ON THE STATUS OF KNOWLEDGE FOR USING PUNISHMENT: IMPLICATIONS FOR TREATING BEHAVIOR DISORDERS Lerman 2002

7/30/2019 ON THE STATUS OF KNOWLEDGE FOR USING PUNISHMENT: IMPLICATIONS FOR TREATING BEHAVIOR DISORDERS Le

1/34

431

JOURNAL OF APPLIED BEHAVIOR ANALYSIS 2002, 35, 431464 NUMBER 4 (WINTER 2002)

ON THE STATUS OF KNOWLEDGEFOR USING PUNISHMENT:

IMPLICATIONS FOR TREATINGBEHAVIOR DISORDERS

DOROTHEA C. LERMAN AND CHRISTINA M. VORNDRANLOUISIANA STATE UNIVERSITY AND

THE LOUISIANA CENTER FOR EXCELLENCE IN AUTISM

In this paper, we review basic and applied findings on punishment and discuss the im-portance of conducting further research in this area. The characteristics of respondingduring punishment and numerous factors that interact with basic processes are delineatedin conjunction with implications for the treatment of behavior disorders in clinical pop-ulations. We conclude that further understanding of punishment processes is needed todevelop a highly systematic, effective technology of behavior change, including strategiesfor improving the efficacy of less intrusive procedures and for successfully fading treat-ment.

DESCRIPTORS: behavior disorders, functional analysis, punishment, treatment

Punishment is generally defined as an en-vironmental change contingent on behaviorthat produces a decrease in responding overtime (Michael, 1993). Numerous proceduralvariations of punishment have been devel-oped for clinical use. Results of research con-ducted over the past four decades haveshown that punishment is effective in reduc-ing problem behavior in clinical popula-tions, and in some cases, may be an essential

component of treatment (see Kazdin, 2001,and OBrien, 1989, for reviews of this lit-erature). However, more knowledge is need-ed about factors that may influence the ef-fects of punishment on problem behavior.Few strategies have been identified for en-hancing the effectiveness of less intrusivepunishment procedures, for attenuating un-desirable aspects of punishment, or for suc-cessfully fading treatment with punishment.

The direct and indirect effects of punish-

We thank Don Baer, Alan Baron, Linda LeBlanc,Tony Nevin, and the anonymous reviewers for theircomments on earlier versions of this paper.

Reprints may be obtained by contacting DorotheaC. Lerman, 236 Audubon Hall, Louisiana State Uni-versity, Baton Rouge, Louisiana 70803 (e-mail:[email protected]).

ment have been studied extensively in thelaboratory. Nevertheless, basic research onpunishment has been declining rapidly de-spite substantial gaps in knowledge (Baron,1991; Crosbie, 1998). The generality of basicfindings to clinical populations and problemsalso is questionable (Hayes & McCurry,1990). Most studies evaluated the effects ofintense, unconditioned punishers (e.g., elec-tric shock), and a number of important re-

lations have not yet been replicated with hu-mans or clinically relevant punishers.

The purpose of this paper is to review ba-sic and applied research findings on punish-ment, identify gaps in the literature, and dis-cuss the implications of these findings for theuse of punishment in clinical settings. Basicfindings that contradict common assump-tions about punishment effects found in text-books and review papers and that help elu-cidate inconsistent results on punishment inthe applied literature are highlighted. Themain premise of this paper is that further un-derstanding of punishment processes maylead to an improved technology of behaviorchange. We extend the most recent reviewpapers on punishment (Matson & Di-Lorenzo, 1984; Van Houten, 1983) by (a)


2/34

432 DOROTHEA C. LERMAN and CHRISTINA M. VORNDRAN

providing a broader overview of the directand indirect effects of punishment and fac-tors that influence basic processes (e.g., his-tory), (b) identifying areas in need of further

research from both basic and applied litera-tures, and (c) discussing recent research find-ings on punishment within the context of ad-vances in the functional analysis of behavior.1

Some authors have suggested that addi-tional applied research on punishment is un-necessary in light of refinements to the func-tional analysis methodology and treatment

with reinforcement (Donnellan & LaVigna,1990; Guess, Helmstetter, Turnbull, &Knowlton, 1987). Results of numerous stud-ies conducted over the past 15 years have

shown that the function of problem behav-ior often can be determined and that thisinformation can be used to develop treat-ments based on extinction, reinforcement,and other processes such as establishing op-erations (e.g., Iwata, Pace, Dorsey, et al.,1994). Nevertheless, punishment may becritical to treatment success when the vari-ables maintaining problem behavior cannotbe identified or controlled (for further dis-cussion, see Axelrod, 1990; Iwata, Vollmer,& Zarcone, 1990; Vollmer & Iwata, 1993).

Punishment also may be preferable to rein-forcement-based treatments when problembehavior must be suppressed rapidly to pre-vent serious physical harm (Dura, 1991; seealso Iwata et al.; Vollmer & Iwata). Moreimportant, results of several studies indicatethat treatments derived from functionalanalyses (e.g., differential reinforcement ofalternative behavior [DRA]) may not alwaysreduce behavior to clinically acceptable levels

without a punishment component (e.g.,Grace, Kahng, & Fisher, 1994; Hagopian,

Fisher, Sullivan, Acquisto, & LeBlanc, 1998;Wacker et al., 1990).

1 Much of the applied research on functional anal-ysis and treatment of behavior disorders has been con-ducted with individuals diagnosed with developmentaldisabilities. Thus, this review reflects this emphasis.

Knowledge about punishment also is im-portant because common treatments that areassociated with other processes may in factreduce problem behavior through the mech-

anism of punishment. For example, proce-dures such as response blocking, guidedcompliance, and the application of protec-tive equipment are often presumed to reduceproblem behavior by terminating the rein-forcement contingency that maintains theresponse (i.e., through extinction; e.g., Reid,Parsons, Phillips, & Green, 1993; Rincover,1978). Research findings suggest that theseprocedural variations of extinction mayfunction as punishment instead of, or incombination with, extinction (e.g., Lerman

& Iwata, 1996b; Mazaleski, Iwata, Rodgers,Vollmer, & Zarcone, 1994). Some authorsalso have suggested that the contingent lossof reinforcement associated with differentialreinforcement procedures (e.g., differentialreinforcement of other behavior [DRO])may constitute a form of punishment (e.g.,Rolider & Van Houten, 1990). Thus, theprocess of punishment may underlie a num-ber of popular function-based treatments.

Functional analysis methodology nowpermits more precise investigations of im-

portant environmentbehavior relations inthe area of punishment. Basic findings in-dicate that various parameters of reinforce-ment influence the direct and indirect effectsof punishment and interact with nearly ev-ery other factor that has been found to in-fluence responding during punishment (e.g.,punishment schedule and magnitude). Suchinteractions have important clinical impli-cations because punishment is more likely tobe used when the responsereinforcer rela-tion cannot be terminated completely. In ap-

plied studies, important reinforcement vari-ables identified via functional analysis couldbe manipulated prior to and during punish-ment, even when the behavior is maintainedby nonsocial consequences (i.e., the behavioris maintained in the absence of socially me-


3/34

433PUNISHMENT

diated reinforcers, such as attention, tangibleitems, and escape from instructions; see, e.g.,Lerman & Iwata, 1996b).

Important issues related to the ethics andacceptability of using punishment to treatproblem behavior in individuals with devel-opmental disabilities have been the subject ofnumerous articles over the past 30 years. Anoverview of these issues is beyond the scopeof this paper but can be found in a varietyof sources (see Donnellan & LaVigna, 1990;Emerson, 1992; Guess et al., 1987; Jacob-Timm, 1996; Sidman, 1989; Van Houten etal., 1988). Suggestions for further appliedstudies on punishment are made throughoutthis paper, with the assumption that pertinent

guidelines and cautions about the applicationof punishment will accompany published re-search findings (see Alberto & Troutman,1999; Lovaas & Favell, 1987; Matson &DiLorenzo, 1984).

Ultimately, the consumers of behavioraltechnologies (e.g., clinicians, caregivers) willdetermine which treatments are used withindividuals with developmental disabilities(see Iwata, 1988, for a cogent discussion ofthis issue). These decisions are at least partlyguided by information generated by the sci-

entific community. Safe, acceptable, andhighly effective technologies of behaviorchange should be available to consumers

who request them, including procedures thatare based on punishment.

Basic and applied research findings onclinically relevant factors that influence thedirect effects of punishment will be discussedin the first half of the paper. Other charac-teristics of punished responding, includingmaintenance, generalization, and side effects,

will be discussed in the second half.

FACTORS THAT INFLUENCETHE DIRECT EFFECTS

OF PUNISHMENT

Much of the basic research on the directeffects of punishment was conducted more

than 30 years ago with nonhumans (seeAzrin & Holz, 1966, for a review of thisliterature). Procedural variations of punish-ment examined in the laboratory have in-

cluded the delivery of stimuli, often calledpositive punishment, and the removal of stim-uli, often called negative punishment(i.e., re-sponse cost, time-out from positive rein-forcement).2 The majority of studies, how-ever, employed contingent electric shock. Inthe earliest basic studies, the effects of pun-ishment were evaluated while the punishedresponse was undergoing extinction (e.g.,Estes, 1944; Skinner, 1938; Thorndike,1932). However, in most subsequent studies,punishment contingencies were introduced

with no change in the prevailing reinforce-ment schedule. This latter arrangement as-sured a certain level of responding by whichto examine the effects of punishment inde-pendently of those produced by extinction(Azrin & Holz, 1966). Complex interactionsbetween reinforcement and punishment pro-cesses also could be evaluated. A number ofauthors suggested that this laboratory ar-rangement may be more pertinent to appli-cation because punishment is most likely tobe used when the reinforcer that maintains

problem behavior cannot be identified orcontrolled (Azrin & Holz, 1966; Dinsmoor,1952).

In fact, the function of problem behaviorwas not determined prior to treatment inmost clinical studies on punishment. Pun-ishment thus was superimposed on an un-known schedule of reinforcement that likelytook the form of extinction when the be-havior was maintained by social consequenc-

2 Consistent with previous articles and chapters on

punishment, few distinctions will be drawn betweenpositive and negative punishment in this review. Cur-rent research findings suggest that the procedures areassociated with similar direct and indirect effects onresponding. Nevertheless, the literature on positivepunishment far exceeds that on negative punishment.Basic processes may differ in important ways underthese two forms of punishment.


4/34


es (Iwata, Pace, Cowdery, & Miltenberger,1994). That is, social consequences that mayhave maintained responding during baseline(e.g., verbal reprimands, escape from in-

structions) often were removed with the in-troduction of punishment. A substantialportion of applied findings thus may havelittle generality to contemporary treatmentapproaches because punishment is most like-ly to be used when problem behavior con-tinues to produce reinforcement. Basic find-ings suggest that reinforcement parameterscan influence the effects of punishment inimportant ways.

THE DIRECT EFFECTSOF PUNISHMENT

Basic research findings have shown thatresponse-contingent shock, noise, blasts ofair, response cost, and time-out can producea rapid decrease in the frequency of behaviorand, in some cases, may lead to completeresponse suppression in rats, pigeons, mon-keys, and humans (e.g., college students,psychiatric patients; Azrin, 1960; Crosbie,

Wil liams, Lattal, Anderson, & Brown,1997). Several studies with humans andnonhumans also found that the initial re-

ductive effects of punishment with shock orpoint loss occurred more rapidly, or to agreater extent, than those produced by ex-tinction, satiation, and differential reinforce-ment (e.g., Holz & Azrin, 1963; Johnson,McGlynn, & Topping, 1973; Rawson &Leitenberg, 1973).

The potential benefit of using punish-ment to treat intractable behavior problemsled to the development of numerous punish-ment procedures for clinical use. Results ofresearch have shown that treatment with a

wide variety of punishers (e.g., verbal repri-mands, restraint, water mist, lemon juice,shock, removal of reinforcing activities orconditioned reinforcers) can produce an im-mediate, substantial suppression in problembehavior (see Kazdin, 2001, and Matson &

DiLorenzo, 1984, for reviews of this litera-ture). Applied findings also indicate that theeffects of punishment are superior to thoseobtained with less intrusive procedures

alone, such as differential reinforcement(e.g., Barrett, Matson, Shapiro, & Ollen-dick, 1981; Favell et al., 1982; Scotti, Evans,Meyer, & Walker, 1991). Although resultsof such comparisons are consistent withthose obtained in the laboratory, the findingsare difficult to interpret because numerousparameters likely influence the effects ofthese behavior-reduction procedures. For ex-ample, a dense schedule of differential rein-forcement may reduce behavior more effec-tively than a mild punisher.

More important, the relative efficacy oftreatment with reinforcement versus punish-ment likely depends on a variety of factors(e.g., history; use of extinction; type,amount, and schedule of the consequence).These complex interactions need to be eval-uated to generate more definitive findingsabout the suppressive effects of punishmentrelative to other procedures. Further researchon strategies to improve the efficacy of pun-ishment would be more pragmatic over thelong run than additional, complex compar-

ative studies of reinforcement versus punish-ment.

Several authors have suggested that treat-ment with punishment is so effective be-cause punishment usually can compete suc-cessfully with reinforcement contingenciesthat maintain problem behavior (e.g., VanHouten, 1983). Although punishment often

was confounded inadvertently with extinc-tion in applied research, recent studies havedemonstrated that common punishmentprocedures (e.g., time-out, brief manual re-

straint) can be effective in the absence ofextinction (Fisher, Piazza, Bowman, Hago-pian, & Langdon, 1994; Lerman, Iwata,Shore, & DeLeon, 1997; Thompson, Iwata,Conners, & Roscoe, 1999). Reductions inbehavior were obtained even after unsuc-


5/34

435PUNISHMENT

cessful attempts to treat the behavior withless intrusive procedures (e.g., Fisher et al.,1993; Lindberg, Iwata, & Kahng, 1999).

Nevertheless, the generality of these find-

ings may be limited because data on effectivetreatments are more likely to be publishedthan those that show unsuccessful outcomes.Potentially important reinforcement param-eters also were unspecified in these studies.Results of pretreatment functional analysesindicated that the behavior was maintainedindependent of social consequences but didnot isolate the precise reinforcer. Methods toidentify the type of nonsocial reinforcement(often called automatic reinforcement) that isfunctionally related to problem behavior

have been examined in a number of studies(e.g., Goh et al., 1995; Kennedy & Souza,1995; Patel, Carr, Kim, Robles, & Eastridge,2000; Piazza, Adelinis, Hanley, Goh, & De-lia, 2000). Although further methodologicalrefinements are needed, these strategies maybe useful for identifying and manipulatingvarious reinforcement parameters (e.g., re-inforcer schedule and magnitude) whiletreating automatically reinforced problembehavior with punishment (e.g., Lerman &Iwata, 1996b). As described in more detail

below, results of studies employing thesetypes of manipulations with behavior main-tained by either social or nonsocial conse-quences could lead to a greater understand-ing of punishment processes and improvedtreatments.

As discussed in the following sections, anumber of factors directly relevant to the de-velopment of an applied technology havebeen found to influence the direct effects ofpunishment. These factors include historicalvariables (e.g., prior experience with the

punishing stimulus or intermittent reinforce-ment); the use of conditioned punishers; re-inforcement variables (e.g., schedule, avail-ability of alternative sources of reinforce-ment); and punishment variables (e.g., mag-nitude, immediacy, schedule). However,

much of the research on these factors hasbeen conducted in the basic laboratory, andour knowledge of some important complexrelations is relatively incomplete (Baron,

1991).History

Basic findings indicate that previous ex-posure to certain factors can alter respondingduring punishment, a phenomenon that isespecially relevant to the application of pun-ishment because clinical populations typical-ly have diverse learning histories. Results ofnumerous basic studies have shown that pri-or experience with the punishing stimuluseither contingently or noncontingently can

decrease a behaviors sensitivity to punish-ment (e.g., Capaldi, Sheffer, Viveiros, Da-vidson, & Campbell, 1985; Halevy, Feldon,& Weiner, 1987). For example, researchfindings with rats indicate that exposure tointermittent punishment with shock decreas-es the efficacy of continuous punishment

with shock, even when several days or weekslapse between intermittent and continuouspunishment (Banks, 1967; Halevy et al.;Shemer & Feldon, 1984). Deur and Parke(1970) replicated this effect with normally

developing children and a loud buzzer as thepunishing stimulus.

Although adaptation to the punishingstimulus may account for these findings (Ca-paldi et al., 1985), a similar relation hasbeen obtained with intermittent reinforce-ment. That is, rats and college students witha history of intermittent reinforcement alsoshowed less response suppression under ei-ther continuous or intermittent punishment

with shock than participants with a historyof continuous reinforcement (e.g., Brown &

Wagner, 1964; Estes, 1944; Halevy et al.,1987; Vogel-Sprott, 1967). Moreover, Eisen-berger, Weier, Masterson, and Theis (1989)found that resistance to shock punishmentincreased for one response (lever pressing) inrats even after a topographically different be-


6/34


havior (running) was exposed to intermittentreinforcement.

In clinical settings, an individual is likelyto experience common punishers (e.g., ver-

bal reprimands, time-out) before these con-sequences are specifically arranged to treat aparticular inappropriate behavior. Further-more, exposure to intermittent schedules ofreinforcement and punishment is typical inthe natural environment. Consequences of-ten are delivered on intermittent schedulesbecause it is difficult for parents and teachersto reinforce or punish every occurrence ofbehavior. When initial attempts to treat abehavior with intermittent punishment fail,caregivers may switch to a continuous sched-

ule of punishment in an attempt to improvethe efficacy of the treatment. Basic findingsindicate that a history with intermittentpunishment may complicate treatment suc-cess in these cases, such that more intensepunishers will be required to suppress be-havior effectively (Halevy et al., 1987;Shemer & Feldon, 1984). However, the rel-evance of these findings to the types of pun-ishers that are more commonly used in clin-ical settings is unknown because nearly allbasic studies in this area evaluated the effects

of shock, and no applied studies have repli-cated and extended these findings to prob-lem behavior.

Nevertheless, basic findings in this areasuggest some important guidelines for clin-ical research and practice. First, it may bebeneficial for caregivers to identify novelpunishers when designing treatments and toavoid using common consequences, such asverbal reprimands and time-out, in unsys-tematic or unplanned ways. Second, inter-mittent schedules of punishment should not

be implemented prior to continuous sched-ules. Third, if adaptation to the punishingstimulus accounts for the decreased sensitiv-ity of behavior (Capaldi et al., 1985), briefhiatus from punishment may be useful, asdescribed in more detail below (Rachlin,

1966). Alternating among several effectivepunishment procedures in lieu of using asingle procedure is another potentially usefulstrategy for minimizing exposure to any sin-

gle punisher (e.g., Charlop, Burgio, Iwata,& Ivancic, 1988).

Conditioned Stimuli

Neutral stimuli that are paired with pun-ishing stimuli eventually may acquire prop-erties of the punishing stimuli. Results of ba-sic studies indicate that these conditionedstimuli can function as punishers when de-livered contingent on behavior in the ab-sence of the primary, or unconditioned,stimulus (e.g., Hake & Azrin, 1965). Con-

ditioned punishers may be useful for increas-ing both the efficacy and acceptability ofpunishment in clinical settings. Suppose thata relatively nonintrusive but ineffective con-sequence (e.g., a brief verbal cue) was estab-lished and maintained as a potent condi-tioned punisher via intermittent pairings

with a more restrictive, time-consuming in-tervention (e.g., overcorrection, time-out).

Application of the conditioned punisherwould reduce the individuals exposure tothe intrusive intervention and the degree of

effort required by caregivers to implementtreatment, factors that might circumventproblems with program inconsistency, habit-uation to the unconditioned punisher, andethical issues associated with the use of re-strictive procedures.

Various stimuli have been established asconditioned punishers in the basic labora-tory, including tones, lights, and low-voltageshock (e.g., Crowell, 1974; Davidson, 1970;Hake & Azrin, 1965). With a few excep-tions (e.g., Trenholme & Baron, 1975), the

unconditioned stimulus was electric shockand the subjects were rats or pigeons. Con-ditioned punishers were established via oneof two primary methods. Under one meth-od, the neutral stimulus was presented be-fore the onset of an inescapable stimulus


7/34

437PUNISHMENT

(e.g., shock delivered independent of re-sponding) and then was either removed withthe onset of the unconditioned stimulus(e.g., Hake & Azrin, 1965; Mowrer & Sol-

omon, 1954) or remained in the environ-ment while the unconditioned stimulus wasdelivered periodically (e.g., Orme-Johnson& Yarczower, 1974). Results of several stud-ies on this method indicated that more con-ditioning occurred if the neutral stimulus

was presented prior to the onset of the un-conditioned stimulus rather than simulta-neously with or after its onset (e.g., Evans,1962; Mowrer & Aiken, 1954).

Under the other method, the neutralstimulus was established as a discriminative

stimulus for punishment. That is, the pres-ence of the neutral stimulus was correlatedwith delivery of the unconditioned stimuluscontingent on responding (e.g., Davidson,1970). Although the discriminative stimulusthen was shown to suppress responding

when delivered contingent on behavior, re-sults of Orme-Johnson and Yarczower(1974) indicated that stimuli established asdiscriminative stimuli were much less effec-tive as conditioned punishers than stimuliestablished via the former method. Regard-

less of the conditioning method, researchfindings have shown that the effects of con-ditioned punishers on behavior are tempo-rary unless the conditioned stimulus and theunconditioned punisher continue to bepaired in some manner (Davidson, 1970;Hake & Azrin, 1965).

A few basic studies have evaluated factorsthat appear to influence the conditioningprocess, such as the magnitude of the un-conditioned stimulus (Mowrer & Solomon,1954) and the duration of the conditioned

stimulus (Hake & Azrin, 1965). However,other clinically relevant parameters of con-ditioning, such as the number of pairingsbetween the conditioned and unconditionedstimuli, the type of unconditioned punisherused, and characteristics of the neutral stim-

ulus (e.g., intensity or saliency), should beexamined in further research. In addition,conditioned stimuli typically were estab-lished and maintained independent of re-

sponding, a method that likely would invokeethical concerns if extended to clinical pop-ulations. Although it may be more accept-able to pair conditioned and unconditionedstimuli contingent on problem behavior, op-portunities to condition the stimulus wouldbe severely restricted if the unconditionedpunisher suppressed problem behavior tolow levels.

The use of conditioned punishers in treat-ing problem behavior has been reported insurprisingly few applied studies. More im-

portant, no applied studies have focused ex-clusively on methods to develop and main-tain stimuli as conditioned punishers in clin-ical settings. Lovaas and Simmons (1969)paired the word no with shock contingenton severe self-injury with 1 participant. Thebrief verbal reprimand then was presentedfor self-injury in the absence of shock duringa limited number of sessions, and resultssuggested that the stimulus had acquired thesuppressive properties of the original punish-er. In a more thorough evaluation, Dorsey,

Iwata, Ong, and McSween (1980) paired theword no with contingent water mist for 2participants who engaged in self-injury. Re-sults showed that contingent presentation ofthe verbal stimulus maintained low levels ofself-injury when water mist was withdrawnfrom the original treatment setting. Further-more, the suppressive effects of the verbalreprimand generalized to a setting that hadnot been previously associated with the wa-ter mist procedure, as well as to other ther-apists who had never delivered the water

mist. Finally, Dixon, Helsel, Rojahn, Cipol-lone, and Lubetsky (1989) paired a mild,less effective punisher (visual screen) with amore effective punisher (the odor of am-monia) while treating aggression and disrup-tion exhibited by a young boy with devel-


8/34


opmental disabilities. Problem behavior re-mained suppressed for a short time when thevisual screen was used alone.

Although results of these three studies in-

dicated that conditioned punishers were es-tablished successfully for a clinical problem,the efficacy of treatment was evaluated acrossa limited number of brief sessions. Basicfindings have shown that the effects of con-ditioned punishers on behavior are tempo-rary unless the conditioned stimulus and theunconditioned punisher continue to bepaired in some manner (Davidson, 1970;Hake & Azrin, 1965). In addition, detailsnecessary to replicate the conditioning pro-cedure (e.g., method of pairing, total num-

ber of pairings, rules for determining whento test the conditioned effect) were not de-lineated. The generality of these findingsand those obtained in the basic laboratoryalso may be limited by the use of relativelyintrusive unconditioned punishers (i.e.,shock, water mist, ammonia).

Thus, current knowledge about condi-tioned punishment is fairly incomplete, andprescriptions for the application of condi-tioned punishers should await further re-search. The efficacy of pairing various types

of auditory, tactile, and visual stimuli shouldbe evaluated with more common and so-cially acceptable forms of punishment (e.g.,time-out). The number of pairings necessaryto produce conditioning and factors thatmight alter the outcome of conditioning(e.g., intensity of the conditioned stimulus)could be evaluated by periodically testing thesuppressive effects of the paired stimulus inthe absence of the unconditioned punisher.The durability of conditioning could be de-termined by presenting the conditioned

stimulus without the unconditioned punish-er until the effects on responding dissipate.This strategy also may be useful when de-veloping a schedule for pairing the condi-tioned and unconditioned stimuli to main-tain conditioning over time. For example,

clinicians could determine the maximumnumber of times that the conditioned stim-ulus could be presented before the condi-tioning effect begins to be extinguished. The

conditioned and unconditioned stimulusthen could be paired prior to that numberon a regular basis.

Reinforcement Schedule

Basic findings indicate that the character-istics of responding during punishment maydepend on the reinforcement schedule thatmaintains the behavior, a factor that is es-pecially relevant to application because prob-lem behavior is likely to be maintained bysome form of reinforcement during treat-

ment. Behavior may be concurrently ex-posed to schedules of reinforcement andpunishment when caregivers do not com-pletely withhold social consequences duringtreatment or when the behavior is main-tained by automatic reinforcement. Resultsof basic studies generally showed that theamount of response suppression under pun-ishment was negatively related to the densityof the reinforcement schedule, with extinc-tion producing the greatest decrease in re-sponding (Azrin & Holz, 1966). Various pa-

rameters of punishment (e.g., schedule, in-tensity) also appear to interact with the re-lation between response suppression andreinforcement density (Bouzas, 1978; Brad-shaw, Szabadi, & Bevan, 1977, 1978). Forexample, Bradshaw and his colleagues foundthat the negative relation between reinforce-ment density and responding was muchmore pronounced when human subjects

were exposed to a variable-ratio (VR) pun-ishment schedule of monetary loss than to avariable-interval (VI) punishment schedule.

Intermittent reinforcement schedules ex-amined in the laboratory have includedfixed-interval (FI), fixed-ratio (FR), VI, andVR schedules. Although these reinforcementschedules have been found to interact dif-ferentially with the effects of punishment,


9/34

439PUNISHMENT

this interaction has not been well studiedand likely depends on various factors, suchas the reinforcement density, punishmentschedule, and amount of reinforcement lost

due to a reduction in responding (e.g., Pow-ell, 1970; Scobie & Kaufman, 1969; see alsoBaron, 1991, for further discussion). Thevarious ways in which these schedules influ-ence punishment effects are relevant to anapplied technology because social contingen-cies for problem behavior often approximatethese laboratory arrangements in the naturalenvironment (e.g., Lalli & Goh, 1993; Voll-mer, Borrero, Wright, Van Camp, & Lalli,2001). Such complex interactions betweenreinforcement and punishment schedules

also are likely responsible for some inconsis-tent findings reported in both the basic andapplied literatures on punishment (see fur-ther discussion below). Additional basic re-search in this area is needed to clarify theserelations.

Knowledge about basic processes and pre-scriptions for best practices when using pun-ishment in clinical settings will be incom-plete without further evaluation of potentialinteractions between reinforcement sched-ules and parameters of punishment. Never-

theless, no applied studies have examinedthe effects of reinforcement schedule or den-sity on the outcome of treatment with pun-ishment. Further research should determineif reducing the density of the reinforcementschedule operating in the natural environ-ment would substantially enhance the effi-cacy of commonly used punishment proce-dures. If so, strategies are needed to thin theschedule of reinforcement for problem be-havior during treatment with punishment.The parameters under which reinforcement

schedule is and is not an important factorwhen treating problem behavior with pun-ishment also should be evaluated. Potentiallyrelevant parameters include the schedule,type, and intensity of the punisher.

Although nonsocial sources of reinforce-

ment may be difficult to modify (but seeLerman & Iwata, 1996b, for one approach),results of this research may lead to usefulguidelines for designing effective yet practi-

cal treatments when caregivers will be un-likely or unable to withhold social reinforce-ment for problem behavior. Current knowl-edge indicates that reinforcement for prob-lem behavior should be withheld ordiminished if possible. Thus, for example,

when caregivers cannot withhold reinforce-ment completely during punishment (e.g.,ignore all instances of self-injury), the den-sity and magnitude of reinforcement forproblem behavior should be reduced andpunishment should be delivered on a con-

tinuous schedule (see further discussion be-low).

Availability of Alternative Reinforcement

Most textbooks and literature reviews onapplication highlight the benefits of combin-ing punishment with some type of reinforce-ment procedure, such as DRA (e.g., Cooper,Heron, & Heward, 1987; Matson & Di-Lorenzo, 1984). Basic studies with rats, pi-geons, and psychiatric patients have shownthat the suppressive effects of contingent

shock, noise, or time-out were enhancedwhen reinforcement could be obtained insome manner other than, or in addition to,engaging in the punished response (e.g.,Boe, 1964; Herman & Azrin, 1964; Holz,

Azrin, & Ayllon, 1963; Rawson & Leiten-berg, 1973). Although these findings suggestthat increasing the density of alternative re-inforcement might enhance the efficacy ofmild punishers in clinical settings, few basicstudies have evaluated clinically relevantpunishers or the parameters under which re-

inforcement may provide optimal benefitsduring punishment.

In a study with pigeons reported by Azrinand Holz (1966), for example, an FR 25schedule of alternative reinforcement did notincrease sensitivity to punishment when a re-


10/34


sponse maintained by the same reinforce-ment schedule was punished with low-volt-age shock (less than 50 V). It is possible,however, that a denser schedule of alterna-

tive reinforcement would have enhanced theefficacy of this mild punisher. Results of astudy by Fantino (1973) indicated that thebeneficial effects of alternative reinforcement

were compromised when the reinforcementrate provided by a concurrent VI scheduleprior to punishment could not be obtainedvia exclusive responding on the unpunishedalternative. Punishment parameters such asschedule and delay and various reinforce-ment parameters also likely modify the ef-fects of alternative reinforcement. Thus, ba-

sic findings suggest that various factors (e.g.,type or intensity of the punisher, density ofavailable reinforcement prior to punish-ment) must be considered when combiningreinforcement with punishment in clinicalsettings.

However, other commonly used reinforce-ment arrangements, such as noncontingentreinforcement (NCR) and differential rein-forcement of low response rates (DRL), havenot been evaluated in the context of a con-current punishment contingency in the lab-

oratory. More important, basic findings onalternative reinforcement may have limitedgenerality to application because clinicallyrelevant factors (e.g., response topography;reinforcement quality, schedule, delay, andmagnitude) usually were held constantacross available response options. Treatment

with punishment and differential reinforce-ment typically will incorporate different re-sponse, reinforcement, and punishment pa-rameters across the targeted behaviors, es-pecially when problem behavior is main-

tained by unknown or uncontrolled sourcesof reinforcement. In such cases, alternativereinforcement may neither suppress the pun-ished response nor increase adaptive behav-ior. Results of basic studies in which sched-ules of reinforcement and punishment were

arranged for both response options also sug-gest that the amount of suppression pro-duced by punishment for a given behaviorcan be influenced by contingencies that op-

erate on other behavior, including the rela-tive schedule, delay, and magnitude of re-inforcement and punishment (e.g., Deluty,1976, 1978; Farley, 1980).

Despite obvious clinical implications,only one applied study has evaluated the re-lation between punishment effects and theavailability of alternative reinforcement.Thompson et al. (1999) examined the sep-arate and combined effects of punishmentand reinforcement on self-injury after resultsof a functional analysis indicated that the

behavior was maintained by automatic re-inforcement. Reinforcement was arrangedfor an alternative behavior (toy manipula-tion) by giving the participants access to pre-ferred toys (thereby establishing automati-cally reinforced toy play) or by deliveringfood contingent on toy manipulation. Re-sults for the 4 participants indicated that al-ternative reinforcement enhanced the effi-cacy of relatively mild punishers (e.g., briefmanual restraint). Furthermore, reinforce-ment alone was fairly ineffective for all par-

ticipants, and punishment alone was ineffec-tive for 1 participant. As noted by the au-thors, however, the combined treatment wasdifferentially confounded with an additionalcontingency (i.e., time-out from positive re-inforcementaccess to programmed rein-forcement was withheld during punishmentdelivery) that may constrain the generalityof the findings.

Further research is needed on clinicalstrategies to enhance the efficacy of mildpunishers through the use of DRA, DRO,

DRL, and NCR procedures. Thus far, basicfindings in this area suggest a number of ten-tative prescriptions for application. Natural-istic reinforcement schedules for targetedproblem and alternative behavior should beconsidered first when developing treatments


11/34

441PUNISHMENT

that combine reinforcement and punish-ment. The type, schedule, and magnitude ofreinforcement maintaining problem behav-ior should be identified, so that a larger

amount of the same reinforcer could be pro-vided independent of undesirable respond-ing or contingent on alternative behavior(Fantino, 1973). When the functional rein-forcer cannot be identified or delivered byothers, selecting reinforcers that compete

with or substitute for maintaining reinforc-ers may be critical to effective treatment(e.g., Shore, Iwata, DeLeon, Kahng, &Smith, 1997).

As many of the available sources of rein-forcement as possible also should be deter-

mined prior to treatment, so that steps canbe taken to ensure that the total amount ofobtainable reinforcement can be sustained orexceeded despite a reduction in the punishedbehavior. To this end, differential reinforce-ment procedures should target simple free-operant responses or adaptive behavior thatis already in the individuals repertoire. Pro-cedures such as NCR, DRO, and DRL,

which do not require an alternative responsefor reinforcement delivery, may be preferableduring the initial stages of treatment with

punishment to insure a sufficient density ofavailable reinforcement. In fact, multiple re-sponses, reinforcers, and reinforcement con-tingencies (e.g., DRO plus NCR) should beincorporated into treatment such that pun-ishment for a restricted number of responsesis implemented within the context of a rich-ly reinforcing environment.

Punishment Magnitude

The relation between the effects of pun-ishment and the magnitude or amount of

punishment delivered for responding is es-pecially germane to the efficacy and accept-ability of punishment in clinical settings.From an ethical and practical standpoint,the least amount of punishment that is ef-fective (i.e., lowest intensity, shortest dura-

tion) should be used to treat behavior prob-lems, and strategies that increase the effec-tiveness of mild punishers should be incor-porated into treatment. Basic studies on

magnitude have shown that response sup-pression is positively related to the intensityand duration of shock with rats, pigeons,monkeys, and college students (Church,1969; Deluty, 1978; Scobie & Kaufman,1969), the duration of time-out with normalhumans (Kaufman & Baron, 1968; N. B.Miller & Zimmerman, 1966), and the num-ber of points lost as part of response cost

with normal humans (Weiner, 1964). Infact, recovery during punishment and fol-lowing the termination of the punishment

contingency was most likely to occur withmild punishers, such as a bar slap (Skinner,1938) and low-voltage shock (Hake, Azrin,& Oxford, 1967). Furthermore, punishment

with high-intensity shock or forceful airblasts was found to be ineffective with ratsif the intensity of the punisher was initiallylow and then gradually increased over time(Cohen, 1968; N. E. Miller, 1960; Terris &Barnes, 1969).

On the basis of these findings, numerousauthors have recommended using moderate

or high-intensity punishers to treat problembehavior and cautioned against increasingthe intensity of punishment gradually overtime (e.g., Cooper et al., 1987; Martin &Pear, 1996; OBrien, 1989). These guide-lines may be difficult to reconcile with eth-ical mandates to identify the least restrictiveprocedure that is effective. Moreover, closeexamination of basic findings in this area in-dicates that the relation between respondingand punishment magnitude is more complexthan frequently assumed. As a result, strat-

egies that are based on recommendations de-lineated in applied textbooks (e.g., Cooperet al.; Martin & Pear) and literature reviewsmay not influence behavior as expected.

As previously noted, the basic relation be-tween response suppression and punishment


12/34


magnitude may be influenced by the avail-ability of alternative reinforcement (e.g.,Holz et al., 1963) and the type of reinforce-ment schedule that maintains behavior (e.g.,

Powell, 1970; Scobie & Kaufman, 1969).Other variables (e.g., immediacy; Cohen,1968) also may alter the relation betweenpunishment magnitude and responding.Such complex interactions may be respon-sible for some contradictory findings onpunishment magnitude that have been re-ported in the applied literature. The extentto which basic findings are directly compa-rable to applied findings also is limited be-cause most basic studies examined the mag-nitude of shock, whereas applied studies in

this area have focused on more clinically ac-ceptable forms of punishment (e.g., overcor-rection, time-out).

In one of the few applied studies to eval-uate the magnitude of electric shock, D. E.

Williams, Kirkpatrick-Sanchez, and Iwata(1993) compared the efficacy of treatmentfor self-injury under two shock-intensity lev-els (3.5 mA vs. 18.5 mA). Results were con-sistent with those obtained in basic studies.The high-intensity shock produced larger,more immediate decreases in behavior than

the low-intensity shock. The generality ofthis finding, however, is somewhat limitedbecause punishment was combined with ex-tinction. In addition, sequence effects couldhave influenced the outcome because theparticipant was exposed to the lower inten-sity shock prior to the higher intensity, anda reversal to the low-intensity condition wasnot implemented.

Results of studies examining the relationbetween punishment magnitude and treat-ment efficacy using other types of punishers

(e.g., physical restraint, unpleasant smells,time-out) have been inconsistent and oftenappeared to be confounded with other var-iables (e.g., Altman, Haavik, & Cook, 1978;Cole, Montgomery, Wilson, & Milan, 2000;Marholin & Townsend, 1978; Singh, Daw-

son, & Manning, 1981). For example, Coleet al. found that treatment with overcorrec-tion produced similar decreases in stereotyp-ic behavior, regardless of whether the inter-

vention lasted 30 s, 2 min, or 8 min. Theeffects of overcorrection, however, may havebeen confounded with those of extinctionand verbal reprimands. Results of studies onthe duration of time-out have shown a pos-itive relation (e.g., Burchard & Barrera,1972; Hobbs, Forehand, & Murray, 1978),a negative relation (e.g., Kendall, Nay, & Jef-fers, 1975), and no relation (e.g., White,Nielsen, & Johnson, 1972) between dura-tion length and treatment effects. Thesefindings are difficult to interpret because the

function of problem behavior was not iden-tified (thus, time-out may have been contra-indicated for some participants), and se-quence effects may have confounded the re-sults (Matson & DiLorenzo, 1984).

Further research on the relation betweenpunishment magnitude and response totreatment, as well as on factors that can alterthis relation (e.g., reinforcement schedule,punishment delay), may be useful for rec-onciling inconsistent findings in the litera-ture and for developing more comprehensive

prescriptions for application. The commonassumption that a larger magnitude of a giv-en punisher will be more effective than asmaller magnitude is not strongly supportedin the current literature, with the exceptionof findings on contingent shock. Magnitudeshould be manipulated with a variety ofpunishers and in a variety of ways that havenot been examined in basic research. For ex-ample, the amount of reinforcement avail-able during time-in is another potentiallyimportant dimension of magnitude when

treatment with time-out is implemented(e.g., Solnick, Rincover, & Peterson, 1977).

Strategies to enhance response suppressionand maintenance under less effective valuesof punishment magnitude also should be ex-plored. Basic findings indicate that smaller


13/34

443PUNISHMENT

magnitudes of punishment may be more ef-fective if the punisher is delivered immedi-ately following the behavior (e.g., Cohen,1968) and if reinforcement is available for

an alternative response (e.g., Holz et al.,1963). Basic studies with shock also indicatethat a less intense punisher may be effective,at least temporarily, if a high intensity levelis decreased gradually over time (e.g., Azrin,1960; Cohen, 1968; Hake et al., 1967). Anapproach that involves periodically inter-spersing less intense punishers with more in-tense punishers may be useful for maintain-ing treatment effects while the magnitude ofpunishment is gradually reduced. As dis-cussed above, further research on condi-

tioned punishers and treatments combiningreinforcement with punishment also maylead to methods for increasing the effective-ness of mild punishers.

Finally, further studies should evaluatemethods for identifying the most appropri-ate magnitude of a given punishment pro-cedure prior to treatment implementation inthe natural environment. The typical trial-and-error approach to punishment selectionis inefficient and may be counterproductiveif an individual receives prolonged exposure

to ineffective procedures (e.g., N. E. Miller,1960; Terris & Barnes, 1969). Efficientstrategies for identifying the least restrictive,effective treatment are surprisingly absentfrom the applied literature. In two studiesconducted by Fisher and colleagues (Fisher,Piazza, Bowman, Hagopian, & Langdon,1994; Fisher, Piazza, Bowman, Kurtz, et al.,1994), the potential suppressive effects ofvarious procedures (e.g., time-out, facialscreen, contingent demands) were rapidly as-sessed by exposing participants to the puta-

tive punishers while negative vocalizations(e.g., yelling, crying) and avoidance or es-cape responses (e.g., dropping to the floor)

were measured. Each punishment procedurewas delivered noncontingently across fivedifferent durations, ranging from 15 s to

180 s. Results of subsequent treatment anal-yses indicated that the procedure associated

with the highest levels of negative vocaliza-tions, avoidance, or escape responses was the

most effective punisher for problem behav-ior. However, results of the initial assessmentdid not differentiate among the various du-ration lengths for any participant, possiblybecause the procedures were alternated rap-idly across a limited number of trials. Theutility of such assessments should be evalu-ated in further studies. For example, a sep-arate assessment of punishment magnitude,similar in design to that conducted by Fisheret al., might be useful after an initial assess-ment has identified a potent punisher.

Until further applied research on magni-tude is conducted, practitioners should selectmagnitudes that have been shown to be safeand effective in clinical studies, as long asthe magnitude is considered acceptable andpractical by those who will be implementingtreatment. Punishment should be combined

with some type of reinforcement procedure,and the punisher should be delivered as im-mediately as possible following occurrencesof problem behavior (see further discussionbelow). If the punisher fails to suppress be-

havior over time, alternative proceduresprobably should be considered instead of in-creasing the magnitude of the punisher un-der the presumption that this strategy willimprove the efficacy of treatment.

Immediacy of the Punisher

Consequences for problem behavior arefrequently delayed in the natural environ-ment. Caregivers and teachers often are un-able to monitor behavior closely or to deliverlengthy punishers (e.g., 15-min contingent

work) immediately following instances ofproblem behavior (Azrin & Powers, 1975).Punishment also may be delayed when theindividual actively resists application of theprogrammed consequences by struggling

with the punishing agent or running away.


14/34


In some cases, problem behavior occurs pri-marily in the absence of the punishing agent,necessarily delaying programmed conse-quences until the behavior is detected

(Grace, Thompson, & Fisher, 1996; VanHouten & Rolider, 1988).For these reasons, research on delayed

punishment is especially pertinent. Labora-tory findings with rats indicate that thelengthier the delay between the occurrenceof the response and delivery of contingentshock, the smaller the amount of responsesuppression under punishment (e.g., Baron,Kaufman, & Fazzini, 1969; Camp, Ray-mond, & Church, 1967). Even brief delaysof 10 s or 20 s have been found to seriously

compromise the effects of contingent shockwith rats and college students (e.g., Banks &Vogel-Sprott, 1965; Goodall, 1984) and ofreinforcement loss with college students(Trenholme & Baron, 1975).

Stimuli that might bridge the interval be-tween a response and its consequence havebeen notably absent from laboratory ar-rangements involving delayed punishment.Results of at least one study suggest that fac-tors such as the presence of a conditionedpunisher and delivery of instructions can al-

ter the efficacy of delayed punishment. InTrenholme and Baron (1975), delays of 10s, 20 s, and 40 s were equally effective withcollege students when a brief noise that waspaired with reinforcement loss also occurredimmediately following the behavior. A sub-sequent experiment showed that delayedpunishment was just as effective as imme-diate punishment when the participants re-ceived instructions about the delay. The gen-erality of these findings to clinical popula-tions, such as individuals with developmen-

tal disabilities, has not been determined. Inaddition, no basic studies have evaluated theeffects of numerous other potentially impor-tant factors on delayed punishment (e.g.,history, reinforcement schedule, availabilityof alternative reinforcement).

Surprisingly few applied studies have eval-uated the efficacy of delayed punishment orstrategies to improve treatment effects whenconsequences do not occur contiguous to

the behavior. In one of the few studies tocompare immediate and delayed punish-ment, Abramowitz and OLeary (1990)found that immediate verbal reprimands

were much more effective in decreasing off-task behavior in school children than werereprimands that were delayed by 2 min.These results are somewhat difficult to in-terpret, however, because delayed repri-mands were delivered only if off-task behav-ior had occurred continuously for 2 min,during which time the students had varied

opportunities to interact with other studentsand non-task-related objects. Thus, the ef-fects of punishment delay were not separatedfrom those of reinforcement schedule andpunishment schedule.

Results of just two studies have delineatedconditions under which delayed punishmentmay produce effective outcomes. Roliderand Van Houten (1985) and Van Houtenand Rolider (1988) demonstrated the effi-cacy of delayed punishment using variousmediated consequences with children with

emotional and developmental disabilities.One form of mediation involved playing au-diotape recordings of the childs disruptivebehavior that were collected earlier in theday. The punishing consequence (physicalrestraint, verbal reprimands) then was deliv-ered. In some cases, the tape recorder wasclearly visible to the child while the record-ings were being collected, and a verbal ex-planation of its role in the delivery of de-layed punishment was provided. These fac-tors may have served to bridge the temporal

gap between inappropriate behavior and itsconsequence (e.g., by functioning as dis-criminative stimuli for punishment; Tren-holme & Baron, 1975). However, for 1 par-ticipant, neither instructions about delayedpunishment nor an immediate consequence


15/34

445PUNISHMENT

designed to signal that punishment wasforthcoming (a mark placed on the childshand) was as effective as the audiotape pro-cedure.

In Van Houten and Rolider (1988), care-givers physically guided 2 participants to en-gage in the problem behavior (aggression ortheft) after occurrences of the behavior weredetected or reported. Caregivers then deliv-ered the punishing consequence (physical re-straint) immediately following the guided re-sponse. Although treatment was effective,the efficacy of delayed punishment withoutthe guided response component was not ex-amined. The length of the delay and care-givers immediate response to the problem

behavior also were not specified.Current knowledge indicates that the

mild punishers typically used in clinical set-tings will be ineffective unless the conse-quence immediately follows problem behav-ior. Thus, further research is needed on fac-tors that might enhance treatment effectsunder delayed punishment, especially pro-cedures or stimuli that would bridge thetemporal gap between a response and itsconsequence. Research on the utility of de-

livering conditioned punishers, instructions,and other types of stimuli associated withdelayed consequences is needed with clinicalpopulations.

Until further applied research is conduct-ed, teachers and caregivers should be con-cerned with selecting punishers that can bereadily delivered as soon as the behavior oc-curs. Consequences that do not require theclose proximity of the caregiver (i.e., stimulithat can be delivered or removed from a dis-tance) and technology to increase the prac-

ticality of immediate punishment may be es-pecially useful in this regard. Electronic de-vices that detect occurrences of problem be-havior and either alert caregivers or deliverconsequences automatically might circum-vent the problems of delayed punishment

(e.g., Linscheid, Iwata, Ricketts, Williams,& Griffin, 1990).

However, the timing of punishment in re-lation to reinforcement delivery also should

be considered, because some basic studieshave found that immediate punishment wasless effective than delayed punishment if theimmediate punisher preceded reinforcementdelivery but the delayed punisher followedit (e.g., Epstein, 1984; Rodriguez & Logan,1980). It is conceivable that diligent caregiv-ers may respond to problem behavior by firstdelivering the prescribed punisher (e.g., con-tingent work, time-out), followed (inadver-tently) by the maintaining social reinforcer(e.g., access to materials). In a similar man-

ner, automated punishment may be deliv-ered immediately prior to social or nonsocialconsequences for problem behavior. Resultsof other basic studies, in which the avail-ability of reinforcement for one response wasperfectly correlated with the delivery of mildpunishment for an immediately precedingresponse, indicated that pairing punishmentand reinforcement in this manner estab-lished the mild punisher as a conditionedpositive reinforcer (e.g., Murray & Nevin,1967; D. R. Williams & Barry, 1966). Thus,

the timing of punishment and reinforcementin the natural environment should be eval-uated carefully as part of treatment.

Schedule of Punishment

The effects of punishment schedules onresponding have important implications forthe efficacy and acceptability of treatment

with punishment. Intermittently deliveredconsequences that successfully reduce prob-lem behavior are easier to use, less time con-suming, and less intrusive than consequences

that must follow each occurrence of behav-ior. Results of basic research with pigeonsand rats suggest that punishment with shockor time-out will not produce acceptable re-sults unless the punisher follows nearly everyoccurrence of the behavior in situations in


16/34


which no alternative is available or when thedensity of reinforcement is not reduced (Ap-pel, 1968; Azrin, Holz, & Hake, 1963; De-luty, 1976; Farley, 1980; Thomas, 1968).

Although a number of applied studieshave examined the efficacy of intermittentpunishment for treating problem behavior,results have been inconsistent, and the con-ditions under which intermittent punish-ment might be effective remain unclear. Insome studies, for example, intermittent pun-ishment schedules were associated with a so-cially significant reduction in behavior, par-ticularly if responding was already sup-pressed to low levels via continuous punish-ment (e.g., Clark, Rowbury, Baer, & Baer,

1973; Rollings & Baumeister, 1981; Ro-manczyk, 1977). Conversely, even dense in-termittent punishment schedules were inef-fective for some individuals in other studies(e.g., Calhoun & Matherne, 1975; Lermanet al., 1997). Basic findings on factors thatinteract with the effects of punishmentschedules (e.g., reinforcement schedule) mayexplain why clinical applications have pro-duced inconsistent results.

First, punishment appeared to be con-founded with extinction and other potential

punishers (e.g., verbal reprimands) in ap-plied studies that showed significant treat-ment effects under thin punishment sched-ules (e.g., Barton, Brulle, & Repp, 1987;Clark et al., 1973; Romanczyk, 1977). Sec-ond, important parameters of punishment(i.e., type, intensity, and schedule) variedconsiderably among these studies. Basicfindings indicate that these variables alter therelation between intermittent punishmentand response suppression. Increasing the in-tensity of a punisher, for example, can either

enhance or degrade the efficacy of intermit-tent schedules, depending on other factors(e.g., Appel, 1968; Lande, 1981). Certainpunishment schedules (e.g., VI) also havebeen associated with greater decreases in re-sponding than other schedules (e.g., FR or

FI; Azrin, 1956; Camp, Raymond, &Church, 1966), although the nature of thisrelation is complex (e.g., Arbuckle & Lattal,1992) and appears to be influenced by the

schedule of reinforcement that maintains thebehavior (e.g., Bradshaw et al., 1977, 1978;Powell, 1970; Scobie & Kaufman, 1969).

In the only applied study that examinedthe interaction between type or intensity ofpunishment and intermittent punishmentschedules, Cipani, Brendlinger, McDowell,and Usher (1991) found that a VR 4 sched-ule of punishment with contingent applica-tion of lemon juice was just as effective as acontinuous schedule in reducing a childsstereotypic behavior. A manual guidance

overcorrection procedure (i.e., physicallyguiding the childs arms over the head andto the sides 10 times) also was effective whenthe procedure was delivered under a contin-uous schedule. Unlike the lemon juice, how-ever, a VR 4 schedule with the overcorrec-tion procedure did not produce clinicallysignificant reductions in behavior.

Finally, interactions between schedules ofreinforcement and punishment may accountfor the idiosyncratic effects of intermittentpunishment on problem behavior. For ex-

ample, Lerman et al. (1997) treated 5 par-ticipants self-injurious behavior (SIB) witha continuous schedule of punishment afterresults of a functional analysis indicated thatthe behavior was maintained by automaticreinforcement. Initial application of inter-mittent punishment (FI 2 min or FI 5 min)

was ineffective for 4 of the 5 participants.The continuous punishment schedule then

was successfully thinned to FI 5 min for 2of these participants. In contrast, continuouspunishment was necessary to suppress SIB

for the other 2 participants, despite repeatedattempts to thin the schedule. Although thefunction of SIB had been identified prior totreatment, important parameters of themaintaining reinforcers were unknown (e.g.,schedule, density, magnitude). These param-


17/34

447PUNISHMENT

eters, which likely varied across participants,may have been responsible for the inconsis-tent success of the schedule-thinning proce-dure.

Further research on interactions betweenpunishment schedule (e.g., VR vs. VI) andother potentially important parameters ofpunishment and reinforcement is needed toclarify the conditions under which intermit-tent punishment would and would not beeffective. Few studies have directly evaluatedstrategies to systematically thin punishmentschedules or to utilize highly variable (andthus unpredictable) schedules. Combining athin schedule of punishment with a richschedule of conditioned punishment is an-

other potential approach for increasing theefficacy of intermittent punishment. De-pending on the nature of the conditionedand unconditioned punishers, this arrange-ment may be more practical than using arich schedule of unconditioned punishmentalone. A dense schedule of alternative rein-forcement also may promote the efficacy ofintermittent punishment.

Other types of punishment schedules ex-amined in the basic laboratory, such as thedifferential punishment of high (DPH) or

low (DPL) response rates, also may be usefulin clinical settings. These schedules do notspecify a direct contingency between the de-livery of the punisher and the occurrence ofa response. For example, under DPH orDPL, punishment is delivered contingent onthe pause length that immediately precededa response (i.e., the selective punishment ofcertain lengths of interresponse times). Re-sults of basic studies on DPH and DPLschedules showed that overall responding in-creased when relatively long interresponse

times (DPL) were punished and decreasedwhen short interresponse times (DPH) werepunished (e.g., Galbicka & Branch, 1981;Laurence, Hineline, & Bersh, 1994). DPHschedules may be more beneficial than con-tinuous punishment when treating behavior

that is considered problematic only becauseit occurs at high rates or in bursts. Furtherunderstanding of these schedules also is im-portant because they may commonly operate

in the natural environment. For example,caregivers may be more likely to deliver pun-ishment when problem behavior occurs in-frequently (i.e., is characterized by long in-terresponse times) than when behavior oc-curs at high rates (see Arbuckle & Lattal,1992, for a discussion of this issue). Such anarrangement could compromise the efficacyof treatment by increasing the frequency ofshort interresponse times.

Current knowledge about punishmentschedules suggests that parents and teachers

should punish each occurrence of problembehavior unless the behavior is simulta-neously exposed to extinction. Until furtherresearch is conducted, clinicians should beextremely cautious when attempting to thinthe punishment schedule, utilize DRHschedules, or evaluate other strategies for im-proving the effects of intermittent punish-ment (e.g., employing variable schedules orconditioned punishers). A continuous sched-ule of punishment always should be imple-mented initially, and intermittent schedulesshould be considered only if the continuousschedule remains effective in suppressingproblem behavior to low levels over a con-siderable amount of time.

FACTORS RELATED TOMAINTENANCE, GENERALIZATION,

AND INDIRECT EFFECTS

A much smaller proportion of basic andapplied studies on punishment have evalu-

ated the long-term maintenance, generaliza-tion, and side effects of punishment relativeto those on direct effects. The extent to

which punishment effects are maintainedover time, transfer across settings and con-texts, and produce changes in other behavior


18/34


has significant implications for treatmentsinvolving punishment.

Maintenance

The durability of treatment with punish-ment is one of the most important consider-ations for practitioners, teachers, and caregiv-ers of individuals with behavior disorders. Anumber of authors, however, have suggestedthat the clinical effects of punishment are rel-atively short-lived, even when the treatmentremains unchanged over time (e.g., Parsons,Hinson, & Sardo-Brown, 2001; Walker &Shea, 1999). In basic studies with both hu-mans and nonhumans, various punishers havebeen associated with continued response sup-

pression under punishment, including relative-ly intense levels of electric shock with pigeonsand rats (Azrin, 1960; Crosbie et al., 1997),point or monetary loss with normal humans(Crosbie et al.; Weiner, 1962), and time-outfrom positive reinforcement with squirrelmonkeys (McMillan, 1967). Response recov-ery, however, has been associated with less in-tense punishers, such as low-voltage shock

with pigeons (Rachlin, 1966), a bar slap withrats (Skinner, 1938), and noise with pigeons(Holz & Azrin, 1962). Basic findings on the

maintenance of response suppression follow-ing the termination of the punishment con-tingency also generally showed that responserates immediately returned to prepunishmentlevelssometimes even temporarily exceedingbaselineunless intense punishers were used(e.g., high-voltage shock; Azrin, 1960).

These results suggest that sufficiently in-tense punishers, including some commonlyused clinical procedures (e.g., time-out),may produce lasting reductions in problembehavior as long as the punishment contin-

gency remains in effect. Nevertheless, basicfindings may not be applicable to treatmentoutcomes in clinical settings because thetime periods evaluated in the laboratory(e.g., 30-min to 60-min punishment sessionsacross 10 to 20 days) may have little relation

to the numerous months (and sometimesyears) over which problem behavior requirestreatment. Moreover, few studies have ex-amined factors that may influence the du-

rability of punishment effects.Although brief treatment evaluations arepredominant in the applied literature onpunishment, an increasing number of stud-ies have examined the long-term efficacy ofpunishment over the past 10 years. Treat-ment effects have been examined for 1 to 60months after punishment was initiated andcontinued with minor changes to the pro-cedure (Duker & Seys, 1996; Ricketts,Goza, & Matese, 1993; D. E. Williams,Kirkpatrick-Sanchez, & Crocker, 1994), and

after the original punishment componentwas withdrawn (Arntzen & Werner, 1999;Foxx, Bittle, & Faw, 1989; Rolider, Wil-liams, Cummings, & Van Houten, 1991).Results have shown varying success in main-taining the reduction in behavior, yet poten-tial reasons for the inconsistent outcomeshave not yet been identified.

For example, D. E. Williams et al. (1993)observed a relapse in treatment with contin-gent electric shock 6 months after punish-ment was initiated. Conversely, Linscheid,

Hartel, and Cooley (1993) found that con-tingent electric shock continued to suppress2 individuals self-injurious behavior for 5years. Duker and Seys (1996) examined thelong-term efficacy of contingent shock with12 individuals by obtaining information onthe degree of physical restraint each requiredfrom 2 to 47 months after the initiation ofpunishment. Results at follow-up suggestedthat treatment remained effective for 7 par-ticipants, including 1 individual who wasevaluated at 36 months and another who

was evaluated at 47 months.Conclusions about applied findings on

maintenance are difficult to draw for a num-ber of reasons. First, the majority of studiesexamined the long-term effectiveness of con-tingent electric shock, so results may not be


19/34

449PUNISHMENT

applicable to other (or more mild) punishers(Azrin & Holz, 1966). Second, the reinforc-ing consequences of problem behavior werenot identified prior to treatment in most cas-

es. Long-term maintenance may have beenmore likely to occur if the maintaining re-inforcer was withheld contingent on prob-lem behavior or readily available for engag-ing in more appropriate behavior (Estes,1944). Third, other factors potentially re-sponsible for both successful and unsuccess-ful cases of treatment maintenance may havevaried widely across the studies (e.g., pun-ishment schedule, availability of reinforcersthat competed with or substituted for themaintaining reinforcer). In fact, components

of the original intervention were modifiedover time in some studies (e.g., additionalbehavioral procedures or drugs were intro-duced; Duker & Seys, 1996), and it is dif-ficult to determine which, if any, treatmentmodifications may have been responsible forthe outcomes. Furthermore, the lengthytime period required to conduct these stud-ies increased the likelihood that unplannedchanges or other uncontrolled factors inter-acted with the efficacy of the original treat-ment in either desirable or undesirable ways.

Finally, the number of treatment relapse cas-es reported in the literature may not accu-rately reflect the prevalence of this problemin applied settings because such cases are lesslikely to be submitted or accepted for pub-lication than successful cases of treatmentmaintenance.

Identifying factors or processes associatedwith long-term maintenance is key to thedesign of a systematic technology for pre-venting and remediating treatment relapse.Several authors have suggested that adapta-

tion, or habituation, to the punishing stim-ulus accounts for instances of recovery (i.e.,repeated exposure decreases the aversivenessof the punisher; Goodall, 1984). Moreover,adaptation is more likely to occur with mildpunishers, which are typically employed in

clinical settings. One strategy that may de-crease the likelihood of habituation is the useof hiatus from punishment. In several basicstudies with pigeons, response suppression

under shock punishment was enhanced fol-lowing brief time periods during which thesubject was removed from the punishing sit-uation or exposed to reinforcement only(e.g., Rachlin, 1966). Further research isneeded, however, because the beneficial ef-fect of this procedure was found to waneacross repeated punishmenthiatus cycles(e.g., Orme-Johnson, 1967). Other strate-gies that may prevent or attenuate habitua-tion, such as using intermittent, varied, orbrief punishers (e.g., Charlop et al., 1988),

should be evaluated in further studies.Research also is needed on strategies tomaintain punishment effects while the inter-vention is systematically faded. Basic find-ings with pigeons and monkeys have shownthat responding will remain suppressed un-der low-intensity shock if an initially intenseshock is reduced very gradually (e.g., Hakeet al., 1967). Further applied research isneeded to determine if treatment effects willbe maintained while the intensity or dura-tion of a punishment procedure is altered

very gradually or less intrusive proceduresare simultaneously introduced. For example,it may be possible to reduce a 5-min time-out to a 1-min time-out over time. The useof conditioned punishers may enhance thelikelihood of fading certain dimensions ofintrusive punishers while treatment effectsare maintained over the long run. Moreover,basic studies have found that response re-covery is more gradual following the with-drawal of intermittent shock punishmentthan following the removal of other punish-

ment schedules (e.g., Azrin et al., 1963;Camp et al., 1966). Thus, strategies to in-crease the utility of conditioned punishersand intermittent punishment for routineclinical practice also may promote the long-term efficacy of punishment.


20/34


Several authors have suggested that com-bining punishment with differential rein-forcement may increase the likelihood thatpunishment can be faded successfully (e.g.,

Kazdin, 2001). Although this clinical strategyhas not been evaluated directly, one studyfound that differential reinforcement wasmore effective in reducing problem behaviorafter a participant had been exposed to a pe-riod of punishment (contingent work) than

when differential reinforcement precededpunishment (Fisher et al., 1993). Researchfindings on the indirect effects of punishmentsuggest that punishment may increase res-ponsivity to reinforcement (see below for fur-ther discussion). Thus, punishment may en-

hance the efficacy of reinforcement for estab-lishing appropriate behavior that competeswith or replaces inappropriate behavior, anoutcome that in turn may increase the like-lihood that punishment can be withdrawn.

Until additional research on long-termmaintenance is conducted, practitioners andcaregivers should not assume that punish-ment will remain effective over the long run.Strategies for increasing the likelihood ofmaintenance should be employed from theoutset of treatment. Although basic findings

suggest that relatively intense punishers maybe associated with successful long-term out-comes, the use of analogous procedures totreat problem behavior probably would raiseethical concerns for all but the most seriouscases. Caregivers instead should focus on theuse of reinforcement to insure that alterna-tive behavior is at high strength in the rep-ertoire of individuals who are exposed topunishment. Systematic reinforcer assess-ments and functional analyses of problembehavior always should precede treatment

implementation (Fisher et al., 1992). The ef-fects of punishment may last longer if ap-propriate behavior is maintained by the samereinforcers that maintain problem behavioror by reinforcers that are effective substitutesfor maintaining reinforcers. The reinforcing

consequences of problem behavior alsoshould be minimized or withheld if possible.

Potential problems with habituation maybe curtailed by limiting exposure to the pun-

isher in various ways. For example, caregiverscould schedule brief vacations from punish-ment on a regular basis (Rachlin, 1966) orrestrict the use of specific procedures to oneor two problem behaviors (e.g., those ofgreatest concern) instead of applying thesame treatment for a variety of responses.Comprehensive punisher assessments alsoshould be employed to identify clinically ac-ceptable procedures that produce the great-est reduction in behavior and, hence, wouldlead to the least amount of exposure to the

punisher (see Fisher, Piazza, Bowman, Ha-gopian, & Langdon, 1994). If the assess-ment identifies more than one effective formof punishment, caregivers could alternateamong several procedures to minimize ex-posure to any single punisher.

Finally, practitioners and caregivers shouldhave a plan for dealing with treatment re-lapse when it occurs during punishment orfollowing the removal of punishment. Thefirst step is to identify and rectify other fac-tors that may be responsible for treatment

failure. Many instances of relapse likely areattributable to problems with treatment in-tegrity (D. E. Williams et al., 1993), espe-cially when the punishment procedure iscomplex or time consuming (e.g., Foxx &Livesay, 1984). The next step is to reassess a

wide range of stimuli and activities that mayfunction as potent reinforcers for appropri-ate behavior and to target additional, mul-tiple responses that might compete with thepunished behavior. Alternative forms ofpunishment should be considered only after

determining that the current punisher is in-effective within the context of a richly rein-forcing environment. At this point, anothercomprehensive punisher assessment shouldbe conducted to identify other effective pun-ishers. Increasing the intensity or magnitude


21/34

451PUNISHMENT

of the ineffective punisher is not recom-mended, as discussed above (see Reinforce-ment Magnitude).

Stimulus Generalization

The transfer of treatment effects across dif-ferent settings and contexts (i.e., stimulusgeneralization) is another critically importantoutcome for individuals with behavior dis-orders. (Response generalization under pun-ishment, or a concomitant reduction in un-punished behavior when a punished responsedecreases, will be discussed under the headingIndirect Effects of Punishment.) Integrationinto the community may be restricted even

when problem behavior is responsive to treat-

ment if the procedure cannot be implement-ed everywhere the behavior occurs (e.g., dur-ing transitions at school or in public placessuch as stores and buses). Basic findings onstimulus generalization, however, suggest thatthe effects of punishment on problem behav-ior may transfer to untreated settings andcontexts. Results of several basic studies withpigeons showed that the suppression in re-sponding produced by shock punishment oc-curred in the presence of antecedent stimulithat were not used in the training situation,

even though punishment was withheld dur-ing tests for generalization (e.g, Hoffman &Fleshler, 1965; Honig & Slivka, 1964). Theamount of response suppression (i.e., level ofstimulus control) was a function of the phys-ical similarity between the generalizationstimuli and the stimuli present during train-ing with punishment, a finding that is anal-ogous to basic findings on stimulus general-ization and reinforcement effects (Guttman& Kalish, 1956).

Research findings with pigeons and shock

punishment, however, may not be directlyapplicable to humans in clinical settings. Inthe only two studies to examine stimulusgeneralization with humans in the labora-tory, generalization was relatively difficult toobtain with college students and a more

clinically relevant form of punishment(point loss; ODonnell & Crosbie, 1998;ODonnell, Crosbie, Williams, & Saunders,2000). Further basic research is needed with

both humans and nonhumans to identifyfactors that influence the degree and dura-bility of stimulus generalization during pun-ishment. Such factors may include parame-ters of punishment or reinforcement (e.g.,intensity, amount, schedule) and features ofthe generalization stimuli (e.g., saliency).

In fact, results of numerous applied stud-ies indicate that punishment effects rarelytransfer to settings or contexts that are un-associated with punishment delivery (e.g.,Corte, Wolf, & Locke, 1971; Doke & Ep-

stein, 1975; Marholin & Townsend, 1978;Rollings, Baumeister, & Baumeister, 1977).Surprisingly few studies have evaluated strat-egies to promote generalization since Matsonand Taras (1989) lamented this gap in a 20-year review of the applied literature on pun-ishment. In early studies, factors such as thepresence of the therapist (Risley, 1968) andproximity of the individual to the therapistor treatment setting (Lovaas & Simmons,1969; Rollings et al.) were found to influ-ence generalization. These findings are con-

sistent with those of basic studies showing apositive relation between amount of re-sponse suppression and the degree of simi-larity between punishment and generaliza-tion contexts. This relation may in fact ex-plain why generalization has rarely been ob-served in applied studies. In most cases, thephenomenon was tested via abrupt alterationof both the stimulus context and the pun-ishment contingency. Participants who con-tinued to exhibit at least some instances ofthe target behavior would readily detect the

transition from a continuous schedule ofpunishment to the removal of punishment(Azrin & Holz, 1966). In basic studies, re-sponding eventually recovered when thegeneralization stimuli were repeatedly pre-sented in the absence of punishment.


22/34


Thus, it is not surprising that punishmenteffects typically failed to generalize in the ab-sence of procedures designed to promotetransfer (Stokes & Baer, 1977). Several au-

thors have suggested that techniques foundto enhance generalization under reinforce-ment may be similarly effective under pun-ishment (e.g., Matson & DiLorenzo, 1984;Miltenberger, 2001). A few applied studieson punishment have evaluated generalizationstrategies analogous to those used to promotereinforcement effects. For example, commonstimuli were introduced into treatment andgeneralization settings (e.g., a discriminativestimulus for punishment was presented in thegeneralization setting; Birnbrauer, 1968),

stimuli that might acquire discriminativecontrol over the behavior were removed fromthe treatment setting (e.g., the therapist washidden from view; Corte et al., 1971; Tate& Baroff, 1966), and training was conducted

with multiple stimulus exemplars (e.g., sev-eral different therapists delivered shock; Lo-vaas & Simmons, 1969). In nearly all cases,however, these strategies were ineffective un-less punishment was delivered in the gener-alization context.

Other potential tactics drawn from the lit-

erature on reinforcement include pairing thepunisher with naturalistic consequences(e.g., verbal reprimands), varying the stim-ulus conditions during initial treatment withpunishment, providing instruction on self-management, and using delayed or intermit-tent punishment (i.e., indiscriminable con-tingencies; Stokes & Baer, 1977; see alsoODonnell & Crosbie, 1998, Experiments 3and 4). Generalization also may be achievedby implementing a modified form of thetreatment in generalization contexts (e.g.,

delivering a smaller amount of the punisheror a single component of a multicomponenttreatment procedure).

Current knowledge about punishment,however, is insufficient to guide the appli-cation of such strategies. For example, many

factors that are useful for promoting gener-alization under reinforcement, such as de-layed or intermittent contingencies, havebeen found to undermine the efficacy of

punishment (Azrin et al., 1963; Goodall,1984; Trenholme & Baron, 1975). The sup-pressive effects of naturalistic consequencesand component derivatives of complex in-terventions likely depend on the process ofconditioned punishment, an area that re-quires further study. Finally, the develop-ment of stimulus control under punishmenthas been evaluated in few basic or appliedstudies (see Rollings & Baumeister, 1981,for a notable exception). Successful general-ization may hinge on the presence of stimuli

that have acquired tight control over re-sponding, such that few responses occur inthe absence of the punishment contingency.Relative to reinforced responding, it may bedifficult to establish control over punishedresponding with stimuli that are not per-fectly correlated with the delivery of punish-ment. Thus, further research in the areas ofconditioned punishment, stimulus control,and intermittent or delayed punishment ap-pears to be critical for developing a technol-ogy of generalization.

The current literature indicates that pun-ishment must be delivered consistently in allrelevant contexts. Nevertheless, various gen-eralization strategies described by Stokes andBaer (1977) may be useful for promotingtreatment generality when the procedure isextended beyond the initial treatment set-ting. For example, a variety of stimulus con-ditions could be arranged in the initial treat-ment setting (e.g., different caregivers andpeers could be present, diverse activitiescould be scheduled, physical features of the

environment could vary). Stimuli commonto other settings and contexts in which pun-ishment will be applied could be introducedin the initial treatment setting before the in-tervention is widely implemented. Treat-ment generality also may be enhanced by en-


23/34

453PUNISHMENT

suring that reinforcement is implementedconsistently across settings, incorporatingcertain aspects of self-management intotreatment (e.g., self-monitoring), and estab-

lishing salient discriminative stimuli for pun-ishment in all settings and contexts (seeStokes & Baer for further discussion of gen-eralization procedures).

Indirect Effects of Punishment

The effects

Documents

ON THE STATUS OF KNOWLEDGE FOR USING PUNISHMENT: IMPLICATIONS FOR TREATING BEHAVIOR DISORDERS Lerman 2002