Reading perspectives moderate text-belief consistency effects in eye movements and comprehension

ABSTRACT

Readers often prioritize processing and comprehension of information perceived as relevant to a particular intention. Using a repeated-measurement study, we investigated how readers’ prior beliefs and external reading perspectives influence processing and comprehension of belief-relevant texts on two socioscientific controversies. University students read belief-relevant texts from a belief-consistent perspective in one experimental session and from a belief-inconsistent reading perspective in another. Eye tracking was used to measure immediate and delayed processing and a sentence verification task was used to measure comprehension. Results revealed longer first-pass reading times for belief-inconsistent claims compared to belief-consistent claims, especially in the belief-inconsistent reading perspective. Longer lookbacks on belief-consistent claims were found in the belief-consistent reading perspective but similar lookback times for both types of claims in the belief-inconsistent reading perspective. We further found better comprehension for belief-consistent information in the belief-consistent reading perspective but balanced comprehension levels in the belief-inconsistent reading perspective.

The Internet increases the exposure to conflicting information about socially debated controversial scientific topics, for example, about the risks and benefits of vaccinations, or alternative fuels (socioscientific issues; Sadler, 2011). Successfully handling socioscientific issues calls for “a greater public need for scientific literacy” (Britt et al., 2014, p. 104). Scientific literacy denotes that readers can successfully comprehend information spread across diverse and even conflicting texts, can critically evaluate the credibility and relevance of socioscientific information as part of a goal directed activity, and can explain and defend their views on current scientific topics based on solid grounds. To this end, readers should pay attention to all credible information on a socioscientific issue, even if it is conflicting.

However, readers might not always follow this maxim. We argue that prior beliefs, as “a position about the truth value of a proposition” (Wolfe & Griffin, 2017, p. 3) are activated effortlessly and nonstrategically during comprehension and serve as standards of relevance, which in turn influence which information readers primarily attend to (Richter & Maier, 2017). One consequence of such a biased processing is the text-belief-consistency effect (e.g., Abendroth & Richter, 2020; Braasch et al., 2022; Maier & Richter, 2013; for a systematic overview see, Richter & Maier, 2017). Extending existing research on text-belief consistency effects, we conducted an eye tracking experiment to examine the combined effects of text-belief consistency and reading perspective. Reading and comprehending socioscientific information as goal-directed activity is not only influenced by internal reading intentions but also by external reading tasks or perspectives (McCrudden & Schraw, 2007). Thus, we tested the general hypothesis that an external reading perspective that increases the relevance of belief-inconsistent information is able reduce the text-belief-consistency effect in processing and comprehension of conflicting socioscientific information. To this end, university students read competing arguments on two socioscientific issues (alternative medicine and gender roles) in two experimental sessions in which they were either instructed to take a belief-consistent or a belief-inconsistent reading perspective. Eye tracking was used to examine moment-to-moment comprehension processes during reading, and a verification task was used to assess comprehension outcomes.

In the following sections, we elaborate on the idea of reading as goal-directed behavior and discuss how external reading perspectives can influence the processing and comprehension of texts on socioscientific issues. We then argue how and when prior beliefs can be used by readers as relevance standards in the processing and comprehension of socioscientific information.

Relevance perceptions as mental frames

The reading of texts—whether in books, magazines, or on the Web—is a central pathway to socioscientific information. The reasons for searching and reading information on socioscientific issues are manifold, for example, for participation in social or private debates, when preparing for academic training, or simply for epistemic curiosity. Such reading goals can exert a strong influence on the evaluation of information’s importance and relevance and how information is processed (McCrudden & Schraw, 2007; Rouet et al., 2017). For example, the goal-focusing model of relevance proposes that given and personal intentions jointly affect the formation of reading goals and relevance judgments (McCrudden & Schraw, 2007). Given intentions refer to externally provided cues on how the learning situation should be approached and describe the specific perspectives or goals that should be pursued. Personal intentions refer to internally generated standards that the readers bring to the learning situation. Such personal intentions could, for example, rest on readers’ standard of coherence as a crucial internal reference point that determines whether they are satisfied with their reading outcome or should continue with more (effortful) processing (Van den Broek et al., 2015), or these intentions could be internal reference points non-strategically set by readers’ prior beliefs (as described by Richter & Maier, 2017).

Together, given and personal intentions influence readers’ actual reading goal, which in turn influences how they approach the information. Such a reading goal—whether externally given or as a consequence of internal intentions or standards or both—influences the evaluation of information deemed to be relevant for the reader, and it influences the way in which readers process, comprehend, and reproduce information (e.g., Anderson & Pichert, 1978; Kaakinen & Hyönä, 2007; McCrudden & Schraw, 2007). For example, in one experiment by Kaakinen and Hyönä (2007), university students read an expository text that contained information on two different diseases. Ten sentences covered each of the two target diseases in the text. Prior to reading, participants received external perspective instructions to increase the relevance of information on either one of the two target diseases. Kaakinen and Hyönä found an effect of the reading perspective on immediate reading. The external reading perspective modified the first-pass reading of the text in such a way that perspective relevant sentences were fixated longer during initial reading, were skipped less often, and received more regressions. Perspective effects were also found in delayed processing such as producing more look-backs to relevant text segments.

In sum, relevance perceptions are a consequence of readers’ interpretation of externally provided cues or tasks, plus their internally generated intentions or goals and work as a mental frame that guides the processing, comprehension, and memory of information. In the next section, we discuss how prior beliefs can be used as standards of relevance that guide the comprehension of science-related information.

Processes involved in comprehension

Text comprehension entails that readers construct a situation model as a mental representation of what the text is about (Kintsch, 1988). The construction of situation models depends on activation, integration, and validation, which are construed as largely passive and nonstrategic cognitive processes (O’Brien & Cook, 2016; Richter & Singer, 2017). First, incoming text information passively activates associated information stored in long-term memory such as information from previous text, prior knowledge, or prior beliefs (activation, e.g., Albrecht & O’Brien, 1993; Beker et al., 2016). The text information is then enriched and integrated with the activated memory information, for example, by adding information from long-term memory to text information to fill coherence gaps (integration, for an overview of relevant theories, see McNamara & Magliano, 2009). Finally, through the validation process, readers continuously check the consistency of new information with the information activated from long-term memory (Richter, 2015; Richter & Singer, 2017).

Validation can be based on all information stored in long-term memory that becomes activated during comprehension, including prior beliefs (Richter, 2015). When readers possess strong prior beliefs, research on argument processing has shown that claims activate prior beliefs passively and quickly during reading (Voss et al., 1993). A claim is a statement that the author of an argument wants the reader to accept as true. Together with data or reasons, a claim forms the minimal structure of an argument (Toulmin, 1958). Reasons represent the support or the information that justifies the claim. Prior research indicates that prior beliefs are not only activated quickly by a claim during reading (e.g., Voss et al., 1993), but that agreement with an argument is mainly driven by agreement with the claim, whereas features of reasons influence how argument quality or strength are perceived (e.g., Wolfe, et al., 2009).

In addition, once activated by reading a claim, prior beliefs are used by readers to validate new textual information quickly and efficiently (Gilead et al., 2019; Maier et al., 2018; Voss et al., 1993). For example, in Gilead et al. (2019), participants read single statements written grammatically correct or grammatically incorrect that were either consistent or inconsistent with participants’ beliefs, for example, “Alternative medicine is a silly field” or “Alternative medicine is a serious field.” Participants’ task was to provide a binary judgment about the grammatical accuracy of the statements. The results revealed that the belief-consistency of the sentences influenced response times for the grammatical judgment. Participants were faster in judging the grammatical accuracy of the statement when they agreed with it but were slower when they disagreed with the statement. Additional evidence for differences in the processing of belief-consistent and belief-inconsistent scientific claims comes from an eye tracking study that used multiple texts presenting opposing arguments on two socioscientific controversies (Maier et al., 2018). In this study, Maier et al. (2018) found slow-downs in immediate reading for belief-inconsistent information, especially for readers with strong prior beliefs. Longer immediate readings of sentences are usually considered to indicate difficulties in the reading process when encountering the material for the first time (e.g., Kaakinen et al., 2015).

In sum, prior beliefs seem to be easily and quickly activated during comprehension and used for nonstrategic validation, especially for topics that are relevant for everyday decisions. In addition, beliefs seem to play a critical role in the immediate processing and comprehension of claims, but they tend to be not so important for reasons (e.g., Maier et al., 2018; Voss et al., 1993). In the next section, we argue how the prior-belief effects during comprehension translate into effects on comprehension outcomes.

Effects of prior beliefs on comprehension outcomes

The Two-Step Model of Validation (Richter & Maier, 2017) makes predictions of how prior beliefs that are used in validation processes influence the comprehension of belief-consistent and belief-inconsistent information. Step 1 of the model proposes that readers in most circumstances rely on routine validation that occurs relatively independent of readers’ strategic control. The model proposes further in Step 2 that readers will engage in a strategic validation of inconsistencies only if they are motivated and able to do so. We elaborate on the effects of strategic validation in the next section and first describe how routine validation affects processing and comprehension. In Step 1, prior beliefs serve as a kind of epistemic gatekeeper which determines the amount of attention and level of cognitive resources that information receive during comprehension. Belief-inconsistent information is assumed to lead to immediate interruptions during reading (visible in longer initial readings), but no efforts are made to strategically resolve inconsistencies or problems of understanding for this type of information. In terms of relevance effects, belief-inconsistent information might be perceived as less relevant during normal comprehension. In contrast, belief-consistent information leads to no interruptions during reading because the validation process results in a positive implicit plausibility judgment for this information. In the present context, we assume that such information is also perceived as more relevant by readers, based on their internal standards set by prior beliefs. However, given the nonstrategic character of validation, this effect is likely to occur without strategic control of the reader.

The belief bias highlighted by the Two-Step Model of Validation may cause comprehension differences between belief-consistent and belief-inconsistent information. The model assumes that the negative validation outcome for belief-inconsistent information results in poorer comprehension of this information. This assumption is supported by a wide range of empirical research that has found impaired processing, understanding and memory of belief-inconsistent information for different subject groups, topics, and measures (for a systematic overview see Richter & Maier, 2017). For example, in a recent study, Tarchi and Mason (2020) found biased argumentative essays after university students read multiple texts on flu vaccination. A similar biasing influence of prior beliefs has been found for adolescents (Braasch et al., 2022).

Direct support for the role of prior beliefs as an epistemic gatekeeper and the postulated effect of prior beliefs on comprehension comes from a recent study that experimentally varied readers’ prior beliefs via videos and investigated the role of plausibility judgments on the comprehension of conflicting socioscientific texts (Abendroth & Richter, 2021). The authors varied beliefs with videos that differed on their argumentative stance (either pro or con) and found that the effect of belief manipulation (i.e., the video versions) on comprehension was fully mediated by the perceived plausibility of the controversial texts. In particular, belief-consistent information was judged as more plausible during comprehension, which lead to a better understanding of this information. In contrast, belief-inconsistent information was judged as less plausible and was consequently less comprehended.

Prior beliefs and reading perspectives

In most reading situations, relying on prior beliefs as an epistemic gatekeeper for comprehension might be convenient and functional for readers. Among other benefits, a shallow comprehension of belief-inconsistent information relieves readers of the need to revise their beliefs despite disproving information (a type of selective exposure effect, Festinger, 1957). Drawing from the idea of relevance (e.g., McCrudden & Schraw, 2007) or perspective effects (e.g., Anderson & Pichert, 1978), prior beliefs activated during nonstrategic validation serve as standards of relevance leading to a belief-consistent reading perspective. The Two-Step Model of Validation (Richter & Maier, 2017) implies that such belief effects might be nonstrategic and might lead to differences during immediate and delayed processing, which in turn yield a memory advantage for belief-consistent information over belief-inconsistent information.

However, in some circumstances, readers might need to comprehend and memorize belief-inconsistent information well, for example, when following an external reading perspective or goal that enhances the relevance of belief-inconsistent information (e.g., Bohn-Gettler & McCrudden, 2018; McCrudden & Sparks, 2014). In support of this idea, a study by Bohn-Gettler and McCrudden (2018) found that external reading perspectives can affect the comprehension of belief-relevant material. Before reading a dual-position text, undergraduates received specific task instructions that required them to either focus on pro or contra arguments. Based on the assessment of participants’ prior beliefs, the authors coded a match or mismatch between participants’ prior beliefs and the task-relevance instruction. Recall data indicated that task-relevant information was recalled better compared to task-irrelevant information, independent of readers’ prior beliefs. In addition, the authors found a main effect of prior beliefs on strategic processing, indicated by think-aloud protocols, and this effect was independent of the specific reading task instruction (Experiment 2). In strategic processing, readers seemed more motivated to protect and confirm their beliefs, for example, by providing more refutational evaluations while reading belief-inconsistent information and more supportive evaluations while reading belief-consistent information (similar to the belief-protection profile, McCrudden & Sparks, 2014). Hence, recall was modified by task-relevance instructions, but the way in which readers strategically approached the texts was influenced by prior beliefs. An important point is that Bohn-Gettler and McCrudden (2018) did not investigate how the specific task instructions influenced immediate processing of belief-relevant information and used recall for explicit text information as the dependent variable. Therefore, conclusions regarding the combined effects of beliefs and relevance instructions on readers’ comprehension of belief-relevant texts cannot be drawn from their study.

Step 2 of the Two-Step Model of Validation (Richter & Maier, 2017) predicts that under specific circumstances, readers engage in elaborative processing that focuses on resolving the inconsistencies between textual information and prior beliefs detected in Step 1. Such a strategic elaboration of inconsistencies requires that readers a) are aware of the inconsistencies and b) possess the relevant cognitive, metacognitive, and motivational resources to strategically resolve the inconsistencies. Inducing belief-inconsistent reading perspectives before reading, for example, might be one way to increase the relevance of belief-inconsistent information and thus, might motivate readers to strategically evaluate belief-inconsistent information. If readers engage in such strategic elaboration, a better comprehension of conflicting scientific information should result, especially when this information conflicts with readers’ prior beliefs. In line with this assumption, Maier et al. (2018) found that more strategic processing of belief-relevant material strengthens readers’ understanding of socioscientific controversies. Moreover, a higher number of lookbacks (as indicated by eye tracking) was associated with a reduction of the text-belief-consistency effect in comprehension.

Rationale and overview of the present study

Students often read texts on the Internet as their primary source of socioscientific information during informal learning. We assume that in this process, prior beliefs activated through nonstrategic validation serve as standards of relevance, leading to an unconscious belief-consistent reading perspective. This reading perspective, in turn, might lead to differences during immediate and delayed processing of belief-relevant texts. The consequence of such a nonstrategic validation process based on prior beliefs is a text-belief consistency effect in comprehension, that is an overrepresentation of belief-consistent and an underrepresentation of belief-inconsistent information in readers’ understanding of the issue. To attenuate the text-belief consistency effect in the comprehension outcome, reading instructions directing participants to invest more cognitive resources for the strategic elaboration of belief-inconsistent information seem promising.

Thus, the overarching goal of the present study was to investigate how prior beliefs as personal intentions and external reading perspectives as given intentions jointly influence the processing and comprehension of belief-relevant information on two socioscientific controversies (alternative medicine and gender roles). An eye tracking study was conducted with university students who participated in two separate sessions. In each session, participants read texts with conflicting claim-first arguments either on the effectiveness of alternative medicine or on the effects of genes on gender roles. First- and second-pass fixation measures were collected as indicators of immediate and delayed processing for claims and reasons, which are the central functional elements of an argument (Toulmin, 1958). Comprehension outcomes were additionally assessed with a sentence verification task (adapted after Schmalhofer & Glavanov, 1986).

Hypotheses for comprehension outcomes (verification task)

Based on the assumptions made in the Two-Step Model of Validation (Richter & Maier, 2017) and on earlier research on the effects of readers’ belief and external reading perspectives on comprehension (e.g., Kaakinen & Hyönä, 2007; Maier et al., 2018), we expected the external reading perspective to moderate the effects of readers’ beliefs on the comprehension of conflicting socioscientific arguments (Hypothesis 1). In detail, we expected that an external belief-consistent reading perspective would boost readers’ natural tendency to better comprehend belief-consistent information compared to belief-inconsistent information (text-belief-consistency effect). In contrast, an external belief-inconsistent reading perspective should attenuate the text-belief-consistency effect leading to similar levels of comprehension for belief-consistent and belief-inconsistent texts because such a reading perspective should enhance elaborative processing for belief-inconsistent information.

Hypotheses for processing (eye movements)

When readers monitor belief-consistency routinely during reading, their initial reading of belief-inconsistent information should be longer compared to their initial reading of belief-consistent information. In the present study, first-pass reading times (summed duration of all first fixations of unread words for a sentence) were computed as indicators of initial processing during reading (cf. Hyönä et al., 2003). Longer first-pass reading times are perceived to indicate that readers experience processing difficulties when reading the material for the first time (e.g., Kaakinen et al., 2015). Hence, we predicted longer first-pass reading times for belief-inconsistent compared to belief-consistent sentences (Hypothesis 2).

Moreover, the effect of belief-consistency on first-pass reading times as indicators of immediate reading should be moderated by argument part and reading perspective (Hypotheses 3). Belief-consistency is especially important during claim processing (Voss et al., 1993). If this assumption holds, the belief-consistency effect for first-pass reading times should especially occur for claims compared to reasons. Second, we expected the belief-consistency effect in claims to be moderated by the external reading perspective. For immediate reading of belief-consistent claims, we expected no effects of the external reading perspective because the validation process leads to no disruption in the comprehension of belief-consistent information (e.g., Maier et al., 2018). In contrast, first-pass reading for belief-inconsistent claims should vary as a function of the reading perspective. For these sentences, we expected a perspective effect, that is longer first-pass reading times for belief-inconsistent claims in the belief-inconsistent reading perspective compared to the belief-consistent reading perspective and vice versa for the belief-consistent claims.

We investigated the duration of lookbacks (summed duration of all backward and forward fixations) for sentences that received at least one lookback as indicator of delayed processing during reading. A preferential focus on claims was expected and consequently, longer lookback times for claims compared to reasons were predicted (Hypotheses 4). We also predicted a perspective effect for lookback times on claims (Hypotheses 5). In contrast to indicators of immediate processing, lookback time directly depends on readers’ strategic decision which type of information they prefer to process. Thus, we predicted this decision to be a function of the external reading perspective (e.g., Kaakinen & Hyönä, 2007). In particular, lookbacks should be longer for belief-consistent compared to belief-inconsistent claims in the belief-consistent reading perspective, whereas lookback time for belief-inconsistent compared to belief-consistent claims should be longer in the belief-inconsistent reading perspective.

Methods

Participants

Participants were recruited via posters and by using a mailing list of the Institute of Psychology at the university where the experiment took place. Thirty university undergraduates majoring in Psychology participated in the study. However, two participants attended only one of the two experimental sessions. In addition, eye tracking data of one participant could not be used because calibration of the eye-tracker failed for this participant. Data from these three participants could not be analyzed. The remaining 27 participants were on average 24.74 years old (SD = 6.73 years). Most participants (n = 21) were in their second semester (M = 2.74 semesters, SD = 1.48 semesters). Gender of the participants was not recorded correctly because of a technical error.

Apparatus

A desk-mount EyeLink 1000 eye tracker manufactured by SR Research Ltd. (Ontario, Canada) with a chin–forehead rest was used to record participants’ eye movements. Pupil location was sampled at a rate of 1000 Hz. Eye movements were calibrated with a nine-point grid (average fixation error: <0.5 degrees of visual angle following validation). We recorded movements of participants’ dominant eye (in most cases the right eye), but participants were able to view the texts with both eyes that was presented on a 22-inch TFT monitor with a resolution of 1024 × 768 and a refresh rate of 75 Hz in a viewing distance of about 86 cm.

Text material

Six claim-first arguments (three pro arguments, three contra arguments) plus a short introductory text each were created for the two socioscientific topics. Arguments were based on information published on the web appearances of reputable German magazines (e.g., Spiegel Online, http://www.spiegel.de) and were written by the first author. The writing style, structure, and length of the arguments were held strictly parallel (see Table 1 for a synopsis of argument characteristics). For each socioscientific topic, participants first read an introductory text (alternative medicine: 160 words, gender roles: 159 words) that provided some general background knowledge on the topic and highlighted the controversial nature of the socioscientific issue. Afterward, three arguments for each argumentative position in the controversy (pro vs. contra) were presented in a blocked fashion. A blocked presentation was used to reduce the beneficial influence of an alternating presentation mode on processing and comprehension (see, Maier et al., 2018). The alternative medicine text consisted of three pro arguments that argued for the effectiveness of alternative medicine and the three contra arguments that argued against the effectiveness of alternative medicine. The socioscientific text on the controversy about gender roles consisted of three pro arguments claiming that gender roles are genetically inborn and the three contra arguments claiming that gender roles are socially learned.

Table 1. Characteristics, quality, and comparability of the arguments.

			Understandability	Plausibility	Interest	Stance
Text topic	Argument	Length (number of words)	M (SD)	M (SD)	M (SD)	M (SD)
Alternative medicine	Pro 1	192	4.68 (1.02)	4.35 (1.11)	4.15 (1.35)	5.54 (0.78)
	Pro 2	178	4.04 (0.88)	3.94 (0.82)	4.00 (1.08)	5.46 (0.78)
	Pro3	171	4.46 (0.94)	4.08 (1.01)	4.62 (1.66)	5.46 (0.88)
	Con 1	193	4.15 (0.85)	3.62 (1.00)	4.08 (1.32)	1.85 (0.99)
	Con 2	187	3.28 (0.76)	3.63 (0.88)	3.08 (1.07)	2.31 (1.38)
	Con 3	197	4.26 (0.72)	3.71 (0.87)	4.31 (1.31)	2.08 (0.86)
Gender roles	Pro 1	165	4.15 (0.92)	3.50 (0.95)	4.18 (0.18)	5.82 (0.41)
	Pro 2	181	4.17 (0.73)	3.09 (0.84)	3.91 (1.04)	4.09 (1.87)
	Pro3	184	3.82 (0.89)	3.48 (0.95)	4.45 (1.13)	5.18 (0.87)
	Con 1	180	4.65 (0.66)	4.50 (0.81)	4.55 (1.29)	1.27 (0.65)
	Con 2	190	4.29 (0.91)	4.05 (1.35)	4.73 (0.79)	1.91 (0.83)
	Con 3	190	4.00 (0.52)	3.95 (0.69)	4.27 (1.01)	1.45 (0.52)

Note. Each entry represents the average judgments of university students in the pilot study (n = 11 for gender roles; n = 13 for alternative medicine). Response categories ranged from 1 = do not agree at all to 6 = fully agree. The plausibility scale consisted of four items (Cronbach’s α ≥ .70) and the understandability scale consisted of six items (Cronbach’s α ≥ .74). Interestingness (response categories ranging from 1 = do not agree at all to 6 = fully agree) and stance (response categories ranging from 1 = alternative medicine ineffective/gender roles socially learned to 6 = alternative medicine effective/gender roles genetically determined) were assessed with one item each.

Each argument presented one unique claim in the first sentence written in bold, followed by 8 to 11 additional sentences that presented the reasons. A sample argument is presented in Table 2. Total number of words of the arguments were held as similar as possible. In addition, the comparability of the arguments with regard to understandability, plausibility, interest, and argumentative position were investigated in a pilot-study with an independent sample of university students (Table 1). Wilcoxon signed-rank test with Bonferroni correction (Type I error probability: .05/9 tests = .006) revealed no significant differences between the arguments in understandability, plausibility, and interest. However, as expected, pro and contra arguments differed with regard to their argumentative position in the Wilcoxon signed-rank test, p < .001, with one exception. For one argument (Pro argument 2 on gender roles), the argumentative position was slightly less clear and Wilcoxon signed-rank tests comparing this argument to the three contra arguments were not significant, all p ≤ .01.

Table 2. Examples of test sentences used in the verification task for the controversy of gender roles for one pro argument (English translation).

Type of test item	Wording
Original argument	Gender roles are an evolutionary result. People whose genetic predisposition is best adapted to the environmental conditions have a higher chance of survival. This process is known as natural selection and has ensured that in the long term the behavior of men and women that is optimally adapted to the environment has developed in the form of genetically fixed gender-typical behavioral preferences. On average, women were and are physically weaker than men. In order to be able to protect themselves despite physical inferiority, it was helpful to notice emotions quickly so that they could flee as early as possible. Women with these characteristics had a survival advantage and were more likely to pass on their genes. Over thousands of years, the gene pool has thus developed toward an overall more empathetic woman. For men, on the other hand, it was evolutionary advantageous to be dominant and strong in order to consolidate their own ranking and act as a good breadwinner. Such gender-specific preferences in the choice of partner can still be found today. For example, a speed-dating experiment showed that women attached more importance to the intelligence and status of men. For men, the physical attractiveness of women was more important.
Paraphrase	For protection in case of a physical disadvantage, it helped to detect emotions at an early stage in order to enable a quick escape.
Inference	Women who do not show gender-typical behavior are less likely to pass on their genes.
Distracter	Behaviors that are gender-specific are, for example, the increased use of cosmetic products by women.

Note. The paraphrase refers to the italicized sentence in the original paragraph. In the original texts, the sentences were not italicized. Note that distracters did not directly belong to either the pro or the contra arguments but are presented only for illustration.

Comprehension measure

We used a verification task (modified after Schmalhofer & Glavanov, 1986) with 54 test items (sentences) per socioscientific controversy to assess comprehension of the arguments. Thirty-six paraphrases (nine pro and nine contra paraphrases for each of the two socioscientific controversies) and 36 inferences (nine pro and nine contra inferences for each of the two socioscientific controversies) were created. Paraphrases conveyed information presented in the arguments. To construct paraphrases, sentences from the arguments were chosen and varied in such a way that the word order and key content words of the sentences were modified. Inferences presented information that was not explicitly stated by the arguments but needed to be inferred by the participants during comprehension to build an adequate mental representation of the arguments. As a third type of test items, we created 18 distractors for each socioscientific controversy as sentences that contained information related to the overall socioscientific controversy but that were not presented in the arguments and could not be inferred from the arguments. Table 2 provides an example for each sentence type for one pro argument of the socioscientific controversy on gender roles. We used the accuracy of responses to the paraphrase and inference items as an indicator of comprehension. In addition, the proportion of yes responses to the distracters was used as an indicator of response bias and was controlled statistically in the analyses.

Reading perspectives

In the same pilot test used to assess the quality and comparability of the arguments, participants evaluated five reading perspectives for each argumentative position (pro vs. contra) for the two socioscientific controversies (response scale ranging from 1 = do not agree at all to 6 = fully agree). Each reading perspective required participants to imagine themselves as a person that pursues a specific goal with regard to the socioscientific controversy (e.g., “Imagine you are a physician at a university and are planning a seminar in which you want to report on the formative influence of culture on typically male and female gender roles”). For each perspective, participants in the pilot test evaluated the plausibility of the reading perspective and the credibility of the person mentioned. In two additional items, participants indicated the extent to which the reading perspective matches the pro stance (e.g., the person takes the perspective that gender-specific behavior is exclusively genetic) or the contra stance (e.g., the person takes the perspective that gender-specific behavior is exclusively socially learned) in the socioscientific controversy. Based on these pilot study results, two reading perspectives that could be clearly assigned to one of the two controversial points of view in each scientific debate were selected for each socioscientific controversy (Table 3). Care was taken to ensure that the other characteristics had comparable mean scores for all reading perspectives used in the experiment.

Table 3. Evaluation of the reading perspectives.

		Argumentative Position – Pro	Argumentative Position – Con	Plausibility of Perspective	Credibility of Person
Text Topic	Reading Perspective	M (SD)	M (SD)	M (SD)	M (SD)
Alternative medicine	Pro = “Imagine you have a practice as an alternative practitioner and organize an information event to convince patients of the advantages of alternative medicine.”	4.92 (0.28)	1.00 (0.00)	3.54 (1.51)	2.54 (1.05)
	Contra = “Imagine you are a general practitioner, and you are planning a training course for staff to educate about the ineffectiveness of alternative medicine.”	1.08 (0.28)	4.92 (0.28)	2.23 (0.93)	3.00 (1.00)
Gender roles	Pro = “Imagine you are a neurologist and want to draw attention to the unchangeable differences between men and women.”	4.30 (0.68)	1.40 (0.52)	2.90 (0.74)	3.10 (0.74)
	Contra = “Imagine you are an educator, and you are planning a parent evening to explain the influence of education on gender-typical behavior.”	1.70 (0.68)	3.80 (0.63)	3.50 (0.53)	2.80 (0.92)

Note. Each entry represents the average judgments of university students in the pilot study (n = 11 for gender roles; n = 13 for alternative medicine). Response categories ranged from 1 = not at all to 6 = totally.

Reader characteristics

Prior beliefs

Participants’ prior beliefs about the two socioscientific controversies were assessed with 10 statements for alternative medicine and eight statements for gender roles (response categories ranging from 1 = do not agree at all to 6 = fully agree).¹ For alternative medicine, five statements assessed participants’ agreement to the pro argumentative stance that alternative medicine is effective and better than conventional medicine (e.g., “I think that alternative medicine is better than conventional medical treatments,” Cronbach’s α = .72) and five statements assessed participants’ agreement to the contra argumentative stance that alternative medicine is ineffective and should not be used (e.g., “I am convinced that alternative medicine is ineffective,” Cronbach’s α = .81). For gender roles, four statements assessed participants’ agreement that gender roles have genetic causes (e.g., “I think that gender-specific behavior patterns can be traced back to the genetic make-up of men and women,” Cronbach’s α = .61) and four statements assessed the belief that gender roles are socially learned (e.g., “I think that gender-specific behavior patterns are learned socially,” Cronbach’s α = .76).

Participants were additionally required to indicate their preference for one of the two competing argumentative stances for each socioscientific controversy as a binary choice. For the controversy of alternative medicine, participants were asked, “Are you a supporter or a sceptic of alternative medicine?,” and for the scientific topic of gender roles, participants were asked, “Do you consider gender-specific behavior to be innate or learned?” Participants indicated their response by pressing one of two keys marked for supporter/innate (“d”) and skeptic/learned (“k”).

Reading skills

To control for effects of reading skills on processing and comprehension, the German reading test ELVES was used (Richter & van Holt, 2005, Cronbach’s α = .87). In this test, participants are instructed to judge the correctness of assertions about abstract and concrete concepts with a binary response (true or false). Combining the accuracy and the speed of the answers for each assertion, the test indicates the efficiency of the respondents to use propositional strategies for reading comprehension at the sentence level (i.e., lexical access, syntactic and semantic integration; for details, see, Richter & van Holt, 2005).

Manipulation check

At the end of each session, participants were asked to write down the reading perspective they had received for reading the arguments. They were further asked to try to remember the wording as accurately as possible. In addition, participants indicated whether they read from this perspective (yes vs. no) and were required to write down the appropriate alternative perspective if they answered “no” to this question.

Procedure

Prior to the study, participants gave their informed consent. The study consisted of two individual sessions (approximately 60 min each) for each participant with a time lag of approximately two weeks between sessions. Each session focused on one socioscientific issue, that is, either on alternative medicine or on gender roles, with the order counterbalanced across participants. Participants were informed at the beginning of each session that they would be asked to read the texts from a given perspective and to remember this perspective at the end of the session. At the beginning of each session, participants indicated their beliefs about the corresponding issue on the belief measure. Afterward, they indicated the dichotomous preference for one of the two argumentative stances. The belief measure was provided before the binary choice to make participants aware of their prior beliefs and their stance in the controversy.

After providing the binary choice, which indicated participants’ stance in the socioscientific controversy, participants received the reading perspective on the computer screen. Thus, belief-consistency of the reading perspective was varied depending on participants’ preference for the two competing argumentative stances in each controversy, not on the finer-grained belief scales. Each participant read the arguments on one socioscientific controversy with a belief-consistent reading perspective in one experimental session and the arguments on the other socioscientific controversy with a belief-inconsistent reading perspective in the other experimental session. The reading perspective and the text topic were counterbalanced across sessions and participants.

After participants had read the reading perspective on screen, they were told to read the arguments from this perspective as thoroughly as possible and to read the texts as if they were the corresponding person described in the reading perspective. To facilitate the adoption of the reading perspective, they were also asked to speak the reading perspective aloud.

In the next step, the eye tracker was calibrated. In each session, a nine-point calibration grid covering the entire screen was used to calibrate the eye tracker before reading the arguments. Initial calibration was considered good when an average angular error of less than 0.5° was reached. If the accuracy of the measurement decreased during the experiment, calibration was repeated.

For each controversy, participants read the introduction and the six arguments on individual screens at their own pace. At each screen, participants were asked to fixate a cross presented at the position where the first word would appear. Arguments were presented with a maximum of 17 double-spaced lines on the screen in black Courier New font (13.5 point, claim presented in bold) on a light gray background. When participants had finished reading the information on one screen, they could proceed by pressing the space bar. After participants had read the introductory passage and all six arguments for one issue, they worked on the verification task. At the end of the second session, participants were thanked and debriefed. Participants received either a small monetary reward (8 Euros per hour) or course credit for their participation.

Design

For the verification responses, the experimental design was a 2 (belief-consistency: consistent argument vs. inconsistent argument) × 2 (reading perspective: belief-consistent perspective vs. belief-inconsistent perspective) within-subjects design. For the eye tracking measures, the experimental design was a 2 (belief-consistency: belief-consistent argument vs. belief-inconsistent argument) × 2 (reading perspective: belief-consistent perspective vs. belief-inconsistent perspective) × 2 (argument part: claim vs. reason) within-subjects design. In addition, the order of the belief-consistent and belief-inconsistent arguments (belief-consistent argument first vs. belief-inconsistent argument first) and the order of the reading perceptive (belief-consistent reading perspective first vs. belief-inconsistent reading perspective first) was counterbalanced across participants. Text topic (alternative medicine vs. gender roles), measurement time (Session 1 vs. Session 2), the arguments, and participants’ reading skills were included as control factors or continuous covariates in all analyses.

Results

Type I error probability for all significance tests was set to .05. We only report effects relevant for the hypotheses and the main effects of the two critical predictors, reading perspective and belief-consistency. Parameter estimates (fixed effects and variance components) for the generalized linear mixed model (GLMM) for the verification task are provided in Table 4 and for the linear mixed model for the eye tracking measures in Table 5. Table 6 further provides the means (and standard errors) for first-pass reading times, and lookback times as a function of belief-consistency, reading perspective, and argument part. In addition to parameter estimates, we report predicted (conditional) probabilities of the responses for the GLMM analyses in the text (back-transformed from the logit-link model with estimated standard errors).

Table 4. Fixed effects and variance components of the GLMM for comprehension (verification task).

	Fixed Effects
Parameter	β (SE)
Intercept	1.72 (0.16)***
Text belief-consistency^a	0.09 (0.07)
Reading perspective^a	0.02 (0.07)
Topic^a	−0.19 (0.13)
Reading order^a	0.06 (0.08)
Session^a	−0.01 (0.07)
Reading ability^b	0.30 (0.13)*
Yes-responses to distracters^b	−0.05 (0.11)
Text belief-consistency × Reading perspective	0.14 (0.07)*
	Variance Components
Parameter	Variance (SD)
Subjects	0.30 (0.55)
Test items	0.64 (0.80)

Note. N = 1,836 observations (27 participants x 72 items minus missing values)

^aContrast coded

^bGrand-mean centered. Text-belief consistency (−1 = belief-inconsistent argument, 1 = belief-consistent argument), reading perspective (−1 = belief-inconsistent perspective, 1 = belief-consistent perspective), topic (gender roles = −1, alternative medicine = 1), reading order (−1 = belief-inconsistent arguments first, 1 = belief-consistent arguments first), session (first session = −1, second session = 1). Comprehension (outcome variable): Yes responses to paraphrases and inferences in the verification task (0 = incorrect answer, 1 = correct answer).

* p < 0.05, ** p < .01, *** p < .001 (two-tailed).

Table 5. Parameter estimates (fixed effects and variance components) in the mixed models for the eye tracking measures.

	First-Pass Reading Time		Lookback Time
	Fixed Effects
Parameter	β (SE)		β (SE)
Intercept	287.05 (11.43)***		282.14 (17.30)***
Text-belief consistency^a	−5.97 (2.36)*		41.15 (12.63)**
Reading perspective^a	−3.00 (2.24)		−3.44 (12.40)
Argument part^a	51.47 (7.07)***		208.52 (15.81)***
Topic^a	−0.77 (4.14)		13.54 (8.06)
Reading order^a	−9.69 (1.73)***		10.04 (6.51)
Session^a	−5.29 (1.26)***		−4.75 (5.29)
Reading ability^b	−48.87 (9.05)***		−7.07 (8.62)
Text-belief consistency × reading perspective	6.95 (2.12)**		42.32 (12.17)***
Text-belief consistency × argument part	−7.67 (2.36)**		40.62 (12.59)**
Reading perspective ×argument part	−3.53 (2.11)		1.43 (12.06)
Text-belief consistency × reading perspective ×argument part	6.78 (2.12)**		39.27 (12.13)**
	Variance Components
Parameter	Variance (SD)		Variance (SD)
Subjects	2174 (46.63)		1110 (33.32)
Test Items	1981 (44.51)		4377 (66.16)

Note. N = 3,184 observations (27 participants x 127 sentences minus missing values due to data cleaning) for first-pass reading time and N = 947 observations for lookback time.

^aContrast coded

^bGrand-mean centered. Text-belief-consistency (−1 = belief-inconsistent argument, 1 = belief-consistent argument), reading perspective (−1 = belief-inconsistent perspective, 1 = belief-consistent perspective), topic (gender roles = −1, alternative medicine = 1), reading order (−1 = belief-inconsistent arguments first, 1 = belief-consistent arguments first), session (first session = −1, second session = 1). Comprehension (outcome variable): Accuracy of responses to paraphrases and inferences in the verification task (0 = incorrect answer, 1 = correct answer).

* p < 0.05, ** p < .01, *** p < .001 (two-tailed).

Table 6. First-pass reading time and lookback time as a function of belief-consistency, reading perspective and argument part.

		First-Pass Reading Time				Lookback Time
		Belief Consistent Text		Belief-Inconsistent Text		Belief-Consistent Text		Belief-Inconsistent Text
Reading Perspective	Argument Part	M	SEM	M	SEM	M	SEM	M	SEM
Belief-consistent	Claim	332.07	17.21	331.88	17.25	652.01	47.46	325.29	48.50
	Reason	237.98	10.18	234.23	10.17	72.33	13.09	65.17	13.59
	Total	285.02	11.85	283.06	11.86	362.17	25.15	195.23	25.74
Belief-inconsistent	Claim	317.68	18.23	372.41	18.29	492.85	48.86	492.50	62.19
	Reason	236.57	10.39	233.53	10.39	75.98	15.48	81.01	15.50
	Total	277.13	12.28	302.97	12.30	284.42	26.20	286.76	32.58
Total	Claim	324.87	16.78	352.14	16.81	572.43	37.30	408.89	42.28
	Reason	237.27	10.09	233.88	10.09	74.16	11.99	73.09	12.15
	Total	281.07	11.67	293.01	11.68	323.29	20.19	240.99	22.59

Prior beliefs

In the difference score of the belief scales (pro minus contra) for gender roles, we found more participants (n = 22) favoring the argumentative stance that gender roles are socially learned. One participant was neutral and four participants had a difference score indicating greater agreement with the belief that gender roles are innate. Likewise, most participants (n = 22) indicated that they believed gender roles to be socially learned in the binary belief choice, whereas the remaining five participants indicated that they believed gender roles to be innate.

Investigating the difference score of the belief scales for alternative medicine (pro minus contra) indicated that seven participants agreed more with the contra belief scale, two participants were neutral and 18 participants agreed more with the pro belief scale. Likewise, for this controversy, 16 participants indicated that they consider themselves to be a skeptic of alternative medicine and 11 participants indicated that they consider themselves as supporters of alternative medicine in the binary belief choice measure.

The binary belief choice for one of the two argumentative stances in each controversy was used to provide participants with one belief-consistent and one belief-inconsistent perspective each in the two experimental sessions.

Manipulation check

All participants correctly reported the reading perspective at the end of both sessions. In addition, most participants reported that they had read the arguments with the given reading perspective in mind. In 77.8% (n = 42) of the 54 experimental sessions (27 participants × 2 sessions), participants followed the reading instruction, whereas 22.2% (n = 12) reported that they had not followed the reading perspective. For the latter data points, 75% (n = 9) reported to have read with their own argumentative stance in mind and 25% to have read from a neutral perspective (n = 3). Given this result from the manipulation check, we recoded the reading perspective to align to what participants had in mind during reading. In detail, data for the topics for which individual participants reported to have read from their own point of view were recoded as belief-consistent reading perspective for these participants. Data points of participants reading with a neutral perspective were treated as missing values in the analyses.

Effects of belief-consistency and reading perspective on comprehension

A GLMM analysis with a logit link function was used to analyze responses in the verification task using the lme4 package with bobyqa as the optimizing function (version 1.1–21, Bates et al., 2019) and the lmerTest package (version 3.1–0, Kutznetsova et al., 2019) in R (version 4.0.3., R Core Team, 2020). Fixed effects were specified for the contrast-coded independent variables belief-consistency (−1 = belief-inconsistent argument, 1 = belief-consistent argument) and reading perspective (−1 = belief-inconsistent perspective, 1 = belief-consistent perspective) as well as their interactions. In addition, to control for ordering and topic effects, the order of the arguments (−1 = belief-inconsistent arguments first, 1 = belief-consistent arguments first), the session (first session = −1, second session = 1), and the topic of the arguments (gender roles = −1, alternative medicine = 1) were included as additional contrast-coded predictors. Reading abilities and participants’ response bias were further controlled by including participants’ score of the German reading test ELVES (z-standardized) and their average responses to the distracters in the verification task (z-standardized) as continuous covariates in the analysis. Subjects and test items of the verification task were included as random factors.

The data set used to analyze the verification task was checked for outliers with the R package influence.ME (version: 0.9.9, Nieuwenhuis et al., 2012). We computed and investigated Cook’s D (Cook, 1977) as indicator of the combined effect of leverage (extremity in the independent variables) and discrepancy (extremity in the dependent variables) and DFBETAS_ij as a local measure of influence of individual data points on specific regression coefficients. Following the suggestions of Cohen et al. (2003), the cutoff value for Cook’s D was set to 1.0. and the cutoff value for DFBETASij, was set to ±1. The outlier check indicated that no participant exceeded the cutoff values for Cook’s D and for DFBETAS_ij.

In support of Hypothesis 1, we found a significant interaction of belief-consistency and reading perspective, z = 2.09, p = .037. In the belief-consistent reading perspective condition, participants’ verification responses were more accurate for the belief-consistent test items (p = .88, SE = .02) compared to the belief-inconsistent test items (p = .82, SE = .03), β = 0.23, z = 2.65, p = .008, OR = 1.26 (95% CI: 1.06, 1.50). In contrast and as expected, participants’ verification responses showed similar accuracy scores for the belief-consistent (p = .84, SE = .03) and the belief-inconsistent test items (p = .85, SE = .03) when participants had received the belief-inconsistent reading perspective instruction, β = −0.05, z = −0.44, p = .657, OR = 0.96 (95% CI: 0.78, 1.18). Figure 1 displays the predicted (conditional) probabilities of verification response accuracies, back-transformed from the logit-link model with estimated standard errors. No significant main effects of belief-consistency (z = 1.32, p = .187) and reading perspective (z = 0.29, p = .771) were found.

Figure 1. Interaction effects of belief-consistency and reading perspective on comprehension (accuracy of responses to paraphrases and inferences in the verification task). Probabilities were back-transformed from the GLMM analysis with logit link function and are provided with estimated standard errors.

Effects of belief-consistency and reading perspective on eye-movement measures

Measures of readers’ eye movements were computed on the sentence level using Eyelink Analysoija (Hyönä et al., 2016). First-pass reading times were computed as the summed duration of all first fixations of unread words for a sentence, and lookback times were computed as summed duration of all backward and forward fixations. These indicators of immediate and delayed reading times were always divided by the number of words per sentence. Approximately 2% of the trials were excluded from the analyses because of tracking losses, blinks, and incomplete trials (62 of 3,429 trials).

We conducted linear mixed model analyses for the reading time measures and similar to the verification task included fixed effects of belief-consistency, reading perspective, and argument part (−1 = reason, 1 = claim) as well as their interactions as main predictors. In addition, fixed effects for the order of the arguments, the session, the topic of the arguments, and participants’ score of the German reading test ELVES as well as random factors for subjects and sentences were included.

First-pass reading time

As predicted by Hypothesis 2, we found a main effect of belief-consistency, indicating that first-pass reading time of belief-inconsistent sentences (M = 293.01, SE = 11.68) was longer compared to the first-pass reading time of belief-consistent sentences (M = 281.07, SE = 11.67), t(3129) = −2.52, p = .012, d = −.13. We further found a main effect of argument part, t(126) = 7.28, p < .001, d = 1.11. First-pass reading time for claims (M = 338.51, SE = 16.18) was longer compared to the first-pass reading time of reasons (M = 235.58, SE = 9.98). No main effect of reading perspective was found, t(3039) = −1.34, p = .18.

As expected, we also found a significant interaction of belief-consistency and argument part, t(3129) = −3.24, p = .001. First-pass reading time for belief-inconsistent claims (M = 352.14, SE = 16.81) were longer compared to the first-pass reading time for belief-consistent claims (M = 324.87, SE = 16.78), t(3128) = −3.03, p = .002, d = −.29. In contrast, first-pass reading time for reasons was not affected by belief-consistency, t(3134) = 1.15, p = .248.

The interaction of belief-consistency and argument part was further moderated by reading perspective as indicated by a significant three-way interaction of belief-consistency, reading perspective and argument part on first-pass reading time, t(3032) = 3.19, p = . 0.001 (Figure 2). First-pass reading times for belief-consistent claims were similar in the belief-inconsistent reading perspective (M = 317.68, SE = 18.23) and the belief-consistent reading perspective (M = 332.07, SE = 17.21), t(3031) = 1.26, p = .208. For belief-inconsistent claims, however, reading perspective exerted a significant effect on first-pass reading times, t(3031) = −3.51, p < .001, d = 1.06. First-pass reading times for belief-inconsistent claims were longer in the belief-inconsistent reading perspective (M = 372.41, SE = 18.29) compared to the belief-consistent reading perspective (M = 331.88, SE = 17.25). This pattern of results is in line with Hypothesis 3. Reading perspective for reasons had no effect on first-pass reading times, as shown by the similar results for first-pass reading times for the belief-consistent reasons (t(3045) = 0.35, p =.726) and for belief-inconsistent reasons (t(3042) = 0.18, p = .859) in both reading perspective conditions.

Figure 2. Three-way interaction of belief-consistency, reading perspective, and argument part on first-pass reading time with estimated standard errors.

Lookback time

The duration of lookbacks was investigated for sentences that received at least one lookback (947 sentences, 29.7%). The analyses for the lookback time revealed a main effect of argument part, t(145) = 13.19, p < .001, d = 2.53. In line with Hypothesis 4, claims received longer lookbacks (M = 490.66, SE = 31.37) than reasons (M = 73.62, SE = 10.71). We also found a main effect of text-belief consistency for lookback time, t(923) = 3.26, p = .001, d = .50. Overall, belief-consistent sentences (M = 323.29, SE = 20.19) received longer lookbacks than belief-inconsistent sentences (M = 240.99, SE = 22.59). The main effect of belief-consistency was further qualified by an interaction of belief-consistency and reading perspective, t(899) = 3.48, p < .001, and a three-way interaction of belief-consistency, reading perspective, and argument part, t(897) = 3.24, p = .001 (Figure 3). For claims, a significant interaction of belief-consistency and reading perspective was found, t(900) = 3.44, p < .001. As predicted for the belief-consistent reading perspective, belief-consistent claims received longer lookbacks (M = 652.01, SE = 47.46) compared to belief-inconsistent claims (M = 325.29, SE = 48.50), t(910) = 5.21, p < .001, d = 1.98. In the belief-inconsistent reading perspective, however, belief-consistent and belief-inconsistent claims unexpectedly received equally long lookbacks, t(897) = 0.01, p = .996. Hence, only a perspective effect for claims was found for participants reading with a belief-consistent reading perspective but not for participants reading with a belief-inconsistent perspective in mind. This pattern of effects only partially supports Hypothesis 5. We found no differences in lookback times for reasons as a function of belief-consistency and reading perspective, t(846) = 0.57, p = .567.

Figure 3. Three-way interaction of belief-consistency, reading perspective, and argument part on lookback time with estimated standard errors.

Discussion

The goal of the present study was to investigate the extent that external reading perspectives moderate the influence of prior beliefs on the processing and comprehension of controversial information on two socioscientific issues. Eye tracking data revealed that belief-consistent and belief-inconsistent information is processed differently during initial reading. In detail, we found longer first-pass reading times for belief-inconsistent information compared to belief-consistent information. This effect occurred especially for claims but not for reasons, which is in line with research indicating that beliefs are more important for claim processing compared to the processing of reasons (e.g., Voss et al., 1993). Longer first-pass reading times can be viewed as an indicator of comprehension difficulties during initial reading (Kaakinen et al., 2015). In this regard, the belief-consistency effect in first-pass reading is consistent with earlier research that provided evidence for an initial monitoring process during text comprehension (Albrecht & O’Brien, 1993; Maier et al., 2018). The result is also in line with the Two-Step Model of Validation (Richter & Maier, 2017) that informed the present research. In this model, the passive validation process that is part of comprehension is assumed to lead to a positive judgment for belief-consistent information. Put differently, no disruptions during immediate reading should occur for this type of information. In contrast, belief-inconsistent information is associated with a negative validation judgment and is thus viewed as less plausible during the immediate validation process. This causes comprehension difficulties, which are visible in longer immediate processing times. The empirical results found in the present study for first-pass reading times fully match the postulated pattern of differences in immediate reading times.

Our study is also able to extend prior research on immediate monitoring effects in comprehension in two important ways. First, a disruption in reading of belief-relevant claims has been found in earlier eye tracking research only for first-pass rereading time (Maier et al., 2018). Similar to first-pass reading time, first-pass rereading time can be considered an indicator of immediate reading difficulties (Kaakinen et al., 2015). Both indicators of immediate processing are closely related, and disruption of reading is likely to become evident in only one of the two indicators. Second, our study also extends prior research on text-belief consistency effects in processing of controversial information by providing insight into how external reading perspectives can influence immediate processing of controversial and belief-relevant texts. First-pass reading times for belief-inconsistent claims were affected by an external reading perspective. The disruption during initial reading for belief-inconsistent claims was greater in the belief-inconsistent reading perspective compared to the belief-consistent reading perspective. This finding is in line with research on perspective effects that show longer initial readings of perspective relevant material (e.g., Kaakinen & Hyönä, 2007). However, the finding can also be interpreted as a stricter monitoring process for belief-inconsistent information in the belief-inconsistent reading perspective. We propose that the perspective effect found in the first-pass reading time for belief-inconsistent claims was likely the result from a shift in readers’ standards of coherence, which in turn influenced the extent of the passive validation process. Implicitly embedded in several theories of text comprehension, standards of coherence indicate the critical point during reading that determines when readers are satisfied with their reading outcome (Van den Broek et al., 2015). The RI-Val Model (O’Brien & Cook, 2016), for example, describes that the coherence threshold, as one type of standard of coherence, determines the time in point when readers finish reading one sentence and proceed to the next sentence. An external reading perspective that encourages engagement with belief-inconsistent information might have modified the coherence threshold for belief-inconsistent claims, which results in longer immediate reading times for this information.

We also found only a partial perspective effect for lookback time as an indicator of delayed processing. When reading with a belief-consistent reading perspective, participants focused more on belief-consistent claims in delayed reading as shown by the longer lookback times for belief-consistent claims compared to the lookback times for belief-inconsistent claims. When reading with a belief-inconsistent reading perspective, however, lookback times for both types of claims were similar. Reading goals are influenced by given and personal intentions (McCrudden & Schraw, 2007), or stated differently, external processing instructions are interpreted based on individual intentions and interests (Rouet et al., 2017). A belief-consistent reading perspective is likely to be generally in line with the personal intentions of the reader, which leads to a mental frame that prefers a strategic engagement with belief-consistent information compared to belief-inconsistent information. This interpretation is in line with the assumptions of the Two-Step Model of Validation (Richter & Maier, 2017) and also with earlier results on perspective effects (e.g., Kaakinen & Hyönä, 2007). In contrast, a belief-inconsistent reading perspective might be at odds with readers’ personal intentions (i.e., to preferentially process belief-consistent information), which leads to a mental frame in which both types of information are strategically processed. Such an interpretation also is consistent with the results for the comprehension outcome that showed a reduced text-belief consistency effect in the belief-inconsistent reading perspective (we elaborate on this result later). Hence, an external reading perspective that calls for participants to read texts from a belief-inconsistent position (e.g., from the opposing side of a controversy) seems to be sufficient to increase strategic processing of belief-inconsistent information.

The pattern of findings in relation to the lookbacks is not in agreement with results from Bohn-Gettler and McCrudden (2018) who found no evidence that external reading perspectives modified the effects of prior beliefs on strategic processing, as indicated by think-alouds. The perspective instructions used in the present study included a description of a specific person who follows a specific reading goal (e.g., an alternative practitioner organizing an information event favoring alternative medicine). In contrast, Bohn-Gettler and McCrudden (2018) used general perspective instructions by asking participants to focus on pro or contra arguments. Our reading instruction might have increased readers’ possibility to put themselves into the person’s mind, which might have made it easier to process belief-inconsistent information. Such an interpretation coheres well with research on transportation and identification that have, for example, been found to influence the effect that narratives have on readers (e.g., Richter et al., 2014). The character description might have created a context that made it easier for participants to adopt a reading perspective that was not their own, compared to a relatively abstract general instruction to focus on either side of a controversy.

The present study also investigated the effects of prior beliefs and external reading perspectives on comprehension outcomes. We found that participants had better comprehension for belief-consistent information in the condition that provided participants with a belief-consistent reading perspective. We interpret from this finding that the belief-consistent reading instruction matched readers’ personal intentions based on their prior beliefs. This effect fits nicely to existing research on the text-belief consistency effect (e.g., Abendroth & Richter, 2020; Maier et al., 2018) and also coheres well with research on perspective effects (e.g., Kaakinen & Hyönä, 2007). In the belief-inconsistent reading perspective, however, the text-belief consistency effect was attenuated as indicated by no difference in comprehension between belief-consistent and belief-inconsistent information. This pattern of findings is fully in line with our expectations. Considering that personal and given intensions together form reading goals that influence processing and comprehension, the null finding can be explained quite well. Similar to the findings for lookback times, the mental frame resulting from given intentions (i.e., the belief-inconsistent reading perspective) and personal intentions (i.e., the natural tendency to favor belief-consistent information) might have resulted in a good comprehension of both types of information. Note that the balanced comprehension of belief-consistent and belief-inconsistent information resulted in an intermediate level of understanding for these two types of information, which is probably a result of a more balanced allocation of limited cognitive resources and processing capacities to both text types.

The results for the lookback times and the comprehension outcomes are in accordance with the results reported by McCrudden and Sparks (2014). In their study, external task instructions were also found to positively affect the use of belief-inconsistent information in a written task, whereas the use of belief-consistent information was not affected by the task instruction. In contrast, Bohn-Gettler and McCrudden found only a perspective effect but no belief effect on their measure of comprehension. Our study, however, assessed comprehension on the level of the situation model, whereas Bohn-Gettler and McCrudden used recall for explicit text information that might be considered to be an indicator of memory for text. Thus, in line with the assumptions of the Two-Step Model of Validation (Richter & Maier, 2017), the effects of external perspectives on comprehension outcomes seem to be modified by prior beliefs for the situation model but not necessarily for text memory.

A major advantage of the current study was the repeated-measurement design. All participants received both reading perspectives, that is, the belief-consistent and the belief-inconsistent perspective. This method allowed us to control for individual differences in eye movements, which enhanced the precision of the design. In addition, the external reading perspective was varied as a function of participants’ beliefs. Thus, the external reading perspective was individually varied for participants based on their responses to the binary belief choice measure in each experimental session. We also used a manipulation check to control for whether participants were actually following the external reading perspective that was given prior to reading. Results of this manipulation check revealed that slightly more than 20% of participants had not followed the external reading perspective. In these cases, reading was mostly driven by readers’ own argumentative stance on the issues (75.0%). This manipulation check allowed us to use the perspective actually taken by the participants as a predictor in the analyses. Nevertheless, the results of the manipulation check also shed light on the difficulties that the participating university students had when reading from a divergent reading perspective. It is therefore important for future studies to strengthen the instruction for the external belief-inconsistent reading perspective, or to pay more attention to the participants’ ability to put themselves in someone else’s position and read from a divergent (that is belief-inconsistent) point of view. As university students already had difficulties with this perspective-taking, this may seem even more difficult for younger target groups.

Some additional limitations of the study need to be discussed. First, using the actual instead of the experimentally induced perspective for some participants lead to a slightly imbalanced experimental design. However, mixed effects models are quite robust and can account for imbalanced designs (Baayen et al., 2008). Therefore, we are confident that the findings are not biased despite the imbalanced design. Nevertheless, future studies should include a screening before the main experiment to ensure that there is a balance of participants leaning toward the pro and the contra argumentative stance on each issue. Second, our study was based on only two socioscientific issues for the experimental text material and a relatively small and selected sample of participants (university students). Especially concerning the rather low sample size, the generalizability of the results should be taken with caution and further studies with larger samples, taken from more heterogeneous populations, and additional socioscientific topics and more texts should be conducted. Nevertheless, using a simulation-based power analyses approach, we performed sensitivity analyses for generalized linear mixed models and found that design and sample size made it possible to detect small effect sizes of .15 in 59% of the simulations and small effect sizes of .2 in 86% of the simulations. Thus, we are confident that the sample size was suitable to reliably detect even relatively small effects such as those found in our study. Third, the present study addressed the influence of validation processes based on readers’ prior beliefs as one type of first-hand evaluation strategy (Barzilai et al., 2020). Secondhand evaluation strategies such as sourcing were not focused. Future studies, for example, integrating source information in belief-relevant texts, might shed light on the interplay of different first- and secondhand evaluation strategies. Fourth, the socioscientific arguments on the controversy about gender roles consisted of arguments claiming that gender roles are genetically inborn and arguments claiming that gender roles are socially learned and thus, they discussed gender difference research. The study population consisted of psychology university students and it can be assumed that gender was distributed among participants similarly to the general population of psychology students (with probably more women compared to men). However, as gender was not correctly recorded because of a technical error, we cannot make any definitive statements about gender distributions and we are also not able to investigate whether gender had an influence on the processing and comprehension of the gender-relevant topic. In addition, the reliability of the prior belief scale assessing participants’ agreement that gender roles have genetic causes was rather low in the present sample. Fifth, participants received the belief-consistent reading perspective in one and the belief-inconsistent reading perspective in the other experimental session based on the binary choice measure. The finer-grained belief measure was given to participants first to so that by answering the items from the belief scales, participants became more aware of their own beliefs and were subsequently better able to make the binary belief choices. Future research should clarify whether or not an assignment to different reading perspectives based on the measured beliefs with a finer-grained scale would results in different effects for processing and comprehension compared to the binary belief choice.

Conclusion

Socioscientific controversies span across numerous disciplines. Discourse on these scientific issues, particularly on the Internet, and the beliefs formed from reading this discourse can have a large impact on public opinion and policy. Although it would be desirable for readers to process and comprehend all credible and relevant information on socioscientific issues in a thorough manner, the findings of the present study suggest that both processing and comprehension are highly dependent on readers’ prior beliefs. In addition, the interplay of external reading perspectives and prior beliefs does not necessarily lead to the intended purpose from the perspective of the external reading instructions. However, our findings also indicate that a belief-inconsistent reading perspective especially promotes strategic processing of belief-inconsistent information and can thereby reduce the text-belief consistency effect. Further studies can use our findings as a starting point for examining the potential of reading perspectives to strengthen a more balanced processing and understanding of socioscientific controversies.

Acknowledgments

We thank Sandra Opoka and Saskia Kramer for their help in collecting and cleaning the eye tracking data.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data files and analysis scripts for the full analyses are available in the Open Science Framework and can additionally be provided by the authors upon request (https://osf.io/ag5vx/?view_only=4aa9919bbd5049b0b1254642254d03ef). The full experimental texts are available from the authors upon request.

Additional information

Funding

The research reported in this article was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, grant RI 1100/5-3).

Notes

^1. Initially, ten items were also provided for the gender role belief scale. However, two items of th prior belief measure on gender roles were dropped (one for the pro gender role scale, one for the contra gender role scale), as they were negatively or not related to the other items of the scales.