Empirical Analysis of Derived Hierarchical Responding

Title

Empirical Analysis of Derived Hierarchical Responding

Full Reference

Villarroel, J., Luciano, C., & Ruiz, F. J. (2024). Empirical Analysis of Derived Hierarchical Responding. International Journal of Psychology and Psychological Therapy, 24(2), 193–219.

Study Type

Laboratory experimental study on derived hierarchical relational responding, framed within Relational Frame Theory (RFT)

---

Context and Objectives

Hierarchical responding is a type of relational response in which classes of stimuli are grouped into broader categories, which can in turn be grouped into even broader categories. From Relational Frame Theory (RFT; Hayes, Barnes-Holmes, & Roche, 2001), a hierarchical network is conceptualized as a derived relational response in which specific stimuli are learned to be related through relational cues (is part of, belongs to, etc.) via multiple exemplars with non-arbitrary functions among stimuli, which are subsequently applied arbitrarily to relate stimuli.

Previous studies had explored derived hierarchical responding using various training strategies. For example, Gil et al. (2012, 2014) trained networks with different levels and branches, and Slattery and Stewart (2014) used simpler networks. However, several aspects remained unexplored: bottom-up and top-down transformation of functions had not been demonstrated jointly, the most inclusive function had not been experimentally isolated, and the networks established in different studies varied considerably.

The present experiment aimed to: (a) train a hierarchical network in which different branches were related through increasingly inclusive functions, (b) assign different functions to different stimuli across the entire network to test for derived responses, and (c) verify that the most inclusive function could be identified within the network. To this end, an intra-subject design was used with four different Multiple-Exemplar Training (MET) protocols, designed to train four arbitrary networks: one analogous to a hierarchical network (based on Inclusion-Distinction) and three control networks with variations (Inclusion-Sameness, Distinction, and Sameness).

---

Method

Participants

Nine undergraduate psychology students (6 females) participated in the experiment. Their ages ranged from 19 to 22 years (M = 19.66, SD = 1.11). They were recruited through an email announcement informing them of the opportunity to receive course credits. None had prior experience with RFT experiments or knew the purpose of the study.

Design

Intra-subject design with one training protocol comprising four different METs, each aimed at training one of four arbitrary networks. The Inclusion-Distinction network was analogous to a hierarchical network, while the other three served as controls with variations in the relational cues used.

Materials and Apparatus

The experiment was conducted in a 3x3 meter room equipped with a table, chair, an HP Pavilion x360 convertible laptop with a 14-inch monitor, and a one-way mirror for observation. A total of 105 stimuli were used in the initial phase, including figures, labels, words, letter pairs, nonsense syllables, numbers, symbols, and effects. The second phase employed 168 labels, nonsense syllables, words, and numbers, along with 26 shapes from the NOUN database, five colors, and seven GIFs. The third phase used 52 labels, 35 letter pairs, and 17 number pairs.

Procedure

The experiment comprised three phases:

Phase 1: Preparatory Phase. Familiarized participants with terminology and trial formats by having them respond to culturally established relations. It included two blocks of 20 trials each with five trial types (selecting the relational cue, selecting stimuli, selecting the inclusive function, selecting the network, and describing stimulus functions of networks). Participants had to demonstrate understanding of the relational symbols (=, ≠, <, >, {).

Phase 2: Relational Cues Training. Four MET protocols were used to train four relational cues as abstract figures: Sameness (Sa), Distinction (Di), Inclusion based on Sameness (IncSa), and Inclusion-Distinction (IncDi). This phase included three parts: (1) Establishing sameness and distinction functions for Sa, Di, IncSa, and IncDi through three training blocks (28, 28, and 22 trials) followed by a 12-trial test block; (2) Training inclusive functions for IncSa and IncDi through three blocks including familiarization, inclusive relation training, and drawing relations; (3) Stimulus substitution with nonsense syllables, comprising 8 substitution trials.

Phase 3: Arbitrary Networks Training and Testing. Divided into two parts: (1) Training and testing the inclusive networks (with upper and lower levels, training relations and functions, and testing derived functions), and (2) Training and testing the non-inclusive networks (Distinction and Sameness). Relations between stimuli in each network were trained and functions (numbers) were assigned to some stimuli. Derived relations and transformation of functions were then evaluated through "describing the stimuli" and "identifying the inclusive function" trials.

Analysis

Individual descriptive analysis of each participant's results. Drawings made by participants to represent relations, results of derived function trials, function inclusivity, and descriptions of relational cue meanings were evaluated.

---

Results

Preparatory Phase and Relational Cues Training. All participants met the mastery criterion for the preparatory phase, although three (P1, P5, and P9) required re-exposure to alternative trials. Regarding relational cues training, all participants completed all three parts. Specifically, eight of nine completed the three blocks of sameness and distinction function training, and all completed the stimulus substitution trials.

Arbitrary Networks: Drawings. Most participants followed a consistent differential response pattern for each network. In the Inclusion-Distinction (IncDi) network, six participants (P2, P3, P4, P5, P7, and P9) correctly represented the complete network using the inclusive symbol ({) for relations among levels and the distinction symbol (≠) between branches. In the Inclusion-Sameness (IncSa) network, eight participants responded accordingly using the sameness symbol (=) between branches. In the Distinction network, all used the distinction symbol. In the Sameness network, eight of nine used the sameness symbol.

Transformation of Functions (ToF). In the Inclusion based on Distinction network, six of nine participants (P2, P3, P4, P5, P7, and P9) correctly identified all functions provided at different levels and branches of the network, showing bottom-up transformation (functions converged to the top stimulus) and top-down transformation (the top stimulus function derived to all stimuli). Two participants (P6 and P8) failed the bottom-up transformation. One participant (P1) failed the test entirely.

In the Inclusion based on Sameness network, six of nine participants showed the same correct pattern of bottom-up and top-down transformation. Two participants (P6 and P8) showed a partial pattern. One participant (P1) failed the test.

Branch Differentiation. In the Inclusion based on Distinction network, eight of nine participants (P2, P3, P4, P5, P6, P7, and P8) described all stimuli at different levels of the network with different functions, showing differentiation between branches. In the Inclusion based on Sameness network, five participants showed complete sameness between branches, and three showed partial sameness.

Inclusivity. In the Inclusion based on Distinction network, six of nine participants identified the function provided to the top stimulus as the most inclusive. This did not occur in the Distinction or Sameness networks, where no participant identified a function as more inclusive—which was the expected result.

Control Networks. In the Distinction network, eight of nine participants described each stimulus as different from the rest. In the Sameness network, six participants described two stimuli with the same (trained) function while others were described differently, and seven of nine showed no bottom-up or top-down transformation, as expected.

Cue Descriptions. All participants described each relational cue differently and consistently with the functions trained through the MET.

---

Discussion and Conclusions

This study advanced the experimental analysis of hierarchical responding by showing, for the first time, all its core features together in an experimental preparation. Specifically, transformation of functions in hierarchical responding involves four features: (1) top-down transformation, where functions of a higher-level stimulus are derived to lower-level stimuli; (2) bottom-up transformation, where functions of lower-level stimuli are contextually transferred to higher-level stimuli; (3) branch differentiation, which prevents function transfer from one branch to another; and (4) a more inclusive or defining function than the rest, identified with the top-level stimulus of the network.

The protocol design was innovative in training four networks in the same participants, using four relational cues trained through MET (Sameness, Distinction, Inclusion-Sameness, and Inclusion-Distinction), so that derived functions could be evaluated in each network. Results showed that eight of nine participants successfully acquired all four networks, and of these eight, six showed a complete pattern of hierarchical responding (bottom-up, top-down, branch differentiation, and inclusivity) in the analog Inclusion-Distinction network, but not in the three control networks.

Among the limitations, the authors note that the procedure was lengthy and demanding, affecting some participants, especially those with more errors during debriefing. Additionally, participants had prior history with verbal and structural tasks, as the preparatory phase included a familiarization phase that could be improved for better coordination of procedures.

In conclusion, the study extends previous research on derived hierarchical relational responding by experimentally demonstrating, for the first time, the core features of hierarchical categorization as a network built with different branches and derived functions across different levels, along with identification of the most inclusive function that keeps network elements integrated. Future research could explore the nature of the non-arbitrary functions that lead to establishing hierarchical relational cues.

---

Importance and Contribution

This study represents a significant advance in research on derived relational responding, one of the most complex behavioral processes within Relational Frame Theory. The joint experimental demonstration of the four defining features of hierarchical responding (bottom-up and top-down transformation of functions, branch differentiation, and inclusivity) constitutes an original contribution to contextual behavioral science and the experimental analysis of complex verbal behavior. The design employing four different networks with the same participants provides robust experimental control that isolates the unique characteristics of hierarchical responding compared to other types of relational responding. These findings have implications both for the basic understanding of category learning and human cognition, and for the design of training procedures in applied contexts, including populations that have not yet developed these complex repertoires.

---

Verification Checklist

• Number of participants verified: 9
• Ages verified: range 19-22, M = 19.66, SD = 1.11
• Design verified: intra-subject, 4 arbitrary networks
• Relational cues verified: Sa, Di, IncSa, IncDi
• ToF results verified: 6/9 complete pattern in IncDi
• Control network results verified
• No data fabricated: all extracted from the PDF