Humans construct knowledge by integrating new information with their existing cognitive structures (Pritchard, 2017). When a learner's prior conceptions clash with instructional materials, misconceptions arise, but resilient ideas that can obstruct understanding and hinder further learning (Duit & Treagust, 2003). Misconceptions in science are well‑documented and notoriously persistent across age groups and domains (Smith et al., 1994; Leonard et al., 2014). For instance, in biology, students tend to think that plants absorb food in the soil or that a human being breathes in oxygen and breathes out only carbon dioxide. These types of myths do not only misrepresent a learner’s understanding of the phenomena of science, but also interfere with the learner’s ability to think in an integrated manner, thus, stifling learning (Vosniadou, 2020).

Within the realm of human physiology, the digestive system is an area where misconceptions flourish. Previous studies have identified that middle‑school and high‑school students believe digestion occurs exclusively in the stomach, that chemical and physical digestion never occur concurrently, or that nutrients are completely absorbed in a single organ (Gul et al., 2024; Uğur, 2010). Educators occasionally reinforce analogous misconceptions when textbooks present vague explanations or when their own mastery of the subject matter is inadequate (Hewson & Hewson, 1983). Such inaccuracies endure within undergraduate cohorts as well as among pre-service teachers, thereby revealing the institutional rather than the incidental character of the difficulty (Griffard & Wandersee, 2001; Dry, 1998). Recognizing and addressing students' misconceptions is therefore a crucial task for science educators (Gil‑Perez & Carrascosa, 1990; Suprapto, 2020). To do so effectively requires reliable tools that can diagnose misconceptions and monitor conceptual change over time.

Assessing misconceptions poses several challenges. Misconceptions are often nuanced and context‑dependent; thus, instruments must be carefully designed to capture variations in understanding (Treagust, 1988). Because correct responses may not always reflect deep conceptual understanding, some students may guess or memorize definitions, many researchers advocate for two‑tier or two‑stage diagnostic tests that require students not only to select an answer but also to justify their choice or to select a reasoning statement (Chandrasegaran et al., 2007; Lin, 2004). The equal measure of content validity, diagnostic cognitive data, and psychometric rigor is non negotiable in every form of sound measurement. However, outlining the more recent misconception scales demonstrates the use of validity and reliability in the scales is fundamentally lacking. Even the scales of internal reliability, factor structure, and content and construct validity were found to be low and exceedingly poor affirming sets of deficiencies regarding validation (Johnson & Morgan, 2016). This study addresses these shortcomings by using modern psychometric techniques, factor analysis and Rasch modeling, to evaluate the DSCS as a diagnostic instrument.

Numerous research have documented the misconceptions concerning the digestive system. Dry (1998) showed that students do not see the contribution of the mouth as an organ of digestion, seeing the liver as a digestion organ. Gul et al. (2015) showed that even after being taught, a student may hold the misconception that the entire digestion process happens in the small intestine or may confuse the mechanical and chemical stages of digestion. Recent studies have pointed out the inability to explain the functions of an enzyme, the contraption of bile, and absorption tissue (Lin et al., 2016; Roswati et al., 2019). These issues may stem from a lack of appreciation of the banal day to day activities and an oversimplistic textbook portrayal. Digestion is often illustrated as a simplistic and unnecessarily simplistically routed through the stomach. Such misconceptions may arise from the cognitive load of complex phenomena such as the process of peristalsis, hydrolysis and the absorption of nutrients (Karpudewan et al., 2017). Such misconceptions necessitate tests that target poorly conceived notions instead of simply gauging lack of depth in ideas.

The DSCS was developed to detect misconceptions among 7th‑grade students concerning the digestive system. It consists of 20 multiple‑choice items, each designed to probe a specific aspect of digestion (e.g., the sequence of digestion, the role of enzymes, the distinction between mechanical and chemical digestion, nutrient absorption, and organ functions). In this study, the items were assessed as correct, scored as ‘1’, or incorrect, scored as ‘0’. The items were taken from the 7th Grade Science Curriculum, and then the clarity and content of the items were reviewed the experts and pilot tested. While the DSCS originally contained fewer items, the current version expands coverage across the domain to improve reliability and to permit robust factor and Rasch analysis.

The goals that we set for ourselves are two. First, we conduct an exploratory and confirmatory factor analysis of the DSCS’s factor structure to consider the possibility of one latent trait(s) (the severity of misconceptions) and how well one or more factors could structurally describe the phenomenon. Second, we use Rasch modeling to examine item functioning, person reliability, local item independence, and general scale reliability. All these approaches together contribute to adequate psychometric reasoning to interpret the DSCS scores and formulate the appropriate instructional strategies.

To understand why misconceptions stick, one must study constructivist learning theory where knowledge is added to what is already known. Children, according to Piaget, add new knowledge to their mental structures as long as it is new until it is shown that they must change the structures in order to accommodate new knowledge. Posner and the rest built on this idea to conceptual change and argued that learners must become dissatisfied with their conceptions before they can change to more scientifically accurate ones (Posner et al., 1982). This dissatisfaction is more than often triggered by cognitive conflict situations where the current reality does not match what is believed. Cognitive conflict alone is insufficient. Therefore, learners have to deal with new explanations that are understandable, believable, and practical. The long-lasting presence of misconceptions is due to the absence of explanations, or the explanations that are provided do not fulfill requirements. (Taber, 2006). In the case of the digestive system, “digestion happens where food is stored” and “enzymes get used up like fuel” are naive heuristics that children have as explanations, and are used because they are similar to real-life. Such intuitions are hard to change because they are practical in real life and scientific phenomena that are below the surface level of life are complex and invisible. According to diSessa's phenomenological primitives theory (diSessa, 1993), such fragments of knowledge (or "p-prims") are contextually activated and need not form coherent theories. Effective conceptual change thus requires not only introducing correct information but also restructuring underlying intuitions. Strategies such as refutation texts (Tippett, 2010), which explicitly state a misconception and then refute it with evidence and other conceptual change texts can help in this process. Diagnostic tests like the DSCS provide teachers with the information necessary to tailor such interventions.

Another perspective is the resources framework, which suggests that misconceptions are not monolithic ideas but rather consist of multiple semi‑independent resources that can be activated differently across contexts (Hammer, 2000). From this viewpoint, assessment instruments need to sample understanding in varied contexts to capture the variability of student thinking. The DSCS therefore includes items that probe the same underlying concept in different ways, for example, multiple items assess understanding of chemical digestion in both the mouth and the stomach to ensure that we capture stable misconceptions rather than isolated responses. This design helps reveal whether students hold consistent misconceptions across scenarios or whether their answers vary with surface features.

Each item was scored dichotomously: correct responses received a score of 1 and incorrect responses 0 (Hammond, 1994). Omitted answers were treated as incorrect. Thus, total raw scores ranged from 0 to 20. For factor analysis, we computed a tetrachoric correlation matrix among the items. Tetrachoric correlations estimate the relationships between the underlying continuous propensities that lead to the dichotomous item responses, under the assumption of an underlying bivariate normal distribution for each item pair. These correlations are appropriate for binary data and provide a foundation for factor extraction.

For Rasch modeling, we constructed a person–item response matrix (with rows as participants and columns as items). We examined the distribution of total scores and calculated the person separation reliability. To facilitate interpretation of Rasch results, we converted each student's logit ability estimate into a standardized T‑score (mean = 50, SD = 10). These transformed scores place students on a familiar scale while preserving the interval properties of the Rasch logit scale.

We adopted a sequential analysis plan that combined classical test theory indices, factor analyses, and item response theory. All analyses were conducted using R (version 4.2.1) with relevant packages (e.g., psych, lavaan, eRm, and mirt).

Descriptive statistics and item analysis. We first examined classical item statistics. For each item, we computed the proportion of students answering correctly (item difficulty) and the corrected item–total correlation (point-biserial correlation between the item and the total score excluding that item, often called item–rest correlation). These indices helped identify any potentially problematic items, such as those that were extremely difficult or easy (difficulty < 0.20 or > 0.80) or those that did not correlate with the overall test (item–total correlation < 0.20). Items showing extreme difficulty or poor discrimination were flagged for further scrutiny.

Suitability for factor analysis. To ensure that factor analysis was appropriate for our data, we calculated the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and performed Bartlett's test of sphericity on the item correlation matrix. A KMO value ≥ 0.60 and a significant Bartlett's test (p < 0.001) would indicate that the data are factorable and justify proceeding with factor analysis.

Exploratory factor analysis (EFA). Using the tetrachoric correlation matrix, we conducted EFA to explore the underlying factor structure without imposing a predetermined model. We employed principal axis factoring (which does not assume multivariate normality and is well-suited to non-normal or binary data) and used oblimin rotation (an oblique rotation) because we expected any factors to be correlated (Costello & Osborne, 2005). To decide how many factors to retain, we triangulated multiple criteria: parallel analysis (which compares empirical eigenvalues to those from random data), the scree test (visual identification of an "elbow" in the plot of eigenvalues), and Velicer's Minimum Average Partial (MAP) test. Notably, parallel analysis can tend to over-extract factors with dichotomous data due to sampling error, so we cross-checked its suggestions with the MAP and the nScree method (which aggregates several criteria). In our case, the MAP test favored a single factor whereas parallel analysis suggested as many as eight factors, and the nScree indicated two factors. Based on these results and theoretical expectations (that a general misconception trait might have subdomains), we decided to examine both one-factor and two-factor solutions. We considered an item to load saliently on a factor if its loading was ≥ 0.30 on that factor and it did not simultaneously load > 0.20 on another factor. We interpreted any extracted factors in light of the content and conceptual focus of the items grouping on them.

Confirmatory factor analysis (CFA). We next tested the dimensionality indicated by EFA using CFA. We fit two CFA models with lavaan: a one-factor model (all 20 items loading on a single latent construct) and a two-factor model (items loading on two correlated latent factors, based on the EFA groupings). Given the dichotomous nature of the data, we treated items as ordered categorical variables and used the WLSMV estimator, which is robust for categorical outcomes. Factor variances were fixed to 1 for model identification. Model fit was evaluated with standard indices: CFI and TLI (values ≥ 0.90 or 0.95 for good or excellent fit) and RMSEA and SRMR (values ≤ 0.08 for acceptable fit, with ≤ 0.05 indicating close fit). We compared the one- and two-factor models on these indices. We also checked modification indices for any local misfit (e.g., suggesting correlated residuals) and considered whether allowing any additional covariance or cross-loading was substantively justified. In addition to overall fit, we calculated composite reliability (CR) and average variance extracted (AVE) for each factor in the two-factor model to assess internal consistency and convergent validity of the factors. Generally, CR > 0.70 and AVE > 0.50 are desired (Fornell & Larcker, 1981). We report these values and discuss whether the factors were sufficiently distinct (discriminant validity) or essentially measuring the same construct.

Rasch modeling. Finally, we performed Rasch analysis by fitting a one-parameter logistic (1PL) model using the eRm package. This yielded estimates of item difficulty (the location of each item on the latent scale) and person ability for each student. All items were constrained to equal discrimination (slope) in this model. We examined item fit via Infit and Outfit mean square statistics; we considered values in the range 0.70–1.30 as indicative of acceptable fit. Items with fit statistics outside this range or with standardized fit residuals |z| ≥ 2.0 were flagged for review. (We note that with N = 412, the standardized fit statistic can be sensitive, so we emphasize mean-square fit and substantive content review of flagged items rather than rigidly dropping items on the basis of z alone.) We also examined person fit to identify any students with aberrant response patterns (e.g., high ability students getting many easy items wrong, or vice versa). Person separation reliability and the separation index were computed to evaluate the precision of measurement on the logit scale. We generated a Wright map (person–item map) to visualize the distribution of student abilities against item difficulties; this helps identify whether the test items appropriately target the range of student misconceptions (e.g., whether very high or very low ability ranges are underrepresented by items). Finally, we assessed local independence by computing Yen's Q3 statistic for residual correlations between item pairs. Item pairs with Q3 > 0.20 would suggest local dependence (Yen, 1984), potentially indicating redundant items or shared stimuli, but as reported below we found no such violations.

This study sought to validate the 20‑item DSCS using robust psychometric methods. As a whole, the DSCS showed strong technical attributes for a concept inventory. The instrument showed sufficient reliability (Cronbach's α = 0.77; Rasch person reliability = 0.706) and exhibited range for student understanding mastery. Both EFA and CFA indicated that student responses were driven primarily by a single latent trait (general understanding of digestion concepts vs. misconceptions). However, a two‑factor model provided a marginally better fit in CFA and offered pedagogical insight by grouping misconceptions into two interpretable facets (digestive processes vs. organ functions). Rasch modeling confirmed that items functioned appropriately (no significant misfit after accounting for content considerations) and that local item independence was satisfied. The person–item map revealed that the scale was somewhat easier for our sample than expected, suggesting the need for a few more difficult items to better challenge high-understanding students. Taken together, these results support using the DSCS as a valid and reliable instrument to diagnose misconceptions about digestion among 7th‑grade students.

This study provides robust evidence for the validity and reliability of the 20‑item Digestive System Concept Scale as a diagnostic tool for student misconceptions. Through rigorous analyses that combined factor analysis and Rasch modeling, we demonstrated that the DSCS is essentially unidimensional, with two strongly correlated facets that offer interpretive value. The scale can reliably differentiate students by their misconception severity and provides detailed diagnostic information at the item level. The DSCS thus serves as a powerful instrument for both researchers and educators in the effort to understand and address misconceptions in science education.

In synthesizing the broader implications of our findings, the DSCS emerges not only as a test of student knowledge but also as a platform for improving science teaching, informing curriculum design, and guiding professional development. Continuing to refine the instrument (for example, by revising ambiguous items and adding more challenging ones), extending its use across different cultures and languages, and integrating it into adaptive digital learning systems will further amplify its impact. Ultimately, tools like the DSCS contribute to a more nuanced understanding of learning by making misconceptions visible. They help us see that misconceptions are not roadblocks but rather stepping‑stones on the path to deeper understanding: if assessments, instruction, and content are constructively aligned to confront and replace misconceptions, we can foster enduring conceptual change in our students.