Investigating the Roles of Document Presentation and Reading Interactions on Different Aspects of Multiple Document Comprehension

Abstract The present study examined reading interactions and multiple document comprehension of 108 university students who read five partly conflicting documents on a health-related issue. Documents were presented on a large touch interface that either enabled a simultaneous or imposed a sequential presentation of the documents. None of the three multiple document comprehension measures (number of intertextual connections in essays, score in source-content mapping task, and number of source names recalled) was affected by document presentation. However, in the sequential condition, the number of revisits to documents was positively related to memory for source names and source-content integration, but not to intertextual integration. Conversely, in the simultaneous condition, participants who grouped documents during reading showed greater intertextual and source-content integration, but no greater memory for source names than those who had not grouped documents. To conclude, reading interactions played a more important role in readers’ multiple document comprehension than document presentation.


Introduction
When trying to understand complex and controversial health-related issues, such as potential positive or negative health effects of ultraviolet (UV) radiation, individuals often turn to the Internet. There, they are usually faced with multiple documents, which stem from a variety of sources, such as online newspapers, magazines, and health institutions, among others. Some documents might provide consistent or complementary information on the issue at hand, while other documents might stand in conflict with each other . Thus, building a comprehensive understanding of such a controversial subject matter requires comparing, evaluating, and integrating information across those documents (e.g., Anmarkrud et al., 2014;List et al., 2021;Mahlow et al., 2020;Primor & Katzir, 2018;Wineburg, 1991).
Complex reading tasks (e.g., on the Internet) as outlined above are usually conducted on conventional computer screens, laptops, or tablets, where screen space is limited, and document reading applications (e.g., PDF Readers, Web Browsers) by default often present only one document at a time (Lombard et al., 2021). However, a sequential presentation of multiple documents may not be ideal for building a comprehensive understanding of a complex issue (Leroy et al., 2021;Lombard et al., 2021;Wiley, 2001). In contrast, as assumed by various researchers from both human-computer interaction research and text-comprehension research, a simultaneous presentation of multiple documents may better support comparing and integrating information across documents (Andrews et al., 2010;Ball & North, 2005;Benshoof et al., 1995;Bi & Balakrishnan, 2009;Leroy et al., 2021;Wiley, 2001). Yet, empirical research that examines differences in readers' comprehension of multiple documents when documents are read on an interface presenting them simultaneously as compared to sequentially is very limited. The present work aimed to contribute to this line of research by examining the effects of a simultaneous as compared to sequential document presentation on three different aspects of multiple document comprehension. For this purpose, half of the participants read five partly conflicting documents on an interactive touchinterface that enabled a simultaneous presentation of all documents, whereas the other half read the documents on an interactive touch interface that allowed only for a sequential presentation, that is, with only one document presented at a time.
Furthermore, when presenting texts by means of an interactive reading interface it might be crucial for readers' multiple document comprehension how they use the capabilities of the interface to interact with documents as a tool for document comparison and integration (e.g., Leroy & Kammerer, 2022;Delgado et al., 2020;Leroy et al., 2021;List & Alexander, 2018;Lyu & Li, 2020). Thus, a second major goal of the present research was to investigate how specific interactions with documents (i.e., reading interactions) that might reflect cross-document information comparison are related to different aspects of readers' multiple document comprehension in the respective presentation condition. Specifically, we assessed participants' engagement in revisiting documents (sequential condition) and in grouping the partly conflicting documents on the screen (simultaneous condition) during reading. In the following, we first describe what constitutes an adequate mental representation of multiple documents that skilled readers are assumed to form during reading. Subsequently, we discuss which roles simultaneous as compared to sequential document presentation and/or specific reading interactions performed might play in constructing this mental representation.

Readers' mental representation of multiple documents
According to the Documents Model Framework Perfetti et al., 1999), in addition to the core information of single documents (Kintsch, 2004), an adequate mental representation of multiple documents, a socalled documents model, comprises the following components: (a) information on how statements from different documents relate to each other or how documents as a whole relate to each other, respectively (i.e., intertextual connections), (b) information about the origin, that is, the source of the documents (e.g., publication venue or name of the author), and (c) information about which content stems from which source (i.e., connections of source and contents).
Building on the assumptions of the Documents Model Framework, the quality of readers' mental representation of multiple documents (and, thus, their multiple document comprehension) can be assessed in tasks that address the specific components of readers' documents models. For instance, these can be (a) essay tasks that allow to assess the number of intertextual connections made across documents (Kobayashi, 2009;Linderholm & Van den Broek, 2002;List, 2019;Salmer on et al., 2018a), (b) source name recall tasks to assess memory of source information (Bråten, Ferguson, et al., 2014;Kammerer, Kalbfell, et al., 2016;Salmer on et al., 2018b), or (c) source-content mapping tasks that address readers' source-content integration (Delgado et al., 2020;Kammerer, Meier, et al., 2016;Stang Lund et al., 2019;Strømsø et al., 2010; for recent reviews also see Barzilai et al., 2018;Primor & Katzir, 2018). While Britt and colleagues  mainly focused on describing and assessing readers' mental representations that result from reading multiple documents, other theoretical work has addressed the processes required to construct such mental representations, such as comparing, evaluating, and integrating information from and across multiple documents (e.g., Afflerbach & Cho, 2009;Braasch & Bråten, 2017;Cho & Afflerbach, 2017;Rouet et al., 2017;Rouet & Britt, 2011;Stadtler & Bromme, 2014). For instance, Cho and Afflerbach (2017, p. 120) provided a taxonomy in which they described various important strategies involved in comprehending multiple documents. These include (a) "comparing and contrasting the content of the text being read with the content of related texts to develop a coherent account of cross-textual contents," (b) "rereading and linking text segments that were previously regarded as unrelated to finalize cross-textual meaning structures," (c) "perceiving that multiple texts related to the same topic can provide diverse and contrasting views about the topic, complementary information about the topic, or both," and (d) "using information about a present source to evaluate and interpret text content." In line with this taxonomy, several empirical studies showed a positive relation between readers' multiple document comprehension and their engagement in cross-document information comparison during reading as reflected by their think-aloud comments Hilbert & Renkl, 2008;Wineburg, 1991), or their re-reading behavior (Goldman et al., 2012;List, Du, et al., 2019).
However, from the above description of the many-faceted strategies involved in comprehending multiple documents it becomes clear that constructing a documents model is a challenging task for most readers (e.g., Kiili & Leu, 2019;. Therefore, it has become an important research goal to identify characteristics of a reading environment that support readers in performing such strategies (Barzilai et al., 2018;. As we will outline in the following, one such characteristic of a reading environment that has been considered to support cross-document information comparison and integration is whether documents are presented simultaneously or sequentially.

Effects of a simultaneous presentation of documents on multiple document comprehension
Previous findings from observational studies suggest that a simultaneous presentation of multiple documents can support readers in tasks that require cross-document information comparison (Andrews et al., 2010;Ball & North, 2005;Hutchings & Stasko, 2004;O'Hara et al., 2002). Furthermore, experimental research from the field of human-computer interaction revealed positive effects of a reading environment enabling a simultaneous presentation (as compared to imposing a sequential presentation) of documents for analytic problem-solving tasks that require information integration across documents (Jang et al., 2011Takano et al., 2015), and experimental research from the field of multiple document comprehension indicated respective positive effects on crossdocument information comparison (Olive et al., 2008), on the overall quality of readers' multiple document comprehension (Wiley, 2001), and on readers' intertextual integration (Leroy et al., 2021). However, it remains unclear from previous work whether different aspects of multiple document comprehension are differentially affected by document presentation.
As argued by previous work, the advantage of a simultaneous over a sequential presentation of documents might be due to an easier re-accessing of or switching between multiple documents (Andrews et al., 2010;Bi & Balakrishnan, 2009;Haber et al., 2014;O'Hara & Sellen, 1997;Takano et al., 2015), which, in turn, might support cross-document information comparison and integration through increased re-processing of the respective information. Still, results of experimental studies investigating the effect of a simultaneous vs. sequential presentation of documents on multiple document comprehension suggest it requires more than simply providing a simultaneous presentation of documents to readers for it to have a beneficial effect. For instance, in a study by Wiley (2001), participants read ten documents either on a splitscreen simultaneously presenting two out of ten documents, or in a sequential presentation with only one document visible at a time, and wrote argumentative essays on the topic after reading without the documents available. Results showed that when participants received explanations for why they were given a split-screen interface before reading, they wrote essays reflecting better integration compared to participants who had read the documents in the sequential interface. However, the split-screen interface did not yield better integration than the sequential interface when no explanations for the split-screen interface were given. Similarly, Lombard et al. (2021) found that the provision of a tool that enabled a partly simultaneous presentation of (up to three out of five) documents yielded better integration than a sequential presentation of documents only for participants who had received explicit instructions to use all features of the tool.
The simultaneous conditions of the aforementioned studies, however, only allowed a partly simultaneous presentation of two or three documents. In contrast, a fully simultaneous presentation of all documents might yield greater support for cross-document information comparison and integration, and thus foster readers' multiple document comprehension even without explicit instructions. To the best of our knowledge, to date, the study by Leroy et al. (2021) is the only one investigating the effect of a simultaneous presentation of all documents provided to readers (as compared to a sequential presentation of these documents) on multiple document comprehension. In their study, participants read six partly conflicting documents on a multitouch table (i.e., a large horizontal display) presented either simultaneously or sequentially, with the possibility to freely re-arrange documents on the screen in both conditions. Participants received instructions on how to use the respective interface but no explicit instructions to use all features of the interface. Results showed that participants in the simultaneous condition spontaneously were more likely than those in the sequential condition to spatially organize (i.e., move to similar locations) the partly conflicting documents on the screen. Most interestingly, the likelihood to engage in this reading interaction fully mediated the positive effect of document presentation on intertextual integration. That is, similar to previous findings suggesting that a (partly) simultaneous document presentation might only be beneficial when readers are instructed to use the interface, Leroy et al. (2021) found that the simultaneous presentation of documents is particularly beneficial for readers who use it to group documents during reading. Hence, merely providing readers with a simultaneous presentation of documents per se might not be sufficient to support multiple document comprehension. Rather, how readers interact with an interface during reading might play a crucial role in the effectiveness of the kind of document presentation on their multiple document comprehension.

The role of reading interactions in forming an understanding of multiple documents
The findings by Leroy et al. (2021) point out the importance of assessing reading interactions reflecting readers' engagement in comparing information across documents during reading (cf. Anmarkrud et al., 2014;Bråten, Ferguson, et al., 2014;Britt et al., 2004;Britt & Aglinskas, 2002;Ferguson et al., 2012;Wineburg, 1991). In addition to having analyzed whether participants had grouped the partly conflicting documents during reading, in the sequential condition, Leroy et al. (2021) also assessed the number of revisits to documents from user interaction logfiles as a second measure of reading interactions. Revisits to documents were assumed to reflect cross-document information comparison. In the following, we will briefly outline previous work explaining the relation between the reading interactions of grouping documents or revisiting documents, respectively, and readers' multiple document comprehension.

Grouping documents
Instructional research has previously suggested that strategical re-organization of information provided by multiple documents is positively related to multiple document comprehension (Barzilai & Ka'adan, 2017;Hilbert & Renkl, 2008;Lombard et al., 2021;Lundstrom et al., 2015). That might be because the respective layout of re-organized information provides visual signals that highlight the relations between pieces of information (Barzilai & Ka'adan, 2017). In their observational studies, O'Hara and Sellen (1997) and O'Hara et al. (2002) furthermore reported that readers also spontaneously re-organized printed documents spatially during reading to "gain a sense of overall structure" and to "check on or to relate specific pieces of information across documents" (p. 346; see also Haber et al., 2014). When documents are partly conflicting such that some documents provide arguments in favor of a particular issue, while other documents provide information against that issue, such rearrangement might take the form of grouping documents according to their overall stance, and the resulting layout of documents might bear contextual information regarding the relation of documents (Andrews et al., 2010b;Bi & Balakrishnan, 2009;Leroy et al., 2021). Of note, to group documents according to their overall stance, readers need to be already aware of the fact that some documents provide positive arguments for a subject matter, while others provide negative arguments for it. Yet, the resulting layout of documents (i.e., their arrangement in groups) can make intertextual relations even more salient (cf. Barzilai & Ka'adan, 2017;Skuballa et al., 2018), and the spatial proximity of documents might further support comparison and integration of information across documents (cf. Ginns, 2006).
To our knowledge, however, the study by Leroy et al. (2021) is the only one that investigated the role of readers' engagement in grouping partly conflicting documents during reading in their multiple document comprehension. They found that participants' intertextual integration (as assessed from essays written after reading) was better when they had grouped documents according to their overall stance during reading than when they had not grouped documents. Yet, Leroy et al. (2021) had assessed participants' intertextual integration as the sole measure of multiple document comprehension, thus leaving the question unanswered whether grouping documents is also related to other measures of multiple document comprehension. The present research aimed to address this question by assessing different measures of multiple document comprehension.

Revisiting documents
Readers' intent to compare or re-evaluate information across documents might be reflected in their engagement in reaccessing previously read documents (Goldman et al., 2012;Leroy et al., 2021;List & Alexander, 2018). For example, List and Alexander (2018) found that participants who had revisited at least one document were more likely to include source citations in the essays written after reading. Regarding other measures of multiple document comprehension, Delgado et al. (2020) found a positive relationship between accessing source information (provided as pop-up windows in a hypertext) and source-content integration. Similarly, Leroy & Kammerer (2022) found that the number of revisits to documents was positively related to source-content integration, but was unrelated to intertextual integrationunless participants were able to highlight text (as it was also found by Leroy et al., 2021). Overall, the findings by Leroy & Kammerer (2022), Leroy et al. (2021), and List and Alexander (2018) thus suggest that (when text-highlighting is not possible) revisits in particular play a role in multiple document comprehension measures related to source information. This is particularly interesting since the documents in these studies were partly conflicting and readers usually are more likely to return to source information when they contain conflicting information . Hence, revisits might rather be initiated by readers to re-evaluate information in the light of its source (cf. List & Alexander, 2018;Wineburg, 1991) than to link information between documents.
However, previous work has examined only one or two aspects of multiple document comprehension (e.g., only intertextual connections or source-content integration or source citations). The present work thus contributes to this line of research by assessing three measures that address different aspects of multiple document comprehension.

Present study
In this study, we built on previous work by Leroy et al. (2021) to examine the effect of a reading interface enabling a simultaneous presentation of multiple documents as compared to one that imposes a sequential presentation of documents on different aspects of multiple document comprehension. Further, we aimed to investigate how reading interactions, namely revisiting previously read documents in the sequential presentation and grouping partly conflicting documents in the simultaneous presentation, are related to readers' multiple document comprehension. To this end, we presented five partly conflicting documents on the topic of the health effects of UV radiation on a large multi-touch table (i.e., a large horizontally oriented touchdisplay). In the simultaneous condition, the documents were initially presented on a stack and could be flexibly rearranged on the interface by participants to generate a simultaneous presentation (without overlap, if desired) similar to when interacting with printed documents (O'Hara et al., 2002). In the sequential condition, similar to a regular computer screen or tablet interface, the documents were presented statically (i.e., not allowing any re-arrangement of documents on the screen) and only one at a time.
We had the following three main research questions (RQs) in the present study: RQ1: Does an interface that enables a simultaneous presentation of multiple documents result in an increased use of intertextual connections in argumentative essays as well as in better memory for source names and better source-content integration than an interface that imposes a sequential presentation?
RQ2: Do readers in the simultaneous condition who spontaneously (i.e., without being prompted to) group the partly conflicting documents according to their overall stance show better intertextual integration, better memory for source names, and better source-content integration?
RQ3: Is the extent of revisiting documents in the sequential condition positively related to readers' intertextual integration, memory for source names, and source-content integration?
Regarding RQ1, based on previous findings by Leroy et al. (2021), we expected to find a beneficial effect of the simultaneous compared to the sequential condition for readers' intertextual integration. Furthermore, we explored whether the presentation condition also had an effect on participants' source-name recall or their source-content integration.
Regarding RQ2, based on findings by Leroy et al. (2021), we expected participants' intertextual integration to be positively related to grouping documents. Furthermore, we explored whether participants who had grouped documents during reading in the simultaneous condition also showed better memory for source names or better source-content integration than those who had not grouped documents during reading.
Regarding RQ3, as in previous research, we did not expect participants' intertextual integration to be related to the number of revisits made (Leroy & Kammerer, 2022;Leroy et al., 2021). However, based on prior research that found a positive relation between revisiting documents and source citations in essays (List & Alexander, 2018) as well as source-content integration (Leroy & Kammerer, 2022;Delgado et al., 2020), we expected the number of revisits to be positively related to participants' memory for source names as well as to their source-content integration.

Participants and experimental design
One hundred and eight university students from different majors at a large German university participated in this study. However, two participants had to be excluded from the analyses since they did not fulfill the requirements (i.e., they stated their mother tongue was not German in the questionnaire and their essays did not reflect a good level of German). This resulted in a final sample of N ¼ 106 participants (75.47% female; M ¼ 22.53 years, SD ¼ 2.75 years). Participants were compensated with 8e for their participation. The study was approved by the local ethics committee and participants gave their written consent at the beginning of the study.
Participants were randomly assigned to one of two experimental conditions, which differed in whether the reading environment enabled a simultaneous or imposed sequential presentation of documents on the multi-touch table. Fiftyfour participants were serving in the simultaneous condition and 52 participants in the sequential condition.

Task and documents
Participants' task was to inform themselves about the health effects of UV radiation by reading five documents that were provided on the multi-touch table to, afterward, write an argumentative essay about the topic as well as answered several questions. The partly conflicting documents were slightly adapted (and translated) versions of documents used in previous studies (Leroy & Kammerer, 2022;Leroy et al., 2021;Lombard et al., 2021;Strømsø et al., 2016). The source of each document was indicated using a source logo at the top of each document. The sources of the two documents reporting positive health effects of UV radiation (through the promotion of vitamin D production) were a reputable liberal-conservative German newspaper ("FAZ -Frankfurter Allgemeine Zeitung") and a medical science magazine (" € Arztezeitung"); the sources of the two documents reporting negative health effects of UV radiation (e.g., promoting skin cancer) were a university research magazine ("Mundo -Wissenschaftsmagazin der TU Dortmund") and the German national cancer association ("DKG -Deutsche Krebsgesellschaft"). The source of the neutral document providing general information about UV radiation was a German school book publisher ("Cornelsen"). Each document was structured into two or three sections, which were separated by subheadings. The documents contained an average of M ¼ 352.40 words (SD ¼ 27.32) and M ¼ 20.80 sentences (SD ¼ 1.64). The LIX scores (cf. Bj€ ornsson, 1968) computed for each text revealed comparable readability and that documents were not too easy to read (M ¼ 52.06, SD ¼ 2.65), which, however, can be presumed appropriate for university students' reading skills.

Documents presentation
Participants read the documents on a multi-touch table (ARV EDGE 55 00 4 K; resolution: 3840 Â 2160 px, 1.88 Â 1.06 meter; comparable to Leroy et al., 2021, Figure 1, but without a highlighting function). The interface either enabled a simultaneous or imposed a sequential presentation of the five documents. All documents were initially displayed in A4 size but could be freely scaled by participants using a zooming gesture (i.e., pinching). However, in the sequential condition, each document was displayed in A4 size again upon re-opening. In the simultaneous condition, documents could furthermore be re-arranged freely by using a dragging gesture. While in the sequential condition documents were always displayed in the center of the bottom of the screen, documents were initially stacked in the simultaneous condition with the text of only one document (the topmost document) being visible (see Figure 1). Specifically, the document initially presented as the topmost one in the simultaneous condition was presented in the same position as the documents in the sequential condition when they were opened. The purpose of initially presenting documents in a stack in the simultaneous condition rather than distributing them on the screen with all documents being initially visible (cf. Leroy et al., 2021) was to investigate a situation where multiple documents can be presented simultaneously, but are not presented simultaneously per default. In both presentation conditions, counterbalanced across participants, documents were presented in one of two alternating document orders. In both document orders, the document presented in the third position provided general information about UV radiation without taking a stance. Documents in the first and fifth position took a positive stance in one document order, and a negative stance in the other document order, whereas documents in the second and fourth position took the respective opposing stance. Overall, 27 participants in the simultaneous and 27 participants in the sequential condition were presented with a positive document first, and 27 and 25 participants in the respective conditions were presented with a negative document first.

Multiple document comprehension measures
We assessed participants' multiple document comprehension with four dependent variables. First, we assessed participants' intertextual integration through the number of intertextual connections included in their argumentative essays (e.g., Kobayashi, 2009;Leroy et al., 2021;List, Stephens, et al., 2019;Salmer on et al., 2018a). Second, we assessed participants' spontaneous sourcing in essays as a measure of their source memory (e.g., List & Alexander, 2018). Third, since spontaneous sourcing might underestimate participants' source memory, participants' memory for document sources was furthermore assessed in a free recall task, in which they were asked to list as many of the five document sources (i.e., the names provided by logos at the top of each document) as they remembered (Kammerer, Meier, et al., 2016). Fourth, to assess participants' source-content integration we used a source-content-mapping task, that is, a task that requires participants to map which information stemmed from what source (e.g., Delgado et al., 2020;Kammerer, Meier, et al., 2016;Stang Lund et al., 2019;Strømsø et al., 2010).
As in previous work (Leroy & Kammerer, 2022; Leroy et al., 2021), we coded one intertextual connection in essays for each statement that connected information clearly stemming from different documents. Such connections could be made between conflicting (i.e., information stemming from one positive and one negative document) or complementary (i.e., information stemming from two documents taking the same stance or from the neutral and any other document) pieces of information, and could also span several sentences when the connection was semantically apparent. For examples of intertextual connections, see Leroy & Kammerer (2022). Two raters independently coded 20 (18.87%) essays for intertextual connections and reached an interrater agreement of 80.92%, which was calculated in a point-wise manner (i.e., the agreement was coded when both raters gave a point for the respective statement, and disagreement was coded when only one rater had coded an intertextual connection for the respective statement). Disagreements were discussed and resolved before one rater proceeded with the coding of the remaining 86 essays.
Two independent raters judged the correctness of the source names listed by all 106 participants (i.e., they evaluated whether the given name was close enough to the correct, full name of a document's source). Overall, the 106 participants wrote down 309 source names and the interrater agreement of their correctness was 93.83%. Disagreements were resolved through discussion.
Finally, for the source-content mapping task, participants were presented with a 2-column table containing source logos of all five documents in alphabetic order in the first column and main statements of each document in the second column (see Figure 2). The statements were initially presented in one defined order in which none was in its correct position. Participants' task was to re-arrange (i.e., drag and drop) statements, such that source logos and main statements of each document were positioned in the same row of the table. Because once four statements were correctly positioned, the fifth was automatically correct too, the maximum score in the source-content mapping task was 4.

Reading interactions
Regarding the reading interactions assessed for participants in the simultaneous condition, based on screen recording videos that were recorded for the whole task processing time of each participant, two independent raters evaluated whether the participant had grouped documents according to their overall stance during reading (i.e., whether the two positive documents and/or the two negative documents were grouped). Interrater agreement of all 54 screen recordings was 92.59%, and disagreements were resolved through discussion.
Regarding the reading interactions assessed for participants in the sequential condition, we used interaction log files to assess the number of revisits to documents. As in previous research (Leroy et al., 2021), we used a threshold for document openings of 1 s to exclude short, presumably accidental (re-)openings from our analyses. Consequently, a revisit of a document was counted for each opening that lasted at least 1 s after a first opening that lasted at least 1 s. Take, for example, the following partial document opening sequence: A (52.0 s) -B (0.9 s) -A (3.2 s) -B (73.4 s) -A (11.8 s). In the first step, we cleared the sequence of all openings with a duration <1 s, resulting in A (52.0 s) -A (3.2 s) -B (73.4 s) -A (11.8 s). In a second step, we checked for "openings" of the same document in direct succession, which were then accumulated into one opening (i.e., one first reading or one revisit). In the given example, this results in the following final sequence of A (52.0 þ 3.2 s) -B (73.4 s) -A (11.8 s) with one revisit of document A.

Control variables
To ascertain comparability across experimental conditions, we assessed participants' working memory capacity (15-item computer-based reading span task; cf. Kane et al., 2004), prior topic knowledge (see below), as well as their topic interest (1 item, from 1 ¼ very low to 7 ¼ very high) and prior topic beliefs (2 items; one addressing positive and another addressing negative effect of UV-radiation on health; from 1 ¼ totally disagree to 7 ¼ totally agree) on 7point Likert scales. To gain one score for participants' prior topic beliefs, we reverse-coded the negative item and averaged both values (Cronbach's alpha ¼ .76).
Prior topic knowledge was assessed with an essay task before reading the documents, in which participants were asked to write down within 5 min everything they knew about the potential effects of UV-radiation on human health. Participants were awarded one point for each relevant and correctly reported information (e.g., UV-radiation stems from the sun; sunbeds use UV-radiation; there are three types of UV-radiation) or argument concerning the health effects of UV-radiation on humans (e.g., UV-radiation causes skin cancer; UV-radiation promotes the production of vitamin D in the body) mentioned in their essay. Two independent raters scored all 106 prior knowledge essays. Interrater agreement for point-wise scoring was 95.56% (i.e., 95.56% of the points were awarded by both raters). Disagreements were resolved through discussion.

Procedure
Participants were tested in single sessions (of $1 h) in the lab. They first completed a demographic questionnaire and the topic knowledge measure using a laptop. Subsequently, participants received a written explanation of the interaction possibilities on the multi-touch table (i.e., how to open and scale documents in the sequential condition, and how to scale and re-arrange documents in the simultaneous condition) and practiced interaction with five blank documents in the respective experimental condition. Once they felt comfortable in operating the interface, they received written task instructions on the screen. Specifically, they were told that in the following they would be provided with five documents from the Internet, which they should read carefully, to subsequently write an argumentative essay about the potential health effects of UV radiation without having the documents available. They were also informed that they had a maximum of 15 min to read the documents and that within this time frame they could read the documents as often as they wanted. After reading, participants were asked to write an argumentative essay on a laptop within 15 min. They were told to include arguments from the documents in their essay to justify their statements. Afterward, also on the Figure 2. Items of the source-content-mapping task (translated from German). The initial order of source logos (left) and statements (right) was as shown here. Participants' task was to drag the statements vertically such that their position matched that of the source logos. Arrows indicate the correct assignments.
laptop, they were asked to recall the documents' sources and to complete the source-content mapping task.

Results
Overall, we assessed four different measures of multiple document comprehension as dependent measures, namely the number of intertextual connections in essays, the number of source citations in essays, the number of correctly recalled document source names in a free recall task, and the score in a source-content mapping task. However, analyses revealed that only six participants (11.11%) in the simultaneous condition and five participants (9.62%) in the sequential condition cited at least one document source in their essays. Thus, we refrained from analyzing this measure in the present study. In the following, the results regarding the remaining three measures of multiple document comprehension are reported.

Differences regarding control variables and multiple document comprehension (RQ1) between experimental conditions
Regarding control variables, Welch's t-test with presentation condition as an independent variable showed that participants in the simultaneous and sequential condition did not differ with respect to prior topic knowledge, t(98.65) ¼ 0.76, p ¼ .447, or working memory capacity, t(99.14 ¼ 1.36, p ¼ .177. They also did not differ in their topic interest, t(101.79) ¼ 0.19, p ¼ .848, or in their prior beliefs regarding health effects of UV radiation, t(102.43) ¼ 0.16, p ¼ .874, which on average were rather negative (i.e., participants, on average, were more aware of negative health effects of UV radiation). Furthermore, neither the overall time participants spent reading the documents, t(103.71), p ¼ .357, nor the length of their essays, t(101.57) ¼ .217, differed between presentation conditions. Table 1 shows the means (and standard deviations) for these measures for each presentation condition.
To answer RQ1, we also conducted Welch's t-tests with document presentation as independent variable and the three multiple document comprehension measures as dependent variables (see Table 1 for means and standard deviations). However, no significant differences between the simultaneous and sequential condition were shown for any of the three measures of multiple document comprehension (number of Subsequently, to answer RQ2 and RQ3 about the relation between participants' interaction behavior and their multiple document comprehension for the two presentation conditions separate analyses were conducted and are reported in the following.

The role of grouping documents during reading for multiple document comprehension (RQ2)
Twenty participants (37.04%) in the simultaneous condition had grouped documents according to their overall stance during reading. Eighteen out of those twenty participants had grouped the two positive as well as the two negative documents, one participant had only grouped the positive documents, and one participant had only grouped the negative documents. Table 2 provides the means and standard deviations of all control variables as well as the three multiple document comprehension measures for participants who had grouped and not grouped documents during reading, and respective results of the mean comparisons (Welch's t-tests).
Regarding the three multiple document comprehension measures, participants who had grouped documents during reading included significantly more intertextual connections in their essays and scored significantly higher in the sourcecontent mapping task. However, there was no significant difference in the number of source names correctly recalled between participants who had and had not grouped documents (see Table 2).
Regarding control variables, participants who had grouped documents during reading had significantly higher prior topic knowledge than those who had not grouped documents during reading. However, neither measure of multiple document comprehension correlated significantly with prior topic knowledge (number of intertextual connections: r(52) ¼ .14, p ¼ .323; number of source names recalled: r(52) ¼ .04, p ¼ .779; source-content mapping score: r(52) ¼ .18, p ¼ .193). There was no significant difference between those who had and had not grouped documents in any of the remaining control variables (also see Table 2). Also, participants who had grouped and not grouped documents during reading did not differ in their overall time spent reading the documents. Yet, participants who had grouped documents during reading wrote significantly longer essays than those who had not grouped documents during reading (also see Table 2).

The role of number of document revisits for multiple document comprehension (RQ3)
Five participants (9.62%) in the sequential condition did not make any revisits. The remaining 47 participants, on average, made 10.28 revisits (SD ¼ 8.22). For the complete sample of 52 participants, Table 3 provides correlational analyses between the number of revisits to documents during reading, on the one hand, and control variables and multiple document comprehension measures, respectively, on the other hand.
Regarding multiple document comprehension measures, there was a significant positive correlation between the number of revisits and both the number of source names correctly recalled and the score in the source-content mapping task. In contrast, there was no significant correlation between the number of revisits and the number of intertextual connections in essays (see Table 3). The number of revisits was not significantly correlated to any of the control variables. Finally, the number of revisits was positively correlated to the time participants took for reading, but not to essay length (also see Table 3).

Discussion
With the present study, we aimed to extend previous work examining whether a simultaneous as compared to a sequential presentation of multiple documents fosters multiple document comprehension (Andrews et al., 2010;Ball & North, 2005;Bi & Balakrishnan, 2009;Czerwinski et al., 2003;Hutchings & Stasko, 2004;Leroy et al., 2021;Olive et al., 2008;Wiley, 2001). We assessed different aspects of multiple document comprehension, that is, intertextual integration, source memory, and source-content integration, and also investigated the role that reading interactions played in these aspects. To our knowledge, only one previous study (Leroy et al., 2021) had investigated the effect of document presentation on multiple document comprehension by also taking reading interactions into account. However, readers' intertextual integration, that is, the number of statements in their essays that combined information from two documents, was assessed as the sole measure of multiple document comprehension in that study. Thus, multiple document comprehension measures reflecting readers' mental representation of source information or of source-content integration were not considered in Leroy et al.'s (2021) study.

Multiple document comprehension when reading multiple documents in a reading environment (not) enabling simultaneous presentation
Other than expected, the simultaneous condition did not foster participants' intertextual integration. Furthermore, exploratory analyses revealed that also participants' memory for source names and their source-content integration did not differ across presentation conditions. Regarding intertextual integration, these findings are in contrast to those by Leroy et al. (2021), who found that reading documents in the simultaneous condition yielded a significantly higher number of intertextual connections in essays written after reading (i.e., the same measure of intertextual integration as was used in the present work) than reading documents in the sequential condition.
One potential post-hoc explanation for the inconsistent findings might be that other than in Leroy et al.'s (2021) study, in the simultaneous condition documents were initially presented on a stack in the present study rather than spread out on the screen from the start. That is, the simultaneous condition of the present study enabled a simultaneous presentation of documents, yet readers had to actively "create" a simultaneous presentation of documents by spreading out the initially stacked documents on the screen. Thus, only for those readers who actually made use of the possibility of creating a simultaneous document presentation, the potential beneficial effects of a simultaneous presentation could have been brought to bear.
Second, while documents were presented in random order for each participant in the study by Leroy et al. (2021), they were presented in a specific alternating order in the present study, with conflicting documents being presented in direct succession. While the document order was only initially given in the simultaneous condition (because documents could be re-arranged, which might have changed their order), it was rather static in the sequential condition, in which documents had to be opened from a menu. Hence, the given order might have played a more important role in the sequential condition than in the simultaneous condition. As previous work showed, the order of partly conflicting documents can affect readers' mental representation of multiple documents and their relations (Braasch et al., 2021;Maier & Richter, 2013). Thus, reading documents that take an opposite stance in direct succession may have enhanced readers' conflict awareness, hence supporting their intertextual integration. In consequence, this might have diminished any detrimental effect of a sequential as compared to simultaneous presentation of multiple documents. In the same vein, an increased conflict awareness in the sequential condition due to the alternating document order might also have fostered participants' attention to sources (cf. Braasch & Bråten, 2017). This, in turn, might have diminished any potential detrimental effects of a sequential as compared to simultaneous condition on readers' memory for sources or their source-content integration. Of course, these assumptions are speculative in nature and will need to be examined in future research.

The role of grouping documents for multiple document comprehension in a reading environment enabling a simultaneous presentation of documents
As in previous work (Leroy et al., 2021), participants' strategic re-organization, that is, their grouping of documents in the simultaneous condition, was positively related to their intertextual integration. This may be due to the spatial contiguity resulting from grouping documents which "can make the relations between multiple [documents] more salient" (Barzilai & Ka'adan, 2017, p. 199; see also Ginns, 2006) and thus support readers in cross-textual information comparison and integration (cf. Leroy et al., 2021). Regarding readers' memory for source information as a second measure of multiple document comprehension, the present study revealed that regardless of whether participants had grouped documents during reading, they only recalled about two to two and a half document sources. This is in line with previous research showing that readers typically only pay little attention to source information (e.g., Braasch & Bråten, 2017;Britt et al., 2004;Le Bigot & Rouet, 2007;Rouet et al., 1996). However, participants who had grouped documents according to their overall stance were better able to map main statements to the respective source names than were participants who had not grouped documents. In conclusion, readers' spontaneous engagement in grouping documents might not be related to their overall attention to sources, but rather support integration processes within and across documents (cf. Ginns, 2006), yielding better intertextual integration as well as source-content integration (i.e., having mentally represented which source said what). However, further research is needed to understand the differences in integration processes between readers who do and do not (spontaneously) group documents during reading.

The role of document revisits for multiple document comprehension in a reading environment imposing a sequential presentation of documents
A different pattern of results was shown for the relation between multiple document comprehension and the number of revisits to documents in the sequential condition. As in previous work (Leroy & Kammerer, 2022;Leroy et al., 2021), the number of revisits to documents was not related to participants' intertextual integration. In contrast, the number of revisits to documents was positively related to both multiple document comprehension measures related to source information, that is, to participants' memory for source names, and to participants' source-content integration, that is, their performance in the source-content mapping task. That is, the more revisits readers made during reading, the more document sources they recalled after reading and the better they performed in the source-content mapping task. These findings corroborate previous work by List and Alexander (2018) regarding source citations and by Leroy & Kammerer (2022) and Delgado et al. (2020) regarding source-content integration and indicate that during revisits readers (also) pay additional attention to source information. Since the document set in the present study was partly conflicting, this is in line with previous work showing that conflicting information fosters readers' attention to source information (e.g., Braasch et al., 2012;Braasch & Bråten, 2017). That is, when encountering information that stands in conflict with previously read information, readers might revisit the previously read document, possibly in an attempt to restore coherence in their overall mental representation of the subject matter Braasch & Bråten, 2017). This reasoning is also in line with previous assumptions that revisit might be initiated in an attempt to re-evaluate information in the light of information found in another document (cf. Anmarkrud

Conclusion
In conclusion, document presentation did not affect multiple document comprehension. However, the two reading interactions assessed in the present study (i.e., grouping documents in the simultaneous condition and the number of revisits to documents in the sequential condition) were differentially related to different measures of multiple document comprehension. These findings considerably extend previous work. Specifically, while re-organizing partly conflicting documents into groups according to their overall stance was positively related to integrative processes within and across documents rather than to readers' attention to source information per se, the number of revisits to documents (which might be initiated to restore coherence after encountering conflicting information) was positively related to readers' mental representation of source information, but not to integrative processes. Hence, the findings of the present study stress the importance of human-computer-interaction: While specific features of user interfaces might pave the way for different kinds of interactions (e.g., necessitating a re-access of a document to re-read it or enabling grouping of documents to facilitate information comparison), how users interact with the interface might play a (more) crucial role in how well they accomplish the task. As for practical implications, thus, the present findings suggest that users might benefit from interfaces that provide strong affordances to execute particular actions, such as revisiting or grouping documents, or from respective instructions that inform or train users (how) to perform these actions.

Limitations and future work
We acknowledge that the present research comes with certain limitations. First, we examined a homogenous sample of university students, which can be regarded as proficient readers who might engage more strategically in reading interactions than non-proficient readers (cf. Goldman et al., 2012;Rouet et al., 2017). Hence, whether our findings extend to other samples, such as readers with lower reading skills, remains an open question that we would like to see addressed in future researchespecially when keeping in mind that, for example, school students could benefit from training in how to engage with reading materials in effective ways.
Second, each reading interaction could only be assessed in one presentation condition. In the simultaneous condition, it was not possible to assess the number of revisits to documents (i.e., from logfiles), since a re-inspection of information could have also occurred without overt interaction with the document. Furthermore, participants' engagement in grouping documents could not be assessed in the sequential condition, since the re-arrangement of documents was not possible in this condition in the present study, just like in most sequential reading environments in which documents usually cannot be spatially re-arranged on the screen either. Regarding the latter, one could also think of a sequential presentation condition in which re-ordering of document icons in the menu is possiblejust like tabs in a web browser can be re-ordered. Future work is needed to understand whether readers' re-ordering behavior is related to measures of multiple document comprehension in the same way as grouping documents spatially.
Third, it remains unclear from the present study which intentions participants had in engaging in the respective reading interactions. For instance, eye-tracking data and (concurrent or retrospective) think-aloud protocols (e.g., Muntinga & Taylor, 2018) could help shed light on whether readers who group documents during reading do so directly after realizing that documents support or contradict each other. Furthermore, eye-tracking data might also help shed light on how the grouping of documents affects readers' further reading process with regard to information integration within and across documents. The use of eye-tracking methodology would also allow to investigate revisits of documents that are presented simultaneously on the screen. In that case, both in a simultaneous and in a sequential document presentation, think-aloud methodology could also help shed light on readers' intentions in revisiting documents (e.g., Anmarkrud et al., 2014;Goldman et al., 2012;Wineburg, 1991).
Finally, we acknowledge that our findings regarding the relationship between reading interactions and multiple document comprehension measures are only correlational in nature. Thus, no conclusions regarding causality can be drawn. While the positive relations between reading interactions and specific aspects of multiple document comprehension suggest that training readers in strategically performing such reading interactions might support them in building a comprehensive mental representation of multiple documents, further experimental work is needed to test this assumption. Further, we cannot rule out the possibility that both grouping documents or revisiting documents, respectively, and participants' performance regarding intertextual integration, memory for sources, and source-content integration were causally influenced by a third variable, such as general cognitive ability, analytical thinking skills, or intrinsic motivation.
Notwithstanding these limitations, the findings of the present study indicate that reading interactions do play an important role in multiple document comprehension. Thus, we encourage future research to take into account reading interactions when investigating the effect of different reading interfaces or tools.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was partly funded by the Leibniz ScienceCampus T€ ubingen "Cognitive Interfaces" granted to Yvonne Kammerer (PI), Uwe Oestermeier, and Peter Gerjets.