EVIDENCE FOR CONTEXTUAL CONTROL BY BASELINE REVIEWS BEFORE PROBES

：芭芭拉

REORGANIZATION OF EQUIVALENCE CLASSES: EVIDENCE FOR CONTEXTUAL CONTROL BY BASELINE REVIEWS BEFORE PROBES
Marilice Garotti, Júlio C de Rose. The Psychological Record. Gambier: Winter 2007. Vol. 57, Iss. 1; pg. 87, 16 pgs

Abstract (Summary)
Two experiments investigated baseline reviews as a relevant variable in reorganization of equivalence classes. After formation of three 4-member classes, participants learned reversals of baseline conditional discriminations and expanded the classes to 5 members each. In Experiment 1, 4 students responded on equivalence probes without baseline reviews preceding each test. They seldom reorganized classes consistently with reversed baselines. Their performance was sometimes consistent with the original baseline and sometimes inconsistent with both baselines. In Experiment 2, 4 additional students received the same training and tests, with baseline reviews preceding each test. These participants always reorganized classes consistently with the reversed baselines. Baseline reviews may have functioned as contextual cues, strengthening the most recent baseline against the conflicting earlier ones.

：芭芭拉

Full Text (4861  words)

Copyright The Psychological Record Winter 2007

[Headnote]
Two experiments investigated baseline reviews as a relevant variable in reorganization of equivalence classes. After formation of three 4-member classes, participants learned reversals of baseline conditional discriminations and expanded the classes to 5 members each. In Experiment 1, 4 students responded on equivalence probes without baseline reviews preceding each test. They seldom reorganized classes consistently with reversed baselines. Their performance was sometimes consistent with the original baseline and sometimes inconsistent with both baselines. In Experiment 2, 4 additional students received the same training and tests, with baseline reviews preceding each test. These participants always reorganized classes consistently with the reversed baselines. Baseline reviews may have functioned as contextual cues, strengthening the most recent baseline against the conflicting earlier ones.


In arbitrary matching-to-sample, each trial displays a sample (e.g., A1, A2, . . . An) together with two or more comparison stimuli (such as B1, B2, . . . Bn). In the presence of An, selections of Bn are designated as correct and usually reinforced, whereas any other selection is considered incorrect and unreinforced. Therefore, matching samples A and comparisons B establishes a conditional relationship (conventionally designated as AB) between each element of set A (sample) and a corresponding element in set B (comparison).

Sidman and Tailby (1982) argued that humans performing arbitrary matching may do something more than conditional discriminations: They may show additional conditional relations, demonstrating that the originally trained relations possess the mathematical properties of reflexivity, symmetry, and transitivity. As Sidman and Tailby argued, conditional discriminations that are reflexive, symmetrical, and transitive are equivalence relations and the related stimuli comprise classes of equivalent stimuli.

Several studies demonstrated equivalence class formation by children and adults with typical development as well as by children and adults with developmental disabilities. Although the nature of stimulus equivalence and the variables responsible for it remain controversial, the finding that equivalence relations emerge after conditional discrimination training is quite robust. Several studies also showed that equivalence classes persist for months, in the absence of any additional training or testing. Typically, such studies conducted additional tests for equivalence several weeks or months after the initial demonstrations. During this time gap, participants did not have additional training or testing (e.g., Rehfeldt & Hayes, 2000; Saunders, Wachter, & Spradlin, 1988; Spradlin, Saunders, & Saunders, 1992; Wirth & Chase, 2002). Rehfeldt and Dymond (2005), however, reported that transfer of functions, a phenomenon related to stimulus equivalence, in the sense that functions acquired directly by one class-member transfer to other class-members, showed considerable less stability when retested after 1 month.

Equivalence relations seem to resist not only the passage of time, but also interferences with the original baseline. For instance, Saunders, Saunders, Kirby, and Spradlin (1988) trained a reversal of one of the original baseline relations after participants showed the emergence of equivalence classes. Performance on equivalence probes, for 2 of the 3 participants, did not show the rearrangement of classes predicted from the reversed baseline, and it remained consistent with the original classes. Pilgrim and Galizio (1990, 1995) also reversed the baseline relations after demonstration of equivalence. They found that the emergent relations of transitivity remained consistent with the original baseline, whereas symmetry relations reversed, according to the reversed baseline. Wilson and Hayes (1996), however, reversed baseline relations after tests for equivalence and found that 19 out of 23 college students reversed symmetry and transitivity relations, consistently with the most recently trained baseline. These authors observed a return to equivalence relations consistent with the earlier baseline only when their probe trials punished all choices consistent with the most recent baseline. Wirth and Chase (2002) conducted tests immediately after reversing baseline conditional discriminations and found that classes rearranged accordingly. However, in retention tests conducted 4 and 10 weeks later, most participants switched back to emergent relations consistent with the original conditional discriminations. Smeets, Barnes-Holmes, Akpinar, and Barnes-Holmes (2003) found, however, that classes usually rearranged after reversal training.

After a reversal of baseline conditional discriminations, participants have conflicting histories on which to base their selections on equivalence probes. The original baseline would lead to selections consistent with the original classes whereas the later reversed baseline would lead to the rearrangement of classes. Because all these studies conducted probes in the absence of differential consequences, it is possible that features of the training procedure determine the history that will prevail. For instance, Pilgrim and Galizio (1995) trained the original conditional discriminations with continuous reinforcement, and the reversal with intermittent reinforcement. They also trained independently each conditional discrimination of the original baseline, whereas reversal training trials were mixed with the trials of the other baseline relations together with test trials for symmetry or transitivity. It is possible, therefore, that the training conditions in the study of Pilgrim and Galizio strengthened the original baseline relative to the reversal, so that the original baseline prevailed over the reversed one when both could serve as the basis for selection in equivalence probes.

Garotti, de Souza, de Rose, Molina, and Gil (2000) replicated the general design of the study of Pilgrim and Galizio (1995), but made reversal training more comparable to the training of the original baseline. In this study both the original and reversed conditional discriminations were taught with continuous reinforcement and in blocks of trials containing only training trials. Also, reviews of the most recent baseline preceded blocks with probes for symmetry and transitivity. With these conditions, 7 of 9 participants rearranged equivalence classes consistently with the reversed baseline.

The results of Garotti et al. (2000) support the hypothesis that performance on postreversal probes depends on conflicting reinforcement histories, so that the prevailing history may depend on procedural details that strengthen one relative to the other. One such procedural variable may be the baseline review before probes. The baseline review may strengthen the relevant history for consistent performance in the probes. To investigate this possibility, the present study conducted two experiments: The first omitted baseline reviews and the second included them. The design of training and probes in this study was similar to that of Garotti et al. (2000), however, the number of comparison stimuli in conditional discrimination trials was increased to three in order to minimize the possibility of control by relations between sample and S-, in an attempt to further reduce the variability of probe performance.

：芭芭拉

☆Experiment 1

Method

Participants

One college student (MALU) and 3 high school students (BARB, ELLE, and NUTA), 15 to 19 years old, served as subjects. The students were recruited by personal contact and did not have experience with derived relational responding research. Participants' native language was Portuguese, and all instructions were presented in this language.

Equipment and Setting

An Apple Macintosh microcomputer Performs CD 6320 presented stimuli and recorded responses using the MTS software (V. 11.1.3; Dube & Hiris, 1997). Each trial displayed five white windows (6 cm x 6 cm) on the gray screen, one at the center and one near each of the monitor's corners. Participants responded moving the computer's mouse to position a cursor on a window and then clicking the mouse's button.

Sessions were conducted in a 4-m x 5-m room in the researcher's home and were approximately 10 to 20 min long. A participant received a credit of R$ 5,00 (about US $2.50 at the time the experiment was conducted) for each experimental day, in which an average of three experimental sessions were conducted with each participant. They collected the full payment at the end of their participation in the experiment. Number of sessions to complete the experiment for BARB, MALU, ELLE, and NUTA was 38, 32, 53, and 68, respectively.

Procedure

Matching-to-sample-trials and sessions. Each trial began with the presentation of a sample on the center window. A response to the sample produced the presentation of three comparisons (except in the initial pretraining trials, as noted below) on the outer windows, and the participant selected one of them. The sample, as well as the positions of the correct and incorrect comparisons, and the blank window, changed from trial to trial according to a prearranged randomized sequence. Correct responses produced a sequence of tones together with a display of stars moving on the screen, and initiated a 1.5-sec intertriai interval (ITI), and incorrect responses produced a 3-sec time-out, during which the screen became completely dark, followed by the ITI. No differential consequences occurred in probe blocks and some baseline review blocks. Responses in these trials produced only the ITI. Any response during the ITI delayed the onset of the following trial for 1.5 sec.

Pretraining. Pretraining trials displayed abstract pictures, different from the ones used in the experimental phases, labeled as P1, P2, P3, T1, T2, T3. The first trial displayed P1 as the sample. A message on the screen instructed the participant to respond to P1 and then to select one of two stimuli displayed on the outer windows (T2 and T3). After they selected either T2 or T3, a message on the screen presented the Portuguese equivalent of "Sorry, your choice was incorrect. After any incorrect choice, the screen would go dark. Try again." The next trial presented sample P1 together with comparisons T1, T2, and T3, and these stimuli remained on the screen until the participant selected the correct comparison stimulus, T1. This selection produced the message: "You made a correct choice. After a correct choice, stars will appear and you will hear tones. Try again." After three additional correct selections, this message was discontinued, and a new message appeared, saying "However, there will be times in which the computer will not tell whether your choices were correct or incorrect. Keep working with attention. If you have any questions, ask now."

A trial block displaying sample (P1) in all its 12 trials followed. The three initial trials displayed two comparison stimuli (T1, T2), and Trials 4 through 12 displayed all three comparison stimuli. The next two blocks repeated this sequence with samples P2 and P3, respectively. The last block consisted of 12 trials with samples P1, P2, and P3, in a randomized sequence, and with the three comparison stimuli. This block repeated until the participant made 100% correct selections.

Sequence of experimental conditions. Each experimental phase comprised a sequence of trial blocks. Table 1 shows the types of trials and number of trials of each type, in each trial block of Experiment 1, as well as the number of presentations of each trial block for each participant. Figure 1 shows the experimental stimuli. Each phase initiated with training of original or reversed conditional discriminations. After the training was complete, a block reviewed all baseline relations. This block, then, repeated without differential consequences, in preparation for probes (blocks without differential consequences are underlined in the table). All blocks that preceded the probes (either with or without differential consequences) had a performance criterion: The block repeated until the participant made two or less incorrect choices in the block. Probe blocks interspersed probe trials (types of probe trials in each block are within parentheses in Table 1) within baseline trials. There was no performance criterion in probe blocks. However, there was a stability criterion for MALU and BARB (see below).

Phase 1 - Initial baseline training and tests for equivalence. The first block taught conditional discrimination AC. The following block trained conditional discrimination BC. Participants then learned the AD conditional discrimination. The next block mixed AC, BC, and AD trials. After the participant met criterion, a message on the screen informed that the computer would no longer signal whether choices were correct or incorrect. The block then repeated without differential consequences. When a participant met criterion, tests for equivalence began.

Reflexivity tests were conducted in a session with three probe blocks. Symmetry tests were conducted in a session with four probe blocks. Transitivity probes were also conducted in a session with four probe blocks. In this phase, transitivity probes included only conditional relations AB, BA, BD, and DB (see Table 1 for details). The stability criterion in probe blocks (for MALU and BARB only) was (a) no more than two incorrect choices per block in baseline trials and (b) no change in the pattern of choices between the last three probe blocks. Pattern of choices were categorized as consistent with the original baseline, reversed baseline, idiosyncratic, or inconsistent (see Results for details). Probe sessions for MALU and BARB repeated whenever performance was considered not stable in a session. For ELLE and NUTA there was a probe session of each type (reflexivity, symmetry, and transitivity) without stability criterion. This schedule was changed for ELLE and NUTA to avoid loss of accuracy in baseline trials. They received one series of probes (reflexivity, symmetry, and transitivity) without differential consequences, followed by two additional series in which selections produced differential consequences on 50% of baseline trials. The point where this change happened for each of these participants is marked with an asterisk in Table 1, right column.

Phase 2 - Reversal of conditional discrimination AD and tests for class reorganization. This phase began with a baseline block with reversal of the AD conditional discrimination (labeled ADr). In ADr trials, selections of D3 were correct in the presence of sample A1, selections of D1 were correct in the presence of sample A2, and selections of D2 were correct in the presence of sample A3. After participants met criterion, the block repeated without differential consequences to prepare for probes, as in Phase 1.

Reflexivity and symmetry probe blocks were identical as in the previous phase, and transitivity blocks included also CD and DC trials, to verify whether any possible tendency to maintain performance in transitivity probes consistent with the original (prereversion) baseline would occur as well in relations not probed before the reversion (cf. Pilgrim & Galizio, 1995).

Phase 3 - Training conditional discrimination DE and testing for equivalence classes. This phase began with a block training conditional relation DE. The next block, added DE trials to the baseline. After participants met criterion, the following block withdrew differential consequences to prepare for probes.

Reflexivity probes added the E stimuli, and symmetry probes now included the ED relation. There were two types of transitivity sessions. The first assessed the same conditional relations tested in the earlier phases (AB, BA, BD, DB, CD, and DC). The second type tested relations involving the E stimuli (AE, EA, BE, EB, CE, and EC).

Phase 4 - Reversal of conditional discrimination BC and tests for class reorganization. In this block comparison stimuli defined as correct for samples B1, B2, and B3 were now C3, C1, and C2, respectively. This was labeled as the BCr conditional discrimination. After the participants met criterion, differential consequences were withdrawn in the next block, to prepare for probes. The following blocks repeated the probes of Phase 3.

Phase 5 - Reinstatement of the AD and BC conditional discriminations. The first block of this phase reinstated the original contingencies for conditional discriminations AD and BC. Differential consequences were withdrawn after students met criterion, and the sequence of probes of Phase 4 was then repeated.

Results and Discussion

Table 1 (rightmost column) shows the number of repetitions of each trial block during training for each participant. The college student, MALU, needed fewer blocks to complete training than the high school students, particularly ELLE and NUTA. Criterion for emergent relations in Phase 1 (before AD reversal) was no more than two responses inconsistent with the probed emergent relation in a probe block. All participants demonstrated formation of equivalence classes. They attained criterion in the first test session for reflexivity and symmetry. In tests for transitivity, MALU, BARB, and ELLE achieved criterion in the first session, and NUTA attained criterion in the second session.

Performance in probes after reversals was considered consistent with the original baseline when no more than three inconsistent choices for each probed relation occurred on the last three blocks of a probe session. Performance was considered consistent with the reversed baseline when no more than two inconsistent choices occurred on the last three blocks of probes. Otherwise, performance was considered inconsistent with both baselines.

In symmetry probes, performance of all participants was consistent with the reversed baseline from Phase 2 through Phase 4, and it was consistent again with the original baseline after reinstatement of this baseline in Phase 5. The only exception was the CB relation for ELLE in Phase 4 (after reversal of the BC relation): ELLE performed in CB probes according to the original baseline, and not the reversed one. In reflexivity probes, ELLE, MALU, and BARB selected the comparison identical to the sample throughout the probes. NUTA, however, after AD reversal, showed inconsistent selections in some probes and eventually came to select a nonidentical stimulus in all probes. The reinstatement of the original baseline in Phase 5 also reinstated selection of the identical stimulus in reflexivity probes.

Table 2 shows outcomes of the last three blocks of transitivity probes in all experimental phases after AD reversal for all participants. In Phase 2, BARB and ELLE performed according to the reversed baseline. MALU and NUTA performed consistently neither with the original nor with the reversed baseline. In Phase 3, after DE training, BARB continued to show transitivity according to the reversed baseline. ELLE showed transitivity consistent with the reversed baseline in all probed relations except those involving the E stimuli, whereas MALU and NUTA again performed consistently with neither baseline. In Phase 4, after reversal of the BC relation, MALU performed according to the reversed baseline in all relations. ELLE performed in some relations consistently with the original baseline and in others consistently with the ADr baseline, but not with the BCr baseline. NUTA's performance was consistent neither with the original nor with the reversed baseline. BARB performed consistently with the reversed baseline in all relations not involving the E stimuli. In the DE baseline, after withdrawal of differential consequences, she maintained a conditional performance different from the trained one, selecting consistently E3 in the presence of D2 and E2 in the presence of D3. Performance in probes involving the E stimuli was always consistent with transitivity of this idiosyncratic DE relation as well as the ADr and BCr relations. After reinstatement of the original AD and BC baselines, no participant performed consistently with the original baselines.

Participants of the present study nearly always reversed performance in symmetry probes consistently with the most recent baseline but, differently from that in the study of Garotti et al. (2000), performance in transitivity probes was variable and often was not consistent with class reorganization. Pilgrim and Galizio (1990, 1995) also found this dissociation between outcomes in symmetry and transitivity tests after reversal of baseline conditional discriminations. It is not clear why symmetry probes remain consistent with the most recent baseline whereas transitivity outcomes are so variable. A possible explanation is that symmetry involves a direct relation between two stimuli, whereas in transitivity the stimuli are indirectly related through a common relation with a third stimulus. Conflicting conditional histories possibly contributed to the variability and loss of conditional responding that occurred in transitivity probes. Experiment 2 investigated whether baseline reviews before probes could strengthen the most recent history relatively to the original one and reduce variability in probes.

：芭芭拉

☆Experiment 2

Method

Participants

Participants were 3 college students, 2 males (PAL and ROG), and 1 female (IRA), and 1 female high school student (KEL). Their ages ranged from 18 to 24 years. They reported no prior experience with derived relational responding research.

Setting and Equipment

Sessions were conducted in a 3-m x 4-m room in a university laboratory and were approximately 10 to 20 min long. All participants completed the study within 23 sessions, with a maximum of 9 experimental days. Other conditions were similar to Experiment 1.

Procedure

The following changes were made in the procedure of Experiment 1. In the initial baseline training in Phase 1, BC training was followed by a block mixing AB and BC trials (see Table 3). Test sessions of symmetry and transitivity always began with a block, repeated until criterion, revising the most recent baseline, without differential consequences. The stability criterion in probes was used for all participants. In Phase 2 (AD reversal), after the participants met criterion, probe conditions similar to those of Phase 1 (probing relations AB/BA, and BD/DB) followed. After performance stabilized, another block followed, adding CD and DC probes. All other conditions were similar to Experiment 1. Table 3 presents the sequence of trial blocks, with the type of trials in each block, number of trials of each type, and number of presentations of each block for each participant.

Results

All participants learned the original baseline and demonstrated equivalence classes. After AD reversal, all participants showed reorganization of the classes according to the ADr relation. They then learned the DE conditional discrimination and continued to perform in probes consistently with the ADr relation. After reversal of the BC relation, all participants again showed reorganization of the classes consistent with the BCr relation (and with the ADr relation as well). When the original baseline was reinstated, all participants showed class reorganization consistent with the original BC and AD relations. Table 4 shows performance of all participants in transitivity probes in Phases 2 through 5.

General Discussion

In Experiment 2, in which baseline reviews preceded probes, classes always reorganized consistently with the most recent baseline. In Experiment 1, when tests occurred in the absence of baseline reviews, reorganization of classes was typically not found, and participants often showed probe performances inconsistent with neither baseline.

There were a few other differences between both experiments, other than baseline reviews. Most participants of Experiment 1 were high school students, whereas most participants of Experiment 2 were college students. The setting of both experiments also differed: Experiment 1 was conducted at the experimenter's home, whereas Experiment 2 was conducted at a laboratory. It is unlikely that this difference in experimental setting could affect a phenomenon as robust as equivalence class formation, because participants maintained baseline accuracy and all of them showed initial class formation. Also, there were not relevant differences between the performance of college and high school students in each experiment. Therefore, it is most likely that presence vs. absence of baseline reviews is responsible for presence vs. absence of class reorganization.

Before baseline reversals, participants have only one experimental history of reinforcement for their conditional discrimination performance. If we consider the AD conditional discrimination, only selections of D1, D2, and D3 have been reinforced in the presence of samples A1, A2, and A3, respectively. After AD reversal, however, participants have conflicting histories in the AD conditional discrimination. In the presence of each A sample, selections of two different D comparisons have been reinforced in different phases. Each of these histories may lead to the emergence of a different equivalence class. In a CD probe, for instance, selections of D1 in the presence of sample C1 are consistent with one (earlier) reinforced baseline, whereas selections of D3 in the presence of C1 are consistent with another (more recent) baseline.

Studies that probed equivalence classes after reversals showed that recency is not always the most important variable determining which conflicting history will control selections in the tests. The more recent baseline usually prevails in symmetry tests, but the earlier baseline may often prevail in transitivity tests (cf. Pilgrim & Galizio, 1995). Conflicting histories may also lead to class disruption, especially in children (Pilgrim, Chambers, & Galizio, 1995). Baseline reviews may thus function as a contextual cue that signals which of the conflicting histories is currently in force. The review may strengthen the most recent history relatively to the earlier one.

Experiment 1, which did not present baseline reviews, showed dissociation of symmetry and transitivity. Most participants performed consistently with the most recent baseline in symmetry probes. However, performance in transitivity probes was rarely consistent with the most recent baseline. This dissociation between symmetry and transitivity was also found by Pilgrim and Galizio (1990, 1995), who also did not make baseline reviews before tests.

The procedure of Experiment 2 replicated that of Garotti et al. (2000), adding a third comparison stimulus in conditional discriminations. Results of Experiment 2 were similar to that study. However, Experiment 2 virtually eliminated the intersubject variability found by Garotti et al. Several investigators (de Rose, 1996; Dube & McIlvane, 1996; Sidman, 1994) argued that variability in the outcomes of equivalence probes may be related to different controlling relations engendered in the baseline, so that participants sometimes "select" the correct comparison stimulus in the presence of a given sample and sometimes "reject" the incorrect comparison in the presence of a sample. Carrigan and Sidman (1992) argued that these "select" and "reject" relations give rise to different emergent relations and, therefore, variability of outcome depends on the variability of control in the baseline. Carrigan and Sidman (see also Sidman, 1987) argue that adding a third choice in conditional discriminations increases the likelihood of "select" relations in the baseline and, therefore, should contribute to decrease intrasubject and intersubject variability. The present results support this conclusion, showing reduced variability with three choice conditional discriminations with baseline reviews. Increased variability found in Experiment 1, even with threechoice conditional discriminations, may be attributed to competing control between two conflicting histories. In Experiment 2, when baseline reviews strengthened the most recent history against the earlier one, variability was reduced.

[Reference]
References
CARRlGAN, P. F., & SIDMAN, M. (1992). Conditional discrimination and equivalence relations: A theoretical analysis of control by negative stimuli. Journal of the Experimental Analysis of Behavior, 58, 183-204.
DE ROSE, J. C. (1996). Controlling factors in conditional discriminations and tests of equivalence. In T. R. Zentall & P. M. Smeets (Eds.), Stimulus class formation in humans and animals (pp. 253-277). Amsterdam: North Holland.
DUBE, W. V., & HIRIS, J. (1997). Match to Sample Program (version 11.1.3) [Computer software]. Waltham, MA: E. K. Shriver Center for Mental Retardation.
DUBE, W. V., & MCILVANE, W. J. (1996). Some implications of a stimulus control topography analysis for emergent behavior and stimulus classes. In T. R. Zentall & P. M. Smeets (Eds.), Stimulus class formation in humans and animals (pp. 197-218). Amsterdam: North Holland.
GAROTTI, M. F., DE SOUZA, D. G., DE ROSE, J. C., MOLINA, R. C., & GIL, M. S. A. (2000). Reorganization of equivalence classes after reversal of baseline relations. The Psychological Record, 50, 35-48.
PILGRIM, C., CHAMBERS, L, & GALIZIO, M. (1995). Reversal of baseline relations and stimulus equivalence: Children. Journal of the Experimental Analysis of Behavior, 63, 239-254.
PILGRIM, C., & GALIZIO, M. (1990). Relations between baseline contingencies and equivalence probe performances. Journal of the Experimental Analysis of Behavior, 54, 213-224.
PILGRIM, C., & GALIZIO, M. (1995). Reversal of baseline relations and stimulus equivalence: Adults. Journal of the Experimental Analysis of Behavior, 63, 225-238.
REHFELDT, R. A., & DYMOND, S. (2005). The effects of test order and nodal distance on the emergence and stability of derived discriminative stimulus functions. The Psychological Record, 55, 179-196.
REHFELDT, R. A., & HAYES, L. J. (2000). The long-term retention of generalized equivalence classes. The Psychological Record, 50, 405-428.
SAUNDERS, R. R., SAUNDERS, K. J., KIRBY, K. C., & SPRADLIN, J. E. (1988). The merger and development of equivalence classes by unreinforced conditional selection of stimuli. Journal of the Experimental Analysis of Behavior, 50, 145-162.
SAUNDERS, R. R., WACHTER, J., & SPRADLIN, J. E. (1988). Establishing auditory stimulus control over an eight-member equivalence class via conditional discrimination procedures. Journal of the Experimental Analysis of Behavior, 49, 95-115.
SIDMAN, M. (1987). Two choices are not enough. Behavior Analysis, 22, 11-18.
SIDMAN, M. (1994). Equivalence relations and behavior: A research story. Boston, MA: Authors CoO p e r ative.
SIDMAN, M., & TAILBY, W. (1982). Conditional discrimination vs. matching to sample: An expansion of the testing paradigm. Journal of the Experimental Analysis of Behavior, 37, 5-22.
SMEETS, P. E., BARNES-HOLMES, Y., AKPINAR, D., & BARNES-HOLMES, D. (2003). Reversal of equivalence relations. The Psychological Record, 53, 91-119.
SPRADLIN, J. E., SAUNDERS, K. J., & SAUNDERS, R. R. (1992). The stability of equivalence classes. In S. C. Hayes & L. J. Hayes (Eds.), Understanding verbal relations (pp. 29-42). Reno, NV: Context Press.
WILSON, K. G., & HAYES, S. C. (1996). Resurgence of derived stimulus relations. Journal of the Experimental Analysis of Behavior, 66, 267-281.
WIRTH, O., & CHASE, P. N. (2002). Stability of functional equivalence and stimulus equivalence: Effects of baseline reversals. Journal of the Experimental Analysis of Behavior, 77, 29-47.

[Author Affiliation]
MARILICE GAROTTI
Universidade Federal do Pará
JÚLIO C. DE ROSE
Universidade Federal de São Carlos

[Author Affiliation]
This study was based on a dissertation presented by Marilice Garotti to Universidade de São Paulo, in partial fulfillment of the requirements for a doctoral degree in Experimental Psychology. The research was supported by MCT/FINEP/PRONEX, and preparation of the manuscript was supported by FAPESP/PRONEX, Grant # 03/09928-4. M. F. Garotti was supported by a CAPES/PICD doctoral scholarship, J. C. de Rose by a Research Productivity Fellowship from CNPq. We thank Maria Teresa Araujo Silva and Renata Molina for their support, and two anonymous reviewers for their thoughtful criticism and suggestions.
Reprint requests and correspondence concerning this article should be addressed to Julio C. de Rose; Departamento de Psicologia; Universidade Federal de Sáo Carlos, Caixa Postal 676; 13565-905 Sáo Carlos, SP; Brazil. (E-mail: djcc@power.ufscar.br).