Review completion

ORIGINAL ARTICLE

Changing Racial Bias by Transfer of Functions in Equivalence Classes

Táhcita M. Mizael1 & João H. de Almeida1 & Carolina C. Silveira1 & Julio C. de Rose1

# Association for Behavior Analysis International 2016

Abstract Several studies used the stimulus equivalence paradigm to investigate attitudes toward socially relevant stimuli. In one of these studies, de Carvalho and de Rose (The Psychological Record 64: 527–536, 2014) used matching-to-sample (MTS) training to establish equiva- lence relations between a positive symbol and faces of individuals of African descent (toward which children showed negative bias prior to the research). Only one of four children showed the intended classes. The present study manipulated training parameters to increase the yield of equivalence classes comprising relations contrary to children’s previous racial bias. Thirteen children learned matching tasks that would potentially establish equivalence relations between Black faces and positive symbols, con- trary to their preexperimental bias. All 13 children showed equivalence class formation, and nine of them maintained relations between Black faces and positive symbols in a different and more stringent test. Children’s evaluations of the faces with the Self-Assessment Manikin (SAM) showed a

pronounced negative bias toward Black faces before training. After class formation, however, the difference between evaluations of Black and White faces was no longer sta- tistically significant. These results showed that procedures based on equivalence and transfer of functions may con- tribute to educational programs designed to decrease racial biases, a significant challenge for our increasingly multi- cultural and multiracial societies.

Keywords Stimulus equivalence . Transfer of functions .

Conflicting relations . Stimulus equivalence and attitudes .

Prejudice

Stimulus equivalence has proven to be a powerful procedure to change behavioral functions of stimuli. Stimuli are said to be equivalent, and comprise a class of equivalent stimuli, when arbitrary relations between them are symmetrical, tran- sitive, and reflexive (Sidman, 1986, 1994; Sidman & Tailby, 1982). Human participants, particularly verbally able partici- pants, readily form equivalence classes with matching-to- sample (MTS) training (e.g., Carr, Wilkinson, Blackman, & McIlvane, 2000; Devany, Hayes, & Nelson, 1986; Sidman & Tailby, 1982). When equivalent stimulus classes are formed, a function imparted to one class- member by direct training is often exhibited by the other class members without any addi- tional training. This transfer of functions has been document- ed in several studies (e.g., Amd, Barnes-Holmes, & Ivanoff, 2013; de Rose, McIlvane, Dube, Galpin, & Stoddard, 1988; Dougher, Augustson, Markham, Greenway, & Wulfert, 1994; Dymond & Barnes, 1994; Hayes, Kohlenberg, & Hayes, 1991; Wulfert & Hayes, 1988).

Research on stimulus equivalence typically uses stimuli, such as arbitrary shapes or nonsense syllables or pseudowords, which are presumably meaningless. Another way to demonstrate

Táhcita Mizael and Carolina Silveira were supported by Graduate fellowships from the Brazilian Ministry of Education (CAPES). João de Almeida was supported by a post-doctoral fellowship from the São Paulo Research Foundation (Grant # 2014/01874-7). Julio de Rose was supported by a Research Productivity Grant from the National Research Council (CNPq). This manuscript is based on the master’s thesis presented by the first author to the Graduate Program in Psychology at Universidade Federal de São Carlos. This research was part of the scientific program of Instituto Nacional de Ciência e Tecnologia sobre Comportamento, Cognição e Ensino (INCT-ECCE), supported by CNPq (Grant 573972/2008-7) and FAPESP (Grant 2008/57705-8).

* Julio C. de Rose [email protected]

1 Departamento de Psicologia, Universidade Federal de São Carlos, Caixa Postal 676, São Carlos, SP, Brazil

Psychol Rec DOI 10.1007/s40732-016-0185-0

transfer of functions between equivalent stimuli, however, is to establish stimulus classes comprising arbitrary stimuli together with a meaningful one. For instance, Barnes-Holmes, Keane, Barnes-Holmes, and Smeets (2000) trained college students to match the printed word HOLIDAYS (labeled A1) to a geometric shape (B1) and the printed word CANCER (A2) to another geometric shape (B2). Then the shapes B1 and B2 were matched, respectively, to two arbitrary pictures (labeled C1 and C2). Appropriate tests confirmed that the arbitrary picture C1 became equivalent to the geo- metrical shape B1 and to the word HOLIDAYS, where- as the arbitrary picture C2 became equivalent to the geometrical shape B2 and to the word CANCER. In tests for transfer, the pictures C1 and C2 served as labels for two containers with identical cola drinks. Participants were asked to taste the drinks in both containers and report which one they preferred. Most participants reported to prefer the drink labeled with C1, which was equivalent to the word HOLIDAYS (see also Arntzen, Fagerstrom, & Foxall, 2016; Smeets & Barnes-Holmes, 2003).

Bortoloti and de Rose (2009) also demonstrated transfer of functions from meaningful to abstract stimuli. Meaningful stimuli were pictures of facial emotional expressions portrayed by adult humans. Training established three equiv- alent stimulus classes. Each class contained abstract pictures and one emotional expression of happiness, neutrality, and anger, respectively. Transfer of function was measured with a semantic differential, in which participants rated the abstract stimuli on a series of Likert scales anchored by opposite ad- jectives. Stimuli equivalent to the happy faces produced pos- itive evaluations whereas stimuli equivalent to the angry faces produced negative evaluations, and stimuli equivalent to the neutral faces produced evaluations approaching neutrality.

These studies showed that a stimulus initially evaluated as approximately neutral can acquire either a positive or a nega- tive valence when it becomes equivalent to a stimulus that already had a positive or negative valence. A recent study by de Carvalho and de Rose (2014) used a similar procedure as a model to investigate racial attitudes in children. Participants were four children that showed a negative bias toward faces of Black1 individuals in preexperimental tests. They were then trained to match positive and negative symbols (A1 and A2), respectively, to two geometric shapes (B1 and B2); these geo- metric shapes were then matched to Black faces (C1) and an arbitrary picture (C2), respectively. Emergence of AC and CA relations would document the formation of equivalence clas- ses, so that a positive symbol would be equivalent to Black faces and a negative symbol to an arbitrary picture. Only one

of four children exhibited the emergent relation between Black faces and positive symbols.

The failure of most participants to respond according to equivalence in the de Carvalho and de Rose (2014) study replicates results of several studies investigating equivalence relations that contradict the preexperimental history of the participants. These studies used a simple design that we will refer to as the Conflicting Relations Paradigm. The first study to use this paradigm was conducted in Northern Ireland, by Watt, Keenan, Barnes, and Cairns (1991). Watt et al. trained university students to match Catholic names (A) to nonsense syllables (B), and the nonsense syllables to Protestant symbols (C). Therefore, training could lead to the emergence of AC and CA relations between Catholic names and Protestant sym- bols, which would contradict preexperimental relations in a nation with a history of religious conflict. Watt et al. found that English participants showed the intended equivalence rela- tions, whereas few Northern Irish participants did so, suggest- ing substantial interference from preexperimental relations. Subsequent replications of this paradigm were conducted with a variety of conflicting preexperimental and potentially emer- gent relations, consistently suggesting that preexperimental relations interfered with equivalence class formation (Dixon, Rehfeldt, Zlomke, & Robinson, 2006; Haydu, Camargo, & Bayer, 2015; Leslie et al., 1993; Moxon, Keenan, & Hine, 1993).

Most of these studies used a similar training design: Relation AB was trained first, followed by BC training. Then, a block of trials mixed AB and BC trials, with feedback only in 50 % of the trials. This was followed by AC and CA tests for transitivity and equivalence (e.g., de Carvalho & de Rose, 2014; Haydu et al., 2015; Leslie et al., 1993; Moxon et al., 1993; Watt et al., 1991). Several features of this design may be less than optimal to yield equivalence classes: insufficient review of the baseline, starting tests with the most complex relations (transitivity and equivalence; see Adams, Fields, & Verhave, 1993), and lack of symmetry tests. The present study was, therefore, a replication of de Carvalho and de Rose (2014) with procedural changes designed to increase the probability of equivalence class formation: a simple-to-complex protocol (cf. Adams et al., 1993) with (a) symmetry tests after training of each baseline relation and (b) mixing AB and BC trials with feedback in 100 % of the trials, followed by a baseline revision with 50 % feedback.

Furthermore, there are indications that training with de- layed matching-to-sample (DMTS) can increase probability of class formation and the magnitude of transfer of function, compared to simultaneous matching (SMTS; Arntzen, 2006; Bortoloti & de Rose, 2009). Therefore, this study compared a group of children trained with DMTS with another trained with SMTS. Additionally, the study evaluated whether the experimenter’s race could influence the results, since social

1 The terms negro (Black) and branco (White) are generally used in Brazil to refer to races, both by the individuals themselves and by official agencies, with no opprobrium. Because the research was done in Brazil, we will use the closest translation of the terms used in the country throughout this paper.

Psychol Rec

influence can be seen as a form of stimulus control (Weatherly, Miller, & McDonald, 1999).

The major purpose of the study, therefore, was to investigate whether an optimization of training parameters would increase the probability of equivalence relations between Black faces and a positive symbol, overriding the preexperimental relations between Black faces and negative attributes.

Method

Participants

Thirteen children, aged between 8 and 10 years, who showed negative bias toward faces of Black individuals in a screening pretest (see BProcedure^) took part in the experiment. They were all students at an elementary public school in the state of São Paulo, Brazil. Table 1 shows participants’ demographic data.

Setting and Equipment

Data collection was conducted in the school’s toy library. A Dell Inspiron 14.550 computer, with Intel Core i3 processor, and MTS III software (Wallace, 2003) presented stimuli and recorded responses. Sessions were conducted with one partic- ipant at a time and an experimenter remained in the room.

Stimuli

Stimuli A1 and A2 were drawings of a hand making a thumbs- up and a thumbs-down symbol, respectively (also used by de Carvalho & de Rose, 2014). B1, B2, and C2 were abstract pictures. C1 and C3 comprised four pictures of faces of Black individuals (C1) and four pictures of faces of White

individuals (C3). Each C1 or C3 presentation displayed only one of the faces, chosen randomly from the set of four. Therefore, C1 and C3 were not individual stimuli (to avoid control by idiosyncratic features of any particular face) but rather common features of two four-stimulus classes, com- prised by, respectively, Black faces and White faces. The pictures of the faces were obtained at http://faceresearch.org/. The selected pictures portrayed faces of fairly average individuals, with an apparent age between 20 and 30 years. These pictures were chosen because they show only the face on a blank background, with no apparent emotional expression. The abstract pictures were obtained from the stimulus gallery of MTS software v.11.1.3 (Dube & Hiris, 1997).

Assessment Instruments

Self-Assessment Manikin (SAM) This instrument measures stimuli or events on three domains: pleasure, arousal, and dominance (Bradley & Lang, 1994). We used only the domain of pleasure in this research. Figure 1 illustrates this instrument, in the dimension used. Participants were asked to rate each stimulus with five manikins side by side, varying from a smil- ing one to an unhappy and frowning one. Participants were initially asked what feelings each manikin was displaying, and corrected if wrong (which almost never occurred). They were then instructed to choose whichever circle (below each man- ikin and between them) best represented what they felt when they looked at the picture displayed on top (see Fig. 1). The experimenter told them that there were no correct or wrong answers, because people have different tastes and preferences for clothes, TV characters, and so forth.

Table 1 Sex, age and race of the participants

Participant Sex Age (in years) Race

P1 Female 9 White

P2 Male 10 Black

P3 Male 9 White

P4 Female 8 White

P5 Male 8 Black

P6 Female 9 White

P7 Male 10 Black

P8 Female 9 Black

P9 Male 9 Black

P10 Female 9 White

P11 Female 8 Black

P12 Female 9 White

P13 Female 9 White Fig. 1 Self-Assessment Manikin (SAM)

Psychol Rec

Implicit Relational Assessment Procedure (IRAP) The IRAP (Barnes-Holmes et al., 2006) evaluates properties of relations between stimuli. Participants are asked to respond as fast as possible to pairs of stimuli, choosing a response option referring to the relation between the stimuli: yes or no in the present study (see Fig. 2). Correct and incorrect choices vary in successive trial blocks, so that selections consistent with the presumed bias of the participants are correct in some blocks of trials and incorrect in others. It is assumed that participants will respond faster on trials consistent with their biases than in those that are inconsistent (Barnes-Holmes, Murphy, & Barnes-Holmes, 2010). The IRAP has been suc- cessfully used to identify biases (Drake et al., 2010) and racial stereotypes (Barnes-Holmes et al., 2010). It has also been successfully used with young children (Rabelo, Bortoloti, & Souza, 2014).

Procedure

Screening Children rated (see BAssessment Instruments^, and Fig. 1) all stimuli used in training and testing and five pictures of cartoon characters, used as distractors, on the SAM. Stimuli were presented in a randomized sequence. Participants were instructed to mark the circle below or between the manikins that best represented the pleasure level elicited by the target stimulus. Participants that rated one or more of the Black faces with scores below those attributed to the White ones were submitted to a second test to confirm the bias. This was an MTS test, labeled AC3, because trials displayed three com- parison stimuli. Each trial displayed either the symbol A1 or A2 as a sample (thumbs up or down, respectively) together with three comparisons: a picture of a Black face (C1), an abstract symbol (C2), and a picture of a White face (C3).

Participants were instructed to select the comparison which they considered to match the sample. The test comprised 16 trials, eight with each sample, in a randomized sequence. There were no differential consequences for any selections. The following equation was used to determine bias: b ¼ Wþð Þ � Bþð Þ½ � þ B�ð Þ � W�ð Þ½ �, where W + corre- sponds to the number of selections of White faces for the thumbs up sign, B + the number of selections of Black faces for the thumbs up sign, B � the number of selections of Black faces for the thumbs down sign, and W � the number of selections of White faces for the thumbs down sign. A value of b=0 indicated absence of bias, whereas positive values indicated negative bias toward Black faces and negative values indicated positive bias toward those faces. Children who yielded values of b equal to or higher than four were selected for the study. Table 2 summarizes the experimental phases, showing the conditional relations trained or tested in each phase, and the number of trials and learning criteria (when applicable).

Pretraining A pretraining was conducted initially to familiar- ize participants with the arbitrary MTS task. Participants were instructed that, when a picture appeared in the center of the screen, they had to click on it and find out which of the pic- tures displayed below was the one that Bwent with^ the center picture. Participants were also told that on the initial trials the computer would present a cue indicating the correct choice, and they needed to pay attention because cues would no lon- ger be presented after these initial trials. On the first trial, the computer showed only the correct comparison, along with two blank squares. On the second trial, the computer showed two of the comparison stimuli (along with a blank square), and on the third trial, the three comparison stimuli appeared.

Fig. 2 Implicit Relational Assessment Procedure (IRAP). The three pictures in the left- upper side were deemed as posi- tive, whereas the three below were deemed as negative

Psychol Rec

Participants were taught conditional relation XY, with three samples and three comparisons. These were arbitrary pictures different from those used in the training itself.

Matching-to-sample trials A trial began with the presenta- tion of the sample in the central window. In DMTS, a mouse click on the sample produced the withdrawal of the sample and, after a 2-s interval, the presentation of two or three com- parison stimuli, depending on the experimental phase. In SMTS, a click on the sample produced the immediate presen- tation of the comparison stimuli and the sample remained on. The intertrial interval was 1.5 s.

Differential consequences were presented in training blocks and not in test blocks. Correct responses produced a clapping sound and a display of moving stars on the screen, and incorrect responses produced a black screen for 1.5 s. After each correct response, the experimenter also dropped a marble into a cup, with each marble corresponding to a point. Children were told that points would accumulate during the experiment, and would be exchanged for prizes at the end of the study. Before participants entered this phase, they were instructed that this task was similar to the previous one (pretraining), but there would be no cues about which choice would be correct, so they had to pay attention to what happened when they made their choices, starting with the first trial.

Experimental conditions Seven participants did all training and testing with SMTS and six did them with DMTS. A White experimenter conducted all sessions for five partici- pants (three in the SMTS condition and two in the DMTS condition). A Black experimenter conducted all sessions for the remaining participants.

Training The learning criterion for both groups during train- ing and testing was a maximum of one error per block. The participants were required to achieve criterion in two consec- utive blocks to advance to the following phase, except on symmetry tests. On symmetry tests, achieving criterion in just one block was sufficient to advance to the following phase.

After the pretraining, the AB relation2 was taught with a 16- trial block (eight A1B1 and eight A2B2). When criterion was met in two consecutive blocks, a symmetry test (BA) followed, also in a 16-trial block (eight B1A1 and eight B2A2). Before symmetry tests, children were informed that the computer would not tell whether responses were correct or wrong. After criterion was met, BC training was conducted (with the same procedure as AB training), followed by CB symmetry tests. The next training block, with 32 trials, mixed AB and BC trials. This was followed by a baseline review block, similar to the previous one, with the exception that differential conse- quences were presented only in 50 % of the trials. Before this block started, children were told that in some trials the comput- er would not tell whether responses had been correct or wrong.

Equivalence tests Two tests were conducted. The first was a standard transitivity and equivalence test verifying AC and CA relations. Only C1 and C2 appeared as choices or samples

Table 2 Training and testing sequence, containing the conditional relations, number of trials, and learning criteria for each phase

Phase Number of Trials Learning Criterion Conditional Relations

Pretest 16 — AC

Pretraining 15 14 of 15 trials X1Y1/X2Y2/X3Y3

AB1 Training 16 15 of 16 trials A1B1/A2B2

AB2 Training 16 15 of 16 trials A1B1/A2B2

BA Symmetry Test 16 15 of 16 trials B1A1/B2A2

BC1 Training 16 15 of 16 trials B1C1/B2C2

BC2 Training 16 15 of 16 trials B1C1/B2C2

CB Symmetry Test 16 15 of 16 trials C1B1/C2B2

ABBC1 Mixed Training 16 15 of 16 trials A1B1/A2B2/B1C1/B2C2

ABBC2 Mixed Training 16 15 of 16 trials A1B1/A2B2/B1C1/B2C2

Baseline Review 1 16 15 of 16 trials A1B1/A2B2/B1C1/B2C2

Baseline Review 2 16 15 of 16 trials A1B1/A2B2/B1C1/B2C2

AC Test 16 15 of 16 trials A1C1/A2C2

CATest 16 15 of 16 trials C1A1/C2A2

Posttest 16 — A1C1/A2C2

2 A simpler training could just reverse the presumed preexperimental relations, in order to establish classes comprised by A1/B1/C1 and A2/ B2/C3 (i.e., Black faces equivalent to the positive symbol and White faces equivalent to the negative symbol). We considered (as did de Carvalho & de Rose, 2014) it ethically unacceptable to expose children to a training that could establish equivalence relations between racial traits and a negative sign. Therefore, we used the same strategy as de Carvalho and de Rose (i.e., providing training that could establish equivalence relations between Black faces and a positive sign, and between a negative sign and an abstract symbol).

Psychol Rec

in this test. The test blocks had 16 AC and 16 CA trials, comprising 32 trials in a randomized order. There were two consecutive applications of this 32-trial block, regardless of participants’ performance. This was followed by the AC3 test, in which AC trials displayed three comparison stimuli: a Black face (C1), the arbitrary symbol (C2), and a White face (C3). This test had a 16-trial block and was applied two times consecutively, regardless of participants’ performance. Before the tests began, chil- dren were informed that the computer would no longer signal whether responses had been correct or wrong. Figure 3 shows examples of trials for standard AC/CA tests, and for the AC3 test.

Posttests All participants rated the same stimuli with the SAM. They were then submitted to IRAP testing. IRAP cate- gories were Black or White faces, the target stimuli were pos- itive and negative images,3 and the response options were Byes^ and Bno.^ The relations White–positive and Black–neg- ative were deemed as consistent with the previous history of the participants, and the relations White–negative and Black– positive were considered inconsistent. Each trial displayed one of the two categories, one target, and the two response options. Children were told they would be playing a game like Simon Says, and thus the experimenter would give them a rule, and the participants had to follow that rule as much as possible, and respond on the computer screen as fast as pos- sible. In the first trial block, one of the possibilities, either consistent or inconsistent, was considered correct, and the correct and incorrect options were reversed at each of the following blocks. The initial blocks were training blocks, and data were excluded from the analysis. Training ended when participants attained a criterion of at least 70 % correct responses in two consecutive blocks. A correct response was either a consistent or an inconsistent response, according to the block, with a maximum latency of 3 s. A maximum of six training blocks could be conducted with each participant. The number of test blocks was also six.

Results

Pretests

Twenty-two children showed lower SAM ratings for Black faces compared to White faces. Nine of these children did

not reach the value of b=4 to confirm the bias in the AC3 test (see Screening, in the BProcedure^ section), and were not included in the study. For the 13 other children, values of b ranged from 4 to 16 (M=7.62, SD=3.91). For three partici- pants (P5, P6, and P10), the value of b was 4. One child (P13) reached the maximum value for b, 16.

MTS Training and Equivalence Tests

Table 3 shows training results. Eleven participants completed the pretraining in one or two blocks. Exceptions were P13 (three blocks) and P11 (five blocks). All participants reached criterion for AB training in two blocks and for BA symmetry in one block, with the exception of P13, who reached symme- try criterion after two blocks. All participants reached criterion for BC training in three blocks or less, with the exception of P13, who required five blocks. All attained CB symmetry criterion in one block, with the exception of P5, who required three blocks. All participants required two blocks to reach criterion in mixed AB/BC trials; they also required two blocks to attain criterion in the baseline review with feedback in 50 % of trials (with the exception of P10, who needed four blocks for this baseline review).

Table 4 shows results for equivalence tests. There were two blocks of AC/CA tests. All children reached criterion on the first block (only P6 did not attain criterion on the first block, falling short for one response). This AC/CA test did not in- clude White faces. When White faces were included as an option, in the AC3 test, nine participants continued to choose the Black face for the positive sample in at least 15 trials. Three other participants fell short of this criterion for a few trials, and P13 showed the lowest scores of this Black–positive relation (below 50 % for the aggregated trials of the two blocks); this was the participant who showed the most extreme negative bias toward Black faces in the screening AC3 test.

SMTS versus DMTS and Black and White Experimenters Participants performed with similar accuracy on both sym- metry and equivalence tests after training with DMTS or SMTS. Kruskal-Wallis Test confirmed the absence of sig- nificant differences in equivalence class formation for these two conditions (K=38.167, p>.05).

Performance in the standard equivalence test (AC/CA) was accurate for all participants, regardless of experimenter. In the AC3 test, four participants did not attain criterion, two of them with each experimenter. Therefore, two of the eight partici- pants (25 %) with the Black experimenter failed to reach cri- terion in the AC3 test. On the other hand, two of five partic- ipants (40 %) with the White experimenter failed to attain criterion in the AC3 test.

3 These images were provided by the International Affective Picture System (IAPS; Lang, Bradley & Cuthbert, 1999). The picture selection was based on a study (McManis, Bradley, Berg, Cuthbert & Lang, 2001) that verified which pictures children aged between 7 and 11 evaluated as positive, negative, and neutral according to SAM results. We used only one image that was not used in McManis et al. study (rotten fruits), which is, nevertheless, present in the IAPS.

Psychol Rec

Posttests

SAM Figure 4 compares pleasure levels elicited by Black and White faces before and after training. Before training, White faces elicited significantly higher pleasure levels than Black faces, according to Mann-Whitney test (U′=1,873.5, p<.001). After training, average pleasure levels increased for Black faces and decreased for White faces. Visual comparison of the averages shows that ratings of White faces are still above ratings of Black faces, but the differ- ence is smaller, and not statistically significant (U′=1,547.0, p>.05).

AC3 Table 5 shows values of the bias index b (see Screening, in the BProcedure^ section). Positive values of b indicate negative bias toward Black faces, and negative values indicate positive bias. All participants showed b values of 4 or more (this was the criterion to be selected for the study) before the training. After training and equivalence class formation, b values were less than 4 for all participants, except P13 in the second test block. Eleven participants showed positive bias toward Black faces (negative values of b) in at least one of the AC3 posttests.

IRAP Latencies on the IRAP were transformed into D-IRAP scores for each of the four stimulus relations assessed. The

Fig. 3 Examples of equivalence test trials. The upper row presents standard AC (left) and CA (right) trials. The lower row presents AC3 trials

Table 3 Participants’ performance on training phases Participant Pretraining AB BC ABBC Mixed Training Baseline Review

P1* 1 (15) 2 (16-16) 2 (15-16) 2 (16-16) 2 (16-16)

P2 2 (13-14) 2 (16-16) 2 (16-16) 2 (16-16) 2 (16-16)

P3 1 (15) 2 (16-16) 2 (15-16) 2 (16-15) 2 (16-16)

P4* 2(13-15) 2(16-16) 3(14-16-16) 2(15-15) 2(16-16)

P5* 2(10-15) 2(15-16) 2(16-16) 2(16-16) 2(16-16)

P6 2(10-15) 2(16-16) 3(13-16-15) 2(16-16) 2(16-16)

P7* 1 (14) 2 (15-16) 3 (8-15-15) 2 (16-16) 2 (16-16)

P8* 1 (15) 2 (15-16) 3 (14-16-16) 2 (15-16) 2 (16-16)

P9 2 (4-14) 2 (15-15) 3 (12-16-15) 2 (16-15) 2 (15-15)

P10* 1 (15) 2 (15-16) 3 (13-16-16) 2 (16-16) 4 (16-13-16-16)

P11* 5(8-8-9-8-15) 2(15-16) 2(16-16) 2(16-16) 2(16-16)

P12 1 (14) 2 (16-16) 2 (15-16) 2 (16-16) 2 (16-16)

P13 3(10-14-15) 2(15-16) 5(14-15-14-16-16) 2(16-16) 2(16-16)

Note. The first value of each cell represents the number of blocks required to reach the learning criterion, whereas the value in parenthesis denotes the number of correct responses obtained in each block. Data in italics were obtained by the White experimenter

* = SMTS

Psychol Rec

mean D-IRAP scores were .67 for Black–positive, -.99 for Black–negative, -.65 for White–positive, and -.99 for White–negative. Scores did not deviate significantly from ze- ro according to a series of one-sample t tests (p=.511 for White–positive, p = .343 for White–negative, p = .529 for Black–positive, and p=.342 for Black–negative). These re- sults confirm the absence of negative bias toward Black faces also evidenced by the SAM. Actually, the IRAP did not reveal positive or negative biases toward either Black or White faces. A Spearman Correlation test revealed strong correlation

between even and odd IRAP trials (r=.676, p<.01), indicating high internal consistency of the IRAP.

Discussion

This study used a variation of the Conflicting Relations Paradigm (training AB and BC matching so that poten- tially emergent AC and CA relations would contradict preexperimental relations). Specifically, children that tended

Table 4 Participants’ performance on symmetry and equivalence tests

Participant BA Symmetry CB Symmetry AC/CATests AC3 Test

AC1 CA1 AC2 CA2 1 2

P1* 1 (16) 1 (16) 16 16 16 16 16 16

P2 1 (16) 1 (16) 16 15 16 15 16 16

P3 1 (15) 1 (16) 16 16 16 16 16 16

P4* 1(15) 1(16) 16 16 16 16 16 16

P5* 1(16) 3(8-11-16) 16 16 16 16 16 16

P6 1(16) 1(16) 14 16 16 16 16 16

P7* 1 (16) 1 (15) 16 15 16 16 15 16

P8* 1 (16) 1 (16) 16 16 16 16 16 15

P9 1 (16) 1 (16) 15 16 16 16 15 15

P10* 1 (16) 1 (16) 16 15 16 16 12 14

P11* 1(16) 1(15) 16 16 16 16 14 11

P12 1 (16) 1 (16) 16 16 16 16 13 12

P13 2(14-16) 1(16) 16 16 16 16 5 10

Note. The first value of each cell represents the number of blocks required to reach the learning criterion, whereas the value in parenthesis denotes the number of correct responses obtained in each block. Data in italics were obtained by the White experimenter

* = SMTS

Fig. 4 Mean pleasure levels for Black and White faces, pre- and postclass formation

Psychol Rec

to relate faces of Black individuals to negative symbols in pretests were taught matching relations that would potentially lead to emergent relations between Black faces and positive symbols. All 13 children responded according to these emergent relations in standard equivalence posttests. When posttests were modified to include White faces as a third comparison stimulus, nine of 13 children continued to select Black faces when the samples were positive symbols (and vice versa).

To our knowledge, the only previous study applying the Conflicting Relations Paradigm to racial biases was reported by de Carvalho and de Rose (2014). This earlier study repli- cated typical results with this paradigm: Only one of four children showed emergent relations between Black faces and positive symbols. The reason why most children failed to respond according to equivalence in that study may be the interference of previous relations that conflicted with poten- tially emergent relations. However, de Carvalho and de Rose also raised the possibility that training parameters were less than optimal for equivalence class formation. Particularly, as had been done in most earlier studies with the Conflicting Relations Paradigm (Leslie et al., 1993; Moxon et al., 1993; Watt et al., 1991), de Carvalho and de Rose tested only tran- sitivity (AC) and combined symmetry and transitivity (CA), without testing symmetry relations. Therefore, it is possible that absence of symmetrical relations (BA and CB) could account for failure to respond according to equivalence. A few other procedural features in the de Carvalho and de Rose study may have also contributed to the test outcomes. There was an abrupt withdrawal of differential consequences for responding in the equivalence tests. In addition, these tests

added White faces, which were not present during training. Perhaps children would respond consistently with equivalence if tests included only stimuli that appeared in training (as the standard AC and CA tests of the present study). One possible interpretation for the results of the present study, therefore, is that the probability of class formation was increased by the use of a more effective training protocol.

There is another possibility to be considered (which does not necessarily exclude the first one). Results of Fields and colleagues (e.g., Doran & Fields, 2012; Fields, Landon- Jimenez, Buffington, & Adams, 1995; Moss-Lourenco & Fields, 2011) showed that members of equivalence classes may vary in their degree of relatedness as a function of nodal distance. A node is any stimulus that is related in training to two or more other stimuli. This has been confirmed by results of Bortoloti and de Rose (2009), who had established equiv- alence classes comprising abstract stimuli and faces express- ing emotions, and showed that Semantic Differential ratings of the abstract stimuli were similar to the equivalent faces, and this similarity decreased as a function of nodal distance. Bortoloti and colleagues also showed that relatedness may be a function of other training parameters, such as the use of simultaneous or delayed matching in training, direc- tionality of training, overtraining, and the nature of the stimuli used (Bortoloti & de Rose, 2011a, 2011b, 2012; Bortoloti, Rodrigues, Cortez, Pimentel, & de Rose, 2013; see also Silveira et al., 2015). Because several parameters may influence the degree of relatedness, the outcomes of the Conflicting Relations Paradigm may depend on how much the training protocol is effective to maximize the degree of relatedness of prospective class members. In the present study, the choice of parameters may have increased the degree of relatedness between Black faces and positive symbols to the extent of overriding preexistent incompati- ble relations. This increased relatedness of equivalent stim- uli may be the reason why SAM evaluations of the Black faces were higher in the posttest than in the pretest. The higher ratings in the SAM are consistent with stronger transfer of positive functions from the positive symbols to the Black faces, compared with the study by de Carvalho and de Rose (2014).

Results consistent with this interpretation were obtained in a study that experimentally established conflicting relations (de Almeida & de Rose, 2015). An initial SMTS training produced three classes of equivalent stimuli, each containing three abstract pictures and a face expressing emotions of happiness, anger, and neutrality, respectively. Semantic Differential ratings then showed a Btransfer of meaning^ from the faces to the abstract stimuli equivalent to them. Participants, at that point, received additional training to shift class-membership of one of the arbitrary stimuli in each class, and this shift was confirmed by equivalence tests. For instance, stimulus D1, initially equivalent to the

Table 5 Bias index b obtained via AC3 test on pre- and posttraining

Participant Phase

Pretraining Posttraining 1 Posttraining 2

P1* 5 -8 -8

P2 14 -8 -8

P3 5 -8 -8

P4* 9 -8 -8

P5* 4 -8 -8

P6 4 -8 -8

P7* 7 -6 -8

P8* 8 -8 -7

P9 11 -7 -7

P10* 4 0 -4

P11* 6 -4 2

P12 6 -2 0

P13 16 1 4

Note. Data in italics were obtained by the White experimenter

* = SMTS

Psychol Rec

happy face, became equivalent to the angry face. This training to shift class membership was conducted with SMTS for one group and DMTS for another group. After the shift in class membership, Semantic Differential ratings of D1 for the DMTS trained group showed a shift from positive to negative evaluations. SMTS training pro- duced a less pronounced shift, from positive to neutral. The authors attributed these differences to the stronger relations generated by DMTS training, documented in earlier studies (Bortoloti & de Rose, 2009, 2012).

Regarding the present study, we may presume that relations between Black faces and negative attributes have been established with some strength in the participants’ preexperimental histories. We may question, consequently, whether a short training in a highly contrived experimental setting could generate relations strong enough to contradict the preexperimental ones. As a result, we may consider yet another alternative interpretation for the apparent success of this training to generate responding consistent with the intended equivalence classes. It is possible that features of the experimental setting functioned as contextual stimuli con- trolling class membership (Bush, Sidman, & de Rose, 1989). If this is the case, then we might speculate that responding according to the intended equivalence classes would occur only in the experimental setting, and children would continue to respond consistently with their preexperimental history out- side the laboratory.

Based on SAM evaluations, however, we may hypothesize that, to the extent that this instrument has some predictive validity outside the context in which they are applied, changes in evaluation extended beyond the experimental context. SAM evaluations before training showed a significant differ- ence in favor of the White faces. After training, this difference decreased and was no longer statistically significant. The IRAP conducted after training also did not reveal significant anti-Black biases. This is consistent with the SAM results showing reduction of anti-Black biases. However, because the IRAP used in this study did not reveal any bias, the pos- sibility that this IRAP was just insensitive cannot be ruled out. In future investigations, it is important to use the IRAP, or any similar instrument, both in pre- and posttests.

Despite the fact that training could change participants’ racial biases, we should not be very optimistic about any last- ing effects, because children would continue to live in the same environment that established the racial biases. Future studies should, thus, verify maintenance of training outcomes in follow-up assessments and assess generalization outside the experimental context.

Based on the results of de Almeida and de Rose (2015), we should expect that DMTS would be more effective than SMTS to produce responding consistent with equivalence classes. This prediction was not confirmed in the present study, possibly due to a ceiling effect, because all participants

attained criterion in the standard AC and CA tests, and most of them attained criterion in the AC3 test.

The attempt to verify experimenter’s influence was incon- clusive. The proportion of participants that failed the AC3 test was higher with the White experimenter (40 %) than with the Black experimenter (25 %). However, the number of partici- pants was small, and the participant with the highest bias (P13) was randomly assigned to the White experimenter. Several children in the present study were themselves Black, and from a relatively low socio-economic status. They may have considered the experimenter to be in a position of pres- tige and power. Meeting a Black person in such a position may have influenced their evaluations. On the other hand, these same children showed a negative bias toward Black faces in the prescreening tests, with the same Black experimenter with whom they worked throughout the experiment. Data from the present study do not, consequently, provide conclusive data about this question, which should be addressed in future research.

Results of several studies support the conclusion that pos- itive evaluations of meaningful stimuli may transfer to abstract stimuli equivalent to them (e.g., Arntzen et al., 2016; Barnes- Holmes et al., 2000; Bortoloti & de Rose, 2009, 2012; Smeets & Barnes-Holmes, 2003; Straatman, Almeida, & de Rose, 2014). The present study extends these findings to a situation involving conflicting relations: Positive evaluations were transferred to stimuli that were negatively evaluated before the experiment, and SAM results indicated an increase in the valence of Black faces after they became equivalent to posi- tive symbols.

It is tempting to speculate that the procedures of the present experiment may be used in training programs to fight racism. Guerin (2005) observed, however, that racial discrimination comprises many different behaviors, possibly unrelated to one another, and interventions that may be effective to change some of these behaviors (such as, for instance, racial jokes) may not be effective to change others (e.g. harassment). Viewing all these various behaviors as instances of something called Bracism^ is an essentialist position, which distracts from finding the determinants of the actual behaviors and consequently intervening on them. Regarding the present ex- periment, the behaviors addressed were evaluations in a re- stricted context: Black faces received lower evaluations than White faces in the pleasure dimension of an instrument (SAM) and were less likely to be related to a positive symbol and/or more likely to be related to a negative symbol. Several of these children were themselves Black. They have likely lived in a social environment in which being Black is less valued than being White. The procedures of this study, in this sense, may have some promise to be used as elements of behavioral in- terventions to change these evaluations in the future. Levin et al. (2016), for instance, found that three variables, psycholog- ical inflexibility, empathic concern, and perspective taking

Psychol Rec

were all independent predictors of generalized prejudice, which can be defined as negative attitudes toward various groups of people. Because both Black and White children demonstrated negative racial biases on pretests, future studies could use those variables in an intervention, aimed at lessen- ing this negative bias.

Additional studies are necessary to replicate the present results, and to extend them, using other types of instruments and measures of racial bias. It is also important to conduct studies with a larger number of participants and to assess maintenance and generalization of the effects. Further re- search should also explore additional variations in training parameters. It would also be worthwhile to isolate the para- metric changes introduced in the present study (symmetry tests, simple-to-complex protocol, gradual withdrawal of feed- back before tests) to identify the possible contribution of each of them to the effectiveness of training. A replication using faces of children rather than adults (or together with adults) as training stimuli could also be carried out. Considering that adults with racial biases have presumably a longer history of reinforcement for prejudiced behavior, it would be worthwhile as well to replicate this research with adult participants. It is conceivable that racial biases of children will prove easier to change, suggesting the importance of early interventions to change racial biases, and the possible effects of the longer reinforcement history of adults on their performances in this racial version of the Conflicting Relations Paradigm.

Acknowledgments We thank Deisy de Souza, leader of INCT-ECCE, for her encouragement and support for this research.

Compliance with Ethical Standards

Funding This study was funded by the Brazilian National Research Council (CNPq, Grant 573972/2008-7) and the São Paulo Research Foundation (FAPESP, Grant 2008/57705-8).

Conflict of Interest The authors declare they have no conflict of interest.

Ethical Approval This research was approved by the institutional Ethical Review Board, in accordance with Brazilian guidelines for ethical conduct in research with human participants.

References

Adams, B. J., Fields, L., & Verhave, T. (1993). Effects of test order on intersubject variability during equivalence class formation. The Psychological Record, 43, 133–152.

Amd, M., Barnes-Holmes, D., & Ivanoff, J. (2013). A derived transfer of eliciting emotional functions using differences among electroenceph- alograms as a dependent measure. Journal of the Experimental Analysis of Behavior, 99, 318–334. doi:10.1002/jeab.19.

Arntzen, E. (2006). Delayed matching to sample: probability of responding in accord with equivalence as a function of different delays. The Psychological Record, 56, 135–167.

Arntzen, E., Fagerstrom, A., & Foxall, G. R. (2016). Equivalence classes and preferences in consumer choice. In G. R. Foxall (Ed.), The Routledge companion to consumer behavior analysis (pp. 65–77). London: Routledge.

Barnes-Holmes, D., Barnes-Holmes, Y., Power, P., Hayden, E., Milne, R., & Stewart, I. (2006). Do you really know what you believe? Developing the Implicit Relational Assessment Procedure (IRAP) as a direct measure of implicit beliefs. The Irish Psychologist, 32, 169–177.

Barnes-Holmes, D., Keane, J., Barnes-Holmes, Y., & Smeets, P. M. (2000). A derived transfer of emotive functions as a means of estab- lishing differential preferences for soft drinks. The Psychological Record, 50, 493–511.

Barnes-Holmes, D., Murphy, A., & Barnes-Holmes, Y. (2010). The im- plicit relational assessment procedure: exploring the impact of pri- vate versus public contexts and the response latency criterion on pro- white and anti-black stereotyping among white Irish individuals. The Psychological Record, 60, 57–66.

Bortoloti, R., & de Rose, J. C. (2009). Assessment of the relatedness of equivalent stimuli through a semantic differential. The Psychological Record, 59, 563–590.

Bortoloti, R., & de Rose, J. C. (2011a). Avaliação do efeito de dica semântica e da indução de significado entre estímulos abstratos equivalentes [Evaluation of semantic priming effect and induction of meaning among abstract equivalent stimuli]. Psicologia: Reflexão e Crítica, 24, 381–393.

Bortoloti, R., & de Rose, J. C. (2011b). An BOrwellian^ account of stim- ulus equivalence: are some stimuli Bmore equivalent^ than others? European Journal of Behavior Analysis, 12, 121–134.

Bortoloti, R., & de Rose, J. C. (2012). Equivalent stimuli are more strongly related after training with delayed matching than after simultaneous matching: a study using the Implicit Relational Assessment Procedure (IRAP). The Psychological Record, 62, 41–54.

Bortoloti, R., Rodrigues, N. C., Cortez, M. D., Pimentel, N., & de Rose, J. C. (2013). Overtraining increases the strength of equivalence rela- tions. Psychology & Neuroscience, 6, 357–364. doi:10.3922/j.psns. 2013.3.13.

Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: the Self- Assessment Manikin and the semantic differential. Journal of Behavior Therapy and Experimental Psychiatry, 25, 49–59.

Bush, K. M., Sidman, M., & de Rose, T. (1989). Contextual control of emergent equivalence relations. Journal of the Experimental Analysis of Behavior, 51, 29–45. doi:10.1901/jeab.1989.51-29.

Carr, D., Wilkinson, K. M., Blackman, D., & McIlvane, W. J. (2000). Equivalence classes in individuals with minimal verbal repertoires. Journal of the Experimental Analysis of Behavior, 74, 101–114. doi: 10.1901/jeab.2000.74-101.

de Almeida, J. H., & de Rose, J. C. (2015). Changing the meaningfulness of abstract stimuli by the reorganization of equivalence classes: ef- fects of delayed matching. The Psychological Record, 65, 451–461. doi:10.1007/s40732-015-0120-9.

de Carvalho, M. P., & de Rose, J. C. (2014). Understanding racial attitudes through the stimulus equivalence paradigm. The Psychological Record, 64, 527–536. doi:10.1007/s40732-014-0049-4.

de Rose, J. C., McIlvane, W. J., Dube, W. V., Galpin, V. C., & Stoddard, L. T. (1988). Emergent simple discrimination established by indirect relation to differential consequences. Journal of the Experimental Analysis of Behavior, 50, 1–20. doi:10.1901/jeab.1988.50-1.

Devany, J. M., Hayes, S. C., & Nelson, R. O. (1986). Equivalence class formation in language-able and language-disabled children. Journal of the Experimental Analysis of Behavior, 46, 243–257. doi:10. 1901/jeab.1986.46-243.

Psychol Rec

Dixon, M. R., Rehfeldt, R. A., Zlomke, K. R., & Robinson, A. (2006). Exploring the development and dismantling of equivalence classes involving terrorist stimuli. The Psychological Record, 56, 83–103.

Doran, E., & Fields, L. (2012). All stimuli are equal, but some are more equal than others: measuring relational preferences within an equiv- alence class. Journal of the Experimental Analysis of Behavior, 98, 243–256. doi:10.1901/jeab.2012.98-243.

Dougher, M. J., Augustson, E., Markham, M. R., Greenway, D. E., & Wulfert, E. (1994). The transfer of respondent eliciting and extinc- tion functions through stimulus equivalence classes. Journal of the Experimental Analysis of Behavior, 62, 331–351. doi:10.1901/jeab. 1994.62-331.

Drake, C. E., Kellum, K. K., Wilson, K. G., Luoma, J. G., Weinstein, H. H., & Adams, C. H. (2010). Examining the implicit relational as- sessment procedure: four preliminary studies. The Psychological Record, 60, 81–86.

Dube, W. V., & Hiris, J. (1997). Match to Sample Program (Version 11.1.3) [Computer software]. Waltham: E. K. Shriver Center for Mental Retardation.

Dymond, S., & Barnes, D. (1994). A transfer of self-discrimination re- sponse functions through equivalence relations. Journal of the Experimental Analysis of Behavior, 62, 251–267. doi:10.1901/ jeab.1994.62-251.

Fields, L., Landon-Jimenez, D. V., Buffington, D. M., & Adams, B. J. (1995). Maintained nodal distance effects after equivalence class formation. Journal of the Experimental Analysis of Behavior, 64, 129–145. doi:10.1901/jeab.1995.64-129.

Guerin, B. (2005). Combating everyday racial discrimination without assuming racists or racism: new intervention ideas from a contextual analysis. Behavior and Social Issues, 14, 46–70.

Haydu, V. B., Camargo, J., & Bayer, H. (2015). Effects of preexperimental history on the formation of stimulus equivalence classes: a study with supporters of Brazilian soccer clubs. Psychology & Neuroscience, 8, 385–396. doi:10.1037/h0101276.

Hayes, S. C., Kohlenberg, B. S., & Hayes, L. J. (1991). The transfer of specific and general consequential functions through simple and conditional equivalence relations. Journal of the Experimental Analysis of Behavior, 56, 119–137. doi:10.1901/jeab.1991.56-119.

Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1999). International Affective Picture System (IAPS): technical manual and affective ratings. Gainesville: University of Florida, Center for Research in Psychophysiology.

Leslie, J. C., Tierney, K. J., Robinson, C. P., Keenan, M., Watt, A., & Barnes, D. (1993). Differences between clinically anxious and non- anxious subjects in a stimulus equivalence training task involving threat words. The Psychological Record, 43, 153–161.

Levin, M. E., Luoma, J. B., Vilardaga, R., Lilis, J., Nobles, R., & Hayes, S. C. (2016). Examining the role of psychological inflexibility, perspec- tive taking, and empathic concern in generalized prejudice. Journal of Applied Social Psychology, 46, 180–191. doi:10.1111/jasp.12355.

McManis, M. H., Bradley, M. M., Berg, W. K., Cuthbert, B. N., & Lang, P. J. (2001). Emotional reactions in children: verbal, physiological, and behavioral responses to affective pictures. Psychophysiology, 38, 222–231.

Moss-Lourenco, P., & Fields, L. (2011). Nodal structure and stimulus relatedness in equivalence classes: post-class formation preference tests. Journal of the Experimental Analysis of Behavior, 95, 343– 368. doi:10.1901/jeab.2011.95-343.

Moxon, P., Keenan, M., & Hine, L. (1993). Gender-role stereotyping and stimulus equivalence. The Psychological Record, 43, 381–394.

Rabelo, L. Z., Bortoloti, R., & Souza, D. H. (2014). Dolls are for girls and not for boys: evaluating the appropriateness of the Implicit Relational Assessment Procedure for school-age children. The Psychological Record, 64(1), 71–77. doi:10.1007/s40732-014- 0006-2.

Sidman, M. (1986). Functional analysis of emergent verbal classes. In T. Thompson & M. Zeiler (Eds.), Analysis and integration of behav- ioral units (pp. 213–245). Hillsdale: Erlbaum.

Sidman, M. (1994). Equivalence relations and behavior: a research story. Boston: Authors Cooperative.

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. doi:10.1901/ jeab.1982.37-5.

Silveira, M. V., Aggio, N. M., Cortez, M. D., Bortoloti, R., Rico, V. V., & de Rose, J. C. (2015). Maintenance of equivalence classes and trans- fer of functions: the role of the nature of stimuli. The Psychological Record. doi:10.1007/s40732-015-0152-1.

Smeets, P. M., & Barnes-Holmes, D. (2003). Children’s emergent prefer- ences for soft drinks: stimulus equivalence and transfer. Journal of Economic Psychology, 24, 603–618.

Straatman, G., Almeida, S. S., & de Rose, J. C. (2014). Computerized assessment of food preferences in adolescents undergoing the stim- ulus equivalence paradigm. Temas em Psicologia, 22, 613–624. doi: 10.9788/TP2014.3-07.

Wallace, B. W. (2003). Match to Sample Program III [Computer software]. Worcester: UMass/Eunice Kennedy Shriver Center's Behavioral Sciences Department.

Watt, A., Keenan, M., Barnes, D., & Cairns, E. (1991). Social categorization and stimulus equivalence. The Psychological Record, 41, 33–50.

Weatherly, J. N., Miller, K., & McDonald, T. W. (1999). Social influence as stimulus control. Behavior and Social Issues, 9, 25–45.

Wulfert, E., & Hayes, S. C. (1988). Transfer of a conditional ordering response through conditional equivalence classes. Journal of the Experimental Analysis of Behavior, 50, 125–144. doi:10.1901/ jeab.1988.50-125.

Psychol Rec