Epistemic network analysis to study an unplugged model-eliciting activity for computational thinking with high school students in Mexico Open Access

Mexico has emerged as one of the prominent adopters of technology globally, with 96.8 million Internet users by the end of 2022, growing at 9.3% annually. People portray Mexico as a technologically inclined nation, but challenges and inequalities remain. Education policymakers are encouraged to focus on education strategies for using and creating technology (International Trade Administration, 2023). The educational landscape in Mexico is shaped by pronounced economic disparities, exacerbated by the overarching challenges of underdevelopment. Despite the efforts in education reforms, the country continues to lag in growth, inclusion, and poverty reduction compared to peers. Mexico requires a tailored approach that considers the country’s diversity and focuses on providing students with opportunities to develop skills for the digital age, including Computational Thinking (CT) (Brida, Risso, Sánchez Carrera, & Segarra, 2021; Campos-Vazquez & Medina-Cortina, 2019).

This paper examines how two groups of high school students from a school in Mexico City elicit Computational Thinking (CT), a set of process dispositions and skills a person needs to solve problems while solving an unplugged Model Eliciting Activity (MEA). Researchers designed the MEA to enable students from all backgrounds and different levels of computer science expertise to engage with CT concepts through the familiar game of tic-tac-toe.

The educational landscape in Mexico is shaped by pronounced economic disparities that affect access to technology. In this context, unplugged activities offer a promising approach to teaching CT concepts without requiring sophisticated technological tools, potentially helping to level educational disparities throughout the country (Alejandro, Rojas, & Islas, 2021; International Trade Administration, 2023). MEAs provide a framework for this approach as real-world problem-solving scenarios that produce shareable, reusable solution models (Lesh, Doerr, Carmona, & Hjalmarson, 2003).

While CT research in resource-limited settings has grown, a critical gap remains in understanding how students from diverse backgrounds engage with and develop CT skills in unplugged environments (Caeli & Yadav, 2020; Grover & Pea, 2013). Previous studies have primarily focused on measuring learning outcomes or analyzing individual student performance (Brackmann et al., 2017). Still, few have examined the interactions that occur during CT activities in resource-limited contexts. In Mexican education, CT studies have primarily concentrated on technology-based interventions rather than unplugged approaches that might serve diverse socioeconomic populations (Basogain, Olabe, Olabe, & Rico, 2018).

The study aims to explore how MEAs can serve as a pedagogical tool to introduce and enhance CT concepts among high school students in Mexico. It focuses on understanding how students learn from complex, unplugged problem-solving tasks that require them to develop and communicate models, fostering key CT skills in the process. Forming and sharing a model of how to solve a new task that uses CT skills helps to develop those skills.

Additionally, the study examines the potential of MEAs to provide equitable learning opportunities in CT, particularly for students who might not have access to advanced technological tools and resources. By leveraging unplugged approaches, MEAs can bridge the digital divide and help to ensure all students have opportunities to develop CT skills.

There is a growing recognition and movement to integrate CT throughout pre-college curriculum and educational curricula, particularly in regions with limited access to technology and innovative teaching methods. This study could provide insights into scalable and effective ways to introduce crucial CT skills in resource-constrained environments.

Two aspects of this study are applied to develop such insights. One involves the use of Epistemic Network Analysis (ENA), a tool that operationalizes quantitative ethnography (QE) and that allows the researcher to capture and analyze complex cognitive connections students make as they complete or reflect on tasks of interest (Shaffer et al., 2016). The researchers apply ENA to the task of unplugged CT activities. Unlike traditional quantitative approaches that rely on performance metrics or qualitative approaches that depend on conjectural ethnography, ENA enables a more nuanced quantitative ethnography, one that enables statistically grounded visualization of the relationships between different aspects of CT as they emerge through student interactions (Bowman et al., 2021). This approach is particularly valuable for understanding how students from various academic and socioeconomic backgrounds engage with computational concepts in collaborative settings.

Moreover, the focus on unplugged MEAs in Mexico City provides insights into how CT can be effectively taught in settings where access to technology may be limited or inconsistent. By analyzing student interactions during the tic-tac-toe MEA (Carmona, Pate, & Galarza, 2019), the study seeks to demonstrate how accessible, low-cost activities can promote sophisticated CT skills while accommodating various learning styles and backgrounds. This study fills a crucial gap in understanding how unplugged MEAs can democratize access to CT education in resource-limited settings while simultaneously using ENA to advance a QE approach to studying cognitive development in collaborative learning.

The study could contribute to educational research by providing evidence of the effectiveness of MEAs in teaching complex, abstract concepts through practical, student-centered learning approaches. It could also offer a framework for other regions with similar challenges. By documenting the challenges and successes of implementing MEAs in a high school in Mexico, the study could offer a model for other schools in similar socioeconomic settings, demonstrating how innovative educational practices can be adapted and scaled.

Computational Thinking (CT) emerged as a formalized concept from Seymour Pappert’s work in the 1980s with Logo programming, building on constructivist “earning by doing” principles (Feurzeig & Papert, 2011). While Wing (2006) later popularized and expanded the term, CT fundamentally represents the systematic approach to decomposing complex problems and formulating structured solutions—a skill set now considered essential across various domains.

The definition and essential elements of CT continue to evolve in academic discourse. Researchers have proposed various frameworks, including Belmar’s (2022) emphasis on systemic abstraction, decomposition, algorithmic design, generalization, and evaluation and Brennan and Resnick’s (2012) computational concepts, practices, and perspectives. For analytical clarity, this study adopts Krauss and Prottsman’s (2016) framework of four fundamental pillars: decomposition, pattern recognition, abstraction, and algorithms. This framework provides observable behaviors in unplugged activities while acknowledging that elements such as generalization and evaluation often emerge naturally from these core components.

CT can be taught through technology-based and unplugged approaches (Caeli & Yadav, 2020). Initiatives like Computer Science Unplugged demonstrate that CT concepts can be effectively conveyed without electronic devices (Huang & Looi, 2021), making these approaches particularly valuable in contexts with limited technological resources. Beyond technical skills, CT fosters creativity, critical thinking, and collaboration (Pozos, Severance, Denner, & Tellez, 2022), encouraging students to experiment with solutions while enhancing communication and teamwork (Popat & Starkey, 2018).

Several initiatives globally, in fact, explicitly seek to build CT through “unplugged” activities that do not require the electronics that might be associated with notions of computation (Caeli & Yadav, 2020), such as the Computer Science Unplugged project (Huang & Looi, 2021). Unsurprisingly, CT is recognized by theorists to foster creativity and innovation by encouraging students to explore novel approaches to problem-solving. Students are prompted to think critically, experiment with different solutions, and develop innovative ideas through computational tasks and projects, cultivating what Brennan and Resnick (2012) refer to as an entrepreneurial mindset conducive to socioeconomic progress. In a similar spirit, CT fosters student collaboration, encouraging students to collaborate to develop solutions and share knowledge (Pozos et al., 2022). This collaborative approach enhances problem-solving skills and promotes effective communication, teamwork, and social interaction (Popat & Starkey, 2018).

As Mexico continues its journey towards a technology-driven future, embracing CT as a core skill set will be instrumental in unlocking the country’s potential for innovation with a workforce that can thrive in the digital age and that attracts foreign investment. Integrating CT into high school curricula is arguably vital for holding potential for equipping students in low- and middle-income countries with relevant and strong problem-solving skills required to address local challenges and foster socioeconomic development. By prioritizing the development of CT, Mexico can position itself as building a prosperous and technologically sophisticated global leader in the era of Industry 4.0 and beyond (Ríos Félix, Zatarain Cabada, & Barrón Estrada, 2020; Rivera-Kumar, Zambrano, & Barany, 2023).

Model Eliciting Activities (MEAs) are rich, open-ended problem scenarios designed to elicit from students’ initial attempts to find a solution – referred to as their initial conceptual models. Problem-solving occurs when students in a team share, test, and revise their conceptual models with peers until they reach a satisfactory solution (Lesh et al., 2003). The philosophy of model elicitation thus focuses on using the conceptual models that the student already possesses as the starting point for problem-solving instead of imposing pre-existing models. Students have the flexibility to create representations that make sense to them, making them meaningful. MEAs allow for multiple entry points and multiple solution paths, which means that they can cater to a diverse range of learners. Students from different cultural backgrounds or learning styles can find ways to engage with the problems that resonate with them. From the perspectives of Lesh et al. (2003) modeling frameworks, MEAs facilitate the elicitation of knowledge regardless of the students’ backgrounds or expertise. Their inclusive nature stems from their focus on real-world problem contexts that resonate with diverse learners.

Lesh et al. (2003) offered principles for designing MEAs, suggesting that they must be grounded in engaging students with complex, realistic problems that require them to develop and apply models – that is, solutions – that can be articulated and reused in other settings. Because they originate in real-world scenarios and are not tied to any academic discipline (in the way real-world problems cut across many disciplines), MEAs furnish opportunities for students to draw upon their prior experiences and individual knowledge, regardless of their specific academic background.

In the context of equity in STEM education, models and modeling perspectives are crucial in promoting inclusivity and accessibility for all students, regardless of gender, cultural background, or socioeconomic status. Literature on MEAs emphasizes the use of multiple representations while engaging students in meaningful problem-solving tasks; these two features of modeling help reduce the stereotype threat that marginalized or low-income students experience. The MEA community explicitly seeks to provide a framework for diverse learners to excel in mathematics and STEM fields (Galarza Tohen, 2023; Lesh et al., 2003; Zawojewski & Carmona, 2001).

MEAs are ideal and equitable instructional components in Mexican Schools. By their nature, MEAs do not require expensive materials or technology, making them accessible in various educational settings across socioeconomic backgrounds. Efficient use of resources and unplugged activities are essential in Mexico, where there may be significant differences or limitations in student resources. Because MEAs are complex and often require teamwork, they foster student collaboration and communication, crucial ingredients in CT and STEM fields, where working with others and communicating complex ideas is critical.

The socio-cultural aspect of MEAs is essential in a setting like Mexico due to the country’s rich cultural diversity and the critical role of education in fostering equitable socioeconomic development. Socio-culturalism tells us that learning is a process of acculturation (Vygotsky, 1978), and it appears to be involuntary. Acculturation, in turn, involves the intersubjective construction of shared knowledge in the classroom, where actors share knowledge (Vygotsky, 1978). Vygotsky observed that the individual is a product of their time and place. In Mexico, integrating socio-cultural factors into MEAs can significantly enhance learning experiences by considering Mexican students’ diverse backgrounds and experiences (Mayer, 2008; Vygotsky, 1978; Zawojewski & Carmona, 2001).

By engaging students in collaborative problem-solving tasks that reflect their socio-cultural realities, MEAs enhance learning experiences and play a significant role in developing a workforce with the skills necessary to innovate and compete in a rapidly changing global economy (OECD, 2019). These results highlight the confidence in the potential of MEAs in shaping the education of the future workforce.

Recent studies from various regions have demonstrated the effectiveness of unplugged CT approaches in addressing educational challenges. In Brazil and Columbia, researchers found that unplugged activities improved CT skills in public and rural schools (Kampylis, Dagienė, & Bocconi, 2023; Ortiz & Pereira, 2019). Rodriguez et al. (2017) demonstrated the potential of unplugged activities to develop CT skills in rural Columbia, demonstrating that contextually relevant tasks can engage students effectively. In Mexico City, where classrooms often reflect diverse socioeconomic backgrounds, leveraging unplugged MEAs can foster equity in CT education.

For several reasons, continuing the discussion on unplugged CT in regions like Mexico City is essential. First, it addresses the global challenge of ensuring equitable access to CT education, particularly in areas with limited technological infrastructure. Second, it situates Mexico City within an international framework of CT research, contributing to a broader understanding of how socio-cultural and economic factors shape learning outcomes. Lastly, it highlights the available role of MEASs in helping to empower educators to teach CT in inclusive and accessible ways, fostering skills essential for students’ participation in the digital age.

Epistemic Network Analysis (ENA) offers unique advantages for understanding how students elicit CT in small-group settings. Unlike traditional methodologies that analyze isolated elements, ENA models the complex interconnections between different aspects of CT as they emerge through collaborative discourse (Shaffer, 2017). This methodological approach is particularly valuable because CT inherently involves multiple interconnected cognitive processes, including decomposition, abstraction, pattern recognition, and algorithm creation, which are difficult to capture using conventional analytical methods.

ENA is especially powerful for studying collaborative learning environments. As Shaffer and Ruis (2017) explain, ENA provides statistical rigor while maintaining contextual sensitivity, producing visual network models that reveal how students co-construct computational knowledge through their interactions. These visualizations make complex learning patterns accessible while enabling researchers to quantitatively compare different groups (Siebert-Evenstone et al., 2017). For research on how CT emerges through social interaction, ENA provides insights into collaborative knowledge construction that alternative methodologies cannot readily capture.

Quantitative ethnography (QE) is a powerful research method that helps us navigate a data-rich world. It enhances the scope and power of ethnographic or other qualitative methods by incorporating statistical techniques (Arastoopour, Chesler, & Shaffer, 2014). QE, in particular, combines statistical inference with qualitative analysis, respecting the insights gained by ethnography and harnessing the power of statistical techniques (Shaffer, 2006; Shaffer et al., 2009). One such technique is Epistemic Network Analysis (ENA). This tool has found practical applications in various fields, including education and sociology, by modeling the association between elements of complex thinking.

ENA is a tool for the analysis of networks. It models the association between elements of complex thinking. ENA models cognitive networks since the connections among cognitive elements are more critical than studying those elements in isolation. ENA is used to examine the connections and uses visualization and statistical techniques to identify patterns; it quantifies the cooccurrence of concepts within a conversation (Bowman et al., 2021; Shaffer, 2017; Shaffer, Collier, & Ruis, 2016). ENA visually models cognitive networks based on ideas of principal components analysis to understand the knowledge and skills of complex thinking.

The cognitive network modeling for a subject is based on analysis of that subject’s discourse patterns, where discourse (such as in an interview or task transcript) is segmented into lines or stanzas that facilitate analysis (Shaffer et al., 2009). Relationships are calculated and depicted graphically by coding for specific constructs. The visualization arises from examining cooccurrences of constructs in students’ conversations while learning a concept. Node size is the frequency of code occurrence, and the frequency of two codes occurring in the same discourse segment or stanza determines the thickness of the edge connecting the two constructs (Marquart, Hinojosa, Swiecki, Eagan, & Shaffer, 2018).

In a 60-min workshop, a team of researchers observed two groups, each comprising three students, as they worked on solving what is known as the Tic Tac-toe MEA (Carmona et al., 2019). The researchers showed a visual presentation of the Turing test before the task to provide students with a relatable background. The students were then organized into small teams and tasked with formulating strategies and processes to enable a computer to win against a human opponent in tic-tac-toe consistently. To devise an unbeatable algorithm, each group actively engaged with the game. The entire process was recorded with audio and video for subsequent analysis.

In this study, the participants were purposefully selected by their teacher to ensure a diverse sample with varying backgrounds, career aspirations, and levels of expertise in computing. This purposive sampling method was employed to capture a range of perspectives and experiences within the context of the study (Patton, 2002). The purposive sampling approach allowed for the inclusion of participants with different characteristics, which can provide insights into how these factors may influence their engagement with and perceptions of the study’s subject matter (Creswell & Creswell, 2017). However, it is essential to acknowledge the limitations of this sampling method and the study’s context. The high school where students worked in the MEA is a small-sized school with a limited number of students.

The study involved two groups of participants. Group 1 consisted of two boys and one girl, while Group 2 comprised three girls. All participants were seniors attending a private school in a middle-income, urban area of Mexico City.

The students exhibited diverse levels of computational experience and expertise. Two students, one from each group, had completed introductory programming courses covering basic coding concepts. One student had participated in an after-school robotics club but had no formal programming education. Three students had no formal exposure to computing concepts beyond basic computer usage.

The participants represented different academic tracks within the school. Three students from the science track, including those with programming experience. One student from the humanities track, another from the business track, and one from the arts track, who had applied to culinary school.

While all students attended the same private school, they represented varying socioeconomic backgrounds within Mexico City’s middle-income spectrum. This aspect was relevant to our context, the reason for using MEAs, and the unplugged nature of the activity, which eliminated potential technology access barriers.

The sample’s non-randomized nature limits the findings’ generalizability to the broader population of students in Mexico City or other contexts (Flick, 2017). The study’s results should be interpreted cautiously, as they may not represent the experiences and perspectives of all students in similar settings. The small sample size, constrained by the school’s limited enrollment, may affect the depth and breadth of the data collected (Bowman et al., 2021; Bressler, Bodzin, Eagan, & Tabatabai, 2019; Nachtigall & Sung, 2019). The school’s relatively small population restricted the number of students available for participation and the possibility of randomizing the sample. A larger sample size could have provided more diverse perspectives and potentially strengthened the study’s findings (Wooldridge, Carayon, Eagan, & Shaffer, 2018).

Future research should address this limitation by conducting larger studies involving students from multiple schools with varied socioeconomic contexts. Expanding the sample size will enhance the generalizability of findings and provide a broader understanding of how different factors influence students’ engagement with CT and MEAs.

Despite these limitations, the purposive sampling method and the study’s context offer valuable insights into the experiences of a specific group of students within a particular educational setting. The method was chosen to ensure a diverse representation of student characteristics, including varying academic tracks, computational experiences, and career aspirations (Patton, 2002). The method allowed us to capture a range of perspectives and explore how these characteristics might influence student learning.

While purposive sampling inherently limits the generalizability of findings, it provides valuable insights into the dynamics of a specific group of students within their unique educational context. The researchers acknowledge that the selection process may introduce bias, as the sample reflects the teacher’s knowledge of the student’s strengths and interests. Addressing this in future studies by employing mixed sampling methods of randomization, where feasible, could help mitigate this potential bias.

Data collection involved observations in the classroom and the use of video and audio recordings as the primary data source. The study used verbatim transcriptions of student conversations from a private high school in Mexico City as the data collection instrument. After transcription, researchers coded the video recordings to categorize constructs related to the research questions (Lotto, 1986).

Table 1 shows the codes assigned to classify conversational utterances based on the categories derived from CT frameworks (Wing, 2006; Grover & Pea, 2013). Each utterance was examined for its content within the conversation to determine the cooccurrences in the conversations.

Conversations were double coded independently by two researchers: the author and a doctoral student with expertise in CT frameworks. To enhance interrater reliability: coder training that included studying theoretical frameworks; bibliography on foundational texts; calibration sessions to discuss and resolve discrepancies in initial coding refining; and social moderation to resolve disagreements and ensure consensus. Inter-rater reliability was calculated using Cohen’s Kappa, achieving 0.9, which indicates strong agreement between coders (Lotto, 1986).

The researchers used ENA with the Web Tool (version 1.7.0) (Marquart, Hinojosa, Swiecki, Eagan, & Shaffer, 2018). The units of analysis are defined as all lines of data associated with a single value of TEAM sub-setted by PARTICIPANT. For example, one unit comprised all the lines related to PARTICIPANT G1.

The ENA algorithm uses a moving window to construct a network model for each line in the data, showing how codes in the current line are connected to codes that occurred previously (Ruis and Lee, 2021; Siebert-Evenstone & Shaffer, 2019) defined as 10 lines (each line plus the nine previous lines) within a given conversation. The resulting networks are aggregated for all lines for each unit of analysis in the model. In this model, the networks are aggregated using a binary summation in which the networks for a given line reflect the presence or absence of the cooccurrence of each pair of codes. The authors used the moving stanza window method, analyzing the connections made in parts of the activity (Arastoopour et al., 2014; Eagan & Hamilton, 2018). While playing Tic-Tac-Toe, conversations can come back and forth on how to move, and the students will take more lines to elicit concepts, as we can see from the following table (Table 2).

The ENA model normalized the networks for all units of analysis before they were subjected to a dimensional reduction, which accounts for the fact that different units of analysis may have different numbers of coded lines in the data. Normalization transforms the raw network data to ensure it is comparable across units and contexts. Normalization helps us remove biases to focus on the connections between nodes rather than the absolute values (Bowman, 2021).

For the dimensional reduction, ENA uses a singular value decomposition, which produces orthogonal dimensions that maximize the variance explained by each dimension (Bowman et al., 2021; Shaffer et al., 2016). SVD captures the variability of the data; higher SVD indicates greater importance in explaining the data variance. Means Rotation (MR) represents the transformed network after alignment. The dimensions identified through MR furnished meaningful insights into the differences in group conversations.

In ENA, a means rotation is used to align the network space to center the analysis around the mean of the data points. This process adjusts the orientation of the ENA networks to ensure that the primary dimensions reflect the most significant patterns of cooccurrence by the data. That is, rotation enables analysis of the construct relationships most relevant to the research questions. In this case, those questions are to what degree the students understand the CT concepts and to what degree all students participate regardless of their background and expertise. Means rotation improves interpretability and can reduce the influence of noise in the visualization (Shaffer et al., 2016).

Means rotation allowed comparison of the different groups and individuals against a baseline, highlighting the deviations or similarities in patterns of connections. It was possible to identify stronger and weaker connections between specific constructs for individual students, such as abstraction and decomposition, compared to the overall average (Shaffer, 2017). By focusing on the mean differences, the rotation allows the visualization of contrasts between groups along this dimension, making it easier to interpret and compare.

In ENA, networks are visualized using graphs where nodes correspond to the codes, and edges reflect the relative frequency of cooccurrence, or connection, between two codes. The result is two coordinated representations for each unit of analysis: (1) a plotted point, which represents the location of that unit’s network in the low-dimensional projected space, and (2) a weighted network graph. The positions of the network graph nodes are fixed and determined by an optimization routine that minimizes the difference between the plotted points and their corresponding network centroids. Because of this co-registration of network graphs and projected space, the positions of the network graph nodes—and the connections they define—can be used to interpret the dimensions of the projected space and explain the positions of plotted points in the space. The model for this study had co-registration correlations of 0.99 (Pearson) and 0.97 (Spearman), indicating a strong goodness of fit between the visualization and the original model.

Figure 1 shows the results of a two-sample t-test comparing the means of two groups (labeled as 1 and 2) along the X-axis of an ENA model. The test assumes unequal variances between the groups, which is appropriate when the sample sizes are small, and the variances are not known to be equal (Shaffer et al., 2016).

Group 1 has a mean of 0.04, a standard deviation (SD) of 0.06, and a sample size (N) 3. Group 2 has a mean of −0.04, an SD of 0.04, and an N of 3. The t-test results show a t-value of 2.03 with 3.77 degrees of freedom (t(3.77) = 2.03), a p-value of 0.12, and a Cohen’s d effect size of 1.66. (Shaffer & Ruis, 2017). However, it is essential to consider the study’s limitations when interpreting these results. The small sample size (N = 3 for each group) may limit the statistical power to detect significant differences, even if such differences exist in the population (Shaffer & Ruis, 2017). Given the different characteristics of each participant, the similarities in the conversations and results support the claim that MEAs promote the participation of all students regardless of background and expertise.

The similarity in epistemic networks in this specific case can help us answer the research question: If the research question aims to explore whether two groups have similar patterns of knowledge and skill development, non-significant differences may support the hypothesis that the groups are comparable in their epistemic networks (Shaffer & Ruis, 2017). When interpreting this figure, the contextual factors were considered. The specific context in which the study is conducted may influence the results, and non-significant differences in one context may not generalize to other settings (Shaffer et al., 2016). When interpreting these findings, it is crucial to consider their practical applications to CT education. The results suggest that MEAs can be leveraged as instructional tools to foster equitable participation in CT education, regardless of students’ prior experience or background. Given these findings, the study provides valuable insights for designing scalable, accessible CT curricula that can be implemented in diverse educational contexts.

According to Figure 1, ENA explains 38.5% of the variance in coding cooccurrences along the x-axis and 46.2% of the variance on the y-axis. In ENA, the proportion of variance captured by the primary and secondary axes—X and Y—reflects how these dimensions encapsulate the observed differences within the data. The X-axis represents the first dimension, representing 38.52% of the data’s variance, indicating its significant role in distinguishing the epistemic networks within the analytical model. Conversely, the second dimension, represented by the Y-axis, explains 46.23% of the variance, denoting an even more significant contribution to the differentiation of connection patterns among the analyzed units (Shaffer et al., 2016; Shaffer & Ruis, 2017). The reported variances along the X-axis (0.3852) and Y-axis (0.4623) demonstrate that both dimensions substantively represent the variability within the epistemic networks, with the latter dimension slightly more pronounced in its explanatory power.

Figure 2 represents the network for Team 1 conversations; it is apparent from the analysis that the team elicited CT constructs while engaged in solving the MEA. Furthermore, it is observed that Team 1 engaged in conversations, characterized by wider lines in ENA, signifying more frequent cooccurrences presented in Table 3.

In Table 3, we can see the students’ conversations while solving the Tic-tac-toe MEA; added examples for the four dyads that had more comprehensive lines, indicating higher cooccurrences. In the first line, which corresponds to the pattern-abstraction dyad, we can observe that the conversation between teammates identifies a pattern: “If you place it here, you can also win.” The student is identifying a pattern of where to place the X or the O so you can win the game; the other teammate answers, “You need to see when you win, not only play foolishly,” she is asking the other student to realize how she is playing to win, the student is thinking on how to repeat the plays in the future so they can always win, the student is looking to generalize the solution, she is abstracting.

From Figure 3, which represents the conversations from Team 2, we can observe that the team elicited CT constructs while engaged in solving the MEA. The conversation between the members of teams 1 and 2 is very similar. There is a slight change; in team 2, the widest lines of cooccurrences in conversation were between pattern-decomposition (3) and pattern-abstraction (1). This team had fewer cooccurrences in the arch abstraction and decomposition (2), followed by pattern – algorithm (5).

The following excerpt from the transcription exemplifies a conversation dyad coded as pattern recognition and decomposition. Two participants in the group discussed final strategies to place their mark (X/O) in the game; the conversation proceeded as follows:

• B1: You first place it here.

• B2: Place it in the upper right corner.

• B1: the next turn would be on the side.

• B1: or you can place in the opposite corner.

This exchange illustrates how participants identified and applied patterns while breaking down their strategy sequentially, demonstrating CT through pattern recognition and decomposition.

The researchers generated six graphs representing each student’s interactions within the Model-Eliciting Activity (MEA) to analyze the differences and similarities in how each participant elicited computational thinking. These visualizations allow a detailed comparison of each student’s cognitive processes while engaging with CT constructs. By examining these graphs, the researchers could identify patterns in participation, assess variations in problem-solving approaches, and determine the extent to which each student contributed to the group discussions. This approach provided more profound insights into individual engagement levels and the equitable distribution of CT skills across participants.

To explore how student background influenced the eliciting of CT while solving the MEA, the team analyzed how diverse student profiles interacted within the MEA. The dyads of conversations between students construct the networks. Both groups demonstrated robust connections across the four constructs of CT, which can be observed in Figures 2 and 3 (conversations of both teams); this indicated that MEAs effectively facilitate the eliciting of CT. In Figure 4, the individual networks of each of the students are presented. All the students recognized and talked about the four constructs analyzed for CT, regardless of prior expertise. Each of the students contributed unique perspectives and strategies to the problem-solving process.

The analysis of the team dynamics revealed a balanced participation; mixed expertise teams exhibited equitable, meaningful participations, and all the members of both teams contributed to each CT construct. Teams naturally leveraged diverse perspectives, enhancing the development of a solution.

The ENA conversation dyads between students (Figure 4) demonstrated that differences in background, academic profiles, and socioeconomic context did not hinder collaborative problem-solving. The teams engaged in meaningful discussions that elicited the four key constructs of CT: abstraction, decomposition, pattern recognition, and algorithm development.

These findings demonstrate the practicality of unplugged MEAs as scalable and accessible tools for CT education. MEAs can be effectively integrated into curriculum design to facilitate CT instruction, particularly in resource-limited settings. By eliminating technological barriers, they provide equitable learning environments, allowing students from diverse backgrounds to engage meaningfully regardless of prior programming experience.

Unplugged MEAs align with global CT efforts by reinforcing problem-solving, abstraction, and algorithmic reasoning without requiring access to digital tools. The balanced participation observed in MEA-driven discussions suggests the potential for retaining underrepresented groups in STEM by providing early success experiences that foster long-term engagement—furthermore, the implementation of targeted teacher training frameworks and national curricula. Since CT is an essential skill across various disciplines, unplugged MEAs offer an inclusive strategy for bridging digital divides in educational settings worldwide.

With the help of ENA, it is possible to see how different team members participate in coming up with a solution to find the algorithm. While working in the MEA, they elicit CT. The process of completing the MEA helps elicit concepts when an unplugged MEA is well-designed. That is, MEA can encourage students in Mexico to understand CT concepts. ENA helps understand the conversations and compare the results of the different participants.

The ENA networks of the teams solving the MEA to elicit the CT constructs confirm that MEAs facilitate equitable engagement across diverse student profiles. ENA allows the analysis of the networks created by dyads of conversation between the participants of each group. Balanced conversations and participation of the three members of the group are observed. By promoting balanced participation and leveraging interdisciplinary interactions, MEAs ensure that all students actively contribute to CT processes, fostering inclusivity, innovation, and problem-solving.

To address the second research question – How can participation be fostered regardless of background and expertise? The ENA results provided valuable insights. The analysis revealed that students’ eliciting of CT constructs was not significantly impacted by their prior programming experience or academic background. Instead, the unplugged MEA appeared to create an equitable platform where students from different academic tracks and varying levels of technical expertise could contribute meaningfully to the problem-solving process.

Despite their different programming experience and academic background composition, the similar network structures between teams suggest that MEAs can foster inclusive participation in eliciting CT. This is particularly significant in the Mexican context, where access to technological resources may vary significantly across different socioeconomic groups. The unplugged nature of the MEA effectively eliminated potential barriers related to technology access, allowing all students to engage fully with the CT concepts.

These findings support the potential of unplugged MEAs as tools for democratizing access to CT education, particularly in resource-limited settings. Expanding research to include larger, more diverse educational settings would enhance the validity of these findings. Further studies should examine long-term influence of MEAs on CT skills and their applicability across different learning environments. While this study highlights the promise of unplugged MEAs fostering CT, additional exploration is needed to determine their flexibility in different educational contexts and among varied student demographics.

This study was conducted in one of many private schools in Mexico, where they differ in characteristics and students. Given the small sample size, while this initial study provides insights that support further research into how Model Eliciting Activities (MEAs) and unplugged Computational Thinking (CT) approaches can enhance student learning, the capacity to generalize these results is limited.

The time and availability constraints in the high school further limited the study’s scope. Consequently, the findings should be interpreted with caution. They can serve as a foundation for further research and inform educational practices and policies to support students’ learning and development in similar contexts.

Future studies should consider larger, randomized samples across diverse contexts to validate and extend these results. This will strengthen the relevance of the research and contribute to a more comprehensive understanding of how MEAs and unplugged CT approaches can support student learning in varied socio-cultural settings.

Additionally, future research should examine the long-term impact of computational thinking (CT) instruction on STEM persistence across different socioeconomic groups. It is important to identify which specific CT concepts are most effectively taught through unplugged activities versus those better suited for plugged, technology-based approaches. Further research should also explore how various cultural factors may influence the effectiveness of model-eliciting activities (MEAs) in different regional or educational contexts.

This paper forms part of a special section “Quantitative ethnography in education research and evaluation in low- and middle-income nations”, guest edited by Drs Eric Hamilton, Danielle Pascual Espino, Seung Lee and Kristina Lux.

Funding: This work was funded in part by the National Science Foundation (DRL-2225240) and supported by the Quantitative Ethnography Fellows Institute at the University of Wisconsin–Madison. Additional support was provided by the National Science Foundation (DRL-2100320, DRL-2201723, DRL-2225240), the Wisconsin Alumni Research Foundation, and the Office of the Vice Chancellor for Research and Graduate Education at the University of Wisconsin–Madison. The opinions, findings and conclusions expressed are those of the authors and do not necessarily reflect the views of the funding agencies, cooperating institutions or other individuals.