Reducing Persistent Discrimination Errors - Journal of Applied Behavior Analysis - PDF

JOURNAL OF APPLIED BEHAVIOR ANALYSIS 2014, 47, 738–748 NUMBER 4 (WINTER) INCREASING THE SALIENCY OF BEHAVIOR–CONSEQUENCE RELATIONS FOR CHILDREN WITH AUTISM WHO EXHIBIT PERSISTEN...

JOURNAL OF APPLIED BEHAVIOR ANALYSIS 2014, 47, 738–748 NUMBER 4 (WINTER) INCREASING THE SALIENCY OF BEHAVIOR–CONSEQUENCE RELATIONS FOR CHILDREN WITH AUTISM WHO EXHIBIT PERSISTENT ERRORS WAYNE W. FISHER, TAMARA L. PAWICH, NITASHA DICKES, AMBER R. PADEN, AND KAREN TOUSSAINT THE UNIVERSITY OF NEBRASKA MEDICAL CENTER’S MUNROE-MEYER INSTITUTE Some children with autism spectrum disorders (ASD) display persistent errors that are not responsive to commonly used prompting or error-correction strategies; one possible reason for this is that the behavior–consequence relations are not readily discriminable (Davison & Nevin, 1999). In this study, we increased the discriminability of the behavior–consequence relations in conditional- discrimination acquisition tasks for 3 children with ASD using schedule manipulations in concert with a unique visual display designed to increase the saliency of the differences between consequences in effect for correct responding and for errors. A multiple baseline design across participants was used to show that correct responding increased for all participants, and, after 1 or more exposures to the intervention, correct responding persisted to varying degrees across participants when the differential reinforcement baseline was reintroduced to assess maintenance. These findings suggest that increasing the saliency of behavior–consequence relations may help to increase correct responding in children with ASD who exhibit persistent errors. Key words: acquisition, autism, conditional discrimination, discrete-trial training, discriminated operant, error correction, response cost, schedule discrimination, second-order schedule The central goal of almost all behavior- in contexts in which other responses produce acquisition programs, regardless of whether other reinforcers, even when the experimenter discrete-trial training or more naturalistic training arranges only a single operant contingency (cf. methods are used, is to establish one or more Herrnstein, 1970). Therefore, they argue that the discriminated operants (e.g., pointing to each fundamental analytic unit should be the concur- correspondingly colored object when a therapist rent discriminated operant. says “Touch the — colored one.”) The discrimi- In a typical discrete-trial training arrangement to nated operant has traditionally been considered teach conditional responding, a conditional dis- the fundamental analytic unit for the scientific criminative stimulus (SD) is presented (e.g., “touch study of behavior (Davison & Nevin, 1999), red”), and two or more response options are consisting of one or more antecedent stimuli (e.g., available (e.g., touching either a red or a green card), “Touch your nose”) that occasion a specific each correlated with a different schedule, thus response (e.g., touching the nose and no other conforming to Davison and Nevin’s (1999) concept body part), followed by the reinforcing conse- of a concurrent discriminated operant. The quence produced by the response (e.g., therapist concurrent schedules that are typically programmed presentation of praise and a preferred edible item). during training are reinforcement (e.g., fixed-ratio However, as Davison and Nevin (1999) have [FR] 1) for the response corresponding to the pointed out, discriminated operants always occur conditional SD (or “correct” response; e.g., touching red after the conditional SD “touch red”) and Correspondence regarding this research should be sent to extinction for any other responses (an “error”; e.g., Wayne W. Fisher, Center for Autism Spectrum Disorders, touching green after the conditional SD “touch 985450 Nebraska Medical Center, Omaha, Nebraska 68198 (e-mail: [email protected]). red”). Skinner (1953), McIlvane and Dube (2003), doi: 10.1002/jaba.172 and others have argued that referring to responses 738  REDUCING PERSISTENT DISCRIMINATION ERRORS 739 that do not produce reinforcement as “errors,” at and behavior–consequence relations may not be least in the evaluative sense, is somewhat of a so easily remedied. misnomer. Skinner’s oft-quoted dictum that “The Most acquisition training procedures used with subject is always right” (as quoted in McIlvane & individuals with autism spectrum disorder (ASD) Dube, p. 211) implies that that the trainer (rather and related disabilities focus on reducing confus- than the trainee) is responsible for “errors,” or more ability among antecedents, behaviors, or anteced- precisely, instances of misarranged stimulus control. ent–behavior relations, usually through the McIlvane and Dube (2003) and Davison and introduction and subsequent fading of controlling Nevin (1999) argue that instances of misarranged prompts. A controlling prompt is an antecedent stimulus control (henceforth referred to as errors for stimulus that reliably occasions the correct response the purpose of brevity) can result from confus- (MacDuff, Krantz, & McClannahan, 2001). As an ability at any point in the three-term contingency. example, a child who does not point to objects That is, errors may occur because the alternative (a) named by a therapist, but who will imitate a SDs are highly similar and easily confused (e.g., pointing response modeled by that therapist, might discriminating between magenta vs. pink), (b) be taught receptive identification skills using a responses are similar and easily confused (e.g., progressive prompt delay (Striefel, Bryan, & putting a golf ball 6 m vs. 7 m), or (c) reinforce- Aikins, 1974). The trainer could initially present ment schedules are similar and easily confused (e.g., the conditional SD (e.g., “point to cup” or “point to variable-interval [VI] 60 s vs. VI 75 s). plate”) and immediately model the correct response In addition, confusability can occur at the level (i.e., therapist points to the specified object), and of the antecedent–behavior relation or the then progressively lengthen the delay between the behavior–consequence relation (confusing which conditional SD and the prompt until the child stimulus is correlated with which response or reliably and independently points to the correct which response is correlated with which reinforc- object in response to the conditional SD. Similarly, er; Davison & Nevin, 1999). With regard to the controlling prompts have been used to teach latter, consider a fixed-interval (FI) schedule in correct response differentiation between similar (or which the behavior–reinforcer relation involves a easily confusable) responses, such as dance steps one-to-one correspondence between a single (Vintere, Hemmes, Brown, & Poulson, 2004). response and immediate delivery of the reinforc- Investigators have used a wide variety of other er. However, the FI that must precede that single prompt and prompt-fading procedures to promote response is often not highly discriminable; skill acquisition among children with ASD and therefore, delivery of the reinforcer increases developmental disabilities (for a review, see resistance to extinction for all of the responses MacDuff et al., 2001). that lead up to that single response (thereby Davison and Nevin (1999) have recommended producing the FI scallop on a cumulative graph). that therapists consider the discriminability of both That is, the effects of reinforcement “spill over” to antecedent–behavior relations and behavior–conse- other responses that are temporally proximal to quence relations when they attempt to analyze and and precede the single reinforced response. These modify discriminated operant responding in spillover effects could easily be eliminated by applied settings. Nevertheless, relatively few applied introducing an SD that effectively signals the studies have addressed errors that may arise from completion of the FI (e.g., a clock or countdown the confusability of behavior–consequence relations timer; see Skinner, 1953), in which case the involved in concurrent discriminated operants. individual should rapidly learn to emit a single Davison and Nevin also suggested that sensitivity to response when the timer goes off. However, other the concurrent schedules involved in concurrent sources of confusability in antecedent–behavior discriminated operants might be increased by 740 WAYNE W. FISHER et al. enhancing the distinctiveness of the behavior– to the left side on consecutive trials or from consequence relations (i.e., increasing the saliency quasirandom guessing (which might occur with a of which response goes with which schedule). child who is attending to the sample and One potential reason why most applied comparison stimuli only partially or sporadically, studies have examined the effect of interven- as is sometimes reported for children with ASD and tions that target antecedent–behavior relations related disabilities; Dube & McIlvane, 1999; more than behavior–consequence relations is Fisher, Kodak, & Moore, 2007). Error patterns that the most common schedule procedure for like these (that result from confusability of the acquisition programs is to arrange dense behavior–consequence relations) should be ad- reinforcement schedules (e.g., FR 1) for correct dressed with interventions that increase the saliency responses and extinction for errors. One might of the behavior–consequence relations for correct assume that when the concurrent schedules are responses and errors. an FR 1 for correct responses and extinction for One set of training methods that targets errors, a learner would easily show clearly behavior–consequence relations in order to reduce differentiated responding between the two errors (resulting from side biases, quasirandom schedules. However, as McIlvane and Dube guessing, etc.) during discrete-trial training has (2003) have pointed out, it is important to been referred to as error correction (McGhan & consider the obtained rate of reinforcement that Lerman, 2013). Results of research that has occurs when a participant displays patterns of compared various error-correction procedures often responding that include persistent errors. have produced inconsistent and idiosyncratic Consider an example in which a young girl findings across and within studies and participants with ASD is highly sensitive to response effort. (see McGhan & Lerman, 2013, for a discussion). The therapist places two cards in front of the An alternative approach to reducing persistent participant and says “Point to the animal” on errors displayed by individuals with ASD and one half of the trials and “Point to the building” related disabilities based on the conceptual and on the other half. The participant consistently quantitative analyses of discriminated operant chooses the picture card closest to her right hand behavior by Davison and Nevin (1999) and without directly looking at either of the two McIlvane and Dube (2003) would be to alter the response options (thus minimizing response training arrangement in ways that increase the effort). This produces an obtained reinforce- saliency and discriminability of the behavior– ment schedule equivalent to a variable-ratio consequence relations. That is, we should look (VR) 2, even though the girl has learned little or for ways to increase the likelihood that partic- nothing about receptively identifying animals ipants will easily discriminate that only correct and buildings. It is easy to envision how this responses produce reinforcement and that side obtained VR 2 schedule could easily maintain biases, quasirandom guessing, and other error persistent errors in someone whose behavior is patterns produce near-zero obtained rates of highly sensitive to response effort. reinforcement or produce an unfavorable conse- Thus, although the experimenter has pro- quence (punishment). grammed an FR 1 for correct responding and In the current study, we attempted to increase extinction for errors, the relevant schedule discrim- the discriminability of the behavior–reinforcer ination in the above example is between an FR 1 relations in conditional discrimination acquisition and a VR 2, which is much less discriminable than tasks for participants with ASD by (a) placing an an FR 1 and extinction. Similarly, an obtained rate edible reinforcer in one of three clear containers of reinforcement equivalent to a VR 2 also would (in direct view of the participant) contingent on be produced by alternately switching from the right each correct response; (b) delivering the three  REDUCING PERSISTENT DISCRIMINATION ERRORS 741 accumulated edible reinforcers to the participant METHOD after each of the three containers held a reinforcer Participants and Setting (i.e., delivering the edible reinforcers on a second- order FR 3 FR 1 schedule); and (c) removing all Three boys who had been diagnosed with an accumulated reinforcers from the clear containers ASD in a multidisciplinary ASD clinic participat- as a response-cost contingency for errors. The ed in the investigation while they attended an combination of the second-order schedule for early-intervention program. These boys were the correct responses and the response cost for errors first three individuals to be exposed to the current meant that strings of at least three consecutive treatment protocol. Zane was 5 years old and correct responses were required to produce displayed frequent stereotypy (e.g., hand flapping) reinforcement (e.g., if the participant made correct and episodic noncompliance. He emitted persis- responses on the first and second trials and an error tent errors on tasks that involved intraverbal on the third, the two containers were emptied and responses (e.g., “Tell me some things about reinforcer accumulation had to restart at zero). The apples”), which were not responsive to a variety goal of the clear containers was to increase the of previously attempted prompting or error- saliency of the accumulation and loss of reinforcers correction strategies (e.g., progressive prompt for correct responses and errors, respectively. For delays; repeated practice of the correct response example, without the visual display, the second- after errors). For example, he had been exposed to order FR 3 FR 1 schedule in combination with 10 sessions of error correction combined with response cost would have been difficult to other procedures, and correct responding averaged discriminate from an intermittent schedule of 70%. Bard was 5 years old and displayed frequent reinforcement. The goal of removal of accumulated motor stereotypies and echolalia. He emitted reinforcers from the clear containers contingent on persistent errors during tasks that involved intra- errors was to create a potent and salient response- verbal responses (“Tell me some animals”). Prior cost contingency for errors that would be easy to unsuccessful attempts to decrease these errors discriminate from the reinforcement contingency included prompting and error-correction proce- for correct responses. The combination of the dures (e.g., modeled prompts with and without second-order schedule and the response-cost reinforcement of prompted correct responses; contingency greatly reduced the probability that repeated practice of correct responses after errors). a response pattern that included errors intermixed He was selected for this study after he had been with inadvertent correct responses (e.g., due to a exposed to 34 sessions of error correction side bias or quasirandom guessing) would be combined with other procedures, and correct followed by reinforcement (i.e., the probability of responding averaged 76%. Don was 6 years old three consecutive correct responses occurring by and emitted single-word tacts and mands, chance with a two- or a three-choice response array sometimes independently but mostly when would be.125 and.037, respectively, producing prompted by an adult. He displayed frequent obtained rates of reinforcement equivalent to a VR stereotypy, noncompliance, and aggression. He 8 and a VR 27, respectively). Finally, the visual displayed persistent errors during receptive iden- display and schedules used in the current study tification tasks (e.g., pointing to named colors). A were informed, in part, by the token-reinforcement wide variety of strategies had been implemented to and response-cost procedures developed by Hack- decrease these errors without success (e.g., enberg and colleagues that used light displays to progressive prompt delays, presenting tasks in signal token accumulation and loss with pigeons massed- and blocked-trial formats; repeated (Foster, Hackenberg, & Vaidya, 2001; Pietras & practice of correct responses following errors). Hackenberg, 2005). For example, he had been exposed to 13 sessions of 742 WAYNE W. FISHER et al. error correction combined with other procedures, scored it as incorrect; otherwise it was scored as a and correct responding averaged 72%. disagreement. Interobserver agreement averaged Sessions were conducted in small treatment 100% for all of the subjects. rooms or cubicles (2.5 m by 2.5 m). The rooms contained tables, chairs, and other relevant session Design materials (e.g., preferred toys, stimulus cards). The treatment package was evaluated using a multiple baseline design across participants. In Response Definitions, Measurement, and addition, after the package resulted in relatively Interobserver Agreement high and stable levels of correct responding, a Paper-and-pencil data were collected on par- reversal to baseline (henceforth referred to as ticipants’ correct and incorrect responses during baseline maintenance) was conducted to deter- each trial of each 12-trial (Zane and Bard) or 18- mine whether correct responding would main- trial (Don) session. For Zane and Bard, a correct tain at relatively high levels after the treatment response was scored if the participant said three was withdrawn. If removal of the treatment correct exemplars of the requested category within package resulted in a clear decrease in correct 5 s of the presentation of the conditional SD. responding, the treatment package was reintro- Zane’s targets were to respond correctly to the duced. The package was reintroduced once with sentence, “Tell me three things about —,” with Zane, twice with Bard, and three times with Don. one of the following four words inserted in the blank: apples, cars, zebras, milk. Each target word Procedure was presented three times in a quasirandom Preference assessments. Before the start of the fashion during each 12-trial session. Bard’s targets experiment, we conducted a paired-choice prefer- were to respond correctly to the sentence, “Tell me ence assessment (Fisher et al., 1992) to identify a some —,”with one of the following two words set of three to five high-preference snack items for inserted in the blank: animals or numbers. Each use as reinforcers. In addition, to account for day- target word was presented six times in a to-day fluctuations in preferences for the snack quasirandom fashion during each 12-trial session. items, we conducted a brief preference assessment If Bard stated three or more unique elements from each day using the procedures described by the category, it was scored as a correct response; DeLeon et al. (2001), and the snack item selected otherwise it was scored as an error. For Don, a by that assessment was used as the reinforcer for correct response was scored if he pointed to or that day’s sessions. We did not restrict access to touched the correct comparison stimulus within these snacks at other times of the day. 5 s of the presentation of the initial sample Baseline. We initiated new baselines using stimulus. His targets were to respond correctly to differential reinforcement of correct responses the instruction, “Hand me —,” with one of the so that (a) we would have consistent baseline following nine words inserted in the blank (and an procedures across participants; (b) we could array of colored blocks in front of him): red, green, collect reliability data for the baseline sessions; blue, brown, orange, yellow, white, black, and and (c) we could assess maintenance by periodi- purple. Each target word was presented twice in a cally returning to the differential-reinforcement quasirandom fashion during each 18-trial session. baseline. During baseline, the therapist presented Interobserver agreement for correct and incor- the conditional SD (i.e., “Hand me —” for Don; rect responses was collected during 53% of Zane’s “Tell me three things about —” for Zane; “Tell sessions, 58% of Bard’s sessions, and 50% of me some —” for Bard) at the start of each trial. Don’s sessions. An agreement was recorded if both The participant was given 5 s to respond, and the observers scored the response as correct or if both therapist delivered praise and the high-preference  REDUCING PERSISTENT DISCRIMINATION ERRORS 743 snack item contingent on each correct response next trial. If the containers were empty (e.g., first (i.e., on an FR 1 schedule). Errors (emission of an trial) when a participant made an error, the incorrect response or the absence of a response therapist simply proceeded to the next trial. within 5 s of the conditional SD) produced no differential consequence (extinction). For Don RESULTS only, the therapist also presented the comparison array simultaneously with the SD (i.e., three Figure 1 shows the percentages of correct colored blocks with one corresponding to the responses during the baseline, treatment, and color specified by the conditional SD). baseline-maintenance phases for the three par- Treatment. During treatment, trials were con- ticipants. Zane (top) displayed moderate levels of ducted as in baseline except for the following correct responding during the initial differential modifications. For Zane and Bard (and Don, reinforcement baseline (M ¼ 61.1%; range, 50% starting in Session 18), three small clear containers to 66.7%), and 100% of his errors were errors of (about 3 cm by 3 cm) were positioned in a commission (saying the wrong response rather horizontal array near the therapist and in clear than not responding). During treatment, correct view of the participant. For Don, the clear responding increased (M ¼ 83.3%; range, 66.7% containers were stacked in a vertical array for the to 91.7%) and remained above 90% for the final first 10 treatment sessions. The therapist presented four sessions. When the treatment was with- the conditional SD at the start of each trial. For Don drawn, correct responding remained higher than only, the therapist also presented the comparison array simultaneously with the conditional SD (i.e., DSR three colored blocks with one corresponding to BL Treatment DSRBL/Maintenance 100 the color specified by the conditional SD). Each 80 participant was given 5 s to respond, and the 60 therapist delivered the high-preference snack items 40 according to a second-order FR 3 FR 1 schedule. 20 Zane That is, contingent on the first correct response, the 0 therapist delivered praise and placed an edible 0 5 10 15 20 25 30 35 Treatment 40 45 50 55 60 DSRBL/Maintenance 65 70 75 80 100 reinforcer in the left-most clear container (from the Percentage of Correct Trials 80 participant’s perspective); contingent on the second 60 consecutive correct response, the therapist deliv- 40 ered praise and placed a second edible reinforcer in 20 the middle container; contingent on the third Bard 0 consecutive correct response, the therapist deliv- 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 Treatment ered praise and placed an edible reinforcer in the 100 right-most container, and then immediately deliv- 80 ered all three accumulated snack items to the 60 participant. For the first 10 treatment sessions 40 Reinforecement conducted with Don, the order of reinforcer 20 Display Changed to Don Horizontal accumulation progressed from the container 0 10 20 30 40 50 60 70 80 nearest the participant to the container furthest Session from the participant; thereafter, it was from left to right (as described above). Contingent on an error, Figure 1. The percentage of trials with correct responses the therapist emptied any accumulated snack items during baseline (DSRBL), treatment, and DSRBL mainte- from the clear containers and then presented the nance phases. 744 WAYNE W. FISHER et al. in the initial baseline, but was somewhat more stabilized at relatively high levels for the last six data variable than during treatment (M ¼ 86.5%; points (Ms ¼ 87.3% and 91.7% for the entire range, 66.7% to 100%). phase and the last six data points, respectively; Bard (middle) displayed moderate levels of range, 66.7% to 100%). When the treatment was correct responding during the initial differential withdrawn for the second time in the fifth phase, reinforcement baseline (M ¼ 56.7%; range, 50% correct responding again remained higher than the to 75%), and 92% of his errors were errors of original baseline but became more variable and commission. During treatment, correct responding showed a gradual decreasing trend over the course increased (M ¼ 77.8%; range, 58.3% to 91.7%) of the phase (M ¼ 78.8%; range, 55.6% to 100%). and averaged 80% for the final four sessions. When In the final treatment phase (Phase 6), correct the treatment was withdrawn for the first time in responding increased again (M ¼ 87%; range, the third phase, correct responding gradually 77.8% to 100%). decreased to levels similar to the original baseline For each of the participants, the intervention (M ¼ 62.5%; range, 50% to 83.3%). When the initially produced a decrease in the rate and treatment was reintroduced in the fourth phase, amount of reinforcement delivered for correct correct responding gradually increased (M ¼ responses (due to the response-cost contingency). 75.9%; range, 33% to 100%) and then stabilized That is, each correct response produced rein- at relatively high levels for the last six data points forcement in baseline (100%). In contrast, Zane (M ¼ 91.7%; range, 83.3% to 100%). Moreover, and Bard emitted a total of 25 correct responses when the treatment was withdrawn in the fifth each but collected just 12 reinforcers each (48%) phase, correct responding remained at high and during the first three treatment sessions; Don relatively stable levels (M ¼ 96.9%; range, 91.7% emitted 40 correct responses but collected just 25 to 100%). reinforcers (67.5%) during the first three Don (bottom) displayed moderate levels of treatment sessions. These data suggest that the correct responding during the initial differential response-cost contingency, the visual display, or reinforcement baseline (M ¼ 56.3%; range, 27.8% both, may have played important roles in the to 77.8%), and 67% of his errors were errors of initial increases in correct responding. Otherwise, commission. During treatment, correct responding one would not expect increases in correct increased during the first six sessions but then responding concomitant with large reductions decreased to baseline levels. However, when the in the amount of reinforcement delivered. clear containers used for accumulation of the Recently, investigators have developed statisti- reinforcers were arranged in a horizontal rather than cal procedures that have increased applicability to vertical array starting with Session 18 (similar to the single-case designs (relative to prior methods) and other participants), correct responses increased and that calculate effect sizes that are equivalent for stabilized at relatively high levels for the final four group-comparison and single-case designs, mak- sessions in the phase (correct responding was at ing it more likely that the results of studies with 88.9% in each of these sessions). When the small numbers of subjects will be included in treatment was withdrawn for the first time in the meta-analyses (for discussions, see Fisher & third phase, correct responding remained higher Lerman, 2014; Shadish, 2014). Therefore, we than the original baseline but became more variable calculated Hedge’s G statistic (a common effect- and showed a gradual decreasing trend over the size measure) and associated Z statistics and course of the phase (M ¼ 77%; range, 44.4% to probability values using the procedures described 94.4%). When treatment was reintroduced in the by Shadish, Hedges, and Pustejovsky (2014). fourth phase, correct responding increased and then Levels of correct responding during the last five decreased during the first half of the phase and then data points from the last treatment phase were  REDUCING PERSISTENT DISCRIMINATION ERRORS 745 significantly higher than during baseline (Hedge’s times the treatment was withdrawn in the baseline- G ¼ 3.3; Z ¼ 4.0; p <.001). Similarly, levels of maintenance phases but remained above the correct responding during the last five data points original baseline levels. These maintenance effects from the final baseline-maintenance phase were of the intervention suggest that increasing the significantly higher than during baseline (Hedge’s saliency of the concurrent schedules in effect for G ¼ 2.7; Z ¼ 3.9; p <.001). We also applied correct responses (FR 3 FR 1) and errors (response the conservative dual-criteria method of visual cost) during treatment may have facilitated the inspection developed by Fisher, Kelley, and participants’ discrimination of the concurrent Lomas (2003) and found that, for each partici- schedules in effect for correct responses (FR 1) pant, the change in the level of correct responding and errors (extinction) during subsequent baseline from the initial baseline to the initial treatment maintenance phases. phase was significant. These results are potentially important because they illustrate a method for increasing the saliency of behavior–consequence relations when an DISCUSSION individual has made persistent errors during Davison and Nevin (1999) proposed a model acquisition tasks and has been exposed to a variety of discriminated operant behavior that has the of typical prompting and error-correction strate- concurrent discriminated operant as the funda- gies. Another procedure that may increase mental unit of analysis and that emphasizes the discrimination speed, accuracy, or both, by importance of considering the discriminability of emphasizing the saliency of behavior–conse- both antecedent–behavior and behavior–conse- quence relations, called the differential outcomes quence relations when differentiated responding procedure, involves correlating each stimulus– is analyzed. In the current study, we evaluated an response sequence with a unique reinforcer intervention that was designed to increase the (Trapold, 1970). For example, Litt and Schreib- saliency (or discriminability) of the concurrent man (1981) found that children with ASD schedules in effect for correct and incorrect learned a two-choice receptive identification task responses on conditional discrimination tasks for in fewer trials when each sample–comparison pair three children with ASD who had a history of was correlated with a specific reinforcer (e.g., emitting persistent errors when similar tasks were correctly responding to “Give me compass” was exposed to a variety of prompting, differential reinforced with a tortilla chip; correctly respond- reinforcement, and error-correction strategies. ing to “Give me solder” was reinforced with a Correct responding increased for all three cookie). Similar results have been produced with participants when this treatment was introduced. children and adults with developmental disabil- Moreover, these treatment effects were main- ities (Malanga & Poling, 1992; Saunders & tained to varying degrees when the treatment was Sailor, 1979; Shepp, 1962). However, one withdrawn and the differential reinforcement limitation of the differential outcomes procedure baseline was reintroduced to evaluate mainte- is that the number of conditional discriminations nance of the treatment effects. that can be trained at a time is limited to the For Zane, correct responding was maintained at number of available specific reinforcers that are relatively high levels when baseline contingencies roughly equivalent in reinforcement value (all of were reintroduced in the baseline-maintenance these prior studies trained two sample–compari- phase following one exposure to the treatment. For son pairs at a time). In the current study, our Bard, correct responding was maintained at high discriminability intervention was applied success- levels after two exposures to the treatment. For fully with two (Bard), four (Zane), and nine Don, correct responding gradually decreased both (Don) sample–comparison stimuli at a time. 746 WAYNE W. FISHER et al. Although the current results are encouraging, ment. Nevertheless, we cannot determine whether additional research is needed to develop ways to some or all of the components were necessary to determine when a pattern of persistent discrimi- produce the observed increases in correct respond- nation errors is due to problems in discriminating ing; future research should address this issue. antecedent–behavior relations, behavior–conse- It is also possible that the visibility of the quence relations, or both. Neef, Bicard, and Endo reinforcers in the containers produced a condi- (2001) developed a brief computer-based assess- tioned reinforcement effect that contributed to the ment to evaluate the relative influence of various effectiveness of the intervention. That is, placement reinforcement and response parameters (reinforc- of the snack items in the containers as they were er rate, immediacy, and quality; response effort) accumulated on the second-order schedule may on participants’ choice responses. This assessment have been analogous to the accumulation of tokens was sensitive to both within-subject and between- in a token economy. However, it is worth noting groups biases for these parameters (e.g., the that the intervention produced fairly rapid response allocation of children with attention increases in correct responding even though we deficit hyperactivity disorder was influenced more did not expose the participants to the types of by reinforcer immediacy and quality than by the training steps that are commonly employed before other parameters). Perhaps an analogous assess- implementation of a token system with children ment could be developed to evaluate sensitivity to with disabilities (e.g., specifically training partic- variations in the saliency of antecedent–behavior ipants to exchange and save tokens before using and behavior–consequence relations during con- tokens to reinforce target responses; see Zimmer- ditional discrimination training. man, Zimmerman, & Russell, 1969). In addition, One mitigating factor of this study was that the it is typically recommended that a token system treatment package consisted of multiple compo- be well established before a response-cost compo- nents, and their independent effects were not nent is added to prevent treatment failure (see isolated and evaluated; thus, we cannot make Miltenberger, 2001), which was not necessary with conclusions about the relative effectiveness of each the our participants, perhaps because the interven- component. The components were introduced as a tion involved the accumulation and loss of primary package because each component was designed to reinforcers. That is, the use of primary reinforcers support the central purpose of increasing the in this manner probably precluded the need to saliency of the concurrent schedules in effect for teach participants the relation between tokens and correct responses and errors so that reinforcement primary reinforcement. of correct responses did not produce spillover Future investigators might test the indepen- effects and also strengthen errors. Placement of dent effects of the visual display by comparing each accumulated reinforcer in a clear container the schedule manipulations conducted in the allowed the participant to see the increasing current investigation with and without the visual magnitude of reinforcement that was then deliv- display. One could test the independent effects ered contingent on a third consecutive correct of the second-order schedule (which was desi- response. In addition, removal of the accumulated gned to reinforce only consecutive correct reinforcers from the clear bins as a response cost responses) by comparing it with a response- increased the saliency of the schedule in effect for cost contingency with a comparable visual errors. Finally, the combination of the second-order display (e.g., a snack item visible in a container schedule and the response-cost contingency at the start of each trial and delivered for a correct markedly decreased the probability that inadver- response and removed for an error). Finally, one tent correct responses (e.g., due to a side bias or could test the independent effects of the quasirandom guessing) would produce reinforce- response-cost contingency by comparing two  REDUCING PERSISTENT DISCRIMINATION ERRORS 747 identical second-order schedules with and Dube, W. V., & McIlvane, W. J. (1999). Reduction of without a response-cost component. stimulus overselectivity with nonverbal differential observing responses. Journal of Applied Behavior A limitation of the study was that the visual Analysis, 32, 25–33. doi: 10.1901/jaba.1999.32-25 display was changed for Don about midway Fisher, W. W., Kelley, M. E., & Lomas, J. E. (2003). Visual through the first phase of treatment. It is worth aids and structured criteria for improving visual inspection and interpretation of single-case designs. noting, however, that this change made the visual Journal of Applied Behavior Analysis, 36, 387–406. doi: display for the accumulation and loss of reinforcers 10.1901/jaba.2003.36-387 the same for Don as that for Zane and Bard. Fisher, W. W., Kodak, T., & Moore, J. W. (2007). Moreover, subsequent reversals of the treatment Embedding an identity-matching task within a prompting hierarchy to facilitate acquisition of condi- with the horizontal display in place showed that tional discriminations in children with autism. Journal treatment consistently produced increases in of Applied Behavior Analysis, 40, 489–499. doi: correct responding and its withdrawal produced 10.1901/jaba.2007.40-489 Fisher, W. W., & Lerman, D. C. (2014). It has been said decreases in correct responding, albeit gradually. that, “There are three degrees of falsehoods: Lies, damn Another limitation of the study was that we did lies, and statistics.” Journal of School Psychology, 52, 243– not include an error-correction procedure in 248. doi: 10.1016/j.jsp.2014.01.001 Fisher, W., Piazza, C. C., Bowman, L. G., Hagopian, L. P., baseline, although all three participants had been Owens, J. C., & Slevin, I. (1992). A comparison of two exposed to error correction for these tasks during approaches for identifying reinforcers for persons with their clinical treatment before initiation of this severe and profound disabilities. Journal of Applied study. Nevertheless, future investigators should Behavior Analysis, 25, 491–498. doi: 10.1901/ jaba.1992.25-491 consider comparing the current intervention with Foster, T. A., Hackenberg, T. D., & Vaidya, M. (2001). error correction. Second-order schedules of token reinforcement with In summary, we attempted to increase the pigeons: Effects of fixed- and variable-ratio exchange schedules. Journal of the Experimental Analysis of discriminability of the behavior–reinforcer rela- Behavior, 76, 159–178. doi: 10.1901/jeab.2001.76-159 tions in conditional discrimination tasks using Herrnstein, R. J. (1970). On the law of effect. Journal of the a unique visual display to signal a second-order Experimental Analysis of Behavior, 13, 243–266. doi: 10.1901/jeab.1970.13-243 FR 3 FR 1 schedule for correct responses and a Litt, M. D., & Schreibman, L. (1981). Stimulus-specific response-cost contingency for errors. All three reinforcement in the acquisition of receptive labels by participants showed increases in correct responses autistic children. Analysis and Intervention in Develop- that were maintained to varying degrees when the mental Disabilities, 1, 171–186. doi: 10.1016/0270- 4684(81)90030-6 intervention was withdrawn. These results should MacDuff, G. S., Krantz, P. J., & McClannahan, L. E. (2001). be replicated with other children with ASD, and Prompts and prompt-fading strategies for people with the independent effects of the treatment compo- autism. In C. Maurice, G. Green, & R. M. Foxx (Eds.), Making a difference: Behavioral intervention for autism nents should be evaluated. (pp. 37–50). Austin, TX: Pro-Ed. Malanga, P., & Poling, A. (1992). Letter recognition by adults with mental handicaps: Improving performance REFERENCES through differential outcomes. Developmental Disabil- ities Bulletin, 20, 39–48. Davison, M., & Nevin, J. A. (1999). Stimuli, reinforcers, McGhan, A. C., & Lerman, D. C. (2013). An assessment of and behavior: An integration. Journal of the Experimen- error-correction procedures for learners with autism. tal Analysis of Behavior, 71, 439–482. doi: 10.1901/ Journal of Applied Behavior Analysis, 46, 626–639. doi: jeab.1999.71-439 10.1002/jaba.65 DeLeon, I. G., Fisher, W. W., Rodriguez-Catter, V., McIlvane, W. J., & Dube, W. V. (2003). Stimulus control Maglieri, K., Herman, K., & Marhefka, J. (2001). topography coherence theory: Foundations and exten- Examination of relative reinforcement effects of stimuli sions. The Behavior Analyst, 26, 195–213. identified through pretreatment and daily brief prefer- Miltenberger, R. G. (2001). Behavior modification: principles ence assessments. Journal of Applied Behavior Analysis, and procedures (2nd ed.). Belmont, CA: Wadsworth/ 34, 463–473. doi: 10.1901/jaba.2001.34-463 Thomas Learning. 748 WAYNE W. FISHER et al. Neef, N. A., Bicard, D. F., & Endo, S. (2001). Assessment of Skinner, B. F. (1953). Science and human behavior. New impulsivity and the development of self-control in York, NY: MacMillan. students with attention deficit hyperactivity disorder. Striefel, S., Bryan, K. S., & Aikins, D. A. (1974). Transfer of Journal of Applied Behavior Analysis, 34, 397–408. doi: stimulus control from motor to verbal stimuli. Journal of 10.1901/jaba.2001.34-397 Applied Behavior Analysis, 7, 123–135. Pietras, C. J., & Hackenberg, T. D. (2005). Response-cost Trapold, M. A. (1970). Are expectancies based upon punishment via token loss with pigeons. Behavioural different positive reinforcing events discriminably Processes, 69, 343–356. doi: 10.1016/j.beproc.2005. different? Learning and Motivation, 1, 129–140. doi: 02.026 10.1901/jaba.1974.7-123 Saunders, R. R., & Sailor, W. (1979). A comparison of three Vintere, P., Hemmes, N. S., Brown, B. L., & Poulson, C. L. strategies of reinforcement on two-choice learning (2004). Gross-motor skill acquisition by preschool problems with severely retarded children. AAESPH dance students under self-instruction procedures. Review, 4, 323–333. Journal of Applied Behavior Analysis, 37, 305–322. Shadish, W. R. (2014). Analysis and meta-analysis of single- doi: 10.1901/jaba.2004.37-305 case designs: An introduction. Journal of School Psychology, Zimmerman, E. H., Zimmerman, J., & Russell, C. D. 52, 109–122. doi: 10.1016/j.jsp.2013.11.009 (1969). Differential effects of token reinforcement on Shadish, W. R., Hedges, L. V., & Pustejovsky, J. E. (2014). instruction-following behavior in retarded students Analysis and meta-analysis of single-case designs with a instructed as a group. Journal of Applied Behavior standardized mean difference statistic: A primer and Analysis, 2, 101–112. doi: 10.1901/jaba.1969.2-101 applications. Journal of School Psychology, 52, 123–147. doi: 10.1016/j.jsp.2013.11.005 Shepp, B. E. (1962). Some cue properties of anticipated rewards in discrimination learning of retardates. Journal Received April 22, 2014 of Comparative and Physiological Psychology, 55, 856– Final acceptance May 9, 2014 859. doi: 10.1037/h0040940 Action Editor, Dorothea Lerman