EVOLUTION OF THE APPLICATION OF REPERTOIRE GRID TECHNIQUE IN EDUCATION

Patarakin E.D.

Moscow City Pedagogical University
Moscow, Russia
e-mail: patarakined@mgpu.ru

 

Introduction

The repertory grid technique was developed by George Kelly in the mid-20th century as a tool for studying the personal constructs underlying human perception and prediction of events and phenomena. From the very beginning, Kelly's work was of interest to education, particularly in areas where a person is viewed not only as an object of learning, but as an active agent—a researcher and builder of their own intellect [6]. This position brought Kelly's ideas closer to the approach of Seymour Papert, who also emphasized the role of the student's active co-creation and the need to provide them with tools for the independent formation of semantic structures [10,11]. G. Kelly formulated a fundamentally new understanding of a person as an active cognitive system, continuously constructing conceptual models for predicting and managing events. The theory of personal constructs is based on the idea of a person as a "scientist-researcher" who forms internal hypotheses about the world and tests them through the experience of interacting with reality. Kelly's fundamental postulate is that personality processes are guided by constructs that serve to predict events [1]. The philosophical foundation of Kelly's theory is the principle of constructivism, according to which reality can be interpreted in a variety of ways based on individual models of reality. In this paradigm, there are no correct or incorrect interpretations of reality—all hypotheses that allow for effective interaction with the environment are valid. The acceptability of a particular conceptual model is determined not by its degree of conformity with some "objective model," but by its practical heuristic significance [2].

The Repertory Grid Test, developed by Kelly, is an experimental method that incorporates specific techniques for planning and conducting research, processing, and interpreting results. A repertory grid is a specially organized matrix, where rows correspond to constructs (bipolar dimensions) and columns correspond to elements (objects of assessment). The construct identification procedure is based on the minimal context method: subjects are presented with triads of elements and asked to name the quality by which two elements are similar to each other and different from a third. The resulting answer constitutes one pole of the construct, and the opposite characteristic of the third element constitutes the other pole. The methodological value of repertory grid technology lies in its ability to identify cognitive structures that are often unconscious to the individual but manifest through a multitude of verbal labels. Subsequent statistical analysis of the grid allows for the construction of cluster trees of constructs, revealing their hierarchical organization, and establishing latent relationships between various dimensions of perception.

The key advantage of the repertory grid technique was its flexibility and versatility. Created by a mathematician, the method was easily adapted for computer implementations, which contributed to its early dissemination in software environments as early as the 1970s, and in the 1990s it became an integral part of educational computer initiatives and experimental platforms [7; 16]. Examples of the integration of repertory grids into computer-based learning for students and schoolchildren in Russia and Eastern Europe include works that used software that allows playing with results and visualizing constructs directly in the learning environment.

Further technological advances have enabled the use of repertory grids as social objects—tools that can be shared among participants in online educational communities, collaboratively shaping and refining a system of personal constructs. In this embodiment, the repertory grid becomes not just a diagnostic tool but a foundation for students' creative and research activities, which is especially important in today's digitalized education environment [4; 12; 14].

This paper updates the understanding of the diverse uses of repertory grids specifically in education. Of particular interest is the evolution of their application: from a diagnostic method to a comprehensive tool for collaborative analysis and design of educational trajectories, including in the study of new phenomena associated with the implementation of generative artificial intelligence. This approach allows us to consider repertory grids as a universal tool for supporting independent learning and research activities, embodying the principles of subjectivity, activity, and systems thinking in educational practices of the new century.

Materials and methods

The study utilized a comprehensive framework encompassing the stages of collecting, processing, and analyzing bibliometric data to map the scientific field of application of repertory grid techniques in education. The primary query for searching publications across all bibliometric databases was the term "Repertory Grid" to ensure a consistent and comparable sample. The objective was to study the evolution of repertory grid use in educational research using bibliometric mapping. This framework is presented in a PlantUML sequence diagram, which shows the sequential flow of data and the analysis process, separating the parallel stages of searching and visualization (Figure 1).

Fig. 1. Bibliometric analysis scheme

In parallel, searches were conducted in three bibliometric databases using the keyword "Repertory Grid": ACM Digital Library (ACM DL), Lens, and OpenAlex. Data was extracted from all databases in RIS format.

The next step was merging the resulting datasets in the R environment using the revtools package for managing large datasets. Data cleaning was performed, removing duplicates and publications with incomplete or erroneous information.

The data aggregation and processing process was as follows: We combined data collected from various sources.

ris_files <- c("acm.ris", "lens.ris", "openalex.ris")

combined_data <- read_refs(ris_files, return_df = TRUE)

As a result, we obtained a table of 2,731 records. From these records, we removed duplicates and records without DOIs.

unique_data <- deduplicate(combined_data, "title",

method = "string_osa",

rm_punctuation = TRUE,

to_lower = TRUE)

filtered_data <- unique_data[!is.na (unique_data$doi) & unique_data$doi != "", ]

At this stage, we received 1,900 records. Of these, only texts published in books and journals were filtered.

filtered_significant <- filtered_data %>%

filter (source_type %in% c("BOOK", "CHAP", "GEN", "JOUR"))

The Bibliometrix package was used for analysis and subsequent visualization, identifying key thematic areas, the frequency of use of specific keywords, and the distribution of authors. The imported data was loaded into VOSviewer, where research maps were visualized. Keyword data was combined using a thesaurus to eliminate synonyms and spelling variations and visualized as clusters. A similar approach was applied to author names to consolidate various spelling variations. The resulting visualizations were analyzed to identify key research areas and groups.

The second stage of the study involved collectively completing a matrix of personal constructs in the WebGrid online platform (Rep Plus V2.0). In the first experiment, students were asked to simultaneously evaluate eight information resources: Google Scholar, eLIBRARY.RU, Cyberleninka, Wikipedia, ChatGPT, Yandex.GPT, ResearchGate, and ConsultantPlus, using dichotomies reflecting key selection criteria. The students used the following polar construct scales: "Unreliability vs. Reliability," "Popularity vs. Scientificity," "Complexity vs. Convenience," "Limited vs. Volume of Information," "Static vs. Dialogue," "Slowness vs. Speed," "Paid vs. Free," and "Obsolete vs. Relevant." Each rating ranged from 1 to 5, with 1 corresponding to the first pole (e.g., "Unreliability") and 5 to the second pole ("Reliability"). The second experiment was similarly designed using the WebGrid platform (Rep Plus V2.0) to identify students' personality constructs when perceiving generative language models. Participants were asked to simultaneously evaluate nine elements—GPT-4, YandexGPT, GigaChat, Qwen, Llama, Bert, Mistral, DeepSeek, and Alisa—using eight dichotomies reflecting key characteristics of interaction with AI systems. The students used the following dichotomies as polar scales: "Slowness vs. Fast Response," "Opacity vs. Transparency," "Untrustworthy vs. Trustworthy," "Routine vs. Creativity," "Lack of Context vs. Context Understanding," "Specialization vs. Versatility," "Closed vs. Open," and "Simply Provides an Answer vs. Shows the Path to a Result." Each pair was assessed on a scale from 1 to 5, where 1 corresponded to the first pole («Slowness,» «Opaqueness of work,» etc.), and 5 to the second («Quick response,» «Transparency of work,» etc.).

Results

Based on the presented bibliometric data, the following characteristics of the research field can be derived. The analyzed literature corpus covers a significant time period from 1967 to 2025, which indicates the long-term development of this scientific field. The total sample size is 1,709 documents published in 1,115 different sources, including journals, books, and other types of publications. The annual growth rate of publication activity is 5.39%, which indicates the sustainable development of the research field. The average age of documents in the sample is 13.3 years, which characterizes the relatively modern state of the scientific knowledge base. A large scientific community of 4,026 authors with a total of 4,868 author contributions participates in the research. This indicates the active participation of many researchers in several publications. The authorship structure demonstrates the predominance of collective research: 351 authors published works individually, while the share of individual publications is 403 documents (about 24% of the total volume). The average collaboration rates are 0.424 documents per author and 2.85 co-authors per document, indicating a moderate level of scientific collaboration in this field.

The temporal dynamics of publication activity shows a gradual development: the initial period (1960-1980s) is characterized by low activity, significant growth is observed since the 1990s, the peak of publication activity occurs in 2010-2016 (maximum 114 publications in 2016), after which some decrease in research intensity is noted in recent years.

An analysis of the most productive authors in the field reveals key characteristics of the scientific community. Guillem Feixas leads the way in total publications (35 articles), demonstrating his central role in the development of this research area. David Winter (15 publications) and Gwo-Jen Hwang (12 publications) follow, suggesting the existence of a small group of highly productive researchers. An analysis of fractionalized publication activity indicators, which take into account each author's share of multi-author papers, reveals a slightly different picture: Guillem Feixas maintains the lead with a score of 9.90, but Steven Pike ranks second (6.83), and Daniel TL Shek (6.58) ranks third. Fractionalized indicators of author publication activity represent a significantly more accurate and fair tool for assessing scientific contribution in bibliometric studies compared to traditional counting. The basic idea behind fractionalized scoring is that each publication is assigned a "weight unit" that is distributed among all co-authors proportionally to their contribution. For example, if an article has three authors, then with equal distribution, each receives 1/3 = 0.33 of the "article share," rather than a full unit, as in traditional scoring. This allows for a more accurate reflection of each researcher's true contribution to scientific production. Fractionalized scoring allows for accurate comparisons of the productivity of researchers in different fields.

Data analysis using the Bibliometrix package in R revealed key characteristics of the scientific community researching repertory grid techniques.

The analysis results were then examined in detail and visualized in VOSviewer. Figure 2 shows clusters of primary authors, with a filter applied to the map to exclude authors with fewer than four publications.

Fig. 2. Clusters of scientific groups using repertory grids

 

We also analyzed the articles selected based on their citation counts, resulting in the following graph (Figure 3). In the figure, the earliest publications are highlighted in purple, and more recent ones in yellow. The graph appears uninformative in both the print and online versions, but readers can access its expanded and interactive web version at:

https://app.vosviewer.com/?json=https%3A%2F%2Fdrive.google.com%2Fuc%3Fid%3D1_Qg-RPnrDr6qD6gnk_T0sBA55D4SoGss

Fig. 3. Graph of the most cited publications on the topic of repertoire grids

 

Discussion

After identifying the most significant authors based on fractionalized indicators, we analyzed how their work could be used in education, particularly in the training of psychologists and psycholinguists.

Guillem Feixas develops the theory of personal constructs in the study of implicit dilemmas—internal conflicts within the cognitive structure of subjects that block personal development or progress in psychotherapy. His primary contribution lies in the development of methods for identifying such conflicts using repertory grids and the creation of protocols for psychotherapeutic work with them. These studies focus primarily on clinical psychology and psychotherapy [5] rather than on the educational applications of this technique.

Gwo-Jen Hwang's work uses repertory grids primarily to understand how students and teachers construct representations of educational technologies and learning environments.[17] Hwang embeds repertory grids in online and mobile applications that allow participants not only to manually record constructs but also to instantly assign names to "elements" (e.g., learning platform interfaces or virtual reality objects) and define their bipolar constructs through a user-friendly graphical interface. This reduces interview time and improves the accuracy of data collection. After participants rate each "element" on selected constructs on a five- or seven-point scale, Hwang combines the repertory grid data with system interaction logs: time spent on each element, navigation patterns, and test scores. This allows for the identification of which constructs actually influence learning effectiveness and the construction of models linking subjective representations with objective performance metrics. Hwang conducts repertory grid-based research for both novice and advanced users of educational systems, enabling comparisons of their constructs. For example, in one of his studies, novices identified "ease of use vs. complexity" as key constructs, while experienced users emphasized "functional flexibility vs. limitations." This helps designers adapt interfaces to different skill levels. Hwang's approach combines the digitalization of the repertory grid process, the integration of qualitative constructs with educational system parameters, and advanced data visualization to create a holistic view of the cognitive models of educational technology users.

Shaw and Gaines (Mildred L. G. Shaw, Brian R. Gaines) developed the repertory grid method in educational and research contexts, primarily through the creation of the RepGrid and WebGrid series of software tools. Interactive web versions (WebGrid V) and desktop applications (RepGrid Plus) allow instructors to quickly deploy student questionnaires, collect data, and visualize cognitive maps of classes or groups. The RepScript scripting language allows for customizing dialogs and exporting results to systems such as LMS or educational data analysis tools. The work of B. Gaines and M. Shaw demonstrates that computer tools that allow the identification and visualization of individual and group conceptual schemas are extremely useful not only for research purposes but also for clarifying and bridging the positions of members of professional and educational communities [6,9,8]. The WebGrid system they developed is a web server for knowledge mining and modeling that uses an extended repertory grid system for knowledge mining [16]. A key advantage of computer-aided analysis tools is their ability to process complex multidimensional data and present the results in a visually accessible form. This allows researchers to quickly extract consistent patterns from relatively small data sets [8].

In the context of contemporary educational technology research, repertory grids are particularly relevant for studying university students' perceptions of generative artificial intelligence. Using Kelly's methodological framework, we can identify polar constructs associated with students' perceptions of generative AI technologies. Interpretation of the results revealed that Google Scholar received the highest average rating for the "Trustworthiness" construct, while ChatGPT and Yandex.GPT led the "Speed" construct. The "Scholarship vs. Popularity" construct revealed an interesting distinction: traditional academic databases such as Google Scholar and ResearchGate were associated with rigor and depth of scholarly content, while Wikipedia and ChatGPT were perceived as more accessible and popular sources. The results are presented in Figure 4.

Fig. 4. Elements and constructs of the field of digital instruments

 

An analysis of aggregated scores across eight constructs revealed that GPT-4 stands out primarily in the "Speed" criterion, receiving an average score of 4 on the "Slowness - Fast Response" scale, suggesting it is perceived as the most responsive tool. However, on the "Opacity - Transparency" scale, GPT-4 scored only 1, indicating a lack of transparency in its algorithmic decisions. However, on the "Mandatory - Creativity" scale, GPT-4 received a high score (4), reflecting its creative potential. YandexGPT, despite relatively low scores on "Speed" (2), demonstrated a high level of "Transparency" (4), indicating a more user-friendly response generation mechanism. GigaChat received average scores across all constructs, not excelling on any criterion, suggesting a balanced but not exceptional user experience. The Qwen model demonstrated good context awareness (4 points on the Context Loss – Context Understanding scale) and high Transparency (4 points), which, together with moderate Creativity (3 points), makes it promising in situations requiring both creativity and context-sensitive approaches. Llama received high scores on Creativity (3 points) and Shows the way to the result (4 points), indicating its ability not only to generate original texts but also to explain the logic of its reasoning. Bert and Mistral, in contrast, were perceived as more specialized (3 and 4 points on the Specialization – Generality scale) and less creative models. DeepSeek distinguished itself with a high score on Shows the way to the result (4 points), although it received low scores on Speed and Transparency. This indicates that students value DeepSeek's ability to justify its answers, despite the slowness and closed nature of the process. Finally, Alice received the highest scores for "Speed" (4 points) and "Transparency of Operation" (4 points), but a low score for "Creativity" (0 points), reflecting its strength as a practical and understandable tool, but less flexible in generating ideas. Overall, the results of the second experiment demonstrate a variety of perceptions of generative language models across key constructs, reflecting a balance between speed, transparency, trust, creativity, and the ability to explain one's decisions. The results are presented in Figure 5.

Fig. 5. Elements and constructs of the field of generative models

 

Research in the field of generative AI shows that its impact on human creativity and brain activity is ambiguous. Repertory grid methodology allows for a detailed analysis of how students construct their understanding of interaction with AI, the criteria they use to evaluate the effectiveness of such interactions, and how these understandings influence their learning. In previous studies, we considered the mental maps and ontologies created by students using repertory grids as important social objects of collaborative activity [13; 3]. These objects, like digital stories in Scratch or models in the NetLogo environment, can be freely exchanged between community participants, becoming a means of collective knowledge construction and reflection on the learning process. Every action a subject takes on a digital object—whether creation, editing, or discussion—forms a connection between the subject and the object, transforming the latter into a social object around which collaboration is built. Repertory grids created by one participant can be filled in and supplemented by others, as well as compared and combined, facilitating the coordination and development of collective conceptual frameworks. Thus, mental maps and ontologies constructed using repertory grids are not simply individual cognitive structures but become the subject of exchange, discussion, and collaborative refinement in online educational communities. This enables the creation of dynamic, shared knowledge models that support collaborative learning, creativity, and reflection.

A collaborative construct mining framework using WebGrid has been shown to be effective in formalizing and visualizing students' collective perceptions of the comparative characteristics of traditional academic resources and generative AI systems, confirming the suitability of repertory grids for studying the perception of modern educational technologies.

 

 

 

Literature

1. Kelly J. Psychology of personality. Theory of personal constructs. St. Petersburg: Rech, 2000. 249 p.
2. Leontiev D.A. A Misunderstood Classic (on the 100th Anniversary of the Birth of George Kelly (1905–1967)) // Psychological Journal. 2005. Vol. 26,
   No. 6. P. 111–117.
3. Patarakin E.D. Studying and Modeling Collaborative Activities in Online Communities. Moscow: Moscow City Pedagogical University, 2024. 140 p.
4. Patarakin E.D. Pedagogical design of joint network activities of educational entities. Moscow: Moscow City Pedagogical University, 2017. 319 p.
5. Feixas G., Geldschläger H., Neimeyer R.A. Content analysis of personal constructs // Journal of Constructivist Psychology. United Kingdom: Taylor & Francis, 2002. Vol. 15, No. 1. P. 1–19.
6. Gaines BR. Humans as Scientists: Scientists as Humans // Journal of Constructivist Psychology. T. 26, No. 3. P. 210–217.
7. Gaines BR. Knowledge acquisition: Past, present and future // International Journal of Human-Computer Studies. Elsevier, 2013. T. 71,
   No. 2. P.135–156.
8. Gaines BR, Shaw MLG. Concept maps as hypermedia components // International Journal of Human-Computer Studies. T. 43, No. 3. P. 323–361.
9. Gaines BR, Shaw MLG. New directions in the analysis and interactive elicitation of personal construct systems // International Journal of Man-Machine Studies. T. 13, No. 1. P. 81–116.
10. Papert S. Mindstorms: children, computers, and powerful ideas. Da Capo Press, 1993. 258 p.
11. Papert S. What's the big idea? Toward a pedagogy of idea power // IBM Systems Journal. T. 39, No. 3.4. P. 720–729.
12. Parandekar SD, Patarakin E., Yayla G. Children Learning to Code: Essential for 21st Century Human Capital. Washington, DC: World Bank,
    P.151.
13. Patarakin E., Burov V., Yarmakhov B. Computational Pedagogy: Thinking, Participation, Reflection // Digital Turn in Schools—Research, Policy, Practice / ed. Väljataga T., Laanpere M. Singapore: Springer, 2019. P. 123–137.
14. Patarakin ED. Wikigrams-Based Social Inquiry // DIGITAL TOOLS AND SOLUTIONS FOR INQUIRY-BASED STEM LEARNING. P. 112–138.
15. Patarakin ED, Travina LL. Knowledge acquisition and Manipulation with the constructs on the Psychological Web-site // The Second International Conference on Distance Education in Russia. P. 2–5.
16. Shaw ML, Gaines BR. WebGrid: Knowledge Elicitation and Modeling on // Industrial Knowledge Management: A Micro-level Approach. T. 335 p.
17. Sung H.-Y., Hwang G.-J. A collaborative game-based learning approach to improving students' learning performance in science courses // Computers & Education. T. 63. P. 43–51.
18. Travina LL, Patarakin ED. Kelly's RG Test in Logo Media — Play with Results // New media and telematic technologies for education in eastern European countries / Ed. By Kommers PAM et al. 1997, pp. 335–339.

 

References

1. Kelly, J. (2000). Theory of personality. Psychology of personal constructs. Saint Petersburg: Rech'. 249
2. Leont'ev, DA. (2005). Misunderstood classic (on the 100th anniversary of the birth of George Kelly (1905-1967)). Psychological journal, 26(6), 111-117.
3. Patarakin, ED. (2024). Studying and modeling joint activities in online communities. Moscow: Moscow City Pedagogical University. 140
4. Patarakin, ED. (2017). Pedagogical design of joint network activities of educational entities. Moscow: Moscow City Pedagogical University. 319 p.
5. Feixas, G., Geldschläger, H., & Neimeyer, R.A. (2002). Content analysis of personal constructs. Journal of Constructivist Psychology, 15(1), 1-19.
6. Gaines, BR. (2013). Humans as Scientists: Scientists as Humans. Journal of Constructivist Psychology, 26(3), 210-217.
7. Gaines, BR. (2013). Knowledge acquisition: Past, present and future. International Journal of Human-Computer Studies, 71(2), 135-156.
8. Gaines, BR, & Shaw, MLG. (1995). Concept maps as hypermedia components. International Journal of Human-Computer Studies, 43(3), 323-361.
9. Gaines, BR, & Shaw, MLG. (1980). New directions in the analysis and interactive elicitation of personal construct systems. International Journal of Man-Machine Studies, 13(1), 81-116.
10. Papert, S. (1993). Mindstorms: children, computers, and powerful ideas. Da Capo Press. 258 p.
11. Papert, S. (2000). What's the big idea? Toward a pedagogy of idea power. IBM Systems Journal, 39(3.4), 720-729.
12. Parandekar, S. D., Patarakin, E., & Yayla, G. (2019). Children Learning to Code: Essential for 21st Century Human Capital. Washington, DC: World
    Bank, 151 p.
13. Patarakin, E., Burov, V., & Yarmakhov, B. (2019). Computational Pedagogy: Thinking, Participation, Reflection. In T. Väljataga & M. Laanpere (Eds.), Digital Turn in Schools—Research, Policy, Practice (pp. 123–137). Singapore: Springer.
14. Patarakin, ED. (2017). Wikigrams-Based Social Inquiry. In Digital Tools and Solutions for Inquiry-Based STEM Learning. pp. 112-138.
15. Patarakin, ED, & Travina, LL. (1996). Knowledge acquisition and Manipulation with the constructs on the Psychological Web-site. The Second International Conference on Distance Education in Russia, pp. 2-5.
16. Shaw, ML, & Gaines, BR. (2012). WebGrid: Knowledge Elicitation and Modeling on. Industrial Knowledge Management: A Micro-level Approach, 335 p.
17. Sung, H.-Y., & Hwang, G.-J. (2013). A collaborative game-based learning approach to improving students' learning performance in science courses. Computers & Education, 63, 43-51.
18. Travina, LL, & Patarakin, ED. (1997). Kelly's RG Test in Logo Media - Play with Results. In PAM Kommers et al. (Eds.), New media and telematic technologies for education in eastern European countries, pp. 335-339.

 

 

Date of receipt of the article by the editors: 09/04/2025.

Date of acceptance of the article for publication: 09.27.2025.