Predict–Observe–Explain (POE) in Science Education: A Narrative Review of Mechanisms, Variants, Evidence, and Disciplinary Adaptations

Abstract

Predict–Observe–Explain (POE) is an instructional strategy used across science-education contexts to structure learning around prediction, observation, and explanation. This narrative review synthesizes research on the canonical POE sequence and its named derivatives. It interprets the three phases as complementary mechanisms: prediction creates epistemic commitment and reveals prior reasoning; observation provides evidence for comparison and may, but need not, generate discrepancy; and explanation supports reconciliation, model refinement, and transfer. The review also examines major POE variants and critically evaluates evidence of effectiveness. Methodologically stronger comparative studies generally report benefits for conceptual understanding, misconception reduction, achievement, or higher-order skills, although findings remain heterogeneous because many interventions combine POE with simulations, games, problem-based learning, or additional phases, rely on intact classes and locally developed measures, and provide limited delayed or transfer evidence. The disciplinary analysis deliberately focuses on physics, chemistry, and biology/biochemistry to examine how POE changes when relevant evidence is directly observable, instrument-mediated, representational, or model-generated. POE is most defensible when predictions require reasons, observations make relevant evidence epistemically visible, explanations provide intelligible alternatives, and follow-up tasks test whether reconstructed ideas remain usable beyond the initial event.

1. Introduction

Predict–Observe–Explain (POE) has been used in science education across different scientific domains, educational stages, and instructional settings. In its canonical formulation, learners state what they expect to occur, observe an event or the output of a model, and explain the relationship between their expectation and the available evidence [1]. The sequence can be implemented through live demonstrations, laboratory investigations, worksheets, interactive videos, computer simulations, virtual or augmented reality, probeware, or remote experiments. This adaptability helps account for the breadth of settings represented in the literature.

The three phases, however, describe more than a procedural order. Prediction elicits learners’ existing explanatory resources before the outcome becomes available, whether through a teacher demonstration, a learner-conducted investigation, or the output of a model or simulation. Observation establishes a basis for comparison, but only when the evidence relevant to the prediction is perceptible, trustworthy, and interpreted rather than merely watched. Explanation requires learners to coordinate their initial prediction, the reasoning that supported it, and the observed result with a scientifically productive account. Haysom and Bowen [2] make this broader instructional grammar explicit by embedding the three phases within orientation, written justification, discussion, presentation of the current scientific explanation, and follow-up. Their elaboration underscores that a brief prediction question followed by a striking demonstration does not, by itself, constitute a complete POE sequence. This distinction matters for conceptual change because learners may interpret observations through persistent alternative conceptions and reject or reinterpret evidence that challenges their prior explanations [3,4].

The literature has expanded well beyond its early concentration on mechanics and thermal phenomena. POE has been applied across different scientific disciplines, educational levels, and instructional settings, from classroom demonstrations and laboratory work to digital and model-based environments. The classical sequence has also generated numerous derivatives that incorporate discussion, revision, diagnosis, writing, reflection, problem-based learning, project work, collaboration, or differentiation. This expansion has widened the instructional possibilities of POE, but it has also created two analytical problems. Activities that differ substantially in duration, scaffolding, media, and pedagogical purpose are sometimes treated as equivalent because they retain the Predict–Observe–Explain label; conversely, favorable outcomes from complex interventions may be attributed to POE even when simulations, collaborative structures, additional phases, or broader pedagogical models are plausible contributors.

These developments justify a synthesis that separates the instructional logic of canonical POE from the functions of its principal variants and from the evidence supporting their effectiveness. The present article therefore adopts a narrative and integrative review design. Its aim is not to calculate a pooled effect size or claim exhaustive systematic coverage, but to construct a conceptually organized and critically informed account of POE in science education. Particular attention is given to the pedagogical work performed by the three phases, the instructional problems addressed by named derivatives, the strength and limitations of the effectiveness evidence, and the ways in which POE is adapted to different forms of scientific evidence and representation.

The breadth of POE in the research literature should not be equated with uniform or pedagogically robust classroom implementation. The apparent simplicity of the three-phase sequence can conceal substantial differences in how predictions are elicited, evidence is made interpretable, explanations are supported, and follow-up is provided. Teacher-focused evidence also suggests that familiarity with POE does not necessarily translate into confident implementation: Mtsi et al. [5] reported that natural-science teachers recognized the approach but identified gaps in procedural knowledge, resources, and professional development. These considerations strengthen the need to distinguish the surface sequence from the instructional conditions that make POE educationally meaningful.

The overarching question is as follows: Under what conditions can POE support science learning, and how do its mechanisms, variants, effectiveness evidence, and disciplinary adaptations inform instructional design? To address this general question, the review considers four specific questions. First, what pedagogical work is performed by the Predict, Observe, and Explain phases? Second, what instructional problems are the main variants of the classical POE sequence designed to solve? Third, what do the methodologically stronger studies allow us to conclude about the effects of canonical POE and its derivatives? Fourth, what do selected disciplinary adaptations in physics, chemistry, and biology/biochemistry reveal about how POE responds to different forms of scientific evidence and explanation? Section 3, Section 4, Section 5 and Section 6 address these four questions in sequence; Section 7 integrates practical design principles and common failure modes, and Section 8 identifies priorities for future research.

2. Materials and Methods

This article adopts a critical narrative review because its questions require an interpretive synthesis across theoretical accounts, empirical comparisons, instructional-design studies, and subject-specific applications. The aim is to clarify mechanisms, qualify causal claims, and draw transferable design implications. Source selection for individual citation in the narrative synthesis was purposive, and the synthesis integrates empirical findings with contextual theoretical literature. The distinction between the bibliographic corpus and the sources individually cited is made explicit below.

The POE-specific search and eligibility criteria described below apply only to publications that explicitly examined POE or a named POE-derived sequence. Theoretical and contextual sources used to interpret prediction, conceptual change, discrepant events, Nature of Science (NoS), teacher preparation, or review methodology were identified purposively according to their relevance to the mechanisms, design conditions, or review procedures being discussed and are not included in the numerical POE corpus. The preparation and reporting of the review were informed by the quality domains proposed in the Scale for the Assessment of Narrative Review Articles (SANRA), particularly transparent search description, critical scientific reasoning, and appropriate presentation of relevant evidence [6].

The POE-specific literature was identified through iterative searches conducted in Web of Science Core Collection, Scopus, ERIC, and Google Scholar. Search-term combinations used the full name of the strategy with common lexical variants, disciplinary terms, and the names and acronyms of the principal POE derivatives discussed in Section 4. Disciplinary and implementation-specific combinations included terms such as “POE physics,” “POE chemistry,” “POE biology,” “POE virtual laboratory,” “POE simulation,” and “POE teacher education.” Backward and forward citation tracking from the canonical chapter and from frequently cited empirical studies was used to identify earlier publications, regional studies, and work on specific variants. The final literature search and retrieval update was completed on 31 July 2026.

Records were deduplicated using title, authorship, year, and DOI. Titles and abstracts were screened initially, followed by full-text examination. Publications were included when they: (a) explicitly implemented or substantively analyzed canonical POE or a named POE derivative in science education; (b) reported empirical findings, provided a sufficiently detailed design study, or constituted a foundational formulation; and (c) supplied enough bibliographic and methodological information to determine the educational context and the nature of the contribution. Eligible publication types comprised peer-reviewed journal articles, scholarly books or book chapters, and selected conference proceedings when they provided sufficient methodological and outcome detail. The literature searches, screening, and initial selection of publications were conducted by one author, while the selection decisions and resulting corpus were reviewed by the other two coauthors.

Records were excluded during screening when POE referred to an unrelated acronym; prediction was examined without an Observe–Explain structure; the study fell outside science education; or the publication duplicated another version of the same work. Of the 215 records identified, 72 were excluded during screening, leaving 143 for full-text eligibility assessment. A further 15 publications were excluded after full-text examination because they did not provide sufficient information to establish the nature of their contribution or, for empirical studies, the intervention, context, or findings. Screening and full-text eligibility assessment yielded a direct POE corpus of 128 eligible publications published between 1992 and July 2026. The complete bibliographic inventory is provided in Supplementary Table S1. Corpus inclusion and citation in the narrative synthesis were treated as distinct decisions. The 128 records were used to map the breadth of POE research. Of these, 57 publications were selected for individual citation in the narrative synthesis because they contributed distinctly to at least one of the four review questions, provided informative design or outcome evidence, elucidated named POE variants, or illustrated discipline-specific evidential problems. The remaining 71 eligible records informed corpus mapping but were not individually cited when they substantially duplicated applications or findings already represented by more informative sources, provided limited methodological or outcome detail, or added little to the conceptual argument. The eligible corpus is therefore broader than the set of publications cited in the article. It spans controlled experiments, quasi-experiments, qualitative studies, action research, design and development studies, diagnostic applications, and classroom-tested activities. The search and selection process, including the number of records retained at each stage, is summarized in Table 1.

Table 1. Search and selection overview.

Stage	Sources/Procedure	Criteria	Output
Identification	Web of Science, Scopus, ERIC, Google Scholar, complemented by citation tracking	Publications addressing POE in science education, including its theoretical foundations, instructional applications, variants, and educational outcomes	Records identified (n = 215)
Screening	Duplicate removal and title/abstract screening	Exclusion of duplicate publications, unrelated uses of the POE acronym, prediction studies lacking an Observe–Explain structure, and publications outside science education	72 excluded. Records retained for eligibility assessment (n = 143)
Eligibility	Full-text examination of the retained publication and thematic fit check	Exclusion of publications providing insufficient information to establish their contribution or, for empirical studies, the intervention, context, or findings.	15 excluded after full text. Eligible publications (n = 128)
Included	Corpus consolidation and selective individual citation in the narrative synthesis.	Individual citation prioritized publications contributing distinctly to the review questions, providing informative design or outcome evidence, elucidating named POE variants, or illustrating discipline-specific evidential problems	Corpus (n = 128); individually cited (n = 57); not individually cited (n = 71)

The search process also has limitations. Regional journals and conference proceedings are unevenly indexed, terminology varies across countries, and some derivatives do not include “POE” in their titles. English-language abstracts made it possible to consider publications written in several languages, but studies without an accessible abstract or verifiable metadata may have been missed. The corpus should therefore be understood as an extensive and traceable body of directly relevant work rather than as a definitive census of all POE-related publications.

3. POE as an Instructional Grammar: What the Three Phases Do

White and Gunstone’s [1] chapter provides the canonical reference for POE, but the instructional integrity of the strategy depends on how the phases are enacted. A prediction without reasons becomes guessing; an observation without attention to evidence becomes watching; and an explanation that consists only of the teacher announcing the answer leaves little room for reconstruction. The phases are therefore best understood as a coordinated grammar for connecting prior thinking, evidence, and scientific explanation. The following sections examine the distinctive instructional work performed by Predict, Observe, and Explain; issues of follow-up, scaffolding, and implementation failure are addressed subsequently in the design principles.

3.1. Predict: Epistemic Commitment, Diagnosis, and Generative Comparison

Prediction is the feature that most clearly distinguishes POE from an ordinary demonstration. Before the outcome is available, learners must take a position and expose at least part of the model on which that position rests. This creates epistemic commitment: the later observation can be interpreted as feedback on the learner’s own account rather than simply as information delivered by the teacher. Prediction also performs a diagnostic function because identical answers may be supported by different mechanisms. Asking “What will happen, and why?” is consequently more informative than asking students to select an outcome without justification [1,2].

Earlier research helps clarify the instructional value of eliciting expectations before observation. Gunstone and White [7] investigated students’ predictive and explanatory reasoning about gravity and showed the persistence of intuitive interpretations. Champagne et al. [8], using an instrument labelled Demonstration–Observation–Explanation in mechanics, also included tasks in which students predicted motion before observing and justifying the event. These studies illustrate why pre-observation commitment is diagnostically useful: students’ responses can reveal organized commonsense reasoning rather than isolated errors. In a robust POE sequence, the expectation and its rationale become explicit objects of subsequent comparison with evidence and instructional discussion.

Research on prediction as a learning strategy reinforces this interpretation. Generating a prediction can orient attention toward later feedback and establish a discrepancy signal between expected and observed information [9]. It also narrows the range of features that must be monitored during observation: once learners have committed to an outcome and a reason, they have a specific evidential question to pursue. Learning benefits are most likely when the prediction is plausible on the basis of prior knowledge and the feedback can be interpreted; a meaningless guess provides little structure for revision, whereas a confident but explicable error can make the relevant contrast salient [10]. POE tasks should therefore be calibrated between trivial recall and opaque guessing.

Prediction also has a social function. When reasons are shared before observation, the class sees that more than one interpretation is possible and that evidence is needed to adjudicate among them. Haysom and Bowen [2] recommend first eliciting individual written predictions and then discussing the range of responses with sensitivity, since public error can inhibit participation. Multimedia POE work similarly shows that externalizing predictions creates material for justification, critique, and negotiation rather than merely recording who selected the correct option [11,12,13]. In chemistry, the same diagnostic logic has been used to distinguish students who produce similar acid–base answers for different reasons and to reveal difficulties connecting concentration, strength, pH, pOH, and particle representations [14].

3.2. Observe: Evidence, Discrepant Events, and the Motivation to Explain

Observation is the evidentiary hinge of POE, but it is not a neutral copy of reality. Learners attend selectively, interpret through prior concepts, and may disagree about what the event shows. The instructional question is therefore not only whether students see the phenomenon, but whether the evidence relevant to the prediction is perceptible, trustworthy, and representable. Digital video, data logging, simulations, and imaging can strengthen this phase by allowing replay, measurement, or access to processes that ordinary perception cannot resolve. Infrared-camera activities, for example, make thermal patterns available for immediate inquiry and help students connect visible evidence with accounts of heat transfer, evaporation, and condensation [15].

Many POE activities use discrepant events because a mismatch between a reasoned prediction and an observation can create a need for explanation. From a Piagetian perspective, such a mismatch can be described as a perturbation capable of initiating equilibration [16]. Dewey’s account of reflective inquiry provides a complementary rather than interchangeable perspective: a problematic situation can prompt the formulation of possible explanations and their examination through observation, reasoning, and further inquiry [17]. In contemporary science education, however, the more operational account is provided by conceptual-change and cognitive-conflict research. Posner et al. [3] emphasized that dissatisfaction with an existing conception is only one condition: a new conception must also become intelligible, plausible, and fruitful. Research on anomalous data likewise shows that contradiction alone is weak. Learners may reject the result, reinterpret it, attribute it to apparatus, or preserve the old idea alongside the new one [4].

POE should therefore not be identified with surprise. A confirmed prediction can refine causal reasoning, coordinate representations, or increase confidence in a model. Conversely, an unexpected result does not guarantee revision. A more defensible account is that learners generate a prediction, compare it with the available evidence, and, when a discrepancy emerges, use supported interpretation to determine whether and how their model should be revised. This conditional sequence avoids the common assumption that a spectacular outcome “teaches itself.” It also explains why active prediction and discussion improve demonstrations relative to passive observation: the event becomes evidence in relation to an articulated expectation [18].

The motivational value of discrepancy belongs within this same process rather than forming a separate theory of POE. A manageable mismatch can attract attention, stimulate epistemic curiosity, and create an impulse to close an explanatory gap. Yet an anomaly that appears arbitrary, inaccessible, or impossible to resolve may produce confusion, frustration, or mere entertainment. The most productive tasks give learners enough conceptual resources and appropriately calibrated support to believe that an explanation can be constructed. This balance between challenge and support is consistent with Vygotsky’s zone of proximal development, which distinguishes what learners can accomplish independently from what they can achieve with guidance or collaboration [19]. At a finer-grained level, POE can also be understood as a scientific micro-narrative: prediction establishes an expectation, observation confirms or disrupts it, and explanation seeks causal resolution [20]. This narrative analogy is useful for understanding explanation-seeking curiosity, but it should not be elevated into an independent foundation for the strategy.

Beyond its role in conceptual comparison, the Observe phase can make several epistemic issues visible. Students may discover that observation is not a neutral record of reality: prior knowledge shapes what is noticed, descriptions may incorporate unacknowledged inferences, and the same event can support different initial interpretations [21]. These features create opportunities to discuss the distinction between observation and inference and the evidential limits of what has been observed. Recent evidence indicates that activity-based instruction can substantially improve students’ understanding of observation and inference when these epistemic issues are addressed explicitly and reflectively [22]. More generally, such learning about NoS should not be assumed to follow automatically from participation in a POE task, as explicit-reflective instruction is consistently more effective than implicit or alternative approaches [23].

3.3. Explain: Reconciliation, Social Reconstruction, and Transfer

The Explain phase requires more than naming the accepted scientific concept. Learners must coordinate the original prediction, the reasons offered for it, and the evidence encountered. Depending on the case, the result may be confirmation, refinement, partial revision, or replacement. The phase is therefore where the sequence becomes explicitly explanatory rather than merely corrective.

Social interaction can support this reconstruction. Haysom and Bowen [2] propose that students discuss their explanations with a partner or small group before writing and before the teacher presents the current scientific account. This arrangement gives learners an opportunity to test causal language, compare evidence, and revise ideas without treating the teacher’s explanation as the first and only intellectual move. Tao and Gunstone’s two complementary reports [24,25], based on the same computer-supported mechanics intervention, similarly show that collaborative discussion can assist conceptual change, while also revealing that change is uneven, context dependent, and sometimes unstable.

The teacher remains essential because dissatisfaction without an intelligible alternative can leave students with a memorable anomaly and no productive model [3]. A scientific explanation should be presented as the best currently warranted account and should show why it explains the focal evidence more successfully than competing ideas. In chemistry, this often means explicitly connecting macroscopic observations to submicroscopic entities and symbolic representations; in biology, it may require a dynamic systems account; and in physics, it may require linking a visible trajectory to forces, energy, or a mathematical representation.

Transfer requires further work after the initial explanation. An observation that remains an isolated anomaly can be forgotten or reinterpreted through the learner’s prior expectations; explanation is more useful when it embeds the event in a causal or model-based network that can be used again. Follow-up predictions, analogous cases, changes in variables, and delayed applications test whether the reconstructed idea has become usable [2]. This need for extension explains why many derivatives add discussion, exploration, revision, writing, diagnosis, or evaluation. Added phases are justified when they strengthen model reconstruction or transfer, not simply because a longer acronym appears more comprehensive.

The Explain phase also offers an opportunity to examine how scientific accounts are justified and revised. When learners compare competing explanations, identify which evidence supports each one, and consider what further observations would discriminate between them, POE can illuminate the tentative, inferential, and theory-laden character of scientific knowledge. Explicit-reflective designs involving argumentation, collaborative comparison, and the integration of different perspectives appear particularly conducive to the development of informed NoS understandings [23]. The accepted explanation should therefore be presented not as an unquestionable final answer, but as the account currently best supported by the available evidence. Nevertheless, these epistemic implications must be deliberately discussed: evidence comparing explicit and implicit inquiry instruction shows that participation in inquiry alone may leave students’ views unchanged, whereas explicit reflection can produce more informed understandings [26].

4. Variants of the Classical POE Sequence

The proliferation of POE derivatives is best interpreted as an attempt to address limitations or instructional needs not made explicit in the basic three-phase sequence. Rather than treating the variants as competing models, it is more useful to ask what pedagogical problem each added phase is intended to solve. Table 2 provides an overview of the principal variants, defines their expanded sequences, and summarizes the instructional purpose of the added components. The acronyms introduced in the table are used selectively in the discussion that follows, which focuses on the contribution and limitations of different families of variants rather than restating their procedural structure.

Table 2. Major POE derivatives and the design problem addressed by their added phases.

Variant	Expanded Sequence	Primary Design Purpose	Illustrative Studies
PDEODE	Predict–Discuss–Explain–Observe–Discuss–Explain	Make reasons public before observation and support post-observation reconstruction.	Coştu [27]; Abdullah et al. [28]; Lathifa [29]
PEODE	Predict–Explain–Observe–Discuss–Explain	Retain explicit pre- and post-observation explanations while removing pre-observation discussion.	Coştu & Bayram [30]; Coştu [31]
PORE	Predict–Observe–Revise–Explain	Make model revision an explicit scientific product.	Lucas et al. [32]
POED—Diagnose	Predict–Observe–Explain–Diagnose	Reveal residual misconceptions after exposure to the scientific account.	Chen et al. [33]
POED—Do	Predict–Observe–Explain–Do	Extend reconstructed reasoning into action or application.	Nurmalita et al. [34]
PEEOR	Predict–Explain–Enact–Observe–Reflect	Add embodied enactment and sustained reflection.	Wondmagegn Getu et al. [35]; Wondmagegn et al. [36]
POE2WE	Predict–Observe–Explain–Elaborate–Write–Evaluate	Deepen explanation through elaboration, writing, and evaluation.	Barkah et al. [37]
PBLPOE/POEP	POE embedded in problem- or project-based learning	Support extended problem solving, metacognition, collaboration, and transfer.	Fitriani et al. [38,39]; Ilma et al. [40,41]
CPEOE	Collaborative–Predict–Explain–Observe–Explain	Foreground team construction and negotiation of explanations.	Ajayi [42]
Differentiated POE	Adapt prompts, resources, pacing, or products	Widen access while preserving authentic prediction and evidence comparison.	Oktavia et al. [43]; Yatmanto et al. [44]

One family of variants addresses a limitation that is primarily social and discursive: learners may move too quickly from a private prediction to observation without articulating or comparing their reasoning. Discussion-enriched designs such as PDEODE make reasons public before observation and return to them after evidence is available; across condensation and acid–base contexts, these designs have been used to expose, compare, and revise misconceptions [27,28,29]. PEODE likewise requires learners to formulate an account before observation and revisit it through discussion afterward, but omits the initial pre-observation discussion phase of PDEODE. Laboratory work with preservice science teachers and gifted students suggests that this structure can support selected science-process subskills and the interpretation of everyday scientific problems, respectively [30,31]. The distinctive contribution of these variants therefore lies less in the additional labels than in creating discursive checkpoints for articulating provisional accounts and reconstructing them after observation. Their effects nevertheless depend on the quality of discussion, task design, and teacher support rather than on the nominal sequence alone.

A second family makes revision, diagnosis, sustained reflection, or external representation more explicit. PORE is particularly aligned with modeling because it treats revision itself as a scientific product; in cellular-respiration modeling, learners updated system models rather than merely replacing a verbal answer [32]. Diagnostic extensions address a different problem: students may retain residual misconceptions even after receiving or constructing a scientific explanation. Chen et al. [33] therefore added a Diagnose phase, whereas Nurmalita et al. [34] used a Do phase to carry reconstructed reasoning into action. Reflection- and writing-oriented variants similarly respond to the risk that Explain becomes a brief oral response: PEEOR incorporates enactment and reflection and has been applied to motion, force, and geometrical optics [35,36], whereas POE2WE adds elaboration, writing, and evaluation, with Barkah et al. [37] reporting stronger conceptual understanding than conventional inquiry. These extensions are most defensible when the added phase makes reasoning more inspectable, revisable, or transferable rather than merely lengthening the sequence.

Model-oriented POE derivatives should, however, be distinguished from broader approaches to scientific modeling. Sensevy et al. [45] emphasize the construction of relationships between theoretical models and observable events across carefully designed instructional sequences. In physics education, Hestenes [46] established a theoretical foundation for model-centered instruction, while Wells et al. [47] described its classroom implementation through sustained cycles of model development, evaluation, and deployment, supported by experimentation, multiple representations, and collaborative discussion. This approach is also promoted and disseminated by the American Modeling Teachers Association (AMTA). Such frameworks encompass instructional practices that are not specified by adding a Revise phase to POE. PORE may therefore support particular episodes of scientific modeling without constituting a complete model-based instructional approach.

A third group embeds POE within broader pedagogical structures rather than simply adding another phase. Problem-based hybrids and POE-based project learning extend prediction–evidence–explanation cycles across longer problems or collaborative products. Fitriani et al. [38,39] associated PBLPOE with critical thinking, scientific attitudes, problem solving, and self-efficacy, while project-based applications have been associated with metacognition and collaboration [40,41]. Other hybrids foreground social organization or access: CPEOE requires teams to construct and negotiate explanations, with Ajayi [42] reporting stronger chemistry outcomes than discussion teaching, whereas differentiated POE modifies prompts, resources, pacing, or products according to learner readiness and has been applied to temperature and heat and to primary science outcomes [43,44]. Differentiated designs additionally illustrate a broader challenge developed in Section 7: support should widen access without predetermining predictions or weakening evidential comparison. These designs broaden the instructional reach of POE, but they also complicate attribution because observed outcomes cannot readily be separated from the surrounding problem, project, collaboration, differentiation, or assessment structures.

POE can also be embedded within inquiry-based science learning (IBSL), which encompasses scientific questioning, investigation, evidence evaluation, and model construction [48]. However, its level of openness depends on how research questions and investigative procedures are determined. Applying the framework proposed by Bell et al. [49], teacher-directed POE tasks may be characterized as structured inquiry when the teacher provides the investigative question and procedure and students analyze observations to construct their own explanations; guided inquiry involves student-designed procedures, while open inquiry additionally involves student-formulated questions. Thus, a three-phase POE task does not automatically provide the broader learner control over questions and investigative procedures often associated with open inquiry.

Across these families, the central criterion is not how many phases a variant contains, but whether the added component addresses a clearly identifiable instructional need. Discussion is useful when reasoning must be externalized and compared; revision when models require iterative improvement; diagnosis when residual misconceptions may remain hidden; writing and reflection when explanations risk remaining superficial; and broader pedagogical structures when sustained inquiry, transfer, collaboration, or differentiation are explicit goals. Variants become less informative when added phases or structures merely increase procedural complexity without changing the quality of reasoning.

5. Evidence for the Effects of POE and Its Derivatives

Having clarified how the main POE derivatives modify the classical sequence and what instructional problem each additional phase is intended to address, the review can now examine the evidence supporting their effectiveness. The question “Does POE work?” is nevertheless less straightforward than it appears. Studies differ in educational stage, scientific content, duration, outcome measures, and the degree to which the intervention represents canonical POE. Some compare POE with lecture or conventional laboratory work; others compare a digitally enriched POE condition with traditional POE; still others combine POE with problem-based learning, projects, formative assessment, concept cartoons, or additional phases. A balanced review must therefore distinguish evidence for the canonical sequence, named extended variants, and broader instructional packages in which POE is only one component; the unit of inference changes with the comparator.

5.1. Interpreting the Comparative Evidence

For the focused effectiveness synthesis, interpretive weight was assigned qualitatively rather than by applying a formal risk-of-bias score. The publications examined in greater detail were selected for the methodological contrasts they illustrate, particularly differences in comparators, study designs, and inferential limitations, rather than as an exhaustive or formally ranked subset of the eligible corpus. Studies were considered more informative when they used random assignment or a pretest–posttest comparison group, described a credible comparator, aligned outcome measures with the targeted concept or practice, and, where possible, included delayed testing or multiple forms of evidence. These features improve the counterfactual but do not make the studies equivalent: the unit of assignment may be a learner, class, or school; baseline equivalence is not always secure; intact-class clustering can be neglected; measures are often researcher-developed; and teacher, setting, or multicomponent effects remain common. Where information about assignment, statistical treatment of clustering, outcome validity, delayed assessment, or fidelity was absent or unclear, it was not inferred from the reported significance of the results.

The clearest comparisons are those in which assignment was reasonably transparent and the instructional contrast was relatively well defined. Hsu et al. [50] and Yang et al. [51] used randomized designs with young learners and found advantages for POE-structured activities, while also documenting residual difficulties after instruction. Akpınar [52] added a six-week delayed test but compared POE-based animation with regular instruction, so POE and multimedia remained bundled. Karamustafaoğlu and Mamlok-Naaman [53] reported random division of university students between POE and traditional instruction with the same chemistry teacher, and Çıngıl Barış [54] compared the same laboratory activities organized through POE or traditional instruction. These studies support a positive interpretation, but modest samples, topic-specific measures, and bundled features still constrain causal precision.

Kibirige et al. [55] reported greater reduction in misconceptions about dissolved salts under POE than under traditional teaching, but the design provides a relatively weak basis for causal inference. The experimental and comparison groups came from two different schools, so treatment was confounded with school and potentially with teacher effects. Moreover, the analysis treated individual students as independent units despite assignment occurring at the school or class level, and the comparison condition relied on traditional lecture-based teaching. The study therefore provides favorable but methodologically limited evidence and was not included in the illustrative comparative table.

Larger samples do not necessarily resolve these design problems. Okah and Agboghoroma [56] reported higher biology achievement under POE in a sample of 280 students, and Djudin [57] reported stronger elementary science understanding in a 2 × 2 quasi-experimental design across ten schools. Both broaden the contexts in which favorable outcomes have been reported, but incomplete information about assignment, classroom clustering, and local instruments limits the precision of the causal interpretation.

Other studies answer a different question: whether a richer POE-based design outperforms a less enriched alternative. Fitriani et al. [38,39] reported the strongest outcomes for a PBLPOE condition across critical thinking, scientific attitudes, problem solving, and self-efficacy, but the hybrid design does not isolate the contribution of POE. Tüysüz and Özdemir [58] found higher physics achievement with simulation-enriched POE than with traditional POE, which supports the value of the simulation enrichment rather than POE versus non-POE instruction. Booth et al. [59] likewise embedded iterative POE within computational systems modeling in biochemistry. These studies are educationally informative, but the unit of inference is the complete instructional package.

Table 3 summarizes eight illustrative comparative entries and the main interpretive caution attached to each. The table highlights differences in methodological design, comparator type, and inferential limitations rather than presenting an exhaustive or formally ranked selection of effectiveness studies.

Table 3. Illustrative comparative evidence on POE and its derivatives: study designs, findings, and interpretive cautions.

Study	Context and Design	Main Finding	Interpretive Caution
Hsu et al. [50]	Random assignment; 50 preschoolers; POE-integrated vs. non-POE computer game; light and shadow.	Stronger understanding of shadow formation and orientation in the POE condition.	Alternative conceptions persisted; game format and POE were bundled.
Yang et al. [51]	Random assignment; 49 preschoolers; POE block play vs. building guide; two-week delayed test.	Improved acquisition and retention of gear concepts.	Gear speed and direction remained difficult.
Akpınar [52]	Quasi-experiment; 30 vs. 27 primary pupils; POE animations vs. regular teaching; six-week delayed test.	Stronger immediate and delayed static-electricity understanding.	Cannot isolate POE from interactive animation.
Tüysüz & Özdemir [58]	Quasi-experiment; 107 ninth graders; simulation-enriched POE vs. traditional POE.	Higher physics achievement in the enriched POE condition; no gender interaction.	Demonstrates added value of simulation, not POE vs. non-POE.
Fitriani et al. [38,39]	Four-group quasi-experiment; 132 Grade 10 students; PBL, POE, PBLPOE, conventional.	PBLPOE obtained the strongest profile for critical thinking, scientific attitudes, problem solving, and self-efficacy.	Hybrid effects cannot be attributed to POE alone.
Karamustafaoğlu & Mamlok-Naaman [53]	Authors report random division; 40 university freshmen (20/20); same chemistry teacher; electrochemistry.	Both posttests favored POE, with fewer misconceptions.	Small sample; one open-ended pretest showed a non-trivial baseline difference despite a nonsignificant test.
Çıngıl Barış [54]	Randomly determined experimental/control groups; 64 preservice teachers; same laboratory activities organized through POE or traditional instruction.	Higher conceptual achievement and more favorable biology-laboratory attitudes.	Single topic and institution.
Booth et al. [59]	Large upper-division biochemistry courses; computational models structured by iterative POE; comparison instruction.	Approximately 8% average learning gain with modules vs. no gain without modules.	Different instructors and a rich systems-modeling package.

Note: Studies are ordered by educational level, from early childhood to higher education. The eight comparative entries correspond to nine publications, as the two reports by Fitriani et al. [38,39] describe the same intervention and are therefore treated as a single entry.

5.2. Conceptual Understanding and Misconception Reduction

Across the better-controlled studies, the most consistent outcome is conceptual understanding. This pattern is theoretically coherent because POE elicits an initial account and makes later evidence relevant to that account. The clearest results occur when observation is tightly aligned with a known conceptual difficulty: static electricity [52], light and shadow [50], gears [51], electrochemistry [53], condensation [28], and photosynthesis and respiration [54]. The advantage is most interpretable when the explanation explicitly revisits the reasoning that produced the prediction.

Conceptual change is rarely complete even when posttest scores improve. Alternative conceptions persisted in Hsu et al. [50], gear speed and direction remained difficult in Yang et al. [51], and Tao and Gunstone [25] documented uneven, context-dependent trajectories during computer-supported mechanics instruction. POE is therefore better understood as a way to initiate and organize conceptual reconstruction than as a one-cycle cure.

Lower-control and diagnostic studies help explain what aggregate scores can conceal. Kala et al. [14] used POE tasks to distinguish different forms of acid–base reasoning, while Widhiyanti et al. [60] identified troublesome knowledge and threshold concepts in chemical equilibrium. Banawi et al. [61] reported pretest–posttest gains among preservice primary teachers at macroscopic, submicroscopic-verbal, submicroscopic-visual, and symbolic levels of states-of-matter understanding, but the study had no comparison group. Kibirige et al. [55] likewise reported reduced misconceptions about dissolved salts, yet its two-school design confounded treatment with setting. Such studies are useful for mapping where reasoning changes or persists, but they should not carry the same causal weight as randomized or well-matched comparisons.

5.3. Achievement, Scientific Practices, and Higher-Order Outcomes

Achievement effects are generally positive but less directly interpretable. Venida and Sigua [62] found that seventh graders taught through POE outperformed a 5E (Engage, Explore, Explain, Elaborate, Evaluate) comparison class in physics and shifted from neutral to positive attitudes. Ojo and Owolabi [63] reported advantages over conventional laboratory instruction for practical performance and attitudes, while Sarah et al. [64] found larger analytical-skill gains than direct instruction. These studies indicate that POE can support outcomes beyond a single concept test, although two-class designs and local measures limit generalization.

Higher-order outcomes are most visible in hybrid and extended models. Fitriani et al. [38,39] found that integrating problem-based learning and POE strengthened critical thinking, scientific attitudes, problem solving, and self-efficacy. POE-based project learning designs have also been associated with metacognitive awareness, cognitive outcomes, and collaboration [40,41]. The findings are educationally valuable, but they should not be used to claim that canonical POE alone produces the full outcome profile. The additional problem, project, collaborative, and assessment structures are likely active ingredients.

Evidence for process skills and explanation quality is promising but methodologically uneven. Coştu and Bayram [30] reported improvements in selected science-process subskills after PEODE laboratory work; Sakyi-Hagan [65,66] reported a shift from predominantly descriptive or everyday explanations toward formal causal explanations among preservice teachers; and Hartanto et al. [67] reported moderate conceptual gains and generally good process-skill performance in a virtual-laboratory POE sequence on direct-current circuits. The latter two studies lacked comparison groups, so they are best interpreted as evidence of feasibility and developmental change rather than definitive causal tests.

5.4. Attitudes, Interest, and Retention

Affective findings are less consistent than cognitive findings, particularly when POE is embedded in a broader media-rich or collaborative instructional package. Çıngıl Barış [54] and Venida and Sigua [62] reported more favorable attitudes, and Hsu et al. [50] found an effective game-based experience alongside conceptual gains. By contrast, Çalış and Özkan [68] found that distance-education POE improved fifth-grade science achievement relative to a comparison group but did not improve attitudes. James et al. [69] found that adding collaborative POE to an already active introductory geology course did not produce greater content gains than a clicker-based comparison but was associated with preserved self-efficacy and interest. These results suggest that POE can support affective outcomes, but it is not intrinsically motivational; the comparison pedagogy, classroom climate, task design, and technological medium matter.

Delayed evidence remains scarce. Akpınar [52] and Yang et al. [51] are notable because they included retention tests, while most studies assessed students immediately after the intervention. Transfer to new phenomena is even less frequently measured. This is a major limitation because the explanatory phase is intended to make a scientific account fruitful beyond the original event.

Across these outcome domains, the comparative evidence supports a cautiously favorable but design-dependent interpretation of POE and its derivatives, particularly for conceptual understanding. Comparisons of canonical POE with non-POE instruction can provide a more direct test of the core sequence when other instructional features are reasonably comparable; studies of explicitly extended variants assess the extended sequence as implemented, whereas hybrid and digitally enriched contrasts examine broader instructional packages or differences between enriched and less enriched POE conditions, depending on the comparator. Confidence is greatest when the target difficulty is clearly specified, the comparator is credible, and the instructional condition is not confounded with setting or additional components. When POE is embedded in simulations, problem-based learning, projects, or other multicomponent interventions, the observed advantage pertains to the complete instructional design rather than to the canonical three-phase sequence alone.

6. Disciplinary Adaptation of POE: Physics, Chemistry, and Biology/Biochemistry

POE has been used in more domains and educational stages than can be examined with equal depth here. This section therefore concentrates on physics, chemistry, and biology/biochemistry, not to catalogue applications but to examine how disciplinary differences reshape POE. Because the Predict phase retains a broadly similar function across these domains—eliciting a reasoned expectation before the outcome is known—the comparison focuses primarily on variation in what counts as accessible evidence during Observe and in the explanatory reconstruction required during Explain. Table 4 summarizes these contrasts.

Table 4. Disciplinary adaptation of POE across physics, chemistry, and biology/biochemistry.

Domain	Characteristic Evidential Problem	Observe	Explain	Characteristic Adaptation
Physics	Relevant behavior may be rapid, invisible, instrument-dependent, or counterintuitive.	Replay, measurement, imaging, simulation, or a sequence of related cases can make the relevant pattern inspectable.	Observed patterns must be connected to forces, energy, circuit models, thermal processes, or mathematical relations.	Interactive video, infrared imaging, animation, and sequenced electricity problems support evidential comparison [11,12,13,15,70,71,72].
Chemistry	Macroscopic changes and measurements do not reveal the particle-level mechanism that explains them.	Visible events are coordinated with quantitative measurements, dynamic representations, or instrument output.	Learners must connect macroscopic evidence with submicroscopic entities and symbolic relations.	Concept mapping, conductivity curves, electrochemical instrumentation, dynamic visualization, and portable kits help bridge representational levels [14,53,60,61,73,74,75,76,77,78].
Biology/biochemistry	Target processes are often distributed, time-dependent, systemic, or inaccessible to direct observation.	Experimental proxies or computational outputs stand in for the underlying process.	Learners construct or revise causal and systems models rather than merely label an observed outcome.	Enzyme investigations, explicit model revision, and iterative computational modeling make dynamic behavior inspectable [32,54,59,79].

Physics provides a useful starting point because the central disciplinary issue is often not whether an event can be produced, but whether the feature relevant to the prediction can be made sufficiently visible and comparable. Early heat-and-expansion tasks used ordinary phenomena to expose intuitive interpretations [70], whereas animation and instrumented representations extend POE to variables that are difficult to inspect directly. Tahir et al. [71], for example, integrated animation into circuit work to address misconceptions without assuming that a static diagram or a single demonstration would make the relevant relationships evident.

Beyond making evidence perceptible, physics applications also show the importance of sequencing. Potvin et al. [72] designed 17 Kasetsu-inspired POE problems targeting 13 possible misconceptions in basic electricity. The significance of this design lies less in any single discrepant event than in its cumulative structure: successive predictions and observations prepare, test, and refine ideas needed for later cases. Physics therefore illustrates a progression from directly observable events to instrument-mediated evidence and, where conceptual systems are particularly persistent, to deliberately sequenced cycles of comparison and revision.

The evidentiary problem changes substantially in chemistry. Here, disciplinary adaptation centers on the gap between what can be observed macroscopically and the submicroscopic or symbolic account needed to explain it. Computer-based POE combined with concept mapping has been used to trace changes in the organization of acid–base knowledge [73], while conductimetric titration tasks require students to predict and interpret the shape of a measurement curve rather than merely notice a visible change [74]. In such cases, representation is not an accessory to observation but part of the explanatory work itself.

This need to coordinate levels of representation is also evident in laboratory and visualization-based designs. Redox POE sequences [75] and electrochemical “mystery box” activities using Arduino-based measurements [76] require learners to infer underlying processes from instrument-mediated evidence. Dynamic visualizations [77] and a portable reaction-rate kit [78] likewise connect observable outcomes with explanatory representations. What these chemistry applications share is not an attempt to make phenomena more spectacular, but a deliberate effort to connect macroscopic evidence with particle-level models and symbolic relationships.

Biology and biochemistry present a third configuration, because many target processes are distributed, time-dependent, or inaccessible to direct observation. Enzyme investigations provide a relatively concrete case in which changing conditions produces observable evidence [79], whereas the cellular-respiration and metabolism studies discussed earlier rely on computational system behavior and explicit model revision [32,59]. The photosynthesis and respiration comparison [54] further shows that POE can reorganize familiar laboratory activity without changing the underlying experimental tasks. In these contexts, the Observe phase increasingly shifts from witnessing a discrete event toward inspecting evidence generated by an experimental or computational system.

Viewed comparatively, the most important disciplinary differences concern less the Predict phase than what counts as accessible evidence and what kind of explanatory reconstruction is required. Physics often benefits from making fleeting or otherwise inaccessible variables inspectable; chemistry requires coordination among macroscopic, submicroscopic, and symbolic levels; and biology and biochemistry frequently rely on proxies and dynamic models for processes that cannot be observed directly. POE is therefore better understood not as a fixed activity template, but as a common reasoning structure whose Observe and Explain phases must be adapted to the epistemic demands of the target science.

7. Design Principles and Common Failure Modes

Across media and disciplines, robust POE designs share four properties. Predictions are made before outcome cues and require reasons; observations are arranged so that evidence bearing on the prediction is visible, repeatable, or credibly represented; explanations revisit the initial rationale and make an intelligible scientific alternative available; and follow-up cases test retention or transfer. These properties matter more than the surface format of the activity.

One failure mode is the spectacle problem. A demonstration may be surprising and memorable while leaving the explanatory model untouched. Crouch et al. [18] showed that demonstrations produce more learning when students predict and discuss rather than passively watch. In POE, prediction and explanation therefore need protected time; asking a quick question and immediately revealing the answer does not constitute meaningful implementation.

Another failure mode is unsupported conflict. Anomalous evidence does not force students to abandon an existing idea [3,4]. A discrepant event needs a viable explanatory alternative, opportunities to inspect the evidence, and discussion of why the alternative accounts for the result more successfully. When learners cannot interpret the mismatch, surprise may end in denial, confusion, or entertainment rather than revision.

Public exposure creates a different risk. Haysom and Bowen [2] stress that students may be anxious about appearing wrong. This concern is consistent with broader learning-from-errors research, which indicates that students engage more productively when errors are treated as ideas to be examined rather than as threats to competence or social reputation, and when classroom responses protect dignity while keeping reasoning visible [80]. The teacher should therefore value the reasoning represented by a prediction while remaining clear that some explanations are scientifically inadequate. Private or anonymous prediction before discussion, neutral recording of alternatives, and language that distinguishes a provisional idea from the best-supported account can preserve intellectual risk taking.

Phase collapse occurs when Observe is reduced to receiving a teacher statement or simulation output and Explain to copying the accepted answer. Robust implementation asks learners to decide what the observation warrants and to construct or critique causal links. In digital environments, interface screens labeled Predict, Observe, and Explain are not enough if the learner’s responses are not conceptually connected.

The one-cycle assumption is especially problematic for persistent conceptions. Gauld [81] reported a striking case in which students who had acknowledged during an electric-circuit demonstration that the observed current contradicted their expectation later remembered the event as though it had confirmed the original view. Immediate recognition of a discrepancy is therefore not evidence of durable conceptual change. Residual and context-dependent ideas in POE studies with young learners and in computer-supported mechanics point in the same direction [25,50,51]. Sequences of related tasks, delayed revisiting, and transfer questions are more defensible than a single “magic” demonstration.

These considerations also argue against treating POE as a self-sufficient discovery procedure. Even when learners recognize that an initial explanation fails, they may not yet possess the conceptual resources needed to construct the scientific alternative. Sequencing, teacher guidance, and later reinforcement are therefore part of the pedagogy. Taber [82] captures this point by treating POE as a tactic within a longer-term instructional strategy rather than a one-shot mechanism of conceptual change.

Scaffolding should be calibrated to learners’ familiarity with POE, their prior knowledge, and the conceptual demands of the task. Learners who are new to the sequence may benefit from explicit prediction prompts, structured observation records, questions directing attention to relevant variables, and comparison or explanation prompts. As familiarity and competence increase, these supports can be progressively faded so that learners assume greater responsibility for formulating predictions, deciding what evidence is relevant, identifying discrepancies, and constructing explanations. The purpose of scaffolding is to make the reasoning process accessible without predetermining either the prediction or the interpretation of the evidence.

Teacher preparation should therefore address not only the three phases but also task selection, discussion orchestration, scaffolding, evidence interpretation, and follow-up. Mtsi et al. [5] further indicate that recognizing POE does not necessarily provide teachers with the procedural knowledge, resources, or professional development needed for confident implementation. Preparation should thus connect procedural knowledge of POE with the epistemic work of distinguishing observation from inference and with practical strategies for supporting learner explanations.

These principles can be translated into a compact planning checklist for classroom use. Table 5 summarizes questions that help preserve the reasoning function of POE while allowing teachers to calibrate support to learners and tasks.

Table 5. Practical checklist for designing POE activities in science education.

Design Decision	Practical Question for Teachers
Predict	Have learners committed to a prediction and a reason before outcome cues become available?
Observe	Is the evidence relevant to the prediction perceptible, measurable, replayable, or otherwise trustworthy?
Compare	Are learners explicitly asked to compare what they expected with what the evidence warrants?
Explain	Do learners have opportunities and conceptual resources to construct or revise an explanation before the accepted account is simply supplied?
Protect intellectual risk	Can learners expose provisional ideas without public labeling of personal failure?
Scaffold and fade	Is support sufficient for the learners and task while allowing responsibility to shift progressively toward the learner?
Follow-up	Does a related but non-identical case test whether the reconstructed idea can be used beyond the initial event?

For teachers seeking classroom-ready examples, Haysom and Bowen [2] provide a particularly extensive and stable practical resource, with more than 100 POE activities spanning a broad range of physical-science, chemistry, and life-science topics. The volume includes student worksheets, scientific explanations, summaries of learner responses, and materials lists, making it useful both for direct classroom implementation and for adapting POE tasks to new contexts.

Research reports also need stronger design fidelity. Authors should specify the prediction prompt, whether reasons were required, what students actually observed, how explanations were elicited, how the teacher supported the scientific account, and whether follow-up occurred. Comparators should be described well enough to determine whether a study tests POE against passive instruction, another active method, or a less enriched version of POE. Without this information, a positive posttest may show that an instructional package was successful without revealing which elements produced the advantage.

8. Conclusions and Prospects

The overarching conclusion is conditional: POE can support science learning when its three phases function as connected forms of reasoning, rather than as labels attached to a demonstration. It compresses a central dynamic of science learning into a manageable instructional sequence. Regarding the first review question, the three phases perform distinct but complementary pedagogical functions. Prediction externalizes a model and creates epistemic commitment. Observation establishes a comparison with evidence and may reveal confirmation, limitation, or discrepancy. Explanation supports reconciliation, revision, and transfer through individual and social reasoning. This three-part account is sufficient to organize the strategy without treating POE as the convergence point of every educational theory.

With respect to the second review question, variants are most useful when their added phases solve identifiable problems. Discussion can make reasoning socially available; revision can support modeling; diagnosis can reveal residual misconceptions; writing and reflection can deepen explanation; project- and problem-based learning can promote transfer and sustained inquiry; and differentiation can widen access. Acronym proliferation becomes unhelpful when the purpose of the added phase is not articulated or evaluated.

Regarding the third question, the effectiveness evidence warrants a cautiously favorable conclusion, qualified by the comparator and by whether the intervention tested canonical POE, an extended variant, or a broader teaching package. The clearest comparative studies indicate benefits particularly for conceptual understanding, while effects on achievement, higher-order outcomes, and attitudes are more variable. The inference is strongest when POE targets a specified conceptual difficulty, treatment and setting are not confounded, and the comparison condition provides a credible counterfactual. It is less secure when POE forms part of a richer intervention whose individual components cannot be disentangled. A more elaborate intervention may be educationally preferable, but that is analytically different from demonstrating the independent effect of the canonical sequence.

Regarding the fourth review question, the disciplinary analysis shows that POE is not a generic shell imposed unchanged on science. In physics, instruments, replayable media, and sequenced problems can make trajectories, circuits, and thermal patterns available for comparison. In chemistry, visible changes and measurements must be coordinated with submicroscopic and symbolic accounts. In biology and biochemistry, proxies and dynamic or computational models make distributed processes inspectable and can turn model revision itself into part of the scientific product. Focusing on these three domains allows the review to examine disciplinary adaptation in greater depth rather than treating breadth of application as evidence in itself.

Several limitations concerning the synthesis and the inferences drawn from it should also be acknowledged. As a narrative and integrative review, the study did not employ a preregistered protocol, a formal study-level risk-of-bias assessment, or meta-analytic estimation; consequently, the relative weight assigned to individual publications involved interpretive judgment. The evidence base also combines canonical POE, extended sequences, and hybrid interventions that differ in educational stage, duration, comparator, outcome measure, and implementation fidelity, which often prevents the contribution of POE from being isolated from simulations, games, problem-based learning, projects, or other components. The disciplinary analysis is deliberately selective: physics, chemistry, and biology/biochemistry were chosen for depth and contrast, so the review does not attempt equivalent treatment of every science domain or educational stage represented in the corpus. Finally, the predominance of positive findings, together with limited reporting of null results, delayed retention, transfer, and implementation fidelity, may yield an overly favorable picture of effectiveness. The conclusions should therefore be read as a critical interpretation of a broad but methodologically uneven evidence base rather than as a definitive causal estimate of the effects of POE.

Future research should move from broad claims that POE “works” toward mechanism-sensitive questions and stronger causal designs. Which prediction prompts produce explanatory commitment rather than guessing? When does a discrepancy support revision and when does it provoke rejection? How should observation be represented when the target process is invisible or model-based? Which added phase improves explanation, retention, or transfer? How many related cycles are needed for durable change, and under what conditions is POE preferable to other active pedagogies? Experimental dismantling studies, comparisons with credible active controls, cluster-aware analyses, delayed measures, transfer tasks, implementation-fidelity data, and publication of null results would substantially improve the field.

For practice, the central recommendation is to protect the integrity of the sequence. Ask learners to predict and justify. Make the relevant evidence observable and discuss what it warrants. Support explanations with conceptual tools, peer comparison, and the current scientific account. Then ask learners to use the reconstructed idea in a new case. When these conditions are met, POE is not merely an engaging demonstration format; it becomes a disciplined classroom grammar for connecting prior thinking, evidence, and scientific explanation.