The Effect of the Augmented Reality Material Development Process on Preservice Science Teachers’ Educational Technology Self-Efficacy and Computational Thinking Skills

Educational environments are being reshaped at a fast pace by developments in information and communication technologies, with growing attention given to tools that make learning content richer and turn abstract ideas into something more tangible. Among these tools, augmented reality (AR) stands out for its capacity to merge computer-generated digital layers with the physical world in real time, allowing users to interact with both simultaneously (Azuma, 1997; Wu et al., 2013). What distinguishes AR from virtual reality is precisely this retained connection to the physical environment: Rather than replacing it, AR overlays it with functional and meaningful digital information.

An increasing amount of research connects AR use to improvements in academic performance, motivation, and student engagement (Akçayır & Akçayır, 2017; Saidin et al., 2015). This appears especially valuable in science education, where AR’s capacity to render abstract or hard-to-observe phenomena — such as internal body systems — in three-dimensional, interactive form has been shown to support meaningful learning and reduce misconceptions (Erbaş & Demirer, 2014; Somyürek, 2014). Yet, most of this evidence concerns ready-made AR applications used by students (Korucu et al., 2016; Küçük, 2015). Far fewer studies have examined what happens when preservice teachers themselves take on the role of designing and building such technologies.

AR’s value for teacher preparation is not confined to the ways the technology is used by others; the act of designing and building it can itself generate substantial learning. Within a design-based learning approach, AR development tasks appear particularly well suited to cultivating computational thinking (CT) — a construct Wing (2006) described as a set of higher order processes, including breaking a problem into algorithmic steps, abstracting its essential features, spotting patterns, and generalizing solutions, now widely regarded as core to 21st-century competence (Aho, 2012).

These situations apply within enhanced material design processes. Evidence for building CT through structured, tool-based design tasks is not limited to secondary or higher education: Even in early childhood settings, deliberately structured programming-tool interventions have been shown to support measurable CT gains (Louka & Papadakis, 2024), suggesting that the mechanism of interest — engaging learners as designers/builders rather than passive users — generalizes across age groups and tools.

Two key applications are involved in the process of designing AR materials: Unity and Vuforia. These two programs form the basis of many of the everyday applications used in this field. Building an AR material gives these processes a concrete outlet. Positioning three-dimensional objects in Unity (Unity Technologies, 2023) — a cross-platform engine for real-time 3D content — and configuring image targets in Vuforia (PTC Inc., 2023) — a development kit that overlays digital objects onto real-world imagery — requires preservice teachers to abstract problems and assign logical sequences of actions to buttons and interactions. Troubleshooting coding errors along the way exercises problem-solving directly.

In this sense, the AR development task translates CT, otherwise a fairly abstract computer-science construct, into something hands-on within a science-education setting. Testing this link experimentally stands to inform both CT research and how teacher preparation programs are designed.

A second outcome the AR development process is expected to shape is self-efficacy, with respect to educational technology standards (ETSSE). Bandura’s (1997) self-efficacy theory holds that a person’s confidence in their own capacity to carry out a task shapes how they actually perform and behave in that domain. Applied to teachers, this lens captures their confidence in planning, designing, and implementing technology pedagogically and is known to strongly predict technology integration behaviors (Şimşek & Yazar, 2016).

Indeed, the Technological Pedagogical Content Knowledge framework also posits, as a prerequisite, teachers’ ability to use technology, pedagogy, and content knowledge (TPACK) in an integrated manner (Mishra & Koehler, 2006). Within this framework, preservice teachers’ production of an actual AR material using professional software such as Unity and Vuforia offers an authentic developer-role experience. Although TPACK is traditionally examined in the context of teaching with technology rather than building it, the process of deciding how to represent science content digitally and making pedagogically motivated design choices within Unity and Vuforia plausibly draws on and strengthens the technological and technological-pedagogical components of TPACK even before these preservice teachers reach an actual classroom.

This developer-role experience also has the potential to leave a lasting mark on ETSSE. International research examining how preservice teachers’ attitudes and self-efficacy toward emerging classroom technologies (e.g., educational robotics) develop through direct, hands-on engagement has similarly found that firsthand design or operation experience shapes confidence more than passive exposure (Papadakis & Kalogiannakis, 2020), reinforcing the rationale for a developer-role design task rather than a ready-made model. AR development experiences among preservice teachers remain scarce internationally, and within Türkiye the literature leans heavily toward ready-made AR applications rather than teacher-led development (Şengün Öztaş, 2024). This gap suggests value in studies that position preservice teachers as AR developers and test the resulting effects on CT and ETSSE experimentally.

Building upon the body of research that highlights the instructional potential of immersive technologies, this study investigated the comparative effects of an augmented reality-supported course design versus a traditional design on preservice teachers’ computational thinking skills and educational technology self-efficacy.

The specific structure of each instrument used here influenced the selection of these particular constructs, rather than more generic 21st-century-skill measures, to evaluate the material development process. The CT scale (Ertuğrul-Akyol, 2019) is organized around three factors — Computational Thinking, Robotics-Coding and Software, and Professional Development and Career Planning — each of which maps onto a distinguishable component of the task preservice teachers performed.

The Computational Thinking factor, capturing abstraction, decomposition, and algorithmic-sequencing tendencies, corresponds to the work of breaking an organ system into interactive, sequentially triggered AR components. The Robotics-Coding and Software factor corresponds directly to the C# scripting and Vuforia image-target configuration the AR-building process required. The Professional Development and Career Planning factor captures whether working with industry-standard tools such as Unity and Vuforia is experienced as building a professionally relevant skill set rather than completing a one-off classroom exercise.

Because all three factors have a plausible, activity-level counterpart in the AR development process, the CT scale offers closer construct alignment with this specific task than a more general-purpose CT instrument would. This situation also applies to the group that does not use AR. In this group, the material development process goes through the steps of obtaining information, researching, and analyzing, using the design steps.

The same logic motivated the choice of ETSSE (Şimşek & Yazar, 2016). Its five subdimensions — facilitating students’ learning and creativity, designing digital-age learning environments and assessment, modeling digital-age work and learning, modeling digital citizenship, and engaging in professional development and leadership (International Society of Technology in Education [ISTE], XXXX [CITE PUBLICATION YEAR AND ADD TO REFERENCE LIST]) — were treated not as an undifferentiated global score but as standards the AR development task was expected to activate through distinguishable mechanisms.

Building an AR material intended to support middle-school learners’ understanding of body systems speaks directly to facilitating students’ learning and creativity and to designing digital-age learning environments. Producing that material with professional software speaks to modeling digital-age work and learning. In addition, the sustained, semester-long engagement with an unfamiliar technology speaks to professional development and leadership. Modeling digital citizenship is the subdimension with the least direct behavioral counterpart in the present task: It suggests that an overall ETSSE gain, if found, is more plausibly driven by the first four subdimensions than by the fifth — a possibility future work could test using subdimension-level rather than total scores.

Addressing this gap, the present study used a quasi-experimental design to examine how a 14-week AR material development process built around the Systems in Our Body unit affected preservice science teachers’ CT skills and ETSSE levels, relative to a comparably structured non-AR material design experience. Beyond filling a gap in the national literature, the study was intended to inform how science teacher education programs are designed, support technology integration policy, and strengthen future teachers’ digital pedagogy.

Purpose of the Study

This study set out to test, primarily through a between-group comparison, whether the AR material development process yielded outcomes in CT and ETSSE that exceeded those associated with a non-AR, design-based instructional material development experience. Both the experimental and control groups engaged in structured material design within the same course, on the same content, and with comparable instructor support. The control condition was, therefore, treated as an active comparison condition rather than a passive, untreated control. Within-group pretest/posttest change was also examined. Such change alone, however, could not isolate the effect of the AR-specific process, since both groups were exposed to a structured design task concurrently. Accordingly, the between-group comparisons were treated as the primary research questions, and the within-group comparisons as secondary and descriptive. The study addressed the following research questions:

Primary (between-group) research questions:

1. Controlling for pretest scores, do the experimental (AR) and control (non-AR) groups differ significantly on posttest CT scale scores?
2. Controlling for pretest scores, do the experimental (AR) and control (non-AR) groups differ significantly on posttest ETSSE scale scores?

Secondary (within-group) research questions:

1. Do preservice teachers in the experimental group show a significant difference between their CT scale pretest and posttest scores?
2. Do preservice teachers in the control group show a significant difference between their CT scale pretest and posttest scores?
3. Do preservice teachers in the experimental group show a significant difference between their ETSSE scale pretest and posttest scores?
4. Do preservice teachers in the control group show a significant difference between their ETSSE scale pretest and posttest scores?

Significance of the Study

The significance of this study rests on a direct, discipline-specific connection between CT, ETSSE, and the particular demands of science teaching, rather than treating either construct as a generic 21st-century skill loosely attached to a science content example. Science instruction routinely requires representing phenomena that cannot be directly observed — for example, the electrical signaling of the nervous system, gas exchange in the lungs, and the layered structure of the digestive tract — and a substantial body of research links students’ misconceptions in these areas to the difficulty of visualizing dynamic, internal, or microscopic processes (Erbaş & Demirer, 2014; Somyürek, 2014). AR is well matched to this specific instructional problem because it allows a static structure to be layered with interactive, three-dimensional, process-level information. A preservice science teacher who has built such a representation — deciding how to decompose an organ system into interactive parts, how to sequence the information a learner encounters, and how to debug a model that does not yet behave as intended — engages in a task that overlaps substantially with the abstraction, decomposition, and algorithmic-sequencing components of CT (Wing, 2006), applied directly to science subject matter rather than to a domain-general programming exercise. Positioning CT this way, as an extension of core science-teaching practice rather than an imported computer-science skill, is the connection that motivated testing CT as an outcome of this specific design task.

ETSSE is tied to the same argument. The ISTE-based standards operationalized in the Şimşek and Yazar (2016) scale used here include facilitating student learning and creativity with technology and designing digital-age learning environments — both of which describe, almost exactly, what a science teacher does when incorporating a self-built AR model into a lesson on body systems. Self-efficacy for these standards is not simply a proxy for general comfort with technology; it is known to predict whether teachers actually adopt classroom technology once they are no longer required to (Birisci & Kul, 2019; Şimşek & Yazar, 2016). Producing a working AR application using professional tools (Unity, Vuforia) is a mastery experience within the specific domain the ETSSE standards described, which is the mechanism Bandura’s (1997) theory identifies as the strongest source of self-efficacy growth. Positioning preservice teachers as designers of instructional technology, rather than as end users of materials someone else built, also models the constructionist, design-based learning stance that these same teachers will later be expected to enact with their own K–12 students in the future (Papadakis & Kalogiannakis, 2020).

Several further aspects distinguish the present study from prior work. First, little science-education research has examined the process of building AR applications for the Systems in Our Body unit specifically. Turning core organs — the heart, brain, lungs, stomach, intestines, and eyes — into AR materials gives preservice teachers an original, content-specific design task while building a resource archive of direct use in middle-school science teaching. Second, most existing national studies examine the effects of ready-made AR applications on learners (Korucu et al., 2016; Tosik Gün & Atasoy, 2017), while experimental work in which preservice teachers take on the developer role themselves remains scarce. This scarcity is not confined to Türkiye — internationally, the CT literature is comparatively dense for early-childhood and general science, technology, engineering, and mathematics (STEM) contexts (e.g., Louka & Papadakis, 2024) but comparatively thin for secondary-science teacher preparation specifically, which is the gap this study targets. Third, few studies examine CT and ETSSE together as outcomes of the same content-specific, technology-development process for science teacher candidates. Doing so offers a more integrated picture of how a single design experience can simultaneously shape a teacher candidate’s confidence with technology standards and their computational problem-solving orientation. Taken together, the findings are intended to inform how science teacher education programs structure hands-on technology coursework, how technology-integration policy in science teaching is developed, and how future teachers’ digital pedagogy competencies are strengthened.

Method

Research Design

This study used a quasi-experimental pretest/posttest control group design, a quantitative approach suited to situations where random assignment of participants is not feasible, yet a cause-and-effect relationship between independent and dependent variables is still sought, much as in true experimental research (Büyüköztürk et al., 2020). Two already-formed class sections served as the experimental and control groups, both completing pretests and posttests. The instructional intervention built around the AR material development process functioned as the independent variable, while CT skills and ETSSE levels served as the dependent variables. To limit the influence of extraneous variables, both groups were drawn from students at the same grade level and in the same program, followed the same 14-week content sequence, and received the same instructor contact time (see Experimental Process).

Because intact class sections, rather than individually randomized participants, formed the two conditions, preexisting differences between groups could not be ruled out by design alone. Accordingly, formal baseline-equivalence tests were conducted before the main analyses (reported in the Working Group section), and ANCOVA was used specifically to adjust statistically the posttest comparison for any preexisting differences on the covariate (pretest score). Where a key assumption underlying ANCOVA — homogeneity of regression slopes — could not be confirmed, this is reported explicitly, and the corresponding ANCOVA result is interpreted with that qualification in mind (see Data Analysis).

Working Group

In the Turkish higher education system, preservice science teachers complete a 4-year undergraduate science education program housed within Faculties of Education. Science Teaching I (Fen Öğretimi I) is a mandatory junior-year methods course in which candidates learn to design and evaluate instructional materials for middle-school science topics prior to their senior-year teaching practicum. The study was conducted with 62 preservice science teachers (48 female, 14 male) taking this course in the Department of Science Education at Dokuz Eylül University, Buca Faculty of Education, during the 2024–2025 academic year.

A quasi-experimental design was used because the students were automatically assigned into two separate class sections by the university’s course registration system, and institutional policies did not allow the researchers to reorganize these classes. Therefore, students could not be individually randomized; instead, a random assignment was made between the two preexisting class sections by drawing lots to determine which would serve as the experimental group (n = 32) and which as the control group (n = 30).

Table 1 reports independent-samples t-test comparisons of the two groups’ pretest scores on both outcome measures. Contrary to what would be expected in a well-matched quasi-experimental design, the experimental group scored significantly higher than the control group at pretest on both ETSSE, t(60) = 2.72, p = .009, d = 0.69, and CT, t(60) = 2.50, p = .015, d = 0.64. Both were medium-to-large differences by conventional benchmarks (Cohen, 1988).

Table 1
Independent-Samples t-Test Comparison of Experimental and Control Group Pretest Scores

Measure	Group	n	M	SD	t	df	p	Cohen’s d
ETSSE	Experimental	32	154.16	17.49	2.72	60	.009	0.69
ETSSE	Control	30	143.59	12.54
CT	Experimental	32	107.25	12.31	2.50	60	.015	0.64
CT	Control	30	99.22	12.98
Note. Pretest comparisons were computed from group means, standard deviations, and sample sizes using independent-samples t-tests (pooled-variance); the pattern was confirmed using Welch’s t-test for unequal variances. ETSSE = Educational Technology Standards Self-Efficacy; CT = Computational Thinking. Cutoff values for Cohen’s d: .20–.49 small, .50–.79 medium, ≥ .80 large effect (Cohen, 1988).

This baseline nonequivalence did not by itself invalidate the quasi-experimental comparison, but it means the two intact sections entered the study with different starting levels of self-reported technology self-efficacy and computational thinking. It also strengthens the rationale for using ANCOVA, rather than a simple posttest comparison, while requiring that the homogeneity-of-regression-slopes assumption underlying ANCOVA be checked explicitly rather than assumed (see Data Analysis).

Data Collection Tools

Computational Thinking Scale

Preservice teachers’ CT skills were measured with the Computational Thinking Scale, developed by Ertuğrul-Akyol (2019). The instrument comprises 30 items organized into three factors (Computational Thinking, Robotics-Coding and Software, and Professional Development and Career Planning) rated on a 5-point Likert scale from strongly disagree (1) to strongly agree (5). Total scores can range from 30 to 150, with higher values reflecting stronger CT, and reliability analysis in the present study yielded a Cronbach’s alpha of .91.

Educational Technology Standards Self-Efficacy Scale

The ETSSE scale, developed by Şimşek and Yazar (2016) based on the ISTE standards [CITE WHICH STANDARDS AND ADD A REFERENCE ENTRY] to measure teachers’ technological competencies, was employed to determine preservice science teachers’ educational technology self-efficacy levels. It contains 40 items in a 5-point Likert format, organized into five subdimensions: facilitating students’ learning and creativity, designing digital-age learning environments and assessment activities, modeling digital-age work and learning, modeling digital citizenship, and engaging in professional development and leadership. In the present study, the reliability analysis yielded a Cronbach’s alpha coefficient of .92.

Both scales were used with permission from their original developers, whose terms do not permit reproduction of the full item sets; consequently, individual items are not listed here or in the accompanying tables, consistent with standard practice for copyrighted, developer-licensed instruments. Both instruments are widely used, validated self-report scales, and their internal consistency in the present sample was strong. Both instruments, however, capture perceived competence and confidence rather than directly observed performance. This distinction matters most for the CT scale: Computational thinking is most rigorously operationalized in the literature through performance-based measures; for example, evaluating the structure, correctness, or complexity of the artifacts learners actually produce when comparing different programming or design environments (e.g., Papadakis & Orfanakis, 2018) rather than through self-report alone.

A change in CT self-report scores over a 14-week design process may reflect genuine growth in computational thinking, growth in participants’ confidence about that thinking, or some combination of both, and the present design cannot fully disentangle these possibilities. A general instructional-material evaluation rubric was scored for the final AR and non-AR products (see Data Analysis and Results), but this rubric assessed overall material quality against pedagogical and technical criteria rather than CT-specific process indicators. No CT-specific performance data (e.g., code complexity metrics, logged debugging behavior, or a rubric keyed to CT subskills such as decomposition or algorithmic sequencing) were collected in this study. This distinction and its implications for interpreting the present findings is revisited in the Discussion and Limitations sections.

Experimental Process

The implementation process of the study was structured within the framework of the ADDIE (Analysis-Design-Development-Implementation-Evaluation) instructional design model, which offers a comprehensive framework for the systematic planning, development, and evaluation of instructional processes through its systematic and cyclical structure (Branch, 2009). ADDIE served here as the project’s foundational instructional-design model. It was applied by the research team to plan, structure, and evaluate the overall 14-week intervention for both conditions, rather than as an analysis PSTs themselves carried out. The main rationale for selecting this model is that it allows every stage of the AR material development process — from needs analysis to evaluation — to be managed systematically and continuously improved through formative feedback.

The implementation was carried out over 14 weeks (90 minutes per week, 28 class hours in total), in line with the university’s official academic calendar. The content was planned based on constructivist and active learning approaches and the TPACK model. Consistent with the study’s focus on the material development process itself, preservice teachers’ work in both conditions was limited to designing and building the instructional materials as a course deliverable. Delivering these materials as lessons to K-12 students was outside the scope of this study and is not part of the quantitative data reported here. The progression of the process according to the ADDIE model and the chronology of quantitative data collection were as follows:

Analysis Phase: Preservice teachers’ academic, technical, and pedagogical needs related to AR, as well as their readiness levels, were determined through a preliminary needs analysis conducted using Mentimeter [THIS NEEDS A CITATION AND REFERENCE LIST ENTRY], providing input for the lesson plans.

Design Phase: Based on the analysis outputs, lesson plans, instructional objectives, and a content hierarchy aligned with the topic/learning-outcome sequencing of the 2024 Turkish Century Education Model Science Curriculum [NEEDS CITATION AND REFERENCE ENTRY] were designed. At this stage, the content validity of the process plan was strengthened by obtaining expert opinions from a total of five faculty members in the fields of Science Education (2), Computer Education and Instructional Technology (1), Measurement and Evaluation (1), and Turkish Language Education (1).

Development Phase: The 14-week lesson plans were structured, and an instructional material evaluation rubric (ranging from 0 to 4) was developed to be used throughout the process for both conditions. In line with expert opinions, simplification guidelines were added to reduce technical complexity, and free, copyright-free 3D model resources (e.g., Pixabay and Free3D) were identified for the experimental group, while an equivalent set of print- and presentation-ready resources (icon sets, diagrams, and template layouts) was identified for the control group. In addition, the technical AR training that project scholars had received through the BTK Academy [DEFINE WHAT THIS IS] was incorporated into this process.

Implementation and Evaluation Phases: These two phases were carried out concurrently within a 14-week chronological flow, combining formative assessments throughout the process with summative quantitative measurements:

Weeks 1–2: The ethical and data collection procedures of the process were explained, informed consent was obtained, and the CT and ETSSE scales were administered as pretests.

Week 3: The experimental group received introductory technology training (an orientation to the Unity editor and Vuforia workflow) and was organized into role-based subgroups of two to three students each (programmer, content coordinator,  and visual editor). During the same class session, the control group received an equivalent orientation to the tools used in its own material-design process (infographics, physical/3D models, brochures, and presentations) and was organized into comparably sized (two to three students), comparably structured role-based subgroups (content researcher, visual designer, and presenter). Thus, both conditions began the content-development weeks with a similar division of labor and the same 90-minute weekly contact time. This resulted in 13 experimental teams and 11 control teams.

This team structure applied only to the collaborative production of the instructional material and its rubric evaluation. The CT and ETSSE scales, by contrast, were completed individually by all 62 participants, so the two data sources operate at different units of analysis (team level for the rubric, individual level for the scales).

Weeks 4–9 (Organ-Based Content Development): In line with middle school-level content, both groups addressed one organ system per week (circulatory → nervous → respiratory → digestive → excretory → sensory organs; specifically, the heart, brain, lungs, stomach, intestines, and eye, respectively), so that content coverage and pacing were identical across conditions. Each week began with the same conceptual groundwork for both groups. The experimental group then proceeded through the five-step AR development cycle (Preparation, Interface Management, Target Definition, Interaction, and Build), summarized in Figure 1, actively using the Unity Editor interface and the Vuforia Developer Portal (see Figure 2).

Figure 1
From Design to Software: The 5 Basic Steps of the AR Development Process

Figure 2
Unity Editor AR Interface and Vuforia Developer Portal Technical Panels

In the same weeks, the control group covered the same organ content in parallel without using AR, producing traditional collaborative materials such as infographics, concept maps, and so on. Both groups received the same amount of in-class instructor support during these weeks. The course instructor and a research assistant circulated between the experimental and control classrooms. Both were evaluated against organ-specific milestones drawn from the same instructional material evaluation rubric described under Development Phase, applied to the AR builds in the experimental group and to the non-AR artifacts in the control group.

Troubleshooting was integral to both conditions, though its content differed: The experimental group’s troubleshooting centered on resolving Unity/Vuforia build errors and nonfunctioning interactions, while the control group’s troubleshooting centered on structural and design problems in its physical/print materials (e.g., a model that did not assemble correctly or a diagram layout that miscommunicated a process). Both types of troubleshooting were supported by the same in-class instructor circulation described previously.

Weeks 10–12: All materials were integrated into a final product (a unified AR application for the experimental group; a compiled set of instructional materials for the control group), final presentations were delivered by both groups, and peer/team feedback was provided in both conditions based on the shared rubric.

Weeks 13–14: To ensure chronological synchronicity of the measurement instruments, the CT and ETSSE scales were administered as posttests to both groups.

Across the 14 weeks, both groups, therefore, received equivalent instructor contact time (28 hours total), worked in comparably structured teams, followed the same organ-based content sequence, and were evaluated with the same rubric criteria. The intended difference between conditions was whether the resulting instructional material was built as an AR application (experimental group) or as a non-AR instructional artifact (control group). This parity in structure and support was intended to isolate the AR-specific component of the intervention. As reported under Working Group, however, the two intact sections were not statistically equivalent on either outcome measure at the outset of the study, which qualifies how this comparison can be interpreted.

Data Analysis

Analyses were carried out in SPSS 27 [ADD CITATION AND REFERENCE ENTRY]. Before proceeding, the data were checked for missing values and outliers and screened for normality using mean-median comparisons, skewness and kurtosis coefficients, and the Shapiro-Wilk test. To assess the suitability of the dataset for analysis, reverse coding transformations of the items were performed first, followed by missing data analysis. Frequency tables and the SPSS Missing Value Analysis (MVA) menu were used to identify missing data in the variables.

The analysis revealed that the number of missing responses in the dataset was limited and distributed across different items. In the first scale (30 items), missing data was detected in four cells in the pretest and two cells in the posttest. In the second scale (40 items), missing data were detected in four cells in the pretest and two cells in the posttest. In both scales, the ratio of the total number of missing values ​​in the pre- and posttests to the overall dataset was well below 1% (negligible). The distribution of missing data among the items was random, and there was no clustering of missing data in any item.

The analyses revealed that the missing data rate in the dataset was below 5% for all variables (the upper limit for assigning/completing missing data is accepted as 20% in the literature; see Tabachnick & Fidell, 2013). To test the distribution pattern of the missing data, Little’s MCAR (Missing Completely at Random) test was applied, and the p-value (sig. > .05) obtained in the EM (Expectation-Maximization) statistics confirmed that the missing data exhibited a completely random (MCAR) distribution. Due to the completely random distribution of the missing data and the fact that its rate was less than 5%, the missing values ​​were assigned (imputed) using the Series Mean method to maintain data integrity and prevent sample loss. Following the missing data assignment process, the scale scores to be used in the analyses were calculated.

Sample size and distributional characteristics guided the choice between parametric and nonparametric tests: Within-group pretest/posttest comparisons used tests suited to dependent groups, while tests appropriate for independent groups were applied in between-group comparisons. The significance level was set at p < .05 for all analyses. The dataset was confirmed to meet the normality assumption using the Shapiro-Wilk test (ETSSE pretest: SW = .982, p = .507; ETSSE posttest: SW = .968, p = .111; CT pretest: SW = .981, p = .464; CT posttest: SW = .978, p = .337). Posttest comparisons used ANCOVA, with each scale’s pretest score entered as a covariate to account for preexisting differences between the groups.

Before running each ANCOVA, the homogeneity-of-regression-slopes assumption (that the relationship between pretest and posttest scores did not differ significantly by group) was tested by entering the Group × Pretest interaction term into a preliminary model for each scale. The Group × Pretest interaction was not statistically significant for either scale, indicating the homogeneity-of-regression-slopes assumption was met: ETSSE, F(1, 58) = .666, p = .418; CT, F(1, 58) = 1.449, p = .234.

Given the baseline nonequivalence, a post hoc sensitivity analysis was also conducted to clarify how large a between-group effect the study was realistically positioned to detect, independent of whether such an effect exists. Using the achieved sample size (N = 62), the ANCOVA error degrees of freedom actually obtained (df = 59), and conventional criteria (α = .05, power = .80), the design was sensitive to between-group effects of approximately partial η² ≥ .12 (Cohen’s f ≈ .36) or larger — a medium-to-large effect by conventional benchmarks (Cohen, 1988). This threshold is used throughout the Results and Discussion to contextualize the ANCOVA findings.

The instructional material evaluation rubric was analyzed descriptively only (means, standard deviations, medians, and ranges), reported at the team level (13 experimental teams, 11 control teams; see Experimental Process). No inferential test was conducted on the rubric scores: The number of independent units at this level of analysis is small, and scores within a team are not independent of one another, since two to three students contributed to a single rated product. For this reason, the rubric findings reported here are treated as descriptive and hypothesis-generating rather than as a formal statistical comparison between conditions.

Implementation Fidelity and Control

To ensure high instructional fidelity and minimize implementation bias between the experimental and control groups, the instructional processes for both conditions were collaboratively conducted and co-attended, session by session, by two doctoral-level experts in science education, ensuring that instructional pacing, guidance level, and treatment conditions remained balanced and equitable across groups. Furthermore, to maintain procedural consistency and external oversight, the entire implementation process was systematically audited and supervised biweekly by a professor in science education.

To control for confounding variables related to the physical learning environment, all sessions for both the experimental and control groups were conducted in the university’s regular science laboratory — the setting where science education courses are routinely held for these participants.

Ethical Considerations and Data Confidentiality

This study was conducted in accordance with ethical standards for human research. To ensure voluntary participation and eliminate any potential power dynamics or coercion regarding course grades, all participants were explicitly informed that their involvement was entirely voluntary and independent of their academic assessment or course performance. To protect participant confidentiality and prevent instructor bias, a specialized individual coding technique (pseudonymization) was utilized for all data collection instruments. This mechanism ensured that the course instructor had no access to individual responses or identifying data during or after the grading period.

Results

This section presents the findings related to the subproblems of the study, in turn. As explained in the Purpose of the Study, the between-group ANCOVA comparisons are the primary results for each outcome, with the within-group paired comparisons reported as secondary, descriptive findings. Findings are addressed under two main headings: ETSSE and CT.

Findings on ETSSE

When ETSSE pretest scores were controlled as a covariate, no statistically significant difference was found between the adjusted posttest scores of the experimental and control groups, F(1, 59) = .050, p = .824, η²p = .001 (see Table 3). The adjusted (estimated marginal) means were 160.26 (SE = 3.15, 95% CI [153.95, 166.56]) for the experimental group and 159.22 (SE = 3.26, 95% CI [152.70, 165.74]) for the control group, an adjusted mean difference of 1.04 (SE = 4.66, 95% CI [−8.29, 10.37]). The extremely small partial eta squared indicates that the group effect attributable to the type of implementation was negligible in this sample; however, as detailed in Data Analysis, the sensitivity analysis indicates the study was realistically able to detect only effects of approximately η²p ≥ .12 or larger. The present result should, therefore, be interpreted as inconclusive regarding a distinguishable AR effect on ETSSE — consistent with a true absence of a between-group difference, but equally consistent with a smaller true effect that this design lacked the power to detect — rather than as evidence that the two conditions produced equivalent outcomes.

For descriptive purposes, Table 2 reports the within-group paired-samples comparisons. A statistically significant difference was found between the ETSSE pretest (M = 154.16, SD = 17.49) and posttest (M = 163.13, SD = 16.83) scores of preservice teachers in the experimental group, t(31) = −3.629, p = .001, d = −.641. A significant difference was also found between the ETSSE pretest (M = 143.59, SD = 12.54) and posttest (M = 156.16, SD = 21.43) scores of preservice teachers in the control group, t(29) = −3.096, p = .004, d = −.565. Because the experimental group began the study with significantly higher ETSSE scores than the control group (see Table 1), these within-group gains should be read as evidence that both groups’ ETSSE increased over the course of the semester, not as evidence that the two groups started from comparable positions.

Table 2
Paired-Samples t-Test Results for the Experimental and Control Groups’ ETSSE Pretest and Posttest Scores

Group	Measurement	n	M	SD	t	df	p	Cohen's d
Experimental	Pretest	32	154.16	17.49	−3.629	31	.001	−.641
	Posttest	32	163.13	16.83
Control	Pretest	30	143.59	12.54	−3.096	29	.004	−.565
	Posttest	30	156.16	21.43
Note. M = mean; SD = standard deviation. Cutoff values for Cohen’s d: .20–.49 small, .50–.79 medium, ≥ .80 large effect (Cohen, 1988).

Table 3
ANCOVA Results for ETSSE Posttest Scores, Controlling for ETSSE Pretest Scores

Source	SS	df	MS	F	p	η²p
Covariate (ETSSE Pretest)	4415.218	1	4415.218	14.726	< .001	.200
Group (Experimental/Control)	14.895	1	14.895	.050	.824	.001
Error	17689.699	59	299.825
Note. SS = Type III sum of squares; MS = mean square; η²p = partial eta squared. Adjusted (estimated marginal) means: experimental group M = 160.26 (SE = 3.15, 95% CI [153.95, 166.56]); control group M = 159.22 (SE = 3.26, 95% CI [152.70, 165.74]); adjusted mean difference = 1.04 (SE = 4.66, 95% CI [−8.29, 10.37]), p = .824. The homogeneity-of-regression-slopes check (Group × ETSSE Pretest interaction) was not significant, F(1, 58) = .666, p = .418, supporting the use of the standard ANCOVA model.

Findings on CT

When CT pretest scores were controlled as a covariate, no statistically significant difference was found between the adjusted posttest CT scores of the experimental and control groups, F(1, 59) = .323, p = .572, η²p = .005 (see Table 5). The adjusted (estimated marginal) means were 113.28 (SE = 2.15, 95% CI [108.99, 117.58]) for the experimental group and 111.48 (SE = 2.22, 95% CI [107.04, 115.93]) for the control group, an adjusted mean difference of 1.80 (SE = 3.16, 95% CI [−4.53, 8.13]). As with ETSSE, this small effect size, combined with the sensitivity analysis reported in Data Analysis, means the present study cannot determine whether the AR material development process produces a between-group advantage in CT beyond that of the non-AR condition.

For descriptive purposes, Table 4 reports the within-group paired comparisons. A statistically significant difference was found between the CT pretest (M = 107.25, SD = 12.31) and posttest (M = 115.28, SD = 12.89) scores of preservice teachers in the experimental group, t(31) = −4.197, p < .001, d = −.742. A significant difference was also found between the CT pretest (M = 99.22, SD = 12.98) and posttest (M = 109.35, SD = 13.97) scores in the control group, t(29) = −3.595, p = .001, d = −.656. As with ETSSE, the experimental group began the study with significantly higher CT scores than the control group (see Table 1), which qualifies how these within-group gains, and the modest difference in effect sizes between groups, should be interpreted (see Discussion). These within-group gains should be read as evidence that both groups’ CT increased over the course of the semester, not as evidence that the two groups started from comparable positions.

Table 4
Paired-Samples t-Test Results for the Experimental and Control Groups’ CT Pretest and Posttest Scores

Group	Measurement	n	M	SD	t	df	p	Cohen's d
Experimental	Pretest	32	107.25	12.31	−4.197	31	< .001	−.742
	Posttest	32	115.28	12.89
Control	Pretest	30	99.22	12.98	−3.595	29	.001	−.656
	Posttest	30	109.35	13.97
Note. M = mean; SD = standard deviation. Cutoff values for Cohen’s d: .20–.49 small, .50–.79 medium, ≥ .80 large effect (Cohen, 1988).

Table 5
ANCOVA Results for CT Posttest Scores, Controlling for CT Pretest Scores

Source	SS	df	MS	F	p	η²p
Covariate (CT Pretest)	2536.615	1	2536.615	18.087	< .001	.235
Group (Experimental/Control)	45.335	1	45.335	.323	.572	.005
Error	8274.613	59	140.248
Note. SS = Type III sum of squares; MS = mean square; η²p = partial eta squared. Adjusted (estimated marginal) means: experimental group M = 113.28 (SE = 2.15, 95% CI [108.99, 117.58]); control group M = 111.48 (SE = 2.22, 95% CI [107.04, 115.93]); adjusted mean difference = 1.80 (SE = 3.16, 95% CI [−4.53, 8.13]), p = .572. The homogeneity-of-regression-slopes check (Group × CT Pretest interaction) was not significant, F(1, 58) = 1.449, p = .234, supporting the use of the standard ANCOVA model.

Descriptive Analysis Results of the Instructional Material Evaluation Rubric

Throughout the research process, the final instructional materials produced by both groups were evaluated using a rubric graded out of a maximum score of 32 points. Because the AR materials and non-AR materials were produced collaboratively by two to three student teams rather than by individuals (see Experimental Process), the rubric was scored at the team level, yielding 13 experimental teams and 11 control teams, rather than the 32/30 individual-level Ns used for the CT and ETSSE scales. The descriptive statistics regarding these team-level product evaluations revealed distinct distributional patterns. The experimental group (AR-based material development) achieved a mean score of 30.62 (SD = 2.50) with a median of 32.00, ranging from a minimum of 24.00 to a maximum of 32.00, which corresponds to 95.7% of the maximum possible score. Notably, nine of the 13 experimental teams attained the maximum score of 32. In contrast, the control group (traditional material development) yielded a mean score of 28.82 (SD = 4.26) with a median of 29.00, ranging from a minimum of 17.00 to a maximum of 32.00, representing 90.1% of the maximum score.

Descriptively, the experimental teams’ scores clustered more tightly near the ceiling, while the control teams showed wider dispersion, including a lower outlier score of 17 that pulled down that group’s mean and standard deviation. As noted in Data Analysis, this comparison was not tested inferentially, given the small number of independent teams and the nonindependence of scores within a team, so it is reported as a descriptive, hypothesis-generating pattern rather than a confirmed between-group difference.

Dıscussıon and Conclusıons

This study examined the effect of a 14-week AR material development process for preservice science teachers on CT skills and ETSSE levels, relative to a comparably structured non-AR design experience. The findings are shaped around three main patterns: (a) statistically significant increases from pretest to posttest were observed in both groups on both outcomes; (b) the experimental and control groups were not statistically equivalent at pretest on either outcome, which qualifies how the within-group gains can be interpreted and the use of ANCOVA; and (c) no statistically significant between-group difference was found in the ANCOVA-controlled comparison for either outcome, though the sensitivity analysis reported in the method section indicates the study was realistically powered to detect only medium to large between-group effects. Taken together, these results support a cautious interpretation: structured instructional material design was associated with gains in both ETSSE and CT regardless of condition, and the present data do not provide clear evidence that the AR-specific process added a distinguishable benefit beyond this general design-based learning experience. This interpretation is deliberately more conservative than a reading built only around the within-group results, in light of the ANCOVA findings and the baseline nonequivalence between groups.

Discussion of the ETSSE Findings

Because the experimental group began the study with significantly higher ETSSE scores than the control group (see Table 1), the within-group gain reported for the experimental group is best read as evidence that its members’ confidence increased further from an already-elevated starting point, not as evidence of a larger absolute change than the control group; the two groups’ gains were, in fact, similar in size (d = −.641 etc. d = −.565). With that qualification, the significant increase in the experimental group’s ETSSE scores points to a strengthening of preservice teachers’ technology-related self-efficacy following the AR development process.

Bandura’s (1997) self-efficacy theory offers a fitting lens here: Successfully mastering a task in a given domain tends to reinforce one’s belief in one’s own competence within that domain. Given that participants spent an extended period building and completing a working AR application in Unity and Vuforia, a positive shift in ETSSE following this mastery experience is unsurprising.

A comparable pattern emerged in Dönmez Usta et al.’s (2024) TÜBİTAK-supported study with Turkish preservice science teachers, where engaging in digital material development likewise raised participants’ self-efficacy regarding educational technology standards. Belda-Medina and Calvo-Ferrer (2022) reported a similar pattern among preservice language teachers, finding that an AR design task framed around TPACK improved digital competence and attitudes toward technology. Papadakis and Kalogiannakis (2020) similarly found that direct, hands-on engagement with an emerging classroom technology shaped preservice teachers’ attitudes more than passive exposure would.

The observation of a similarly significant ETSSE increase in the control group, together with the absence of a significant between-group difference in the ANCOVA, is a noteworthy finding in its own right rather than a null result to be explained away. This can be interpreted from several perspectives. First, both groups’ enrollment in the Science Teaching I course during the same semester, and the cumulative effect of the overall undergraduate education process on self-efficacy, may explain this parallel development. Indeed, Birisci and Kul (2019) found that technopedagogical competency significantly predicted preservice teachers’ technology integration self-efficacy.

Second, because the control group also developed a range of instructional materials — infographics, technology-related materials and physical models — under comparable instructor support and team structure to the experimental group (see Experimental Process), the design-based learning experience itself, rather than AR specifically, may have contributed to ETSSE development. This interpretation aligns with a recent meta-analysis (Zeng et al., 2022), which found that TPACK and information-technology-integration self-efficacy were significantly and positively correlated.

A subdimension perspective helps refine this interpretation further. Because facilitating students’ learning and creativity and designing digital-age learning environments are the ETSSE subdimensions most directly rehearsed by producing an interactive instructional material — AR-based or not — both groups’ shared engagement in structured material design offers a coherent explanation for parallel gains on the total score even in the absence of a between-group AR effect. Modeling digital citizenship, the subdimension least directly exercised by either condition, would be a natural candidate for a smaller or nonsignificant contribution to the overall change. Testing this directly would require subdimension-level analysis that the present total-score design did not undertake, but it offers one concrete, falsifiable extension of the theoretical account proposed here (see Conclusion and Recommendations).

Discussion of the CT Findings

The within-group increase in the experimental group’s CT scores (d = −.742) was descriptively somewhat larger than that of the control group (d = −.656). This small difference, however, should not be read as evidence of a differential AR effect. The same baseline nonequivalence noted for ETSSE was also present for CT (see Table 1), the direct between-group comparison (ANCOVA) found no statistically significant difference once pretest scores were controlled, and the sensitivity analysis reported in the method section indicates limited power to detect a true effect of this magnitude. The modest gap in within-group effect sizes is, therefore, treated here as a descriptive, hypothesis-generating observation rather than as evidence of a distinguishable AR benefit.

It is useful to be explicit about the boundary between what the present data show and what can only be theorized. The data show (a) statistically significant within-group pre–post gains in CT self-report scores for both groups; (b) a numerically larger within-group effect size in the experimental group; and (c) no statistically significant between-group difference once pretest scores were controlled. The data do not show, and were not designed to show, that particular AR-development activities, such as writing or modifying C# scripts to control object behavior, or iteratively debugging a non-functioning AR build, caused any specific component of CT (e.g., abstraction, algorithmic thinking, and decomposition) to develop.

It is theoretically plausible, and consistent with Wing’s (2006) account of CT and with constructionist, design-based learning perspectives, that tasks such as modeling three-dimensional objects, setting up trigger images, and troubleshooting a Unity/Vuforia build exercise algorithmic thinking, abstraction, and debugging more directly than infographic or concept-map design would. Cheng et al.’s (2023) meta-analytic finding that technology-based design activities are generally associated with CT gains is consistent with this account and with the significant within-group increase observed in the experimental group here. However, this remains a theoretical interpretation grounded in the nature of the AR development task, not a claim directly supported by the present data. The study measured self-reported CT scores before and after the intervention, not observed debugging behavior, code or model quality, or any other performance-based, process-level indicator of computational thinking during the design process itself.

Moving this account from plausible theory to empirical claim would require process-level evidence, such as logged debugging behavior, think-aloud protocols, qualitative analysis of the design process, or a rubric keyed specifically to CT subskills rather than to overall material quality, none of which were collected in the present study. A general instructional-material-quality rubric was collected and is reported descriptively, but it does not substitute for these CT-specific indicators (see Limitations).

With respect to the primary between-group comparison, the ANCOVA result did not reveal a significant difference in CT, F(1, 59) = .323, p = .572, η²p = .005. Two considerations help make sense of this finding, independent of the theoretical account. The first is that designing instructional materials, regardless of whether AR is involved, appears to engage algorithmic thinking and problem-solving in its own right; Wang et al. (2022) similarly argued that CT in STEM education emerges from design-based work generally, not only from coding tasks, specifically. Since the control group also went through planning, sequencing, and evaluation while building its own materials, under comparable instructor support to the experimental group, it is not surprising that the between-group comparison failed to reach significance.

The second consideration is statistical power: the achieved sample size (N = 62) and the resulting sensitivity threshold (partial η² ≥ ~.12) mean the study was not well positioned to detect a small effect. Cheng et al.’s (2023) meta-analysis similarly found that CT intervention effects scale with both sample size and implementation length, reinforcing the need for adequately powered future designs before concluding that AR development and non-AR design produce equivalent CT outcomes.

Positioned within the broader literature, this pattern is not unusual. International reviews of design-based and maker-oriented interventions targeting CT in teacher preparation similarly report that structured design tasks — coding-based or not — tend to produce within-group gains that are difficult to attribute to any single tool once compared against an active, similarly structured alternative (Wang et al., 2022; Cheng et al., 2023).

The instructional material evaluation rubric offers a further, complementary vantage point on the intervention, though it should be weighted according to its methodological limits rather than treated as a stronger form of evidence than the scale-based results. Descriptively, the experimental teams’ rubric scores (M = 30.62, SD = 2.50, 13 teams) clustered closer to the ceiling of 32 than the control teams’ scores (M = 28.82, SD = 4.26, 11 teams), which included a lower outlier of 17 (see Results). Because this comparison was not tested inferentially — the number of independent teams is small, and scores within a team are shared across its two to three members — it cannot be treated as confirmatory evidence of an AR advantage in product quality, and it does not, on its own, resolve the non-significant ANCOVA findings on ETSSE and CT.

With that caveat, the pattern is a plausible, hypothesis-generating complement to the self-report results. It is consistent with the interpretation that the AR development cycle’s more rigid technical requirements — a build must compile and run correctly, or it visibly fails — may impose a floor on product quality that the more open-ended non-AR format does not, even where the two conditions do not differ in participants’ self-reported confidence. Testing this interpretation directly would require an adequately powered, individually or randomly assigned design with a validated, inferentially analyzed rubric, a natural next step for future research (see Conclusion and Recommendations).

Limitations of the Study

Several limitations should be kept in mind when interpreting this study. First, the quasi-experimental design relied on intact class sections rather than individual random assignment, and the two sections were not statistically equivalent at pretest on either outcome measure (see Working Group). Although ANCOVA was used to adjust for this, ANCOVA cannot fully substitute for random assignment, and residual confounding by unmeasured section-level differences (e.g., prior coursework, group dynamics, or instructor rapport) cannot be ruled out.

Second, both outcome measures were self-report scales capturing perceived competence and confidence rather than directly observed performance. This is a particular concern for computational thinking, where performance-based indicators (e.g., artifact quality, code structure, or debugging behavior) would provide more direct evidence of skill change than self-report alone. A general instructional-material quality rubric was collected (see Results), but no CT-specific performance data (e.g., code complexity metrics or logged debugging behavior) were collected in the present study, and this should be treated as a substantive limitation rather than a minor caveat.

Third, the post hoc sensitivity analysis indicates that, given the achieved sample size, the study was powered to detect only medium-to-large between-group effects (partial η² ≥.12). Consequently, smaller yet meaningful between-group effects may have gone undetected. Therefore, the nonsignificant ANCOVA results should be interpreted with caution rather than as evidence of equivalence between conditions.

Fourth, the implementation period was confined to 14 weeks within a single semester at a single university’s science education program, which limits the generalizability of the findings. Fifth, the learning curve associated with relatively technical tools such as Unity and Vuforia may have affected some preservice teachers’ individual contribution to the process in the experimental group, and individual-level variation in technical contribution was not captured.

Finally, although the control condition was designed to match the experimental condition in instructor contact time, team structure, and rubric-based evaluation, some unmeasured aspects of the two design experiences (e.g., perceived novelty, motivation, or workload) may still have differed and could have contributed to the observed pattern of results. Because the study spanned a single unit within one semester, the possibility that heightened engagement in the experimental group partly reflects a novelty effect associated with an unfamiliar technology, rather than a stable AR-specific benefit, cannot be ruled out and would need to be examined through longitudinal designs. An additional limitation concerns the instructional material evaluation rubric. Because materials were produced by two- to three-student teams rather than by individuals, rubric scores were available for only 13 experimental and 11 control teams, individual team members’ scale responses are not independent of their teammates’ shared product score, and the descriptive between-group difference in rubric scores was not tested inferentially. This rubric evidence should, therefore, be read as suggestive rather than confirmatory, and future work should analyze such data with methods appropriate to its nested structure (e.g., multilevel models) and an adequately powered number of teams.

Conclusion and Recommendations

This study revealed that both the AR material development process and the non-AR instructional design experience led to statistically significant gains in preservice science teachers’ ETSSE and CT self-report scores. Once preexisting group differences were statistically controlled, no significant between-group difference emerged for either outcome, and a post hoc sensitivity analysis indicates the study was realistically powered to detect only medium-to-large between-group effects. Rather than treating this nonsignificant result as evidence that AR development confers no distinct benefit, the conclusion is that engaging preservice teachers in structured, hands-on instructional material design — independent of the specific technology involved — is associated with gains in technology self-efficacy and computational thinking self-perceptions, and that the present study cannot determine whether AR development added a further, distinguishable benefit beyond this general design-based effect. The slightly larger within-group effect size for CT in the experimental group is a suggestive, hypothesis-generating pattern rather than confirmatory evidence of an AR-specific advantage, given the nonsignificant between-group result and the baseline nonequivalence between groups.

Internationally, teacher-education programs are under increasing pressure to justify technology-integration coursework empirically rather than on the assumption that any hands-on technology experience is beneficial. The present findings support hands-on, design-based instructional-material development broadly, while leaving open, rather than resolving, the narrower question of whether AR specifically confers an advantage over other design-based technologies once implementation quality and instructor support are held comparable.

The descriptive rubric findings reported in the Results and Discussion section (higher, more tightly clustered scores for AR teams versus a wider, outlier-affected distribution for control teams) offer a suggestive, hypothesis-generating complement to the self-report results, but given the small number of independent teams and the nonindependence of scores within a team, they cannot be treated as confirmatory evidence of an AR-specific advantage in product quality, and they do not change the substantive conclusion: Within the limits of the present design, structured hands-on material design is associated with self-reported gains regardless of technology, while a distinguishable AR-specific effect remains unconfirmed.

These findings point to several directions for future research and practice. Future studies should prioritize statistical power when testing the between-group comparison, using larger samples, longer implementation periods, and where feasible, randomization at the level of individual students rather than intact sections. Where intact sections must be used, preregistering the baseline-equivalence and sensitivity analyses reported here is recommended. Incorporating performance-based or process-level measures of computational thinking — rubric-scored artifacts, logged debugging behavior, or qualitative analysis of the design process — alongside self-report scales would allow future work to test the theoretical mechanisms proposed in the Discussion section directly, rather than inferring them from outcome scores alone.

Analyzing CT and ETSSE at the sub-dimension level, rather than as total scores, is a further concrete extension that could reveal a more AR-specific pattern within particular factors (e.g., Robotics-Coding and Software or modeling digital-age work and learning) even where total-score comparisons do not. Mixed-methods designs could also examine, in depth, preservice teachers’ subjective experiences of the AR development process relative to the comparison condition. Comparative studies could be conducted using different science topics, educational levels, and AR development tools, explicitly framing the non-AR condition as an active comparison rather than a passive control, as was done here.

For practitioners, it is recommended that science teacher education programs integrate hands-on courses covering structured instructional-material design, AR-based tools such as Unity and Vuforia among them, as a means of strengthening preservice teachers’ TPACK competencies and computational thinking, while recognizing that the present evidence supports the value of structured design experience broadly rather than establishing AR as uniquely superior to other design-based approaches.

Author Notes

This research was supported by TÜBİTAK 3005 – the Research Project Support Program for Innovative Solutions in Social and Human Sciences (Project No. 323K097). We thank TÜBİTAK for its contributions.

This research was supported by TÜBİTAK 3005 – the Research Project Support Program for Innovative Solutions in Social and Human Sciences (Project No. 323K097).

References

Aho, A. V. (2012). Computation and computational thinking. The Computer Journal, 55(7), 832–835. https://doi.org/10.1093/comjnl/bxs074

Akçayır, M., & Akçayır, G. (2017). Advantages and challenges associated with augmented reality for education: A systematic review of the literature. Educational Research Review, 20, 1–11. https://doi.org/10.1016/j.edurev.2016.11.002

Azuma, R. T. (1997). A survey of augmented reality. Presence: Teleoperators and Virtual Environments, 6(4), 355–385. https://doi.org/10.1162/pres.1997.6.4.355

Bandura, A. (1997). Self-efficacy: The exercise of control. Freeman.

Belda-Medina, J., & Calvo-Ferrer, J. R. (2022). Integrating augmented reality in language learning: Pre-service teachers’ digital competence and attitudes through the TPACK framework. Education and Information Technologies, 27(9), 12123–12146. https://doi.org/10.1007/s10639-022-11123-3

Birisci, S., & Kul, U. (2019). Predictors of technology integration self-efficacy beliefs of preservice teachers. Contemporary Educational Technology, 10(1), 75–93. https://doi.org/10.30935/cet.512537

Branch, R. M. (2009). Instructional design: The ADDIE approach. Springer. https://doi.org/10.1007/978-0-387-09506-6

Büyüköztürk, Ş., Çakmak, E. K., Akgün, Ö. E., Karadeniz, Ş., & Demirel, F. (2020). Bilimsel araştırma yöntemleri [Scientific research methods] (29th ed.). Pegem Akademi.

Cheng, L., Wang, X., & Ritzhaupt, A. D. (2023). The effects of computational thinking integration in STEM on students’ learning performance in K-12 education: A meta-analysis. Journal of Educational Computing Research, 61(2), 416–443. https://doi.org/10.1177/07356331221114183

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates.

Dönmez Usta, N., Turan Güntepe, E., & Durukan, Ü. G. (2024). Examining the self-efficacy of pre-service science teachers concerning digital material development, technology integration, and educational standards. Open Praxis, 16(3), 410–428. https://doi.org/10.55982/openpraxis.16.3.682

Erbaş, C., & Demirer, V. (2014). Eğitimde artırılmış gerçeklik uygulamaları: Google Glass örneği [Augmented reality applications in education: The Google Glass example]. Journal of Instructional Technologies & Teacher Education, 3(2), 8–16.

Ertuğrul-Akyol, B. (2019). Development of computational thinking scale: Validity and reliability study. International Journal of Educational Methodology, 5(3), 421–432. https://doi.org/10.12973/ijem.5.3.421

Korucu, A. T., Gençtürk, T., & Sezer, C. (2016). Artırılmış gerçeklik uygulamalarının öğrenci başarı ve tutumlarına etkisi [The effect of augmented reality applications on student achievement and attitudes] (Paper presentation). XVIII. Akademik Bilişim Conference, Aydın, Turkey.

Küçük, S. (2015). Mobil artırılmış gerçeklikle anatomi öğreniminin tıp öğrencilerinin akademik başarıları ile bilişsel yüklerine etkisi ve öğrencilerin uygulamaya yönelik görüşleri [The effect of anatomy learning with mobile augmented reality on medical students’ academic achievement and cognitive load, and students’ views on the application] (Unpublished doctoral dissertation). Atatürk University, Turkey.

Louka, K., & Papadakis, S. (2024). Enhancing computational thinking in early childhood education through ScratchJr integration. Heliyon, 10(10), e30482. https://doi.org/10.1016/j.heliyon.2024.e30482

Mishra, P., & Koehler, M. J. (2006). Technological pedagogical content knowledge: A framework for teacher knowledge. Teachers College Record, 108(6), 1017–1054. https://doi.org/10.1111/j.1467-9620.2006.00684.x

Papadakis, S., & Kalogiannakis, M. (2020). Exploring preservice teachers’ attitudes about the usage of educational robotics in preschool education. In M. Kalogiannakis & S. Papadakis (Eds.), Handbook of research on tools for teaching computational thinking in P-12 education (pp. 339–355). IGI Global. https://doi.org/10.4018/978-1-7998-4576-8.ch013

Papadakis, S., & Orfanakis, V. (2018). Comparing novice programming environments for use in secondary education: App Inventor for Android vs. Alice. International Journal of Technology Enhanced Learning, 10(1–2), 44–72. https://doi.org/10.1504/IJTEL.2018.088333

PTC Inc. (2023). Vuforia engine developer portal. https://developer.vuforia.com

Saidin, N. F., Halim, N. D. A., & Yahaya, N. (2015). A review of research on augmented reality in education: Advantages and applications. International Education Studies, 8(13), 1–8. https://doi.org/10.5539/ies.v8n13p1

Somyürek, S. (2014). Öğretim sürecinde Z kuşağının dikkatini çekme: Artırılmış gerçeklik [Capturing Generation Z’s attention in the instructional process: Augmented reality]. Eğitim Teknolojisi Kuram ve Uygulama, 4(1), 63–80. https://doi.org/10.17943/etku.88319

Şengün Öztaş, E. (2024). Eğitimde artırılmış gerçeklik teknolojisi kullanımı: 2014-2024 sistematik alanyazın taraması [The use of augmented reality technology in education: A systematic literature review, 2014-2024]. Ufuk Üniversitesi Sosyal Bilimler Enstitüsü Dergisi, 13(26), 33–54. https://doi.org/10.58635/ufuksbedergi.1594889

Şimşek, O., & Yazar, T. (2016). Education technology standards self-efficacy (ETSSE) scale: A validity and reliability study. Eurasian Journal of Educational Research, 63, 311–334. https://doi.org/10.14689/ejer.2016.63.18

Tabachnick & Fidell, 2013

Tosik Gün, E., & Atasoy, B. (2017). The effects of augmented reality on elementary school students’ spatial ability and academic achievement. Education and Science, 42(191), 31-51. https://doi.org/10.15390/EB.2017.7140

Unity Technologies. (2023). Unity real-time development platform. https://unity.com

Wang, C., Shen, J., & Chao, J. (2022). Integrating computational thinking in STEM education: A literature review. International Journal of Science and Mathematics Education, 20(8), 1949–1972. https://doi.org/10.1007/s10763-021-10227-5

Wing, J. M. (2006). Computational thinking. Communications of the ACM, 49(3), 33–35. https://doi.org/10.1145/1118178.1118215

Wu, H. K., Lee, S. W. Y., Chang, H. Y., & Liang, J. C. (2013). Current status, opportunities and challenges of augmented reality in education. Computers & Education, 62, 41–49. https://doi.org/10.1016/j.compedu.2012.10.024

Zeng, Y., Wang, Y., & Li, S. (2022). The relationship between teachers’ information technology integration self-efficacy and TPACK: A meta-analysis. Frontiers in Psychology, 13, 1091017. https://doi.org/10.3389/fpsyg.2022.1091017