The Influence of Comprehensibility on Interest and Comprehension

The Influence of Comprehensibility on Interest and Comprehension

Learning with texts is a very common form of institutionalized learning and is intended to produce a number of desirable outcomes. Above all, instructional texts should evoke comprehension and motivate their target audience to continue reading and to engage further with the topic of the text, e.g. foster interest in their subject among readers (Sadoski et al., 1993). Research on comprehensibility (readability) has often demonstrated that comprehensibility affects the readers' text comprehension. In order to get learners to further engage with a subject, their motivation also needs to be stimulated. The influence of comprehensibility on motivational variables, however, has hardly been investigated, even though these are key learning outcomes (Hidi, 2001). Recently, research has increasingly focused on the interplay between motivation and cognitive learning processes (Moreno & Mayer, 2007) and has shown correlations and sometimes experimental effects of comprehensibility on comprehension, interestingness, and interest. Yet most of the studies lack a grounding in theories of motivation, are merely correlational studies, and have not used validated instruments. Their supporting evidence is therefore disputable.

The aim of the present article is to substantiate the effects of comprehensibility, as a special form of fluency, on comprehension and interest theoretically and to jointly test the resulting hypotheses using validated instruments. Therefore, the following sections give an overview of the desirable qualities and outcomes of learning with texts. Each section briefly introduces the importance of the respective quality or outcome in the context of learning and instruction and gives a theoretical and empirical rationale for the influence of comprehensibility on the particular variables.

Comprehension and comprehensibility

Comprehending a text means building an appropriate mental representation of the content and the communicative function of the particular text. Readers can use this representation to draw conclusions that go beyond what they have read (Kintsch, 1988, 1998; Mayer, 2014; Schnotz, 2014). For this reason, comprehension is a well studied desirable consequence of reading texts. A large number of empirical studies show that text comprehension is influenced by both text and reader variables. Concerning the text, comprehension is influenced by the frequency of words, the complexity of syntax, and the cohesion of the text, among other factors (McNamara et al., 2014; Reed & Kershaw-Herrera, 2016). Concerning the reader, text comprehension is influenced by prior knowledge, the size of working memory, reading competence, self-efficacy, and interest in the subject, for instance (cf. Cromley & Azevedo, 2007; Kintsch, 1998; Taboada et al., 2009). These different variables of texts and readers have been incorporated into different concepts of comprehensibility and instruments for measuring comprehensibility.

Comprehensibility is defined as the ease with which a certain reader can conduct the processes needed to comprehend a certain text (Kintsch & Vipond, 1979). Concepts of comprehensibility and instruments to measure comprehensibility can be differentiated into two groups (Ballstaedt & Mandl, 1988; Friedrich & Heise, 2019; Kintsch & Vipond, 1979). The majority of these concepts and instruments treat comprehensibility as an inherent feature of the texts (Vajjala, 2021). According to this view, the ease of comprehension depends on certain text characteristics, such as word length, the frequency with which the words of a text are used in the language (word frequency), concreteness of the words (in how far the words evoke mental images during reading), the sentence length, syntactical complexity of the sentences, and how strongly and closely (successive) statements are linked (global and local cohesion; McNamara et al., 2014). This notion of comprehensibility is more strongly associated with the term “readability” and is sometimes referred to as “text difficulty” or “linguistic complexity” (Berendes et al., 2017).

The view of comprehensibility as an inherent property of texts has been criticized. As Kintsch and van Dijk (1978, p. 372) point out, “readability cannot be considered a property of texts alone, but one of the text-reader interaction.” According to this view, a text is more comprehensible to a certain reader in a certain situation, the less cognitive resources are required for the execution of text comprehension processes. This view is supported by everyday experience: A statistics book might be hard to comprehend for psychology students in their first year, but easy to comprehend in their senior year. The comprehensibility has changed due to a gain in knowledge while the text has remained the same. Within current conceptualizations, comprehensibility is therefore not considered as an inherent property of a text, but also depends on characteristics of the reader, such as prior knowledge, vocabulary, reading goals, interest, and working memory capacity (Friedrich, 2017). Accordingly, a text is comprehensible to a particular reader if she or he can easily assign meaning to the words of the text, can easily decode the syntax of the sentences, and can easily build an appropriate mental model of the text and its content (Friedrich, 2017; Kintsch & Vipond, 1979). This notion of the ease of processing is more strongly associated with the term “comprehensibility” in the literature (Sadoski et al., 1993) and has been called “relative text complexity” or the “interactionist view of comprehensibility” (Friedrich et al., 2021). In the following, readability formulas and computational linguistic methods for assessing text complexity as well as questionnaires for assessing the interactionist view of comprehensibility are presented.

A simple example of the text complexity approach that leaves out the structure and cohesion of the text is the LIX formula (Björnsson, 1968, as cited in Klare, 1984, p. 699). The LIX relates the text length in words, the number of sentences, and the number of words with more than six letters to each other. High LIX values indicate a high text difficulty and thereby a low ease of comprehension:

The LIX formula has been tested for Danish, English, Finnish, French, German, and Swedish texts and can be calculated easily (Klare, 1984). Like the Automated Readability Index ARI (Smith & Kincaid, 1970), the Fog Count (Air Force Manual, 1953, as cited in Klare, 1984), Flesch's Reading Ease formula (Flesch, 1948) or the Flesch-Kincaid grade level (Kincaid et al., 1975), the LIX formula rates the comprehensibility of texts on the basis of the surface characteristics of word length and sentence length only (Klare, 1984; McNamara et al., 2014). This concordance between the LIX and Reading Ease, for example, is also expressed in a high correlation of r = –.93 (cp. Friedrich & Heise, 2017). Readability formulas are often used, especially in the Anglo-American world, but have also been widely criticized (e.g. Crossley et al., 2017). On the one hand, readability formulas are economical and predict comprehension reasonably well. On the other hand, the various readability formulas have been developed without reference to theories of comprehension. They capture comprehensibility as a pure text feature and omit features of the situation and the reader. In addition, they focus primarily on surface features of texts and disregard deep structures such as the local and global cohesion.

Since the early 2000s, a number of computer based methods have been developed to assess comprehensibility (Benjamin, 2012; Collins-Thompson, 2014; Vajjala, 2021). Computational linguistic methods typically capture a variety of features at different levels, for example, how often the words occur in a language, how many nominal phrases the sentences contain, how far the predicate is separated from its subject, how much the nouns of successive sentences overlap, how many connectors the text contains, etc. Using, for example, a large number of texts, expert ratings of the required reading level for each of these texts, and machine learning procedures, the programs are then trained to determine an optimal combination of the features that can be used to predict the comprehensibility of texts (Collins-Thompson, 2014; Vajjala, 2021). Most of these methods can so far only be applied to English texts (e.g., Crossley et al., 2017; for German texts, see for example, Berendes et al., 2017). Computational linguistic methods thus provide authors with concrete information on which aspects of their texts need to be revised. Many of the newly developed methods still lack validation with the readers' actual comprehension, though (Vajjala, 2021). Readability formulas and computational linguistic tools are typically applied to assess text complexity. To measure the comprehensibility of texts in an interactionist approach, questionnaires are typically used.

Jucks (2001), for example, has developed a questionnaire based on the Hamburg concept of comprehensibility (Ballstaedt & Mandl, 1988; Langer et al., 1974). The concept incorporates the four dimensions simplicity, organization/ordering, concision, and additional stimulant. According to this concept, a text is more comprehensible to a certain reader, the more familiar the words of the text are to the reader, the easier it is for the reader to construct a mental image of the words, and the simpler the syntax is to her or him (simplicity); the more coherently and meaningfully the text is structured for her or him (organization/ordering); the more the text seems balanced between extreme brevity and excessive redundancy or digressions (concision); the more stimulating and varied the reader finds the text, e.g. due to rhetorical figures (additional stimulant). Studies show that simplicity is the most important of these characteristics for measuring comprehensibility (Friedrich, 2017; Langer et al., 1974). The concept has, however, been criticized for its purely empirical development without reference to (psychological) theories of comprehension (Ballstaedt & Mandl, 1988).

Friedrich (2017) reinterpreted Kintsch and Vipond’s (1979) concept of comprehensibility against the background of Kintsch's construction-integration model (1988, 1998). On this basis, he developed and validated a questionnaire that captures six characteristics of comprehensibility, namely word difficulty, sentence difficulty, effort needed for reorganizations, clarity of representation, variety of language use, and subjective comprehensibility. According to this, the following features make a text more comprehensible to a certain reader: the easier she or he can assign meaning to the words of the text (word difficulty); the easier she or he can decode the syntax of the sentences (sentence difficulty); the less effort she or he has to put into correcting misconceptions about the content or the further progression of the text (effort needed for reorganizations); the easier it is for her or him to build a mental model of the text (clarity of representation); and the more varied she or he finds the phrasing of the text (variety of language use). The questionnaire also assesses a global judgment of the text’s comprehensibility for the reader (subjective comprehensibility). These features make it easier for readers to relate the concepts in the text appropriately to one another and to the content of their long-term memory. Comprehensibility for a target group can be assessed using the mean values of the comprehensibility scales data for this particular group of readers (cf. Jucks, 2001). For very heterogeneous groups, however, the average comprehensibility may not be meaningful. With regard to both questionnaires, it is conceivable that the readers’ assessment of comprehensibility also depends, at least in part, on how well they think they understood the text. Yet, according to Reber and Greifeneder (2017), subjective judgments are the only way to assess the ease of processing.

The influence of comprehensibility on recall and comprehension has been demonstrated by numerous studies (Crossley et al., 2017; Kintsch & Vipond, 1979; Reed & Kershaw-Herrera, 2016). Studies show that texts that are too easy or too difficult more often lead to mind wandering. Yet, too difficult texts impair comprehension significantly more than too easy texts (Feng et al., 2013). Research thus primarily argues for a beneficial effect of the ease of the process of comprehension on the product comprehension. Furthermore, there are theoretical reasons to assume that comprehensibility also influences the readers' interest. The following chapter frames the concept of comprehensibility within fluency theory and derives relations to the evaluation of texts and their content.

Fluency

Considering comprehensibility in a larger context, it becomes apparent that it can be considered as a form of fluency (Graf et al., 2017). Fluency describes how fast and accurately a stimulus can be processed by a certain person in a certain situation. Fluency may apply to processes regarding perception and attribution of meaning. High fluency indicates that the individual is making progress and/or that the stimulus is familiar and therefore probably harmless; fluency theory therefore states that objects are judged more positively, the easier it is to process them (Reber et al., 2004). Among other things, fluency depends on the individual's familiarity with the material, the amount of redundancy within the material, and its complexity. Likewise the interactionist view on comprehensibility defines comprehensibility as the ease with which a certain reader can execute the necessary processes to comprehend a certain text in a certain situation.

Fluency theory is supported by empirical studies from various fields. Pharmaceutical drugs with names that can be pronounced fluently are judged more positively, perceived as safer and more likely to be bought than drugs with names that cannot be pronounced fluently (Dohle & Montoya, 2017). A piece of music is judged more positively when presented with a fluent name (Anglada-Tort et al., 2019). Experiments on implicit learning show that test subjects prefer letter strings that follow a grammar that the participants had implicitly learned earlier (Gordon & Holyoak, 1983). Statements that can be processed easily are more likely considered true (Unkelbach, 2007). In all these cases, the evaluation of an object depends on how easily the presentation of the object can be processed. There is a lack of studies regarding fluency and longer texts, though. One can assume that readers judge a text, and thereby possibly its topic, more positively, the easier it is for them to comprehend it. Fluency theory thus suggests influences of comprehensibility on comprehension and interest (Reber & Greifenender, 2017). This assumption is also supported by the four-phase model of interest development.

Interest

According to Renninger and Hidi (2016), interest has two meanings: first, interest is a psychological state of people who are engaged with a certain object; and second, interest is a motivational variable, more specifically the enduring cognitive and affective disposition of a person to engage with an object. In both cases the object can be a concrete item, an event, an activity, a person, an abstract topic or an idea. In their four-phase model of interest development, Hidi and Renninger (2006) distinguish the following four phases: triggered situational interest, maintained situational interest, emerging individual interest, and well-developed individual interest. The first two phases are summarized as less developed interest and their triggering depends more strongly on environmental factors, such as features of a text. The latter two phases are summarized as more developed interest. Developed interest as a trait is a well-known predictor for the state of interest. According to the four-phase model of interest development, interest may disappear especially during the first three phases of the model when a person encounters difficulties. In accordance with this, Renninger and Hidi (2016) assume that interest may decrease if it is difficult for a person to understand a text. Furthermore, they point out that for the development of interest, it is essential that readers make connections between themselves and the content and that this presupposes their comprehension of the content. It can thus be concluded from both the four-phase model of interest development and fluency theory that the ease with which a text can be processed by a person promotes the development of her or his interest. All forms of interest are associated with focused attention, increased engagement, more enduring exploration of a content, better recall, better comprehension, and deeper learning; therefore, interest is a highly desirable learning quality and outcome (Hidi, 2001). The model has been supported by various studies (for an overview see Renninger & Hidi, 2016). A recent qualitative analysis of pupils' behavior suggests that the comprehension of complex contents is beneficial to develop interest (Renninger et al., 2019). This notion is supported by a longitudinal study showing that learners not only perform well in fields they find interesting, but also develop interest in fields that they perform well in (Harackiewicz et al., 2008). Schulz von Thun, Göbel, and Tausch (1973) report effects of text difficulty on interest in the texts' subject for one of four text pairs they had examined. Schmerder and Tausch (1978) report effects of text difficulty on a scale they named “fun-interest”. The effect sizes equal small to medium effects.

Several studies have investigated the relation of different text characteristics to interestingness (see below). Interestingness is often understood as a feature of texts that evokes situational interest in the text or its subject. None of the cited studies, however, provides a clear definition of the construct. Since interestingness was assessed by asking participants to rate sentences or texts on a scale from “not interesting to me” to “very interesting to me” (cf. Sadoski et al., 1993), it seems reasonable to associate interestingness and situational interest closely (cf. Schiefele, 2009).

Wade et al. (1999) conducted a study using think-alouds to examine what makes texts interesting to readers. Their findings coincide with the assumptions on the relationship between comprehensibility and the four-phase model of interest development: Readers found the texts interesting when they found it easy to process them and to build a mental model of the texts' content. The readers rated small but plausible deviations from their previous beliefs as positive, but larger deviations and indications of a lack of credibility as negative. Accordingly, many of the features of interesting texts mentioned by Wade et al. (1999) correspond to features of fluency or comprehensibility respectively: the ease of assigning meaning to words, cohesion or coherence, the ease of building a mental model of the content, and the ease of linking new information with prior knowledge. In addition, literal speech and addressing the readers directly are rated positively (cf. Ballstaedt & Mandl, 1988; Mayer, 2014).

Schraw et al. (1995) constructed two questionnaires, one to assess perceived interest and one to assess readers' judgments of a text. Results showed that, of the features studied, comprehensibility had the highest correlation with interest, with r = .63. Several more studies show correlations of .64 < r < .96 between judgments of comprehensibility and interestingness (Sadoski et al., 1993; Sadoski et al., 2000). In two studies with 8^th to 10^th grade students, Mikk and Kukemelk (2010) found correlations of .48 < | r | < .75 between the readers’s mean interest in the texts and several text features, like word length, average frequency of the nouns within the language, percentage of concrete nouns, and the average sentence length in words. Surprisingly, the repetition rate of the nouns in the text showed a negative correlation of r = –.48 with the readers' interest in the text; the repetition rate of the nouns in the text is similar to the type-token ratio, a common indicator for the cohesion of a text (McNamara et al., 2014). As the authors point out, these results should be further tested in experimental studies.

In an experiment by Soemer and Schiefele (2019), participants read one of three versions of three different texts. The different versions of each text varied with regard to difficulty (easy, medium, or hard). After reading the texts, the subjects answered an interest questionnaire. Results showed a small to medium effect of r = –.21 of text difficulty on topic interest. Moreover, a recent experiment has shown that the use of jargon (i.e. infrequent words) reduces the ease of processing, which in turn affects the interest in the topic and the perceived understanding of the text mediated through the self-concept regarding the topic of the text (Shulman et al., 2020).

Overall, there is a large number of studies that reveal substantial correlations between comprehensibility and situational interest. Most studies were not primarily concerned with comprehensibility and were rather loosely grounded in theories of comprehensibility and motivation. Furthermore, there are only a few experiments on the influence of comprehensibility on interest (Schmerder & Tausch, 1978; Schulz von Thun, Göbel, & Tausch, 1973; Schulz von Thun, Weitzmann et al., 1973; Shulmann et al., 2020; Soemer & Schiefele, 2019) and most of the studies assessed comprehensibility and interest with instruments that were not, or barely, validated (Mikk & Kukemelk, 2010; Sadoski et al., 1993; Sadoski et al., 2000; Schmerder & Tausch, 1978; Schulz von Thun, Göbel, & Tausch, 1973; Schulz von Thun, Weitzmann et al., 1973). Many studies also deliberately used texts that differed greatly in terms of comprehensibility (Sadoski et al., 1993; Sadoski et al., 2000; Schmerder & Tausch, 1978; Schulz von Thun, Göbel, & Tausch, 1973; Schulz von Thun, Weitzmann et al., 1973; Soemer & Schiefele, 2019). This entails the risk of overestimating the effect of comprehensibility on the dependent variables (Fiedler, 2011). Overall, the empirical support for the influence of comprehensibility on interest is rather weak or contestable. The present article is intended to help close this research gap and links comprehensibility with interest within the broader context of fluency theory, readability, comprehension, and the four-phase model of interest development.

Hypotheses

Based on the preliminary studies presented, three hypotheses will be tested. Comprehensibility was defined as the ease with which a certain reader can conduct the processes needed to comprehend a certain text (see “comprehension and comprehensibility”). Although some researchers suggest that texts can be too simple, research so far overall indicates a linear positive influence of comprehensibility (i.e., the ease with which comprehension processes are carried out) on the product comprehension (i.e., an adequate and coherent mental representation of the content). Therefore, we tested a hypothesis regarding cognitive learning outcomes: Comprehensibility positively influences the readers' comprehension of the text (H_{comprehension}). This hypothesis was also tested to examine whether, for example, an improved comprehensibility leads to less investment of resources and thereby diminished comprehension (Reber & Greifeneder, 2017).

The section “interest” stated that comprehensibility decreases difficulties and should facilitate interestingness and help develop or maintain interest. We therefore tested the following hypotheses: Comprehensibility positively influences the readers' judgment of the text's interestingness (H_{interestingness}), and interest in the text's topic (H_interest). (The hypothesis on interestingness will be tested primarily to compare the results of the present study with the results of prior studies.)

Prior knowledge, self-efficacy, and interest prior to reading are well known predictors of comprehension and interest (Cromley & Azevedo, 2007; Kintsch, 1998; Schiefele & Schaffner, 2016; Taboada et al., 2009) and were therefore assessed as possible covariates.

Method

Materials

The topic chosen for the texts under study was a topic of statistics since, on the one hand, knowledge of statistics is important in many subjects and, on the other hand, it is perceived as difficult, abstract and not very interesting by learners (Zieffler et al., 2008). It would hence be particularly promising to promote comprehension and interest in this topic through increased fluency. Fifteen texts on measures of central tendency were examined. Eleven of these texts were taken from introductory statistics textbooks for psychologists; two texts were taken from introductory textbooks that explicitly address struggling students in their titles; one text was taken from a reference book for pupils; one text was taken from Wikipedia. The aim was to examine a range of authentic texts that students are likely to use in order to study statistics. All texts dealt with the calculation of the mean, the median, and the mode. Since not all texts conveyed the same core contents, all texts were shortened to make them more equal in content. Table 1 shows central properties of the texts.

Table 1 Central properties of the texts used

Text	Number of charactersa	Number of wordsb	Number of sentencesb	LIX valueb
aThese values were assessed using MS Word©. bThese values were assessed using Lenhard & Lenhard (2014–2017).
Text 01	9,495	1,289	48	59.9
Text 02	4,454	575	31	54.7
Text 03	5,210	688	40	51.3
Text 04	7,808	1,087	57	53.5
Text 05	2,446	323	21	52.2
Text 06	8,782	1,229	75	50.1
Text 07	6,447	991	47	44.6
Text 08	3,547	495	25	45.4
Text 09	11,152	1,561	80	52.3
Text 10	6,418	862	47	53.1
Text 11	4,235	565	34	52.7
Text 12	5,893	850	54	44.0
Text 13	10,476	1,443	88	46.7
Text 14	4,000	564	26	53.2
Text 15	13,313	1,899	112	44.1

Table 1 Central properties of the texts used

View as image HTML

Data Analysis

The hypotheses were tested using ANOVAs with planned comparisons. In line with the hypotheses, only focused, linear effects were tested. The assumption of linearity was tested using scatter plots with locally weighted polynomial curves (see Figure 1). ANOVAs with planned comparisons, in contrast to ANOVAs without planned comparisons, have the advantage that they can be used to test focused hypotheses in designs with more than two groups without the use of post tests, therefore reducing the number of required significance tests (Field et al., 2012; Rosnow et al., 2000). To conduct the ANOVAs with planned comparisons, a contrast coefficient is assigned to each group, resulting in a set of contrast coefficients (one coefficient for each group). Each of these coefficients reflects the expectation of how the corresponding group will perform compared to the other groups with regard to the dependent variables (Rosnow et al., 2000). In the present study, the contrast coefficients are based on the texts' (mean) comprehensibility. For mathematical reasons the contrast coefficients must add up to zero. Therefore, the (mean) comprehensibility was calculated for all texts; these means were then subtracted from the grand mean and multiplied with a constant to define contrast coefficients that reflect the expected comprehensibility for every text in relation to the other texts. The resulting contrast coefficients are shown in Table 3 and correlate perfectly with the texts' (mean) comprehensibility. Since comprehensibility was assessed using three different instruments (cf. section Comprehension and comprehensibility and section Instruments), all hypotheses were tested three times, i.e. with three different sets of contrast coefficients. The three sets of contrast coefficients were determined based on the texts' LIX values, the means of the subjective comprehensibility scale by Friedrich (2017), and the means of the scale simplicity by Jucks (2001). One-tailed t-tests and the two effect sizes r_{effect_size} and r_contrast are reported to evaluate how well the empirical data correspond to the expectations. r_{effect_size} is the correlation between the contrast coefficients of the groups to which the individuals belong and the particular dependent variable (Field et al., 2012; Rosnow et al., 2000). r_contrast is the partial correlation between the contrast coefficients of the groups and the respective dependent variable, after the elimination of all variation between the groups which cannot be attributed to the contrasts (Field et al., 2012; Rosnow et al., 2000). r_contrast is needed to conduct the power analysis. The ANOVAs with planned comparisons thus test the linear relationship between the (mean) comprehensibility of the texts and the respective dependent variable. The significance level α was set to 5%. We aimed to survey about twenty persons per text, i.e. ten persons per text and comprehensibility questionnaire. An a priori power analysis revealed that with N = 302, small effects of r_contrast = .14 could be detected with a power of 1 – β = .80.

Table 3 Means and standard deviations of the predictor variables and contrasts for each text

Text	n	Interest prior to reading	Self-efficacy prior to reading	Prior knowledge	Subjective comprehensibility	Contrastsub. comp.	Simplicity	Contrastsimplicity	LIX value	ContrastLIX
Note: Abbreviations used for the variables in the columns: sub. comp. = subjective comprehensibility.
Text 01	19	1.82 (0.78)	0.74 (0.53)	1.13 (1.36)	1.33 (0.85)	–177	2.20 (0.65)	–136	59.9	203
Text 02	22	1.61 (0.69)	1.00 (0.88)	1.36 (1.71)	1.40 (0.75)	–162	2.24 (0.46)	–122	54.7	90
Text 03	23	1.77 (0.78)	1.17 (0.83)	1.54 (1.67)	1.56 (1.16)	–128	2.45 (0.93)	–40	51.3	17
Text 04	21	1.68 (0.55)	0.69 (0.77)	1.02 (1.42)	1.89 (1.09)	–57	2.09 (0.58)	–176	53.5	65
Text 05	19	1.94 (0.61)	1.18 (0.97)	1.34 (1.53)	1.96 (1.30)	–42	2.45 (1.00)	–42	52.2	36
Text 06	22	1.75 (0.63)	0.83 (0.59)	1.20 (1.03)	2.07 (0.78)	–19	2.78 (0.60)	82	50.1	–9
Text 07	20	1.61 (0.61)	1.20 (1.00)	1.00 (1.25)	2.23 (0.77)	15	2.72 (0.99)	59	44.6	–128
Text 08	17	1.54 (0.56)	1.45 (0.62)	1.47 (1.60)	2.24 (1.03)	19	2.46 (0.75)	–37	45.4	–111
Text 09	25	1.76 (0.65)	0.85 (0.80)	1.30 (1.42)	2.25 (1.21)	20	2.58 (0.97)	6	52.3	39
Text 10	17	1.57 (0.58)	1.10 (0.74)	1.68 (1.66)	2.30 (0.79)	30	2.81 (0.66)	95	53.1	56
Text 11	20	1.82 (0.79)	0.80 (0.61)	1.45 (1.39)	2.30 (1.14)	30	2.67 (0.71)	40	52.7	47
Text 12	22	1.63 (0.70)	1.29 (0.84)	1.73 (1.87)	2.58 (0.63)	90	3.05 (0.93)	182	44.0	–141
Text 13	17	1.84 (0.64)	1.27 (1.03)	2.06 (2.15)	2.63 (0.59)	101	2.38 (0.99)	–70	46.7	–83
Text 14	20	1.89 (0.54)	1.10 (0.90)	1.70 (1.31)	2.73 (0.88)	122	2.78 (0.85)	82	53.2	58
Text 15	18	1.73 (0.48)	1.40 (0.89)	1.89 (1.82)	2.90 (1.12)	158	2.77 (0.76)	77	44.1	–139

Table 3 Means and standard deviations of the predictor variables and contrasts for each text

View as image HTML

Results

Descriptive statistics

Means, standard deviations, and internal consistencies for the scales and their intercorrelations are depicted in Table 2. With internal consistencies of Cronbach's α between .71 and .91, the scales proved to be acceptable to excellent. Figure 1 shows scatter plots in which the subjective comprehensibility is plotted on the x-axis in each case and the dependent variables on the y-axis. The graphs also contain locally weighted polynomial curves and a 95% confidence region for the regression fits to visually test the assumption of linear relationships between comprehensibility and the dependent variables. Scatter plots with simplicity on the x-axis and corresponding bar graphs look alike. The graphs are in line with the assumption of linear relationships.

Table 2 Psychometric properties and intercorrelations between the variables

Measure (number of items)	M (SD)	N	>α	01	02	03	04	05	06	07	08
Note: The lower diagonal contains the correlations of the variables. The upper diagonal contains the standard errors for these correlations according to Bonnet amd Wright (2000). Scale range: a 0 = not at all to 4 = very. b 0 = not confident at all to 3 = very confident. c The participants could achieve between 0 and 6 points in these tests. d 0 = I disagree to 4 = I agree. e semantic differential 0 = Agreed to 4 = Agreed. *p >< .05, one-tailed.
01. Interest prior to reading (4)a	1.73 (0.65)	288	.84		.06	.06	.08	.08	.05	.05	.06
02. Self efficacy beliefs prior to reading (6)b	1.06 (0.83)	291	.91	.17*		.03	.06	.07	.05	.05	.05
03. Prior knowledge (6)c	1.45 (1.55)	302	.71	.08	.72*		.07	.08	.05	.05	.04
04. Subjective comprehensibility (3)d	2.14 (1.02)	147	.90	.21*	.51*	.45*		–	.06	.06	.06
05. Simplicity (6)e	2.56 (0.81)	152	.83	.07	.33*	.23*	–		.07	.07	.07
06. Interestingness (1)d	1.74 (1.08)	302	–	.30*	.28*	.24*	.52*	.41*		.02	.05
07. Interest after reading (3)a	1.30 (0.81)	302	.89	.38*	.35*	.27*	.54*	.37*	.76*		.05
08. Comprehension (6)c	3.30 (2.09)	302	.80	.10*	.44*	.54*	.54*	.32*	.31*	.34*

Table 2 Psychometric properties and intercorrelations between the variables

View as image HTML

Table 3 shows the means and, where applicable, standard deviations for all predictor variables as well as the contrast coefficients for the comprehensibility measures, depending on the text. In accordance with expectations, the mean subjective comprehensibility and the mean simplicity score (which were obtained from disjoint subsamples) correlate with r = .68; the mean subjective comprehensibility score correlates with r = –.67 with the objective LIX and the mean simplicity correlates with r = –.53 with the LIX. Table 4 shows the means and standard deviations for all dependent variables as well as the contrast coefficients for the comprehensibility measures, depending on the text.

Table 4 Means and standard deviations of the dependent variables for each text

Text	n	Contrastsub. comp.	Contrastsimplicity	ContrastLIX	Interestingness	Interest after reading	Comprehension
Note: Abbreviations used for the variables in the columns: sub. comp. = subjective comprehensibility.
Text 01	19	–177	–136	203	1.21 (1.03)	0.96 (0.87)	2.63 (2.09)
Text 02	22	–162	–122	90	1.41 (1.18)	0.91 (0.77)	2.73 (1.98)
Text 03	23	–128	–40	17	1.70 (1.15)	1.24 (1.03)	2.57 (2.43)
Text 04	21	–57	–176	65	1.57 (1.17)	1.24 (0.68)	2.62 (2.04)
Text 05	19	–42	–42	36	2.16 (1.43)	1.57 (1.10)	2.53 (1.74)
Text 06	22	–19	82	–9	1.73 (0.88)	1.34 (0.64)	2.41 (2.15)
Text 07	20	15	59	–128	1.75 (1.02)	1.41 (0.69)	3.95 (2.06)
Text 08	17	19	–37	–111	1.35 (1.00)	1.16 (0.84)	4.27 (2.29)
Text 09	25	20	6	39	1.72 (1.14)	1.27 (0.76)	3.56 (2.04)
Text 10	17	30	95	56	1.88 (0.99)	1.37 (0.67)	4.12 (1.58)
Text 11	20	30	40	47	1.80 (0.89)	1.31 (0.72)	4.13 (1.89)
Text 12	22	90	182	–141	2.05 (0.79)	1.51 (0.80)	3.77 (1.95)
Text 13	17	101	–70	–83	1.59 (0.87)	1.13 (0.59)	3.59 (2.06)
Text 14	20	122	82	58	2.15 (1.14)	1.55 (0.87)	3.43 (2.15)
Text 15	18	158	77	–139	2.00 (1.14)	1.51 (0.83)	3.72 (1.87)

Table 4 Means and standard deviations of the dependent variables for each text

View as image HTML

Inferential statistics

No significant correlations were found between the experimental condition and course of study (Cramer's V = .21, χ ² = 13.56, df = 14, p = .48), age (d_xy = –.02, χ ² = 75.56, df = 70, p = .30), gender (Cramer's V = .24, χ ² = 17.15, df = 14, p = .25), or level of education (d_xy = –.03, χ ² = 46.07, df = 56, p = .83).

The experimental conditions did not significantly differ with regard to the interest prior to reading, F(14, 273) = 0.62, p = .85, η² = .03, self-efficacy prior to reading, F(14, 276) = 1.67, p = .06, η² = .08, or prior knowledge, F(14, 287) = 0.76, p = .71, η² = .04. The standard deviations of the comprehensibility scales did not differ statistically significantly between the groups and corresponded to those of previous studies (Friedrich, 2017). This suggests that the experimental groups were homogeneously composed in each case.

Subsequently, contrast coefficients for the various texts were generated which reflect the comprehensibility of each text in relation to the other texts. The texts' mean subjective comprehensibility ratings, for example, were 1.33, 1.40, 1.56, 1.89, 1.96, 2.07, 2.23, 2.24, 2.25, 2.30, 2.30, 2.58, 2.63, 2.73, and 2.90. These means were then subtracted from the grand mean (2.16) and multiplied with a constant to define contrast coefficients that reflect the comprehensibility for every text in relation to the other texts (see section “data analysis”). This procedure yielded the contrast coefficients –177, –162, –128, –57, –42, –19, 15, 19, 20, 30, 30, 90, 101, 122, and 158. The same approach was used to generate contrast coefficients based on the scale simplicity (cp. section “data analysis”) and the LIX values (see Table 3). Table 5 shows the results of the ANOVAs with planned comparisons with all sets of contrast coefficients. The results show small to medium effects in the expected direction. All tests were statistically significant, except for the ANOVA with planned comparisons with coefficients based on the LIX values and interestingness as the dependent variable. All hypotheses are therefore regarded as confirmed.

Table 5 Results of the ANOVAs with planned comparisons; contrasts based on mean subjective comprehensibility, mean simplicity, and the texts' LIX values

Scale	Predictor	df	temp	p	rcontrast	reffect_size
Note: The column Predictor indicates whether the contrasts of the corresponding row are based on the subjective comprehensibility scale of Friedrich (2017), the simplicity scale of Jucks (2001) or the texts' LIX values (Björnsson, 1968, as cited in Klare, 1984).
Interest	Friedrich	287	2.81	.003	.16	.67
	Jucks	287	3.07	.001	.18	.73
	LIX	287	–1.85	.033	–.11	–.45
Interestingness	Friedrich	287	2.76	.003	.16	.62
	Jucks	287	3.11	.001	.18	.69
	LIX	287	–1.37	.086	–.08	–.32
Comprehension	Friedrich	287	3.67	>< .001	.21	.66
	Jucks	287	2.91	.002	.17	.52
	LIX	287	–2.88	.002	–.17	–.53

Table 5 Results of the ANOVAs with planned comparisons; contrasts based on mean subjective comprehensibility, mean simplicity, and the texts' LIX values

View as image HTML

Discussion

The present paper incorporated several rather distinct fields of research to examine the influence of comprehensibility as a special form of fluency on comprehension and interest. Comprehensibility is understood as the ease with which a certain reader can conduct the processes needed to comprehend a certain text in a certain situation (Friedrich, 2017; Kintsch & van Dijk, 1978; Kintsch & Vipond, 1979). Based on fluency theory and the four-phase model of interest development, it is assumed that the easier it is for readers to comprehend a particular text, the more interest and interestingness are promoted. These hypotheses were tested with authentic texts, showing substantial effects of comprehensibility on the interestingness, interest after reading, and comprehension. The effect sizes equaled small to medium effects on the individual level (see r_{effect_size} in Table 5) and medium to large effects on the group level (see r_contrast in Table 5). The results of the present study are consistent with prior studies (Crossley et al., 2017; Mikk & Kukemelk, 2010; Sadoski et al., 1993; Sadoski et al., 2000; Schmerder & Tausch, 1978; Schraw et al., 1995; Schulz von Thun, Göbel, & Tausch, 1973; Shulman et al., 2020; Soemer & Schiefele, 2019). The study thus confirms the assumption of Reber and Greifeneder (2017) that teaching materials that are easy to process increase interest. The strengths of the present study lie in the facts that authentic educational texts were used, that a large number of texts comparable in content were examined, and that both motivational and cognitive variables were assessed using theoretically and empirically substantiated instruments. The results thus provide a deeper insight into the interplay of cognition and motivation and demonstrate the importance of fluency, or more specifically, comprehensibility for the design of learning materials. There was no trade-off between the different variables. What helped to foster comprehension also helped to foster interest, and vice versa.

The results of the current study, however, are subject to a number of limitations. Only students of psychology, social or educational sciences participated, and only one specific topic was investigated. Further research should test these relationships within other populations. Since textbooks are a widely used learning resource, the relationships should be tested in particular among pupil populations.

Moreover, this study, like most other studies that have investigated the effects of comprehensibility on interest, exclusively examined expository texts (Mikk & Kukemelk, 2010; Sadoski et al., 1993; Schmerder & Tausch, 1978; Schulz von Thun, Göbel, & Tausch, 1973; Schulz von Thun, Weitzmann, et al., 1973; Soemer & Schiefele, 2019; Wade et al., 1999). Only two studies also incorporated narratives (Sadoski et al., 2000; Schraw et al., 1995). Even though the cited theories should be applicable to all kinds of texts, the empirical support is currently based on text samples that are not representative for all genres. These relationships should therefore also be examined with other text types, e.g. operating manuals and spoken texts.

Another limitation of the present study lies in the use of authentic texts. Since the texts were assigned randomly, the current study constitutes a randomized experiment (Shadish et al., 2002). Yet since authentic texts were used, the texts differed not only in text complexity, but also in text length, parts of the content, didactic approach, etc. On the other hand, the use of authentic texts increases the external validity of the study and the results illustrate the importance of choosing appropriate learning texts. To investigate the influence of comprehensibility in isolation, though, future studies should manipulate the complexity of the texts by itself.

A further limitation of this study as well as previous studies lies in common-rater effects: Since all variables were measured using questionnaires that were answered by the same individuals within the same session, it is possible that the size of the correlations is overestimated. There is a possibility that participants wanted to appear consistent or answered the different questionnaires according to their implicit theories (Podsakoff et al., 2003). Future research should therefore use other types of instruments than questionnaires or assess interest at a later time, for example, one week after reading the text; alternatively, one might assess the comprehensibility of different texts with one group of the target population while assessing the dependent variables with another group of people to rule out common-rater effects. Nevertheless, since the LIX formula could also significantly predict all dependent variables except for interestingness, this is a strong indicator that the effects reported here correspond to actual effects and cannot be attributed solely to method biases. This result is also in line with the results of Mikk and Kukemelk (2010), who examined correlations between text features and the readers' motivation.

Furthermore, the fact that all three instruments showed substantial correlations with all dependent variables raises the question of which instrument(s) are best suited for the assessment of comprehensibility for which purposes. All three instruments could explain considerable amounts of variance of all dependent variables. These results support the validity of all three instruments. The fact that of the instruments used to measure text comprehensibility the LIX formula shows the lowest correlations with the dependent variables indicates that readability formulas are economical and useful, but also limited by their simplicity and that modern approaches to measuring text comprehensibility are more appropriate. It would be interesting to examine in future studies how well the much more elaborate computational linguistic procedures can predict comprehension and interest and which features of the texts prove to be most important. Finally, it would be interesting, how computational linguistic procedures perform in comparison to questionnaires assessing comprehensibility.

So far, there has been a lack of studies that compare the prognostic validity of different instruments for assessing text comprehensibility (Friedrich & Heise, 2017). However, the results indicate that the ease of assigning meaning to the words and of decoding the syntax of the sentences are the most substantial features for comprehensibility. Future research should compare the prognostic validity of different instruments for the assessment of comprehensibility. Hierarchical linear regression models could help clarify how much additional variance instruments based on the interactionist view of comprehensibility can explain compared to instruments that treat comprehensibility as a feature of the text.

Further research is also needed on the interaction of cognitive and motivational learning processes. For example, it is still unclear how long the effects of comprehensibility on interest last. It is also unclear how the comprehensibility of an entire textbook of several hundred pages, being read over a time period of perhaps several weeks or more, affects the development of readers' interest.

If these findings are confirmed in further studies, it would warrant paying special attention to the comprehensibility of textbooks in practice. Textbooks, and particularly schoolbooks, reach a large number of people. The effort by individual authors to improve learning materials would be multiplied by facilitating learning, improving comprehension, and increasing the motivation of large groups of learners. Moreover, more research is desirable regarding the extent to which the results of these and related studies can be applied to other forms of instruction, e.g. oral instructions instead of written texts and other motivational variables. Fluency theory and the four-phase model of interest development seem to be a sound basis for the further investigation of these relations.

At the same time, the findings also have strong practical implications for the design of educational texts and textbooks (cp. Reber & Greifeneder, 2017). The results indicate that comprehensibility is a key variable to foster comprehension and interest. The large effects of word difficulty and sentence difficulty (see table in ESM 1) on all dependent variables suggest paying especially close attention to using syntax and words that are familiar to the readers and easy for them to decode. Cohesion is a central variable regarding comprehension and comprehensibility (McNamara et al., 2014). The present study, however, suggests that the words and the syntax are more basic not just to foster comprehension, but also to increase interest. This is a reasonable result, because no coherent mental model can be built without assigning meaning to the words of a text and decoding the syntax of its sentences. This notion is supported by the large amount of variance explained by the LIX formula, which only includes the texts’ surface characteristics of word length and sentence length.

However, the considerable variance between the authentic texts with regard to comprehensibility shows that the need for highly comprehensible educational texts is too often ignored. Writers possibly need to be better trained or guided to write educational texts.

In summary, this study substantiated hypotheses on the influence of fluency, more precisely comprehensibility, on comprehension, interestingness, and interest. The study thereby highlights that comprehensibility is a changeable key variable in education to influence not only comprehension, but also interest. When an educational text is comprehensible to learners, it not only leads to greater comprehension but also fosters the conditions for learners to feel comfortable with the subject matter of the text and to engage with it in a more lasting and meaningful way.

Electronic supplementary material