms of some requirement were not reinforced.
response property). For example, Schwartz noted that this intermitten-
when Schoenfeld, Harris, and Farmer cy of reinforcement might have en-
(1966) tried to differentially reinforce gendered part of the variability ob-
variable behavior, they first condi- tained under lag n contingency and
tioned a panel-press response in rats. concluded that the lag n contingency
In a second phase, a panel press was failed to engender operant variability.
reinforced only if it ended an inter- Page and Neuringer (1985) used
response time (IRT), or a postrein- a similar procedure but eliminated
forcement pause, that was different the constraint that exactly four
from the IRT (or postreinforcement pecks occur on each key. Any eight-
pause) of the immediately preceding response sequence would be rein-
response. An IRT was considered forced if it met the lag requirement
218 LOURENÇO DE SOUZA BARBA

TABLE 1
Some studies of operant variability that required subjects to emit sequences of
responses across two or three operanda and adopted the lag n procedure or
threshold procedure. They employed the U value as measure of variability
Lag n procedure Threshold procedure
Cohen, Neuringer, and Rhodes (1990) Denney and Neuringer (1998)
Morgan and Neuringer (1990) Neuringer, Deiss, and Olson (2000)
Neuringer (1991) Neuringer, Kornell, and Olufs (2001)
Neuringer and Huntley (1992) Doughty and Lattal (2001)
Hunziker, Saldana, and Neuringer (1996) Grunow and Neuringer (2002)
Abreu-Rodrigues, Lattal, Dos Santos, and Matos (2005)
Odum, Ward, Barnes, and Burke (2006)
Ward, Kynaston, Bailey, and Odum (2008)

of showing a different configuration studies that adopted a lag n procedure


of L and R responses from each of and required subjects to emit sequenc-
the preceding n sequences. Page and es of responses across two or three
Neuringer conducted six experiments operanda.
and demonstrated that the degree of
sequence variability was a function of Threshold Procedure
the lag requirements. The higher the
value assigned to n, the higher the Denney and Neuringer (1998) dif-
levels of obtained variability. Page ferentially reinforced four-response
and Neuringer also programmed a sequences across two operanda, but
contingency that required birds to they adopted a reinforcement criteri-
emit eight-response sequences, but on based on the relative frequency at
without a variability requirement (a which the current sequence had
yoked variable-ratio [VR] schedule). occurred in the recent past. A se-
In the yoke procedure, the reinforcer quence was reinforced only if its
presentations were yoked to reinforced relative frequency, computed on the
trials that occurred during sessions of set of sequences emitted during the
the lag n contingency. Thus, the current and the previous session, was
subjects in the yoke procedure were less than or equal to a certain thresh-
exposed to the same intermittency of old value. After each trial, the relative
reinforcement as in the lag n contin- frequency of each different sequence
gency, but in the former condition the was recalculated and, when the se-
pigeons were not required to vary, quence was reinforced, the calculations
whereas in the latter the subjects were involved an adjustment by a weighting
required to vary response sequences. coefficient. The adjusted value was the
Results showed that the lag n contin- weighted relative frequency of the
gency generated greater variation in sequence. Any current sequence was
response sequences than did the reinforced only if its weighted relative
yoked VR procedure. This finding frequency was less than or equal to the
demonstrated that variability engen- threshold value. This procedure differ-
dered by the lag n contingency was entially reinforced sequences that oc-
not due simply to intermittency of curred infrequently in the recent past.
reinforcement. Finally, Page and Denney and Neuringer’s threshold
Neuringer demonstrated control of procedure engendered more variability
variable behavior by discriminative than did a control procedure in which
stimuli. These results led Page and variability was not required (but in
Neuringer to conclude that variabil- which the reinforcement frequencies
ity is an operant dimension of behav- were equated to those in the variability
ior. The left column of Table 1 lists condition).
OPERANT VARIABILITY 219

The lag n procedure and the responses, the locations of L and R


threshold procedure have a common responses, and the number and loca-
feature that unifies them. In both tions of switches, are properties that
cases, reinforcement depends on the define the sequence configuration.
frequency at which a sequence oc- Some other properties, such as se-
curred in the recent past: the relative quence duration or the force of each
frequency in the threshold procedure response, do not affect sequence
and the absolute frequency in lag n configuration. All of these properties
procedure. The right column of are intrinsic to the sequence, because
Table 1 lists studies that required it is possible to ascertain them by
subjects to emit sequences of respons- examining only the sequence itself.
es across two or three operanda and There are, however, some proper-
adopted a threshold procedure. ties of each individual sequence that
assume values only when the se-
Lag n Procedure, Threshold quence is taken within a particular
Procedure, and the S Distributions series of sequences. Suppose the
following series of sequences LLRR,
Schwartz (1982) raised some prob- RRRR, LRLR, RLRL is emitted.
lems with the response class defini- We can define relational properties
tion as applied to the study of for any sequence. For the last se-
operant variability: ‘‘Suppose we quence of the series (RLRL), for
wanted to use reinforcement to in- example, the property absolute fre-
crease behavioral variability. What quency at which a sequence containing
would define the operant class on the same number of switches occurred
which reinforcement depends? What in the previous three sequences (one),
objective property of responses would or the relative frequency at which a
unite them into a class? It is clear that sequence containing the same number
there is no such property’’ (p. 178). of L responses occurred in the previous
However, both the lag n and the three sequences (two or three), or the
threshold contingency provide rein- absolute frequency at which a sequence
forcement to some sequences and not showing the same configuration oc-
to others. Thus, both procedures curred in the previous three sequences
differentially reinforce the sequences (zero). These are examples of rela-
subjects emit, and differential rein- tional properties of sequences defined
forcement implies a reinforcement only by examining the series of
criterion. Because these contingencies sequences in which the current se-
differentially reinforce sequences, quence is inserted.
reinforcement criteria must be estab- A reinforcement criterion may be
lished on sequence properties. The established on any sequence property.
number of L responses, the number of The lag n procedure establishes a
R responses, and their duration are reinforcement criterion based on the
some sequence properties. If a subject property absolute frequency at which
shifts from one operandum (e.g., the the current sequence occurred in
left key) to another one (e.g., the right the previous n trials. The threshold
key) during the emission of a se- procedure establishes a reinforce-
quence, it executes a switch between ment criterion based on the property
the operanda. A switch involves, weighted relative frequency at which
therefore, an intrasequential alterna- the current sequence occurred in the
tion between the operanda. The past.
sequences LLRR, RRRR, LRRL, Once a reinforcement criterion is
LRLR contain, respectively, 1, 0, 2, defined, a descriptive class of re-
and 3 switches between the operanda. sponse is also defined. This descrip-
Some properties, such as the number tive class may in turn be expressed in
of L responses, the number of R terms of S distribution. The threshold
220 LOURENÇO DE SOUZA BARBA

TABLE 2
Values assigned to properties FP1, FP2, FP3, and FP4 after each sequence
emission if the subject alternates systematically only two different sequences (A
and B). FP1 corresponds to the property absolute frequency at which the
sequence occurred in the immediately preceding trial, FP2 corresponds to the
property absolute frequency at which the sequence occurred in the immediately
preceding two trials, and so on
A B A B A B A B A B A
FP1 0 0 0 0 0 0 0 0 0 0
FP2 1 1 1 1 1 1 1 1 1
FP3 1 1 1 1 1 1 1 1
FP4 2 2 2 2 2 2 2

procedure reinforces sequences (e.g., LLLL) and B is another se-


whose weighted relative frequency is quence (e.g., RRRR). This hypothet-
less than or equal to a threshold ical stable pattern of sequence emis-
value. Thus, the threshold procedure sion may be used to illustrate how
establishes its S distribution on the one could establish the R distribu-
property weighted relative frequency tions on the properties of FP1, FP2,
at which the sequence occurred in the FP3, and FP4. Table 2 shows how
past (WRF). WRF is a relational one could assign values to these four
property of sequences and ranges properties after each sequence emis-
from 0 to 1. If a sequence presents a sion. In the fourth column, the FP1
WRF value less than or equal to the property assumes the value of 0
threshold value it produces reinforce- because the B sequence has not
ment; otherwise, the sequence never occurred in the immediately previous
produces reinforcement. The lag n emission. In the fifth column, the
procedure reinforces only sequences FP2 property assumes the value of 1
whose absolute frequency in the because the A sequence has occurred
previous n trials is equal to zero. once in the immediately two previous
Thus, a lag n contingency establishes emissions. In the sixth column, the
its S distribution on the sequence FP3 property assumes the value of 1
property absolute frequency at which because the B sequence has occurred
the sequence occurred in the previous n once in the immediately three previ-
trials (FPN). FPN is also a relational ous emissions. In the seventh column,
property of sequences and may as- the FP4 property assumes the value
sume entire values between 0 and n. of 2 because the A sequence has
Sequences whose FPN is equal to occurred twice in the immediately
zero are reinforced with a probability four previous emissions, and so on.
equal to 1. Sequences whose FPN is Table 3 shows the frequency cal-
positive are reinforced with probabil- culations on the values taken from
ity equal to 0. Table 2. It shows, for example, that
in the pattern of alternation between
R Distributions two different sequences (A, B, A, B,
A, B, …,), (a) the FP1 property,
It is also possible to establish an Ri which may assume the values of 0 or
distribution on properties WRF and 1, assumes the value of 0 for any
FPN. If a subject emits only two sequence considered (i.e., the relative
different sequences and alternates frequency at which the FP1 property
them systematically, it produces a assumes the value of 0 is equal to 1);
series of sequence like this: A, B, A, (b) the FP2 property, which may
B, A, B, …, in which A is a sequence assume the values of 0, 1, or 2, never
OPERANT VARIABILITY 221

TABLE 3 some sequences into a class, at least


into a descriptive class.
The calculation of relative frequencies
on the values taken from Table 2 DEPENDENT VARIABLES
IN STUDIES OF
Absolute Relative
Property Range frequency frequency
OPERANT VARIABILITY
FP1 0 10 1 According to Neuringer (2002), ‘‘U
1 0 0 value has been the most commonly
FP2 0 0 0 employed measure of operant vari-
1 9 1 ability’’ (p. 683). The U stands for
2 0 0
FP3 0 0 0 uncertainty, and U value derives from
1 8 1 information theory (Attneave, 1959).
2 0 0 The U value essentially measures
3 0 0 levels of uncertainty. Page and Neur-
FP4 0 0 0
1 0 0 inger (1985) calculated a U value at
2 7 1 three levels: U1 was calculated from
3 0 0 the individual relative frequencies at
4 0 0 which the responses L and R oc-
curred; U2 was calculated from the
relative frequencies at which the pairs
assumes the value of 2 (i.e., the of responses (LL, LR, RL, RR)
relative frequency at which the FP2 occurred; and U3 was calculated from
property assumes the value of 2 is the relative frequencies at which the
equal to 0); (c) the FP4 property, triplets of responses (LLL, LLR,
which may assume the values of 0, 1, LRL, LRR, …) occurred. U values
2, 3, or 4, assumes the value of 2 for show how much the responding is
any sequence considered (i.e., the biased, because a perfectly variable
relative frequency at which the FP4 (or unbiased) sequence pattern of L
property assumes the value of 2 is and R responses contains equal fre-
equal to 1). Figure 3 represents graph- quencies of individual responses (L
ically the R distributions of each and R), equal frequencies of pairs of
property. responses (LL, LR, RL, RR), and
Despite Schwartz’s (1982) objec- equal frequencies of triplets of re-
tion to behavioral variability as an sponses (LLL, LLR, LRL, LRR, …).
operant class, the lag n and the U value may also be calculated
threshold procedures select objective, from the relative frequencies at which
although relational, properties of each of all different sequences occur
responses (or response sequences) (Neuringer, 2002). The number of
on which reinforcement depends. operanda employed (two or three)
These contingencies define a rein- and the number of responses per
forcement criterion (and, therefore, sequence define a universe of all
establish S distributions) based on possible different sequence configu-
those properties, which, in turn, unite rations. A procedure that requires

Figure 3. Graphic representations of R distributions established on the sequence properties of


FP1, FP2, FP3, and FP4.
222 LOURENÇO DE SOUZA BARBA

four-response sequences across two pants to draw figures on a computer


operanda defines a universe of 16 (24) screen.
different sequences (different relative Note that neither the lag n nor the
to configurations). In this case, the U threshold procedure establishes their
value is a function of the relative S distributions on U value because
frequency at which each of the neither specifies a single conditional
different sequences occurs and the probability of reinforcement for each
universe of all different sequences. U value.
Therefore, the change in U value
reflects the change in the frequency Differentiation Process and Measures
distribution of all the different se- of Variability
quences. U value ranges from 0 to 1. As previously discussed, a differ-
When all different sequences occur entiation process should be measured
equally often, the U value is equal to only through the change of an Ri
1. When only a single sequence distribution in the same property as
occurs, the U value is equal to 0. that on which the S distribution was
Thus, the more uniform the frequen- established. Thus, even if a contin-
cy distribution of all different se- gency increases levels of behavioral
quences is, the more the U value variability but in fact differentially
approximates 1. Whenever U value is reinforces some sequences on the
mentioned henceforth it refers to U basis of properties other than their
value calculated from the relative variability, the increase in variability
frequencies at which each different levels, whatever it is, does not allow
sequence occurs.1 The percentage or one to measure a differentiation
proportion of reinforced sequences is process. In this case, the increase in
another common dependent variable variability levels represents a second-
employed. It equals the ratio of the ary effect of differential reinforce-
number of reinforced sequences to ment. Machado (1997) demonstrated
number of trials. such an effect. He argued that the lag
All of the studies listed in Table 1 n contingency of Page and Neuringer
employed U value as a measure of (1985) might have differentially rein-
variability. All the studies in the left forced sequences that contain an
column calculated the percentage of intermediate number of switches,
reinforced sequences or the percentage and the obtained variability therefore
of correct sequences (except Hunziker, might have been a by-product of such
Saldana, & Neuringer, 1996). Neurin- differential reinforcement. If Macha-
ger, Kornell, and Olufs (2001) also do’s hypothesis is correct, Page and
calculated percentage of correct se- Neuringer’s findings demonstrated
quences. Ross and Neuringer (2002) that the number of switches per
programmed a threshold contingency sequence is an operant dimension of
and calculated a U value, but they behavior, and the variability generat-
used human subjects and employed a ed is an indirect effect of the rein-
procedure that required the partici- forcement of sequences that contain
an intermediate number of switches.
Machado’s hypothesis implies the
1
U value is given by the following formula: following prediction: If a contingency
directly reinforces sequences that con-
XN
½pi: log2 (pi)
{ , tain intermediate number of switches,
i~1
log2 (N) it will engender high levels of vari-
ability. Machado tested his hypothesis
where N represents the number of all possible
different sequences and pi represents the by establishing a reinforcement crite-
relative frequency at which each of them rion based on the property number of
occurs in a given set of sequences. switches of the sequence. He used
OPERANT VARIABILITY 223

could not enable one to measure a


differentiation process, because the S
distribution was not established on
the property proportion of different
sequences. Note, however, that the
proportion of different sequences
could be taken as a reinforcement
criterion. After the emission of each
sequence, Machado could calculate
the proportion of different sequences
emitted up to that point and reinforce
Figure 4. S distribution Machado (1997) the current sequence only if the
established in Experiment 2. current proportion were higher than
a certain value. In this case, the vari-
pigeons and an experimental chamber ability measure would coincide with
equipped with two keys. An eight- the sequence property on which the
response sequence was required in reinforcement criterion would have
each trial. The procedures of Exper- been defined. If the change in propor-
iments 1 and 2 established a descrip- tion were controlled by such a con-
tive class on the sequence property tingency of reinforcement, Machado
number of switches. In Experiment 1, would hold that the proportion of
the subjects received food only if they different sequences is itself a relational
emitted a sequence that contained at sequence property that is sensitive to
least one switch (Group 1) or at least differential reinforcement. Proportion
two switches between the keys (Group of different sequences can be updated
2). In Experiment 2, Machado pro- after the emission of each sequence.
grammed a probabilistic schedule of Thus, the proportion is a sequence
reinforcement on the property number property in the sense that it is possible
of switches. Figure 4 shows the S dis- to assign to any emitted sequence
tribution Machado established in a specific value (the proportion of
Experiment 2. In Experiment 3, naive different sequences calculated over
pigeons were explicitly required to the set of all emitted sequences).
vary behavior (a Lag 25 contingency Hence, in Machado’s (1997) Ex-
was used). periments 1 and 2, the obtained
Machado (1997) calculated the variability represented, indeed, some
proportion of different sequences indirect effect of differential rein-
per session (that equaled the ratio of forcement. In these experiments, the
the number of different sequences to measure of variability was clearly
the number of trials) and the propor- dissociated from the sequence prop-
tion of reinforced sequences the birds erty on which the S distribution was
emitted. The former was the variabil- established (It should be noted, by
ity measure. The differential reinforce- the way, that the subjects could
ment of switching behavior produced obtain all the available reinforcement
substantial levels of variability. Ma- with no variability at all.) Machado
chado reported, however, that the Lag can, therefore, assert that his proce-
25 contingency engendered variability dure (in Experiments 1 and 2) shaped
levels slightly higher than did the the operant switching between the
differential reinforcement of interme- keys and the variability he obtained
diate number of switches. Barba and was a by-product of the differential
Hunziker (2002) found similar results. reinforcement of switching behavior.
Note that the increases in propor- This rationale holds because Macha-
tions of different sequences in Ma- do’s S distribution was established on
chado’s (1997) Experiments 1 and 2, the property number of switches (and
although they actually occurred, not on the property variability).
224 LOURENÇO DE SOUZA BARBA

A similar analysis seems to apply All of them adopt a variability mea-


to situations in which the experiment- sure that is distinct from the sequence
er differentially reinforces sequences properties on which the S distribu-
of responses with a lag n procedure tions are established. But this is not an
and measures the effects of differen- inevitable feature of studies about
tial reinforcement through the change operant variability. U value can be
in U value. In this case, the experi- itself a property of individual sequenc-
menter establishes the S distribution es and S distributions can be estab-
on the FPN property (and not on U lished on it, as will be seen below.
value). Therefore, the change in U
value does not allow the experimenter CUMULATIVE U VALUE AS A
to measure a differentiation process. SEQUENCE PROPERTY
It might be argued, perhaps, that
changes in U value and changes in According to Neuringer (2002),
the Ri distribution of the FPN different methods, including a lag n
property are correlated events (much contingency, threshold procedures,
as were the properties force and and differential reinforcement of
duration in my hypothetical example). switches versus repetitions, ‘‘increase
However, this correlation measure is or maintain variability by reinforcing
compromised because it depends on it’’ (p. 682). Neuringer did not men-
the n value and on the number of tion a procedure that takes U value as
different possible sequences. Sup- a reinforcement criterion, but such a
pose, for example, a subject emits procedure can be arranged. In a series
only a single four-response sequence of sequences, it is possible to assign a
with no switches under a contingency cumulative U value to each of them.
that does not require variability. If The cumulative U value of any
the subject is exposed to a Lag 3 current sequence is a function of the
contingency and comes to emit four frequencies of each different sequence
different sequences by alternating that had occurred up to that point and
them with equal frequency, the Ri of the number of all possible different
distribution under the property FP3 sequences. For example, after each
will suffer a substantial change (the emission of a complete sequence, a
largest possible change, indeed), but program might compute a cumulative
the U value (calculated on the U value and provide reinforcement
universe of 16 different sequences) only if the current cumulative U value
will reach only one half of its highest reached a minimal level. Before any
value. If this hypothetical case in- sequence is emitted, the program may
volves an eight-response sequence assume that the subject had emitted
(with the universe of 256 different every possible sequence exactly once
sequences), the change in U value will (like the threshold procedure pro-
be even lower. The same holds for grammed by Denney & Neuringer,
threshold procedure. A subject can 1998). To start with, therefore, the
perfectly meet the requirements of subject might have a cumulative U
such contingency and, at the same value equal to 1. This contingency
time, produce a small change in U establishes its S distribution on the
value if the scheduled threshold value sequence relational property cumula-
is high enough (or, in other words, if tive U value. This procedure provides
the contingency is very permissive). an identity between the most common
Therefore, the studies that use lag n measure of operant variability (U
or threshold procedures and employ value) and the sequence property on
U value as the variability measure which the differential reinforcement is
may be compared with Machado’s based. Like FPN or WRF, cumulative
(1997) Experiments 1 and 2 with U value is a relational property of
respect to the following characteristic: sequences. Only when U value is the
OPERANT VARIABILITY 225

property on which the S distribution lag n contingency, the U value is a


is established does the change in U ‘‘relatively molar measure of variabil-
value provide a basis to measure the ity’’ (p. 162), because it is calculated
differentiation process. As far as I from the frequencies at which each
know, a contingency that establishes sequence was emitted during an
the S distribution on the sequence entire session. The percentage of
property cumulative U value has not reinforced trials, on the other hand,
yet been arranged.2 is a measure affected by ‘‘more
molecular aspects of variability’’
Arbitrariness of U Value as a General (p. 162). If n is low (relative to the
Measure of Operant Variability universe of all possible different
sequences), a lag n contingency re-
If a contingency does not establish quires subjects to emit only sequences
its S distribution based on the U that differ from the few more recent
value, the adoption of the U value as ones. This contingency does not
the main dependent variable is to require subjects to emit equally often
a large extent arbitrary. Why, for all the possible different sequences.
example, does the U-value calcula- That is, it demands only local vari-
tion take the entire universe of all ability. Subjects that initially emit
different sequences? If a subject emits only a single sequence can meet the
only 10 different sequences during a requirement of such contingency (and
session, why is the U value calculated earn all the available reinforcement)
on the entire universe (e.g., 16 different with small changes in U value,
sequences, if a four-response sequence because the U value reflects the
is required) and not on the universe of uniformity of the frequency distribu-
10 different sequences? Only if a tion of all possible different sequenc-
contingency establishes its S distribu- es. Therefore, the change in U value
tion on the cumulative U value constitutes a rough measure of the
(calculated from the entire universe effects of differential reinforcement
of all different sequences) can the engendered by one such contingency.
experimenter satisfactorily justify the Indeed, a lag n contingency always
calculation of U value (as a dependent requires local variability, because the
variable) based on the entire universe completion of more recent sequences
of all different sequences. is never reinforced. Only as n increas-
es does the lag n contingency come to
Local Variability and require the subject to emit equally
Molar Variability often a greater number of different
As Odum, Ward, Barnes, and sequences.
Burke (2006) pointed out, under a On the other hand, a contingency
that imposes a minimal cumulative U
2
According to Machado (1997), the subjects value to provide reinforcement re-
exposed to lag n contingency learn to emit quires mainly molar variability. This
sequences with intermediate number of switch-
es, and the change in variability levels is a by-
contingency does not necessarily re-
product of this learning. That is, variability quire high levels of local variability.
obtained under lag n contingency might not be In fact, the same sequence could
a basic behavioral process. The present work produce reinforcement even if it were
does not address this question. This article emitted in consecutive trials if it was
argues that, if variability is an operant dimen-
sion of behavior, then the methods that sup- infrequent enough in the set of all
posedly demonstrate it must allow one to sequences already emitted. Converse-
measure differentiation process on the change ly, a sequence could not be rein-
in the dependent variable that represents vari- forced, even if it did not repeat the
ability levels. It is claimed here that this
measurement possibility is a necessary but not more recent sequences if its cumula-
a sufficient condition to demonstrate that tive U value does not reach the
variability is an operant dimension of behavior. minimal cumulative U value. This
226 LOURENÇO DE SOUZA BARBA

contingency primarily requires the from this hypothesis: If it is correct,


unbiased emission of different se- future research will be able to dem-
quences over extended periods of onstrate (e.g., by employing yoked
time. Under this contingency, the procedures) that contingencies that
change in U value enables one to require only local variability and
measure the differentiation process contingencies that require mainly
and the change in local variability molar variability may produce the
levels represents, in turn, a secondary same level of molar variability (e.g.,
effect of differential reinforcement. the same U value) with different
levels of local variability. This would
CONCLUSION be a tangible consequence of my
analytic investigation. If future re-
Neuringer (2002, 2003, 2004, 2009) search shows that behavior is differ-
has argued that variability is an entially sensitive to a requirement of
operant dimension of behavior, and local variability versus that of molar
has reported that U value has been variability, then Neuringer’s claim
the most common measure of oper- will be strengthened. Such a result
ant variability. If differentiation is the would refine the concept of operant
process that demonstrates an operant variability.
relation, only the change in a depen- As I pointed out in the title, this
dent variable that provides the basis article proposes a conceptual analysis
to measure a differentiation pro- of operant variability. The paper can
cess might unequivocally demon- be thought as a critique of the adop-
strate that variability is an operant tion of U value as a general measure of
dimension of behavior. If U value is operant variability in studies designed
not the property on which a rein- to demonstrate experimentally that
forcement criterion is established, the variability is an operant dimension of
change in U value does not provide behavior. By general measure I mean
the basis to measure a differentiation that U value is often adopted as the
process. In such a case, the U value is main variability measure indepen-
at best a correlate of the property on dently of the sequence property on
which the S distribution is estab- which the reinforcement criterion is
lished. Thus, the change in U value, established. This article provides an
taken as general measure of operant analytic approach to the question of
variability, seems to constitute a operant variability as long as it
crude measure of the effects produced deconstructs the variability, as a
by operant variability contingencies. complex behavioral dimension sensi-
Hence, methods of studying operant ble to environmental consequences,
variability in which the property that into some more elementary compo-
defines an S distribution and the nents (or properties), each one of
property on which the effects of which may be differentially sensitive
differential reinforcement are mea- to differential reinforcement. This
sured are the same might offer more analytic effort does not deny that
convincing evidence that variability is behavioral technologies derived from
an operant dimension of behavior. operant variability research are suc-
It is possible, in addition, that cessfully applied in several fields.
contingencies that require only local
variability and contingencies that
require primarily molar variability REFERENCES
will engender different patterns of
behavior. It is possible that distinct Abreu-Rodrigues, J., Lattal, K. A., Dos
Santos, C. V., & Matos, R. A. (2005).
contingencies select variability at a Variation, repetition, and choice. Journal of
molar level or variability at a local the Experimental Analysis of Behavior, 83,
level. A prediction can be derived 147–168. doi:10.1901/jeab.2005.33-03
OPERANT VARIABILITY 227

Attneave, F. (1959). Applications of informa- Neuringer, A. (2003). Creativity and rein-


tion theory to psychology: A summary of forced variability. In K. A. Lattal & P. N.
basic concepts. New York, NY: Holt, Chase (Eds.), Behavior theory and philoso-
Rinehart & Winston. phy (pp. 323–338). New York, NY: Klewer
Barba, L. S., & Hunziker, M. H. L. (2002). Academic/Plenum.
Variabilidade comportamental produzida Neuringer, A. (2004). Reinforced variability in
por dois esquemas de reforçamento [Behav- animals and people. American Psychologist,
ioral variability produced by two reinforce- 59, 891–906. doi:10.1037/0003-066X.59.9.
ment schedules]. Acta Comportamentalia, 891
10(1), 5–22. Neuringer, A. (2009). Operant variability and
Catania, A. C. (1973). The concept of the the power of reinforcement. The Behavior
operant in the analysis of behavior. Behav- Analyst Today, 10, 319–343.
iorism, 1, 103–116. Neuringer, A., Deiss, C., & Olson, G. (2000).
Catania, A. C. (1998). Learning (4th ed.). Reinforced variability and operant learning.
Upper Saddle River, NJ: Prentice Hall. Journal of Experimental Psychology: Animal
Cohen, L., Neuringer, A., & Rhodes, D. Behavior Processes, 26, 98–111. doi:10.1037/
(1990). Effects of ethanol on reinforced vari- 0097-7403.26.1.98
ations and repetitions by rats under a multi- Neuringer, A., & Huntley, R. W. (1992).
ple schedule. Journal of the Experimental Reinforced variability in rats: Effects of
Analysis of Behavior, 54, 1–12. doi:10. gender, age and contingency. Physiology &
1901/jeab.1990.54-1 Behavior, 51, 145–149. doi:10.1016/0031-
Denney, J., & Neuringer, A. (1998). Behav- 9384(92)90216-O
ioral variability is controlled by discrimina- Neuringer, A., Kornell, N., & Olufs, M.
tive stimuli. Animal Learning & Behavior, (2001). Stability and variability in extinc-
26, 154–162. doi:10.3758/BF03199208 tion. Journal of Experimental Psychology:
Doughty, A. H., & Lattal, K. A. (2001). Animal Behavior Processes, 27, 79–94.
Resistance to change of operant variation doi:10.1037/0097-7403.27.1.79
and repetition. Journal of the Experimental Odum, A. L., Ward, R. D., Barnes, C. A., &
Analysis of Behavior, 76, 195–215. doi:10.
Burke, K. A. (2006). The effects of delayed
1901/jeab.2001.76-195
reinforcement on variability and repetition
Galbicka, G. (1988). Differentiating The
of response sequences. Journal of the Ex-
Behavior of Organisms. Journal of the
perimental Analysis of Behavior, 86,
Experimental Analysis of Behavior, 50,
159–179. doi:10.1901/jeab.2006.58-05
343–354. doi:10.1901/jeab.1988.50-343
Grunow, A., & Neuringer, A. (2002). Learn- Page, S., & Neuringer, A. (1985). Variability is
ing to vary and varying to learn. Psycho- an operant. Journal of Experimental Psy-
nomic Bulletin & Review, 9, 250–258. chology: Animal Behavior Processes, 11,
Hunziker, M. H. L., Saldana, L., & Neur- 429–452. doi:10.1037/0097-7403.11.3.429
inger, A. (1996). Behavioral variability in Ross, C., & Neuringer, A. (2002). Reinforce-
SHR and WKY rats as a function of rearing ment of variations and repetitions along
environment and reinforcement contingen- three independent response dimensions.
cy. Journal of the Experimental Analysis of Behavioural Processes, 57, 199–209. doi:10.
Behavior, 65, 129–144. doi:10.1901/jeab. 1016/S0376-6357(02)00014-1
1996.65-129 Schoenfeld, W. N., Harris, A. H., & Farmer,
Machado, A. (1997). Increasing the variability of J. (1966). Conditioning response variability.
response sequences in pigeons by adjusting the Psychological Reports, 19, 551–557. doi:10.
frequency of switching between two keys. 2466/pr0.1966.19.2.551
Journal of the Experimental Analysis of Behav- Schwartz, B. (1982). Failure to produce
ior, 68, 1–25. doi:10.1901/jeab.1997.68-1 response variability with reinforcement.
Morgan, L., & Neuringer, A. (1990). Behav- Journal of the Experimental Analysis of
ioral variability as a function of response Behavior, 37, 171–181. doi:10.1901/jeab.
topography and reinforcement contingency. 1982.37-171
Animal Learning & Behavior, 18, 257–263. Skinner, B. F. (1938). The behavior of organ-
doi:10.3758/BF03205284 isms. New York, NY: Appleton-Century
Neuringer, A. (1991). Operant variability and Crofts.
repetition as functions of interresponse time. Skinner, B. F. (1969). Contingencies of rein-
Journal of Experimental Psychology: Animal forcement. New York, NY: Appleton-Cen-
Behavior Processes, 17, 3–12. doi:10.1037/ tury Crofts.
0097-7403.17.1.3 Ward, R. D., Kynaston, A. D., Bailey, E. M.,
Neuringer, A. (2002). Operant variability: & Odum, A. L. (2008). Discriminative
evidence, functions, and theory. Psychonom- control of variability: Effects of successive
ic Bulletin & Review, 9, 672–705. doi:10. stimulus reversal. Behavioural Processes, 78,
3758/BF03196324 17–24. doi:10.1016/j.beproc.2007.11.007

You might also like