As mentioned before, both national and international research on reading education display Dutch students’ lack of engagement in reading instruction. One of the explanations for this lack of motivation can be found in the way reading education is conducted in Dutch primary education. A typical feature of Dutch reading instruction is the strong focus on strategy instruction. This focus is in fact so strong that practicing the application of strategies while answering questions about a text has become the main purpose of reading lessons (Bogaerds-Hazenberg et al., 2022). As a consequence, reading tasks hardly promote a transfer of reading skills to other subjects, and do not appeal to a sense of autonomy and engagement in children, which results in a lack of motivation (Aarnoutse, 2017).

Reading motivation is determined by many factors, including reader, task and text characteristics, and it has also been established that reading motivation and text comprehension are reciprocally related (Toste et al., 2020). Therefore, it is important to choose a reading instruction policy that supports motivation (Houtveen et al., 2019). One of the instructional practices that have proven to enhance both children’s reading motivation and text comprehension, is the integration of reading instruction in content-area subjects. In such practices, knowledge acquired through content-area instruction can be viewed as both a foundation and a motive for reading (Cervetti et al., 2012). In curricula integrating literacy and content learning, language education is often integrated in science. This makes sense as scientific inquiry learning provides a meaningful context for reading, and reading can help students construct their understanding of science concepts (Bradbury, 2014). Texts can thus support scientific inquiry in broadening and outlining what is learned in firsthand investigation and, in addition, science and literacy share meaning-making strategies such as metacognitive regulation and making connections (Cervetti et al., 2005). Intervention programs that integrate science and language arts at the elementary level have shown positive effects on both science knowledge and language proficiency as measured via reading assessments and writing tasks. On top of that, integrated approaches facilitate improved attitudes toward both reading and science (Bradbury, 2014). Despite these benefits, an integrative approach of science and literacy is rarely practiced in the Dutch context of primary education (Gresnigt, 2018).

A closer look at the results of PIRLS 2016 but also at the outcomes of assessments of reading proficiency conducted at the national level, reveals that reading tasks appear to be particularly difficult when texts are complex, and students have to make inferences within and beyond the text, make multiple inferences, or summarize and connect text parts beyond the sentence level (Gubbels et al., 2017; Kühlemeier et al., 2014). These outcomes indicate that Dutch children in primary education struggle particularly with higher-order reading skills.

In order to tackle this problem, and learn children how to build a coherent mental representation of the macrostructure of a text, it has been recommended to provide text structure instruction (Bogaerds-Hazenberg et al., 2020; Hebert et al., 2016; Pyle et al., 2017). Text structure concerns the way in which the ideas in a text are organized and related (Pyle et al., 2017, p. 469); it plays a prominent role in the representation that readers make of the information in a text. A clear text structure makes it easier to construct a coherent representation (Sanders & Sanders, 2006). Coherence relations between parts of a text can be found at three levels: between clauses, between sentences, and between larger text parts such as paragraphs, the so called top-level structure of the text (Jones et al., 2016). In other words, coherence relations like cause-consequence and list do not just exist between consecutive sentences; they can also constitute text structures that organize whole paragraphs or texts (Sanders et al., 1992). The five most common text structures in expository texts are: description, sequence, cause-and-effect, compare-and-contrast, and problem-and-solution (Meyer, 1975). Appendix A shows short descriptions and examples of each of these text structures.¹

How knowledge about text structure can facilitate text comprehension during the reading process can be explained using the Construction-Integration Model of text comprehension (Kintsch, 2013). This model pictures reading as an interactive process of construction and integration, involving the activation of prior knowledge and making inferences within and beyond the text in order to form a coherent mental representation of the text: the situation model (van Dijk & Kintsch, 1983; Pyle et al., 2017). The five text structures listed in Appendix A can thus be seen as prototypical rhetorical structures consisting of specific coherence relations (Graesser et al., 2004). Good readers use the so-called structure strategy, which means that they use the organization of ideas in a text to organize their own understanding. Markers of text structure such as signaling words can cue text structures and help readers build a coherent text representation (Meyer & Ray, 2011; Sanders et al., 2007; Sanders & Noordman, 2000; van Silfhout et al., 2015).

Text structure instruction can improve reading comprehension in a number of ways. With knowledge about text structures it is easier to predict what the text will be about and to locate certain information in the text. Understanding how the author presents and organizes information can free up memory and processing resources and allows the reader to focus on the content of the text. It also promotes deciding which information is important, and facilitates the process of creating a coherent mental representation (Hebert et al., 2016; Pyle et al., 2017). Several meta-analyses have indeed shown positive effects of text structure instruction on text comprehension of both narrative and expository texts, even in young children (Bogaerds-Hazenberg et al., 2020; Hebert et al., 2016; Pyle et al., 2017).

Nevertheless, text structure instruction currently receives little attention in Dutch primary education. Activities aimed at text structure are often spent on recognizing signaling words, but hardly any connection is being made between signaling words and the type of inferences they mark, neither is it explained how signaling words can facilitate the reading process. Furthermore, analyses of reading instruction materials reveal that declarative knowledge taught about various text structures is scant and in some reading programs even limited to the classification of introduction, core and conclusion (Bogaerds-Hazenberg et al., 2017, 2022).

Taking into account the large body of positive effects of text structure instruction on text comprehension, and considering the outcome that Dutch students particularly seem to struggle with reading tasks that require text comprehension at the level of the situation model, expanding and improving text structure instruction in Dutch primary education seems a favorable approach.

Given the finding that the combination of literacy and science education benefits reading proficiency, it is worthwhile to examine if and how text structure education can also be integrated in science education.² In addition to the aforementioned opportunities, this approach seems promising because expository text structures appear to match science content very well. Text structure instruction focuses on coherence relations such as sequence and cause-effect, whereas science education teaches crosscutting concepts such as systems and cause and effect (Duschl, 2012). Text structure can thus be deployed to explicate these crosscutting concepts and thereby promote both text comprehension and comprehension of the content being taught.

However, before considering any implementation of evidence-based recommendations for reading instruction, it is important to understand what is already being taught in schools and to learn more about the texts, illustrations and assignments in the textbooks (Wijekumar et al., 2021). The content and quality of textbooks is an ecological component of relevance in the Component Model of Reading proposed by Aaron et al. (2008). Reviews of textbooks in several countries have already shown that evidence-based practices are hardly implemented in textbooks, and that assignments do not promote higher-order thinking skills (Agius & Zammit, 2021; Peti-Stantić et al., 2021). Since little is known about the content of Dutch science teaching materials and the opportunities for text structure instruction in particular, we conducted an analysis of textbooks created for grades 3 to 6. These materials were included in the corpus because students make the transition from ‘learning to read’ to ‘reading to learn’ starting from grade 3 (Fang, 2008), and several studies indicate that text structure instruction should be taught from an early grade level, with a steady increase in text complexity over grade levels (Pyle et al., 2017; Williams et al., 2014). The aim of our analysis was to answer the following research question:

To what extent are current Dutch science textbooks and workbooks used in grades 3–6 appropriate for text structure instruction?

To gain insight into the variety of the teaching materials available, we examined whether and how the science programs differed on the features we analyzed. In order to find out whether the materials show an increase in complexity, we also compared the outcomes of grades 3/4 and grades 5/6. We consider this analysis a case study that could inform researchers and educational experts from other countries as well about the opportunities that can be found for text structure instruction in which science teaching materials are used.

In order to be able to learn how to apply text structure knowledge during reading, students need instruction about text structures such as comparison, cause-effect and problem-solution (Ray & Meyer, 2011). Although it is unclear in what order the different structures should be introduced and if some combinations of text structures are more effective than others, research indicates that it is useful to teach multiple structures concurrently (Bogaerds-Hazenberg, et al., 2020; Hebert et al., 2016). For the introduction of a new text structure, readers need exemplary, well organized single-structured texts as model texts (Jones et al., 2016). Once readers are familiar with several text structures, they should be familiarized with reading multiple structured texts, shaping up to the ability to attend to the complexity of text structures in authentic texts (Dickson, 1999; Jones et al., 2016; Pyle et al., 2017).

Readers’ ability to recognize and use text structure can also be positively influenced by textual signaling devices that explicitly indicate the structure of the text (Ray & Meyer, 2011). Text features such as titles, headings, introduction and conclusion, segmenting and graphical features are organizational features that can help students to identify the structure of a text. Texts being used in text structure instruction should therefore provide clear examples of these features (Jones et al., 2016; Ray & Meyer, 2011). This starts with a well-articulated layout that guides the learners through the resources and enables them to easily identify relevant information (LaSpina, 1998; Pettersson, 2015). Previous research has revealed that in many content-area textbooks it is hard to identify the organizational principles at the global and local level (Armbruster & Anderson, 1988), even though educational publishers have ample opportunity to use page layout, paragraphing and graphic aids to highlight the structure of the texts.

Figure 1 and 2 show two examples of science texts for primary education. In Figure 1, segmenting and layout are clear, whereas Figure 2 displays a colorful and scattered layout with a variety of font types and sizes, which make it hard to find out the reading direction and the hierarchy between text parts.

Introductions to the text can foreshadow the content and provide information about the text organization (Lorch & Lorch, 1996; Ray & Meyer, 2011). For instance, example (1) mentions the main topics of the text and provides insight into the text structure. Based on this introduction, the reader can expect cause-effect relations and a problem-solution structure. By contrast, example (2) only introduces a topic, probably to draw attention to the reader, but hardly refers to the content and structure of the text, which describes geothermal energy and the way this is being used for heating.

Given the importance of the text features mentioned above, our materials analysis will include an analysis of a) the text structures found in science books, b) the layout of these texts, and c) the presence and content of introductory paragraphs.

Over time the number of images and graphical resources in science textbooks has increased significantly. Where language predominated traditional textbooks, modern textbooks are often organized around images. This shift in the proportion of text and images caused a change in the relationship between written texts and images. In traditional textbooks, illustrations were subordinate to the text, whereas in modern texts the combination of text and image serves functions of complementation, comparison, contrast, detail or elaboration. Images can, for example, be used as rhetorical devices and thereby relate to larger patterns of text organization (Martins, 2002; Mayer, 2009; Mikk, 2000).

Pictures can enrich or elaborate the representation of the text. In some cases the picture represents the structure of the text and can thus support the formation of the mental model (Eitel et al., 2013; Schnotz et al., 2014). Figure 3 for example can help the reader to follow the sequential steps mentioned in the accompanying text. Psycholinguistic studies have revealed the multimedia effect, establishing that knowledge acquisition through a combination of visuals and text is more successful than through text or images in isolation (Clark & Mayer, 2016; Mayer, 2009; Serafini, 2022). Thus, illustrations that visualize the rhetorical structure of the text can be a useful tool in text structure instruction, which is why we analyzed the presence of such illustrations in Dutch books for science education.

In learning to recognize the structure of a text and to use it for text comprehension, students may benefit from reading comprehension activities such as summarizing, generating inferences, and monitoring comprehension (Wijekumar et al., 2017). Asking targeted questions about the text structure in order to select the most important elements in the text has proven to be an effective strategy in scaffolding (Williams, 2018). This can for instance be done with inference questions such as (3) or sentence-completion tasks such as (4).

Another powerful tool to help students develop schemata for specific text structures and to comprehend and recall information from texts, is the use of structure-based visualizations, so-called graphic organizers (Bogaerds-Hazenberg et al., 2020; Pyle et al., 2017). It is important that students not only are exposed to these visualizations, but also actively fill out graphic organizers (Bogaerds-Hazenberg et al., 2020). In the Natuurzaken materials for grade 4, for instance, we found a text about the life stages of a human being. In the accompanying assignment, students are asked to connect the concepts ‘toddler’, ‘adolescent’, ‘baby’ etc. to a timeline and, by doing so, become aware of the sequential order of the ideas in the text.

Several scientists have emphasized the importance of including writing as a part of text structure instruction (Hebert et al., 2016; Ray & Meyer, 2011). Again, such writing activities can draw students’ attention to the structure of the text. In our materials analysis, we examined the assignments to find out whether these can be related to the structure of the text.

We selected five science programs produced by the four largest educational publishers for primary education in the Netherlands (Malmberg, Thieme Meulenhoff, Zwijsen and Noordhoff Uitgevers), and added two science programs that apply an innovative approach: Alles-in-1 is a thematically organized teaching program that integrates the content subjects with language arts. Blink Wereld takes inquiry learning as a starting point (see Table 1). For each program, we randomly selected the textbook and workbook materials of three teaching units per grade. Consequently, for every program twelve teaching units were analyzed, resulting in a corpus of 84 teaching units.

We coded features of the texts and of illustrations in the textbooks, as well as features of the assignments in the workbooks. This section explicates how the analysis was conducted.

In order to test whether the coding protocol resulted in consistent categorizations of the content being analyzed 10 % of the corpus was coded by a second annotator (Lacy et al., 2015; cf. Neuendorff 2002). This was conducted in four stages: 1) texts, 2) paragraphs, 3) illustrations and 4) assignments. During each stage the second annotator first engaged in a training phase to make her familiar with the procedure and criteria in the codebook. After this training phase the outcomes and differences were discussed and resolved and, if necessary, decision rules in the codebook were refined. With a few exceptions that will be discussed later in this paragraph, the inter-annotator agreement was moderate to almost perfect (.61<K>.82) (cf. Landis & Koch, 1977) as is shown in Appendix C. Pearson correlation coefficients were calculated for number of texts per unit (N = 14, r = .97, p<.001, 86 % agreement) and for number of paragraphs per text (N = 16, r = .99, p<.001, 69 % agreement).

The structure of problem-solution hardly occurred in paragraphs. As a result, disagreements about this structure caused poor agreement between coders. After discussing the disagreements only three paragraphs were assigned to the structure of problem-solution. In the analysis of the assignments, it turned out quite complicated to determine whether the information required for the assignment could be found in the text or not. In many cases some information was provided but the exact answer was not in the text. This explains the fair kappa score of .34. All disagreements were discussed and resolved and definitions in the codebook were sharpened.

The datasets were analyzed using IBM SPSS Statistics version 25. The analyses were completed via general linear univariate models. In the analysis of text structures at the text level and paragraph level we added number of paragraphs respectively number of sentences per paragraph as covariates.

The number of texts per teaching unit (see Figure 4) differed between teaching programs (F(6,70) = 19.28, p<.001). Differences were relatively small as the numbers of texts ranged from an average of 1.00 text per teaching unit in Natuniek, Natuurzaken, and Naut to an average of 2.08 texts in Blink Wereld. The mean number of paragraphs per text (see Figure 5) also differed between teaching programs (F(6,102) = 24.28, p<.001). Natuurzaken (11.25) and Natuniek (11.08) displayed the highest number of paragraphs per text (all ps<.001), while the number of paragraphs in Blink Wereld (2.72) was lower than that of all other teaching programs (all ps<.003), except Wijzer (p = .88). Throughout the corpus 15.5 % of the texts consisted of one paragraph.

The length of paragraphs (see Figure 6) differed between teaching programs (F(6,695) = 18.33, p<.001), with mean number of sentences per paragraph ranging from 5.07 in Natuurzaken to 11.80 in Wijzer. In addition, an interaction effect was found between program and grade level (F(6,695) = 6.13, p<.001). Argus Clou (F(1, 93) = 18.35, p<.001) and Natuurzaken (F(1,133) = 4.22, p = .04) show an increase in number of sentences per paragraph between grades 3/4 and 5/6, whereas Naut shows a decrease between these grade levels (F(1,131) = 12.37, p = .001).

As Figure 7 shows, sentence length differed both between teaching programs (F(6,694) = 11.24, p<.001) and grade levels (F(1,694) = 39.08, p<.001). In addition, an interaction effect of teaching program and grade level was found (F(6,694) = 7.73, p<.001). Four programs revealed a significant increase in number of words per sentence from grade 3/4 to grade 5/6: Alles-in-1 (F(1,140) = 11.81, p = .001), Natuniek (F(1,74) = 27.37, p<.001), Natuurzaken (F(1,133) = 5.61, p = .02), and Naut (F(1,130) = 78.18, p<.001). In Argus Clou the number of words per sentence decreased between these grade levels (F(1,93) = 5.75, p = .02).

These outcomes show that teaching programs differ strongly in segmentation and text length. Some programs present texts with many relatively short paragraphs, others use fewer paragraphs but with more sentences. In addition, programs do not display an increase in text length in terms of a higher number of paragraphs in grades 5/6 than in grades 3/4. However, some programs do show an increase in paragraph length, either in number of sentences per paragraph or in number of words per sentence. Still, this increase in paragraph length is not consistent over programs, as some programs do not display such an increase (Wijzer), or the increase in number of sentences in a paragraph comes at the cost of sentence length (Argus Clou), or vice versa (Naut).

This section describes the outcomes of the structure analysis of the texts and paragraphs. Texts that consisted of only one paragraph (15.5 % of the texts) and paragraphs consisting of one sentence (4 % of the paragraphs) were not taken into account.

We first determined the proportions of texts that included one or more types of text structures (see Figure 8). In most of the 34 cases this concerned texts in which two adjacent paragraphs were involved in a coherence relation. No differences between programs (F(6,82) = 1.42, p = .22) or grade levels (F(1,82) = 0.00, p =.99) were found. Out of the 34 texts that included one or more text structures, only 19 texts (56 %) were organized around a top-level structure that covered at least 75 % of all paragraphs. In most cases this was sequence (42 %), followed by problem-solution (26 %), cause-effect (16 %) and comparison (16 %).

The proportions of paragraphs with one or more types of structures (see Figure 9) showed more variation, as we found a main effect of teaching program (F(6,625) = 4.63, p <.001), as well as an interaction effect of program and grade level (F(6,625) = 3.35, p = .003). In Natuurzaken the proportion of paragraphs with one or more types of structures was higher in the materials for grade 5/6 than in those for grade 3/4 (F(1,128) = 9.71, p = .002), while the other programs did not show differences between grade levels (all ps>.09). In many cases the structure did not cover the whole paragraph, and clear examples as listed in Appendix A were rare. Out of the 336 paragraphs that were assigned to one or more text structures, 35 % (n = 112) concerned a single-structured top-level. Here problem-solution (32 %) and sequence (28 %) were slightly more frequent than cause-effect (22 %) and comparison (18 %).

Next, we calculated the mean number of different text structures per text (Figure 10), which ranged from .08 (Wijzer) to .83 (Natuurzaken) with a total average of .39. Again, at the text level no differences were found between programs (F(6,82) = 1.31, p = .26) or grade levels (F(1,82) = 0.12, p = .73).

At the paragraph level the mean number of different text structures (Figure 11) did vary, with a main effect of teaching program (F(6,625) = 2.42, p = .03) and an interaction effect of program and grade level (F(6,625) = 2.64, p = .02). Posthoc comparisons revealed that Blink Wereld showed a decrease in mean number of different structures per paragraph between grade levels from .83 to .32 (F(1,51) = 5.03, p = .03), whereas Natuurzaken showed an increase from .39 to .71 between grade levels (F(1,128) = 8.31, p = .005).

In order to establish whether all programs would familiarize students with all types of structures, we subsequently determined the frequency in which the four structures occurred in the text parts (see Figures 12 to 19). At the text level, no differences were found with respect to sequence, problem-and-solution, and comparison (all ps>.09), but the proportion of texts with a cause-and-effect structure differed between programs (F(6,83) = 2.23, p = .048). In four programs no examples of texts with a cause-and-effect structure at the text level were found, while Natuurzaken showed the highest proportion (.25).

At the paragraph level, differences in the occurrence of structures were found for all text structures except sequence. For cause-and-effect we found a main effect of program (F(6,626) = 4.01, p<.001) and an interaction effect of program and grade level (F(6,626) = 3.51, p<.002). In Blink Wereld the proportion of paragraphs with a cause-and-effect structure decreased between grade levels from .61 to .23 (F(1,52) = 9.24, p = .04), while Natuurzaken showed an increase from .26 to .44 (F(1,129) = 4.58, p = .03).

Analysis of the proportion of paragraphs with a problem-and-solution structure revealed a main effect of program (F(6,626) = 4.76, p<.001), grade level (F(1,626) = 12.58, p<.001) and an interaction between program and grade level (F(6,626) = 3.11, p = .005). Both Wijzer (F(1, 46) = 5.55, p = .02) and Natuniek: (F(1, 52) = 2.84, p = .02) showed a decrease between grade levels in paragraphs with a problem-and-solution structure.

In addition, programs differed in their application of the comparison structure at the paragraph level (F(6,626) = 2.39, p = .03). Alles-in-1 had more paragraphs with a comparison than Natuurzaken (p = .049), whereas in Blink Wereld we found no comparisons at all.

Table 3 presents the mean number of texts with a clear layout, the proportion of texts with an introduction, and the proportion of introductions that served as an advance organizer to the content of the text. The type of layout differed between programs (F(6,102) = 14.01, p<.001). Four programs (Alles-in-1, Natuniek, Naut and Wijzer) always presented a clear layout, while two programs (Argus Clou and Blink Wereld) did so in most of the texts. In Natuurzaken almost all pages differed in the way the text parts and illustrations were distributed over the pages, making it hard to the reader to easily create a general overview of the text.

Programs also differed in their use of introductions (F(6,102) = 25.76, p<.001) and the proportion of introductions that functioned as an advance organizer (F(4,52) = 6.45, p<.001). Natuniek and Naut always included an introduction that gives a preview of the content of the text, while Alles-in-1 and Natuurzaken did not include introductions at all, and the other three programs varied in their use and function of introductions.

In the corpus we counted 336 paragraphs that captured one or more text structures. The mean proportion of paragraphs accompanied by an illustration that visualizes the rhetorical structure of the text was .14. Table 3 shows the results per teaching program. No differences between science programs or grade levels were found (all ps<.29). Most illustrations depicted a sequential structure (46 %), or a causal relation (28 %). Only a few illustrations showed a problem-solution relation (2 %) or a comparison (4 %).

In the analysis of assignments we first determined the proportions of assignments that could be related to one of the four text structures (see Figure 20). A Repeated Measures ANOVA revealed a difference between programs (F(6,1060) = 2.72, p = .01) and text structures (F(3,3180) = 61.64, p<.001). A post hoc Tukey test revealed that Alles-in-1 contained a lower proportion of assignments related to one of the four structures than Natuniek (p = .04).

As with the paragraphs, the cause-and-effect structure was by far the most frequent in all programs (.18). The other three structures were found less often (sequence: .03, problem-and-solution: .03, comparison: .04).

In most assignments (72 %), the information needed to answer the question or to carry out the task could be found in the text, most often at the paragraph level (68 %). Sometimes information had to be retrieved from more than one paragraph (23 %) and in some cases the information needed was no more than one sentence (9 %). For the assignments for which the information was contained in the text, the structure of the text corresponded with the structure that we had linked to the assignment in 48 % of the cases. Almost all assignments with a sequential structure concerned the organization of given information, whereas causal relations were often represented in open questions, multiple choice questions, and gap-filling tasks. Dealing with problem-solution was often targeted with a writing task or an open-ended question, whereas for comparison programs mostly used open-ended questions or the organization of given information.