Third graders’ digital and paper text comprehension

Third graders’ digital and paper text comprehension ☆ Costanza Ruffini a , Chiara Pecini a , David Salda˜ na b , Pablo Delgado b,* a Department of Education, Languages, Intercultures, Literatures and Psychology (FORLILPSI), University of Florence, Via di San Salvi 12, 50135, Florence, Italy b Individual Differences, Cognition, and Language Lab (LABDICOLE). Departamento de Psicología Evolutiva y de la Educaci´ on, Universidad de Sevilla. Avda. Camilo Jos´ e Cela s/n, 41018, Sevilla, Spain ARTICLE INFO Keywords: Reading comprehension Screen vs. paper Narrative text Expository text Primary graders ABSTRACT Background: Research suggests differences in digital and paper text comprehension, but studies on primary school children are limited, and the role of textual genre remains unclear. This study examines the effects of reading medium and its interaction with text genre on text comprehension and explores whether these effects vary with comprehension skills, medium preference for academic vs leisure purposes, and computer use, controlling for working memory and attention skills. Method: A within participants design was implemented with 157 third graders (mean age =8.40, SD =0.30, 48.41 % female) reading four different texts: two linear texts (one narrative and one expository) on paper and two other linear texts (one narrative and one expository) on screen. Participants read four texts (two narrative and two expository), with counterbalanced variations in medium (paper-digital; digital-paper) and textual genre (narrative-expository; expository-narrative). Results: Students demonstrated better comprehension of narrative texts compared to expository ones. However, no significant effects on comprehension outcomes were observed for the reading medium or its interaction with text genre, regardless of students’ reading comprehension abilities or their preferred use of computers. Conclusion: The absence of the reading medium effect aligns with some previous research in primary education, although it contrasts with meta-analytic findings on the effect of reading medium. We highlight the importance of promoting the strategic use of digital devices for reading in later school years to prevent potential misuse, especially as unproductive digital activities become increasingly diverse. 1. Introduction Children worldwide frequently interact with digital devices such as computers, tablets, laptops, and smartphones, which have become integral to their daily lives (Mullis et al., 2017). Digital reading for school purposes is increasingly common, with teachers often providing digital texts for instruction (Barzillai & Thomson, 2018; Duncan et al., 2016; Livingstone et al., 2014; Merga & Roni, 2016). This shift is reflected in international assessments like PIRLS and PISA, which are now conducted digitally (Backes & Cowan, 2019; OECD, 2019). Despite the convenience and accessibility of digital materials (e.g., Jere-Folotiya et al., 2014; Picton, 2014), recent meta-analyses show that comprehension outcomes are generally lower when reading on digital devices compared to print (Clinton, 2019; Delgado et al., 2018; Kong et al., 2018). However, this finding primarily applies to adults, as few studies have explored how children comprehend digital versus paper texts. Among the limited studies on children, most focus on older students who have already automated reading and achieved a solid level of print-based comprehension skills (e.g., Aydemir et al., 2013; Golan et al., 2018; Støle et al., 2020). Exceptions include Florit et al. (2022, 2023) and Lenhard et al. (2017). Third grade is a pivotal stage for examining digital tool use in learning. It precedes the “fourth-grade slump”, a period when reading difficulties often emerge (Meichenbaum & Biemiller, 1998; Sweet & Snow, 2003; Terry et al., 2023). At this stage, children transition from learning to read to reading for comprehension and learning (Best et al., ☆ The research team belongs to two research groups that focuse on understanding learning processes and cognitive engagement in children with typical development and neurodevelopmental disorders. Their work explores how various factors—such as executive functions—affect academic achievement. By investigating these cognitive mechanisms, the group aims to identify both strengths and challenges in children’s learning experiences. The ultimate goal is to develop evidencebased interventions that enhance educational outcomes and foster adaptive strategies, improving both classroom performance and overall cognitive development. * Corresponding author. E-mail addresses: [email protected] (C. Ruffini), [email protected] (C. Pecini), [email protected] (D. Salda˜ na), [email protected] (P. Delgado). Contents lists available at ScienceDirect Learning and Instruction journal homepage: www.elsevier.com/locate/learninstruc https://doi.org/10.1016/j.learninstruc.2025.102186 Received 16 October 2024; Received in revised form 23 May 2025; Accepted 3 July 2025 Learning and Instruction 100 (2025) 102186 Available online 1 August 2025 0959-4752/© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/bync-nd/4.0/ ). 2008; McNamara et al., 2011). They also begin encountering expository texts alongside narratives (Bowen, 1999; Snow, 2002) and face more complex texts across various subjects, as well as high-stakes testing. Teachers frequently report a decline in reading performance during this period (Johnson, 2024), underscoring the importance of studying this age group to guide early interventions. Kintsch’s construction-integration model (1998) outlines three levels of comprehension: superficial linguistic processing, understanding semantic content, and forming situation-related analogies. This model highlights the importance of selecting relevant information and deactivating irrelevant details to create a coherent mental representation. Initially, working memory is engaged to organize and integrate information. The resulting situation model, however, can surpass working memory constraints and is stored in long-term memory. The specific application of these processes can vary depending on the cognitive demands of different text genres. Given the role of text genre in comprehension (Cartwright & Duke, 2019; Clinton et al., 2020), this research examines how reading medium and text genre affect comprehension of linear texts. It also explores potential variations based on participants’ reading comprehension skills, medium preferences, and primary use of computers. As digital technology increasingly replaces traditional reading and learning activities, understanding its impact on younger children is critical. 1.1. Digital vs paper text comprehension Recent meta-analyses (Clinton, 2019; Delgado et al., 2018; Furenes et al., 2021; Kong et al., 2018) have consistently demonstrated the “paper advantage” or “screen inferiority,” where reading on screens results in lower comprehension levels compared to paper. Although the effect size of this “screen inferiority” is modest, it is statistically significant across studies (e.g., Clinton, 2019: g = − 0.25; Delgado et al., 2018: g = − 0.21; Kong et al., 2018: effect size = − 0.21; Wang et al., 2007: effect size = − 0.08). Importantly, reading time does not appear to differ significantly between mediums (Clinton, 2019; Kong et al., 2018), but time constraints during comprehension tasks disproportionately impact screen reading performance (Delgado et al., 2018). Notably, Delgado et al. (2018) observed that the digital disadvantage has grown more pronounced in recent studies. Several theories attempt to explain the challenges of digital reading. One leading explanation is the Shallowing Hypothesis (Annisette & Lafreniere, 2017), which suggests that frequent use of digital devices for entertainment and quick information promotes superficial reading strategies. These strategies, characterized by skimming rather than deep reading, negatively affect comprehension in tasks requiring focused and analytical engagement, such as learning from texts. Deep reading involves skills like critical analysis, reflection, and insight, in contrast to surface-level reading that prioritizes speed over understanding (Wolf, Barzillai, & Dunne, 2009). The habitual use of digital media may foster less engaged reading behaviours, making deep cognitive tasks more difficult. Other factors contributing to the digital disadvantage include difficulties in self-monitoring reading comprehension when reading on screen, the cognitive demands of navigating digital texts, challenges posed by scrolling, and unfamiliarity with digital reading formats (Ackerman & Lauterman, 2012; Chen & Lin, 2016; Sanchez & Wiley, 2009; Wylie et al., 2018; but see Fesel et al., 2018). Reduced cognitive and metacognitive engagement when reading on screens may also amplify individual differences in fundamental cognitive resources, such as attention and working memory, further influencing the extent of screen inferiority (e.g., Cain & Oakhill, 2012; Kieffer et al., 2013; Kim et al., 2015; Seigneuric & Ehrlich, 2005). However, most studies in these meta-analyses focused on adult readers, predominantly undergraduates (Clinton, 2019; Delgado et al., 2018; Kong et al., 2018). In contrast, research on younger children often explores shared reading and medium effects, finding that printed narrative books read aloud by adults yield better comprehension than digital narrative books with voiceovers (Clinton-Lisell et al., 2024; Wang et al., 2007; but see Takacs et al., 2015). Comparatively few studies examine digital versus paper text comprehension among primary school children (e.g., Golan et al., 2018; Halamish & Elbaz, 2020; Kerr & Symons, 2006). Studies involving school-age children often focus on those over 10 years old, who are proficient in reading and accustomed to comprehension tasks (Salmer´ on et al., 2021; Aydemir et al., 2013; Eyre et al., 2017; Golan et al., 2018; Grimshaw et al., 2007; Halamish & Elbaz, 2020; Higgins et al., 2005; Ronconi et al., 2022; Støle et al., 2018). Results are mixed, with some studies finding no differences between mediums (Aydemir et al., 2013; Florit et al., 2022) and others confirming a digital disadvantage similar to adults (Eyre et al., 2017; Golan et al., 2018; Halamish & Elbaz, 2020; Kerr & Symons, 2006; Mangen et al., 2013; Støle et al., 2018). Research on younger children’s comprehension is sparse and shows mixed findings. Lenhard et al. (2017) observed pronounced screen inferiority among younger children (1st–6th grade), while Florit et al. (2022, 2023) found no disadvantage for first graders proficient in word reading and familiar with technology. However, a paper advantage emerged in inferential comprehension for children with weaker word reading skills. 1.2. Text genre and its interaction with reading medium Texts can be categorized into narrative and expository genres. Narrative texts are structured stories featuring characters, a setting, a plot, and problem resolution, aimed at entertaining the reader (Weaver & Kintsch, 1991). They follow temporal and causal event relationships, conveying experiences and meanings (Zabrucky & Moore, 1999; Zabrucky & Ratner, 1992). In contrast, expository texts, or informational texts, aim to communicate information about a specific topic, with a hierarchical structure starting with an introduction followed by detailed explanations (Collins & Gentner, 1980; Graesser et al., 1991; Graesser & Goodman, 1985; Medina & Pilonieta, 2006). Text genre significantly impacts comprehension (Duke & Roberts, 2010). According to Kintsch’s construction-integration framework (1998), text genre may influence learning at three levels: surface structure, textbase, and situation model (Clinton et al., 2020). Expository texts tend to feature less familiar vocabulary, more complex syntax, and require more prior topic-specific knowledge than narratives, making them harder to comprehend (Beers & Nagy, 2011; McNamara et al., 2012). Narratives, by contrast, are often easier due to their clear structure, real-life knowledge reliance, and coherence (Clinton et al., 2020; Graesser et al., 2004). Meta-analyses affirm that narrative texts outperform expository texts in comprehension, recall, and inference across all ages (Clinton et al., 2020; Mar et al., 2021), a trend especially evident in school-age children (Best et al., 2008; Olson, 1985; Spiro & Taylor, 1980; Tun, 1989). Early exposure to narratives during primary school likely contributes to this advantage (Duke, 2000). The transition from basic reading skills to academic learning occurs during the shift from primary to middle school (Best et al., 2008; McNamara et al., 2011), emphasizing the need for considering text genre in studies of text comprehension in both adults and children (Mar et al., 2021). Research investigating the medium-genre interaction reveals that digital reading disadvantages are more pronounced with expository texts. For instance, recent meta-analyses found no significant difference between digital and paper comprehension for narrative texts, but digital reading negatively impacted comprehension of expository texts (Clinton, 2019; Delgado et al., 2018). This effect may be due to the higher cognitive demands of expository texts, such as decoding complex vocabulary, processing intricate structures and constructing meaning from information that is less directly connected to real-world knowledge (Clinton et al., 2020; Delgado et al., 2018; Graesser et al., 2004). Digital reading, which often promotes superficial reading strategies (Annisette C. Ruffini et al. Learning and Instruction 100 (2025) 102186 2 & Lafreniere, 2017), may amplify these challenges. Metacognitive strategies, crucial for engaging deeply with expository texts, may also be less effectively deployed on digital platforms (Tibken et al., 2022, 2024). However, the interaction between genre and medium remains underexplored, with mixed results reported in the limited studies available (Margolin et al., 2013; Rasmusson, 2015; Simian et al., 2016). Few studies on school-age children have examined the interaction between text genre and reading medium. Among these, Florit et al. (2022) observed a narrative advantage over expository texts in paper reading among first graders, attributed to children’s limited digital exposure and greater familiarity with paper-based narratives. However, Florit et al. (2023) found no medium-genre interaction, citing structural similarities between the expository and narrative texts used. The mixed results and limited research underscore the need for further investigation into how genre influences medium effects on text comprehension in school-age children. 1.3. Exploratory variables of the relationship between reading medium and text genre The impact of reading medium and text genre on third-grade children’s comprehension of linear texts may depend on three variables linked to the reading medium effect: reading comprehension skills (e.g., Ruffini et al., 2023; Salmer´ on et al., 2021), medium preference (e.g., Ackerman & Goldsmith, 2011; Ackerman & Lauterman, 2012), and primary computer use (Annisette & Lafreniere, 2017; Florit et al., 2023). Comprehension skill levels significantly influence the relationship between reading medium and text genre. Lower-skilled comprehenders struggle with expository texts, which demand more cognitive and metacognitive resources than narratives, and often adopt superficial strategies with limited metacognitive awareness (Magliano et al., 1999; Kraal et al., 2019; Mar et al., 2021). Digital reading, being more cognitively demanding and distracting, exacerbates these difficulties, especially for those with lower working memory and weaker skills (Margolin et al., 2018). Studies showed low-skilled students performed better in print under time constraints, while high-skilled readers remain unaffected (Salmer´ on et al., 2021). Similarly, among undergraduates reading expository texts, skilled comprehenders show comparable performance across digital and paper mediums, while less-skilled comprehenders exhibit a screen inferiority effect (Stiegler-Balfour et al., 2023). However, digital reading may benefit certain subgroups: low comprehenders tend to perform better with digital narrative texts, while high comprehenders show better outcomes in paper mode—an interaction that diminishes with grade level but remains observable across grades 3 to 5 (Ruffini et al., 2023). Medium preference is a key variable to investigate in the relationship between reading medium and text genre, as it reflects individuals’ comfort and motivation when using a specific medium, which could, in turn, shape how effectively they engage with and comprehend different types of texts. While adults generally prefer reading from print, children tend to favor screens (Golan et al., 2018; Huang et al., 2012). Some argue that children’s early exposure to technology allows them to comprehend texts equally well—or even better—on screens than on paper (e.g., Chen et al., 2014). Others suggest this familiarity may lead to faster but more superficial reading, impairing comprehension (e.g., Duncan et al., 2016; Pfost et al., 2013). Only one study, however, has explored medium preference and its role in reading comprehension among elementary education students, focusing solely on fifth graders (Halamish & Elbaz, 2020). This study found no moderating effect of medium preference on comprehension but did not consider whether the effect varied by text genre. Furthermore, medium preference may not only reflect simple liking or motivation but could also stem from familiarity, comfort, or habitual strategies that readers develop with a particular medium. These strategies might transfer—or fail to transfer—across reading media, influencing comprehension differently depending on the context and the nature of the text. There is a gap in understanding how medium preference might influence younger children, who are still developing reading habits and may approach different media differently than older students. Investigating medium preference among younger readers is crucial, as their preferences could interact with text genres to affect both engagement and comprehension in unique ways. Furthermore, when examining the interaction between reading medium and text genre among children, computer use habits is an important factor to consider. In educational research, main computer use typically refers to the primary purpose for which an individual uses a computer, often categorized as either academic (e.g., homework, school tasks) or leisure (e.g., gaming, streaming, socializing; Rocheleau, 1995; Tondeur et al., 2007). Since younger children have limited experience using digital devices for leisure reading or entertainment, they may be less prone to developing superficial digital reading habits (Annisette & Lafreniere, 2017; Florit et al., 2023). Halamish and Elbaz (2020) studied fifth graders and found that the paper advantage in reading comprehension was independent of computer use habits. However, this does not rule out the possibility of a different effect in younger children or with various text genres. Younger children’s technology use tends to be more regulated and focused on educational activities, which may make their reading comprehension more sensitive to their level of computer experience (´ Alvarez et al., 2013; Cadena SER, 2025). Exploring these factors in third graders is essential to understanding how medium, genre, and individual differences interact in early education. Lastly, individual differences in reading comprehension skills have been found in two previous studies to moderate the effect of reading medium among elementary school students (Ruffini et al., 2023; Salmer´ on et al., 2021). Salmer´ on et al. (2021) reported that digital reading hindered comprehension among fifthand sixth-grade students with lower reading skills. In contrast, students with strong reading comprehension performed equally well on paper and tablets. These proficient readers likely possess more effective monitoring and self-regulation strategies, enabling them to adapt more successfully to digital environments. This finding supports the hypothesis that reading comprehension skills can moderate the screen inferiority effect, suggesting that stronger readers are better equipped to manage the cognitive demands of digital reading. However, Ruffini et al. (2023) observed the opposite pattern among students in Grades 3 to 5. They argued that the digital advantage found for low comprehenders may be related to reduced working memory demands in digital formats—an effect more pronounced in this group. By contrast, high comprehenders, who typically exhibit stronger executive functioning, may be better suited to the demands of paper-based reading. These contrasting findings highlight that the role of reading comprehension skills in moderating the medium effect remains an open question, warranting further investigation. In sum, although the effects of digital versus print reading have received increasing attention, the literature has not yet provided a comprehensive account of how reading medium and text genre interact with individual factors such as comprehension skills, medium preference, and primary computer use. This gap is particularly notable with regard to third-grade children, who are at a critical stage in the development of reading comprehension. 1.4. The present study The present study investigates the effect of reading medium (print vs. digital) and its interaction with text genre (expository vs. narrative) on the comprehension of linear text in third-grade children. Additionally, it examines whether these potential effects vary based on three variables previously suggested to relate to the reading medium effect: participants’ reading comprehension skills (e.g., Ruffini et al., 2023; Salmer´ on et al., 2021), medium preference (e.g., Ackerman & Goldsmith, 2011; Ackerman & Lauterman, 2012), and primary use of computers (Annisette & Lafreniere, 2017; Florit et al., 2023), while controlling for students’ attention and working memory capacity. C. Ruffini et al. Learning and Instruction 100 (2025) 102186 3 Two main research questions (RQ) were followed: (1) Do reading medium (paper vs computer) and its interaction with text genre (narrative vs expository) affect reading comprehension performance in third graders? (2) Does the influence of reading medium and its interaction with text genre vary depending on participants’ reading comprehension skills, medium preference, and main use of computers? In line with RQ1, it was hypothesised that (H1a) reading comprehension would be lower when reading in a digital format compared to paper (e.g., Delgado et al., 2018; Golan et al., 2018; Halamish & Elbaz, 2020; Støle et al., 2020) and (H1b) comprehension levels would be higher for narrative texts compared to expository ones (Clinton et al., 2020; Mar et al., 2021). Furthermore, based on recent meta-analyses (Clinton, 2019; Delgado et al., 2018), (H1c) a larger difference in comprehension by medium was expected for expository texts than for narrative texts. With regard to RQ2, it was hypothesised that (H2a) the mediumrelated difference in comprehension would be larger for children with lower reading comprehension skills, with paper being more beneficial (Salmer´ on et al., 2021; Stiegler-Balfour et al., 2023). Additionally, (H2b) no significant relation with medium preference was expected (Golan et al., 2018; Halamish & Elbaz, 2020). Lastly, (H2c) it was expected that children who frequently use computers for educational purposes would show smaller differences in comprehension across media, whereas those who primarily use computers for recreational purposes would benefit more from reading on paper (i.e., the Shallowing Hypothesis; Annisette & Lafreniere, 2017). To account for individual differences, students’ attentional capacity and working memory were measured and statistically controlled in the analyses. 2. Method 2.1. Participants Nine third grade classes took part in the project. All children whose parents provided written informed consent for participation were involved in the study. A total of 184 children initially participated in the study; however, the final sample for analysis consisted of 157 typically developing children (mean age =8.40 years, SD =0.30, 48.41 % female) with no known language or reading difficulties, as verified through teacher reports and classroom observations. All the children spoke and understood Spanish. Among the initial 184 participants, seven children were absent from at least one project session, and 20 were identified as having special educational needs (SEN). These children were officially documented as requiring support due to conditions such as neurodevelopmental disorders, curriculum modifications, or foreign language assistance. To ensure consistency and to avoid potential interference from cognitive or learning challenges in text comprehension, these children were excluded from the analysis. The classes belonged to four different schools in Seville, Spain: one public school attended by children from an upper-middle socio-cultural background and located in a rich neighbourhood and three public schools located in a working-class neighbourhood with a high frequency of foreign children. All schools participating in this project had new laptops, at least one for each child, and Wi-Fi connection. However, the teachers of the classes involved reported verbally to the experimenters that they rarely used computers at school with the children. We performed a priori statistical power analysis to determine the minimum sample size needed for our study. Our hypotheses were tested by means of linear mixed-effect models (LMM; see Data analysis below), but statistical power for this type of analysis need to be based in raw data from prior similar studies (Kumle et al., 2021), to which we did not have access. Thus, we tested a priori statistical power for mixed ANOVA using G*Power 3.1 (Faul et al., 2007). Type I error was set at 0.05 and type II error at 0.20. We expected correlations between within-participant scores in text comprehension to be strong, so we set the correlation value at 0.75. We calculated the minimum sample size required for a 3-way interaction between two 2-level within-participant factors (i.e., medium and genre; four repeated measures in total) and a 2-level between-participant factor (i.e., reading skills groups [low, high], medium preference [print, digital], or main use of computers [leisure, homework]), as this was the most demanding analysis among those planned. As we expected an effect size of d =0.27 for the effect of reading medium on comprehension of expository texts (as found in the meta-analysis by Delgado et al. (2018) and no effect of medium in narrative texts (i.e., d =0), the expected effect size included in the a priori power analysis was set at d =0.135 (the effect size was halved as recommended in Brysbaert, 2019). The results yielded a required sample size of 142 participants. Thus, and given that LMMs offer greater statistical power than ANOVAs (Luke, 2017), the final sample size of 157 students in our study was appropriate for our purposes. 2.2. Experimental materials and measures 2.2.1. Texts Four texts, two narrative and two expository, suitable for school-age children, were used both on screen and on paper. The two narrative texts were taken from an Italian standardised battery for the assessment of text comprehension (Prove MT-Kit scuola, Cornoldi et al., 2017), translated and adapted into Spanish. They deal respectively with the story of a cat and her kittens who are rescued from the frost thanks to a child and the story of the King of England who rests in a worker’s house without being recognised. The two expository texts were created ad hoc by the research team. They deal with the topic of bees and Antarctica, respectively. The characteristics of the texts used are shown in Table 1. In addition to text length, two indices specific to Spanish texts were used to analyse the texts: the readability and grade level indices. The INFLESZ scale (Barrio-Cantalejo et al., 2008) is an index of the readability of a text based on the number of syllables, words, and sentences. Crawford index (1989) measures the grade level, the years of school needed to understand the text. It only applies to primary school children, and it is calculated considering the number of years of schooling, the number of sentences per hundred words, and the number of syllables per hundred words. Based on both indexes, as can be seen in Table 1, the two texts for each text genre were qualified as quite easy (difficulty equivalent to that of primary education textbooks and gossip magazines) and appropriate for 4th grade, and as very easy (equivalent to primary education textbooks and children’s comics) and appropriate for 3rd grade, respectively. Other distinctive features across texts are provided in the Supplementary Materials (see Table A1). 2.2.2. Dependent measure Text Comprehension. After reading each text, children were asked to answer a set of multiple-choice questions on text content by indicating Table 1 Characteristics of the texts used. Text Length INFLESZ scale ( Barrio-Cantalejo et al., 2008) Grade level ( Crawford, 1989) Text A narrative – The Cat 220 74.20 quite easy 4.5 Text B narrative – King Alfred 271 81.43 very easy 3.7 Text A expository - Bee Dance 287 72.04 quite easy 4.6 Text B expository - What is Antarctica? 243 80.34 very easy 3.9 Note: “Quite easy” (65–80) and “Very easy” (>80). C. Ruffini et al. Learning and Instruction 100 (2025) 102186 4 the correct answer among four response options without a time limit. According to the medium assigned, they could mark their answer with the pen or using the touchpad of the keyboard. Half of the questions addressed literal comprehension (i.e., comprehension of single ideas or details explicitly stated in the texts), whereas the other half addressed inferential comprehension (i.e., an idea that needs to be drawn from by bridging two ideas in the texts). An example of a literal question is “What colour was the cat?“, as it seeks an explicit, objective piece of information without any deeper or figurative meaning. An example of an inferential question is “Why do you think scientists usually go to the Antarctic between October and March?“, as the correct answer requires bridging this information with the fact that it is less cold during the summer in Antarctica, which runs between October and March. The aggregate score of literal and inferential questions was used as the dependent variable. The set of questions for each narrative text consisted of the same 10 questions included in the above-mentioned standardised battery, whereas 8 questions were created by the researchers for each expository text. Given the difference in number of questions between both text genres, the scores were transformed to proportion of correct answers, so that minimum and maximum potential scores for each text were 0 and 1, respectively. The reliability of each set of questions in our study’s sample was examined using the omega coefficient (McDonald, 1999), calculated based on a polychoric transformed correlation matrix. The omega coefficient is considered more appropriate than Cronbach’s alpha for dichotomous items (Trizano-Hermosilla & Alvarado, 2016). The reliability of the questions for the texts The Cat, Bee Dance, and What Is Antarctica? was acceptable ( ω =0.72, ω =0.71, and ω =0.71, respectively), whereas it was good in the case of the text King Alfred ( ω =0.86). As the combined use of a standardised test and a researcher-made test is potentially problematic, we further explored the consistency of these results across reading mediums. As expected, reliability for the standardised test was more consistent (The Cat in print: 0.73, on screen: 0.72; King Alfred in print: 0.87; on screen: 0.86) that the researcher-made test (Bee Dance in print: 0.66, on screen: 0.77; What Is Antarctica? in print: 0.63; on screen: 0.66). In addition, we examined the frequency of selection of each response option per question. As can be seen in the Supplementary Materials (see Table A2), some of the distractor options were substantially more selected than others, both for the standardized and the researcher made questions. Both circumstances were discussed as a limitation of our study. 2.2.3. Individual differences We measured three factors related to students’ individual differences that have been suggested in previous literature to influence the reading medium effect on text comprehension outcomes: reading comprehension skills (e.g. Salmer´ on et al., 2021; Stiegler-Balfour et al., 2023), reading medium preference (e.g. Golan et al., 2018; Halamish & Elbaz, 2020), and type of main use of computers (e.g. Annisette & Lafreniere, 2017). Reading Comprehension Skills. The PROLEC-R text comprehension subtest (Cuetos et al., 2009) of the Batería de Evaluaci´ on de los Procesos Lectores is a Spanish standardised test assessing text comprehension through four texts (two expository and two narrative). Children are required to read one text and answer four open questions, then repeat this for all texts without time limits. Children are not allowed to return to the text once the page of questions has been turned. The number of correct answers (CR, from 0 to 16) were recorded. The internal consistency (Cronbach’s α ) of this test, as reported in the original test manual, is 0.72 (Cuetos et al., 2009). Medium Preference for Reading. Each child answered the question “Where do you prefer to read a text?“. Response options were “On the computer”, “In a book”, or “No preference”. Computer Main Use. Each child answered the question “What do you use the computer most for?“. Response options were “Doing homework”, “Playing games”, “Seeing and talking with friends”, “Watching TV series”, “Other”. 2.2.4. Additional control measures As control measures, data was collected on children’s attentional capacity and working memory capacity. Attentional Capacity. The CARAS-R test (Thurstone & Yela, 2014) is a measure of perceptual and attentional skills. It assesses the ability to perceive similarities and differences quickly and correctly among 60 graphic items consisting of schematic drawings of faces. Children are asked to determine which of the three faces constituting each item is different from the other two. Faces can differ in hair, eyebrows, eye, and mouth. All items are placed inside a sheet on which the children mark the answer with a cross on the different face of each item. This test was administered collectively to each class group in 3 min. Number of correct responses (CR, from 0 to 60) and number of commissions (Com, from 0 to 60) were recorded. This test was validated (Thurstone & Yela, 2014) in Spain in a sample of 12,190 students (Cronbach’s α =.91) Working Memory Capacity. The Letters and Numbers reordering subtest of the WISC V measures working memory ability (Wechsler, 2015; adaptation from Capodieci et al., 2019). Children are read a progressively longer series of letters and numbers mixed and are asked to write them by first signing the numbers in ascending order and then the letters in alphabetical order. The children marked their answers on a blank sheet of paper consisting of one line per item. The number of correct responses (CR, from 0 to 24) was recorded. The internal consistency (Cronbach’s α ) of the original version of the test is 0.84 (Wechsler, 2015). 2.3. Procedure For the present study, a within-participant design was implemented. All children took part in an initial session where a standardised text comprehension test (PROLEC-R test), an attention test (CARAS-R test) and a working memory test (Letters and number reordering, WISC V) were administered at class level. In the second and third sessions, all the children read two texts (one expository and one narrative) on paper or on a computer depending on the randomly assigned group. The order of the texts (narrative, expository) and the type of text (paper, digital) were counterbalanced between the children. The computers used for this project were those provided by the schools, in particular all public schools had the same computer provided by the Spanish government (HP Intel Core 13 inches) while the private school had a computer provided directly by the school to each student (HP intel inside 11 inches). Each digital text appeared on one page, while the questions were presented on a separate page that required scrolling to view all of them. The text size was set to the same point size across all devices, maintaining consistent proportional dimensions relative to the screen. The digital texts were presented on the computer via the Qualtrics online digital platform, regularly accessible from any web page. The paper texts were provided to each student in a paper protocol format: each text was presented in one page while questions were grouped on various pages (3 or 4 according to the different texts). After completing the computerised experimental text comprehension task, the children completed a digital questionnaire on their reading medium preference, main use of computers, and basic computer skills. 2.4. Data analysis To test our hypotheses, we performed linear mixed-effect models (LMM) with restricted maximum likelihood estimation including students’ comprehension of the experimental texts as dependent variable. The model building process was as follows. We first constructed the null model by adding student, text, and student school class as random effects. The intercept of these random effects was set as random in all cases, whereas decisions on their slopes were made based on goodness of fit comparisons between different null models, as recommended by C. Ruffini et al. Learning and Instruction 100 (2025) 102186 5 Meteyard and Davies (2020). None of the null models that included random slopes for any of the random effects –whether estimated for reading medium or for text genre– improved model’s fit. Thus, we based on the most parsimonious null model (i.e., the model including all the slopes as fixed) to add the fixed effects for testing our hypotheses. Before adding the factors to test our hypotheses, we performed Pearson pairwise correlations between students’ attentional capacity, working memory capacity, reading comprehension skills, and text comprehension scores. All of these three individual differences significantly correlated with text comprehension scores (see Table 3 in Results), thus, they were included as covariates in the LMMs described below. For testing H1a and H1b we added to the model the main effect terms of reading medium and text genre as fixed factors. To test H1c we included the interaction term of these factors instead of their main effect terms. For H2a (i.e., moderator effect of children’s comprehension skills on the reading medium effect), we split the sample into two groups based on their scores on the PROLEC test. Children who scored below the sample median score (i.e., 10 points) were categorized as low in reading comprehension skills (n =75), whereas those whose score was equal or above the sample median were categorized as high in reading comprehension skills (n =82). We then built a model including the interaction term of reading medium, genre, and reading comprehension skills group as fixed factors. To test H2b and H2c (i.e., effects of children’s medium preference and of children’s main use of computers, respectively), the models included the interaction term of reading medium, text genre, and medium preference (H2b); and the interaction term of reading medium, text genre, and computer main use (H2c), respectively. Medium preference and Computer main use were included as categorical variables which included, respectively, 3 possible values (Preference for computer, For print, or No preference) and 2 possible values (Use for homework, i.e., “Doing homework” responses) or Use for leisure (i.e., any response other than “Doing homework”; see Experimental materials and measures above). All contrasts involving categorical variables were simple contrasts, where each category is compared to the reference category. To aid interpretation, the reference category for each categorical variable is indicated in the footnote of the tables that show the results from the models in the following section. Additionally, given that H2b posited a null hypothesis, a traditional null hypothesis significance test (NHST) would not allow us to conclude equivalence in the case of non-significant results. Therefore, we conducted an equivalence test focused on the interaction between medium preference and reading medium. We estimated the effect of reading medium (i.e., screen minus paper) separately for each combination of medium preference and text genre. We then compared these medium effects across all pairs of medium preference groups (i.e., computer, print, and no preference), within each genre. For each pairwise comparison, we computed the difference in the medium effect between the two preference groups and derived a 90 % confidence interval. These differences were standardized by the residual standard deviation of the model to express the results in Cohen’s d units. Equivalence was evaluated using a predefined equivalence margin of ±0.20 d units, consistent with the conventional threshold for a small effect size (Cohen, 1988). A difference was considered statistically equivalent to zero if its confidence interval was fully contained within the equivalence margin. All the analyses described above were performed using the R software version 4.4.1 in R studio (2024.09.0 +375 version). We performed Pearson correlations using the cor function (stats package v4.4.1; R Core Team, 2024), LMMs using the lmer function (lme4 package v.1.1–35.5; Bates et al., 2024), and model comparisons using the anova function (stats package v.4.4.1; R Core Team, 2024), which evaluates the variance explained by each model based on likelihood ratio testing. We calculated confidence intervals of the models’ estimate with the confint function (also stats package v.4.4.1) and models’ fit using the “r.squaredGLMM” function (“MuMIn” package v.1.48.4; Barto´ n, 2024). To perform equivalence testing for H2b, estimated marginal means for each level of reading medium within combinations of medium preference and genre were obtained using the emmeans function (emmeans package v.1.10.1; Lenth, 2024). Differences in medium effects across preference groups were computed using the contrast function, (also (emmeans package v.1.10.1). Standardized effect sizes (Cohen’s d) were calculated by dividing the estimated contrasts by the residual standard deviation obtained using the sigma function (stats package v.4.4.1; R Core Team, 2024), and confidence intervals were derived using the qnorm function (also stats package v.4.4.1). Prior to performing the analyses described above, we inspected the presence of outlier values (±2.5SD from the sample mean) and checked normality based on Kurtosis and Skewness values of data distribution for all the variables (see Results). 3. Results As shown in Table 2, which presents means and standard deviations of children’s text comprehension scores in each experimental condition across reading medium and text genre as well as covariate variables, all the variables were approximately normally distributed, as Kurtosis and Skewness values fell within ±1 range (George & Mallery, 2010). We identified no outlier values in any of the variables except for the covariate reading comprehension skills, two values (score on PROLEC test =2) were slightly lower than −2.5SD from the sample mean. Following the winsorization method (Field, 2013), these data were replaced by the next highest score that was not an outlier (i.e., score =3). Distribution of comprehension scores per text and reading medium were examined. As can be seen in Fig. A3 in the Supplementary Materials, the scores showed enough variability to explore differences in this measure across the experimental conditions. As already mentioned, students’ attentional capacity, working memory capacity, and reading comprehension skills positively correlated with text comprehension scores in all the experimental conditions, showing r values larger than 0.25 in all cases. As shown in Table 3, all of them remained significant once Bonferroni-correction for multiple correlations was applied (i.e., α =.0017) except for the correlations between attentional capacity and comprehension scores in the screenexpository condition, and between working memory and comprehension scores in the screen-narrative condition. Therefore, all these covariates were included in the LMMs to test our experimental hypotheses. 3.1. The effect of reading medium and text genre on text comprehension As can be seen in Table 4, the model including the main effect terms of both factors (H1a and HIb) showed no significant main effect of reading medium (t = − 0.128, p =.898) and a significant main effect of text genre indicating that text comprehension scores were higher for the narrative texts than for the expository texts (estimate: 0.116, SE =0.018, 95 % CI [0.080, 0.152], p =.024). In addition, the model including the interaction term of both factors (H1c) showed no significant interaction effect of medium and genre on text comprehension scores (t = − 1.671, p =.095; see Table 5). 3.2. Relationship between reading medium, text genre and participants’ reading comprehension skills As described in a previous section, to explore whether the effects of reading medium and text genre varied depending on participants’ reading comprehension skills (H2a), we split the sample into two groups: students with low and high reading comprehension skills, respectively. Results of the model including the interaction term of these three factors revealed that the influences of reading medium and text genre on text comprehension scores were not qualified by children’s reading comprehension skills, as there were no significant interactions between reading comprehension skills group and reading medium, reading comprehension skills group and text genre, or between reading C. Ruffini et al. Learning and Instruction 100 (2025) 102186 6 comprehension skills group, reading medium, and text genre (all ts < 1.148, ps >0.252; see Table 6). 3.3. Relationship between reading medium, text genre and participants’ medium preference, and main use of computers Lastly, we examined whether the effects of reading medium and text genre were related with children’s reading medium preference (i.e., print, computer, or no preference; n =49, n =70, and n =38, respectively; H2b) and main use of computers (i.e., for leisure or for school tasks; n =79 and n =78, respectively; H3c). The models performed showed no significant interactions between these factors, with all ts < 1.246 and ps >0.214 in the case of the model including children’s medium preference (see Table 7), and all ts <1.168 and ps >0.243 in the model including children’s main use of computers (see Table 9). Thus, the effects of reading medium and text genre were also not qualified by these two variables. As explained earlier, to further assess H2b (i.e., null hypothesis about the moderating effect of medium preference on the effect of medium on reading comprehension), we conducted an equivalence test using estimated marginal means and equivalence margin of ±0.20 standardized units (Cohen’s d). The analysis compared reading medium differences across combinations of medium preference and text genre. Importantly, none of the 6 comparisons yielded confidence intervals that fell entirely within the equivalence margins, including three effect sizes that exceeded those bounds (see Table 8). This result indicated that the interaction effects could not be considered statistically equivalent to zero, suggesting that, although no significant moderation was found, the data do not provide sufficient evidence to confirm the absence of a meaningful moderating effect of children’s medium preference. In sum, our results revealed the absence of interaction between reading medium and text genre and better performance on narrative texts in comparison to expository texts. Moreover, controlling for the role of inter-subjects variability in attention and working memory, there was no effect of the reading medium on their text comprehension scores regardless of text genre and children’s reading comprehension skills, medium preference for reading, and main use of computers. Nevertheless, the null hypothesis regarding the moderating effect of medium preference was not supported by the equivalence test. 4. Discussion The present study aimed to investigate the effects of reading medium (print vs. digital) and its interaction with text genre (expository vs. narrative) on comprehension outcomes among third graders (RQ1) and to examine whether these effects vary based on participants’ reading comprehension skills, medium preference, and primary use of computers (RQ2), controlling for attention skills and working memory. Unlike previous research, which has primarily focused on older students or examined these factors in isolation, this study uniquely explores their combined influence in a young population. To our knowledge, it is the first to specifically target third graders while simultaneously considering Table 2 Descriptive statistics of the measured variables. Attention Working memory RC skills Print-Nar. comprehension PrintExp. comprehension Screen-Nar. comprehension Screen-Exp. comprehension All texts comprehension Mean 27.382 10.701 9.586 0.633 0.495 0.613 0.519 0.565 SD 7.972 5.622 2.670 0.227 0.216 0.214 0.230 0.169 Skewness 0.086 −0.358 −0.239 −0.484 0.052 −0.152 −0.021 0.051 Kurtosis −0.332 −0.985 −0.594 −0.467 −0.328 −0.753 −0.418 −0.749 Note. RC skills: Reading comprehension skills. Nar: Narrative. Exp: Expositive. Table 3 Pearson correlations between the measured variables. 1234567 1. Attention –      2. WM 0.251 –     3. RC skills 0.292* 0.338* –    4. Print-Nar. 0.304* 0.303* 0.455* –   5. Screen-Nar. 0.359* 0.214 0.366* 0.524* –  6. Print-Exp. 0.292* 0.392* 0.464* 0.442* 0.358* – 7. Screen-Exp. 0.238 0.281* 0.453* 0.490* 0.490* 0.334* – 8. All texts 0.390* 0.390* 0.571* 0.810* 0.774* 0.695* 0.766* Note. Alpha value was set at 0.0017 after applying Bonferroni correction. *p <.0017. WM: Working memory. RC skills: Reading comprehension skills. Nar.: Narrative. Exp.: Expositive. Table 4 LMM for text comprehension scores including main effect terms of reading medium and text genre. Fixed effects Estimate SE 95 % CI t (df) p Intercept 0.074 0.051 [-0.024, 0.174] 1.480 (103.67) 0.142 Working memory capacity 0.005 0.002 [0.001, 0.009] 2.350 (151.64) 0.020 Attentional capacity 0.004 0.001 [0.001, 0.007] 2.865 (144.14) 0.005 Reading comprehension skills 0.028 0.004 [0.019, 0.036] 6.587 (151.61) <0.001 Reading medium 0.002 0.013 [-0.024, 0.028] 0.128 (467.92) 0.898 Text genre 0.116 0.018 [0.080, 0.152] 6.320 (2.00) 0.024 Random effects Variance SD Student (intercept) 0.009 0.096 School class (intercept) 0.001 0.036 Text (intercept) 0.0001 0.013 Model fit Marginal R 2 Conditional R 2 0.268 0.472 Note. Reference value for reading medium: Print. Reference value for text genre: Expository. C. Ruffini et al. Learning and Instruction 100 (2025) 102186 7 multiple cognitive and contextual variables that may contribute to differences in reading comprehension across mediums. It was hypothesised to find a digital disadvantage in reading comprehension in comparison to paper medium (e.g. Delgado et al., 2018; Golan et al., 2018; Halamish & Elbaz, 2020; Støle et al., 2020) as well as a higher comprehension level of narrative texts in comparison to expository ones (Clinton et al., 2020; Mar et al., 2021). Results partially confirmed this initial hypothesis. Whereas there were no differences between comprehending texts on computer and on paper among third graders, they comprehended better the narrative texts than the expository texts. Although an increased medium effect was expected when reading expository texts (vs narrative), no medium-genre interaction was found. Secondly, it was supposed that low comprehenders could obtain an advantage of paper in comparison to digital medium (Salmer´ on et al., 2021; Stiegler-Balfour et al., 2023), as well as it was expected an absence of effect of medium preference on digital and paper differences (Golan et al., 2018; Halamish & Elbaz, 2020). The results partially support the second hypothesis, as the expected differences based on children’s reading comprehension skills did not arise, whereas medium preference did not influence the effect of reading medium on text comprehension outcomes (as expected). Lastly, contrary to our expectation based on the Shallowing hypothesis (Annisette & Lafreniere, 2017), children who usually utilised computers to perform leisure activities did not show larger differences between digital and paper reading compared to those who used computers mainly for academic tasks. 4.1. The effect of reading medium and text genre on text comprehension 4.1.1. Differences between digital and paper reading comprehension As the first results, the study found no discernible difference in performance between third graders when reading on screen versus on paper. However, this finding warrants further examination within a theoretical framework that can explain why differences in reading performance between digital and paper-based mediums are sometimes observed and sometimes absent. One relevant framework is the Shallowing hypothesis (Annisette & Lafreniere, 2017), according to which the frequent use of digital devices for instant gratification can lead to superficial processing in digital reading. In the context of young children, it is plausible to hypothesize that this mechanism is not yet fully applicable. Specifically, young children may not have sufficient exposure to digital devices for leisure activities such as social media browsing or web surfing—activities that could reinforce a habitual, shallow reading approach. Consequently, they might not have developed the fast, superficial reading patterns typically observed in adults using digital devices, nor generalized such patterns to digital text Table 5 LMM for text comprehension scores including interaction effect term of reading medium and text genre. Fixed effects Estimate SE 95 % CI t p Intercept 0.064 0.051 [-0.036, 0.164] 1.252 (107.48) 0.213 Working memory capacity 0.005 0.002 [0.001, 0.009] 2.350 (151.64) 0.020 Attentional capacity 0.004 0.001 [0.001, 0.007] 2.865 (144.14) 0.005 Reading comprehension skills 0.028 0.004 [0.019, 0.036] 6.587 (151.62) <0.001 Reading medium 0.024 0.019 [-0.013, 0.060] 1.272 (466.02) 0.204 Text genre 0.138 0.023 [0.094, 0.182] 6.118 (4.64) 0.002 Reading medium ×Text genre −0.044 0.026 [-0.096, 0.008] −1.671 (466.02) 0.095 Random effects Variance SD Student (intercept) 0.009 0.096 School class (intercept) 0.001 0.036 Text (intercept) 0.0002 0.013 Model fit Marginal R 2 Conditional R 2 0.270 0.474 Note. Reference value for reading medium: Print. Reference value for text genre: Expository. Table 6 LMM for text comprehension scores including interaction effect term of reading medium, text genre, and reading comprehension skills group. Fixed effects Estimate SE 95 % CI t (df) p Intercept 0.222 0.048 [0.128, 0.315] 4.639 (95.96) <0.001 Working memory capacity 0.006 0.002 [0.001, 0.010] 2.663 (151.98) 0.009 Attentional capacity 0.005 0.001 [0.002, 0.008] 3.558 (145.37) <0.001 Reading medium 0.018 0.027 [-0.035, 0.071] 0.657 (464.99) 0.511 Text genre 0.134 0.030 [0.076, 0.193] 4.517 (14.22) <0.001 RC skills group 0.128 0.032 [0.065, 0.191] 3.980 (464.99) <0.001 Reading medium × Text genre −0.012 0.038 [-0.087, 0.063] −0.322 (464.91) 0.748 RC skills group ×Reading medium 0.011 0.037 [-0.062, 0.085] 0.306 (464.70) 0.760 RC skills group ×Text genre 0.006 0.037 [-0.067, 0.080] 0.171 (464.85) 0.864 RC skills group ×Reading medium × Text genre −0.061 0.053 [-0.165, 0.043] −1.148 (464.62) 0.252 Random effects Variance SD Student (intercept) 0.010 0.102 School class (intercept) 0.002 0.041 Text (intercept) 0.0002 0.012 Model fit Marginal R 2 Conditional R 2 0.242 0.475 Note. RC skills: Reading comprehension skills. Reference value for reading medium: Print. Reference value for text genre: Expository. Reference value for RC skills group: Low. C. Ruffini et al. Learning and Instruction 100 (2025) 102186 8 comprehension tasks. Secondly, and closely tied to the above explanation, it is possible that young children may not yet have associated reading on paper with duty and reading on a computer with pleasure, leading to similar cognitive and motivational efforts across both mediums (Florit et al., 2022). This explanation aligns with evidence suggesting that in Spain, computer and internet use among primary school-aged children is primarily oriented toward academic purposes rather than recreational ones (´ Alvarez et al., 2013). Moreover, parental regulation plays a significant role in shaping children’s digital habits, with active guidance limiting exposure to non-educational activities. The fact that Spanish children typically receive their first mobile phone around the age of 12 further underscores the limited role of personal technology in younger children’s leisure activities (´ Alvarez et al., 2013; Cadena SER, 2025). However, the argument that primary school students are not or minimally exposed to digital devices for leisure activities remains speculative and requires stronger empirical evidence. Moreover, methodological factors, such as the setting of execution, could also play a role, with distractions in group settings (peers and other stimuli present in classrooms or corridors) potentially negating the advantages of paper reading observed in individual tasks (Ruffini et al., 2023). Fourthly, the absence of time constraints in the study may have mitigated differences in performance between digital and paper reading, as time pressure typically favours paper-based comprehension (Delgado et al., 2018; Kerr & Symons, 2006). It is also possible that students compensated for medium-specific processing demands by taking more time when reading digitally. In this case, similar comprehension outcomes across media might reflect differences in processing efficiency rather than equivalent processing. Although time-on-task data were not collected, this possibility should be acknowledged as a nuance in interpreting the findings. Lastly, the novelty and perceived appeal of digital tasks in the school environment might have affected children’s preferences for digital comprehension tasks over paper ones (Ferri, 2014; Florit et al., 2022; Mumtaz, 2001). 4.1.2. Differences between expository and narrative texts As for the second result, comprehending an expository text was more difficult for third graders than comprehending a narrative text. This result is consistent with literature in both adults (Clinton et al., 2020; Mar et al., 2021) and children (Best et al., 2008; Tun, 1989). Lower performance in comprehending an expository text could be explained by considering that expository texts are characterised by more infrequent words and complex syntax, requiring more prior knowledge and world knowledge (Beers & Nagy, 2011; Clinton et al., 2020; McNamara et al., 2012). However, considering that in the present study readability of texts was controlled, it is important to acknowledge other possible reasons for the differences found between the two texts. Indeed, it is important to consider that narratives are easier to process and recall because they align with cognitive processing patterns, mirroring real-life Table 7 LMM for text comprehension scores including interaction effect term of reading medium, text genre, and children’s medium preference. Fixed effects Estimate SE 95 % CI t (df) p Intercept 0.101 0.056 [-0.008, 0.211] 1.811 (148.52) 0.072 Working memory capacity 0.004 0.002 [0.0003, 0.008] 2.100 (150.07) 0.037 Attentional capacity 0.004 0.001 [0.001, 0.007] 2.958 (136.39) 0.004 Reading comprehension skills 0.027 0.004 [0.019, 0.036] 6.544 (149.83) <0.001 Reading medium 0.014 0.034 [-0.052, 0.080] 0.416 (460.09) 0.678 Text genre 0.120 0.036 [0.050, 0.190] 3.347 (27.45) 0.002 Medium preference (computer) −0.059 0.037 [-0.131, 0.013] −1.606 (482.45) 0.109 Medium preference (no preference) −0.031 0.042 [-0.113, 0.052] −0.730 (498.15) 0.466 Reading medium ×Text genre −0.022 0.048 [-0.115, 0.071] −0.465 (461.99) 0.642 Medium preference (computer) ×Reading medium 0.003 0.044 [-0.083 0.090] 0.074 (448.79) 0.941 Medium preference (no preference) ×Reading medium 0.034 0.051 [-0.065, 0.134] 0.678 (461.50) 0.498 Medium preference (computer) ×Text genre 0.023 0.044 [-0.063, 0.109] 0.523 (462.00) 0.601 Medium preference (no preference) ×Text genre 0.032 0.051 [-0.068, 0.131] 0.621 (460.51) 0.535 Medium preference (computer) ×Reading medium ×Text genre −0.001 0.062 [-0.123 0.121] −0.014 (460.13) 0.989 Medium preference (no preference) ×Reading medium ×Text genre −0.089 0.072 [-0.230, 0.051] −1.246 (460.96) 0.214 Random effects Variance SD Student (intercept) 0.009 0.095 School class (intercept) 0.001 0.034 Text (intercept) 0.028 0.166 Model fit Marginal R 2 Conditional R 2 0.283 0.479 Note. RC: Reading comprehension. Reference value for reading medium: Print. Reference value for text genre: Expository. Reference value for medium preference: Print. Table 8 Standardized effect sizes for the medium effect and equivalence test results by medium preference and text genre. Genre Medium preference comparisons Mean Difference SE Cohen’s d90 % CI for dEquivalent Expository No pref. vs Pref. for comp. 0.031 0.048 0.188 [-0.284, 0.660] No Expository No pref. vs Pref. for print 0.034 0.051 0.207 [-0.297, 0.712] No Expository Pref. for comp. vs for print 0.003 0.045 0.020 [-0.423, 0.462] No Narrative No pref. vs Pref. for comp. −0.057 0.047 −0.346 [-0.815, 0.123] No Narrative No pref. vs Pref. for print −0.055 0.051 −0.331 [-0.834, 0.172] No Narrative Pref. for comp. vs for print 0.002 0.044 0.014 [-0.420, 0.448] No Note. No pref.: No preference. Pref. for comp.: Preference for computer. Pref. for print: Preference for print. C. Ruffini et al. Learning and Instruction 100 (2025) 102186 9