Specificity of epistemic beliefs across school subject domains

Abstract

Based on the theory of integrated domains in epistemology, the question of domain specificity of epistemic beliefs was investigated from a comprehensive perspective. We examined intraindividual differences in epistemic beliefs about the natural, mathematical, social, and linguistic sciences that represented almost the entire spectrum of subjects taught in secondary schools. The sample comprised 196 pre-service teachers who four times filled in an epistemic belief questionnaire, which could capture four dimensions of epistemic beliefs in four school subject domains with sufficient reliability. Regarding the core question of domain specificity of epistemic beliefs, all four domains were found to largely differ from each other on all four epistemic dimensions. These findings were supported by correlational analyses and structural equation modelling. In total, they provide strong evidence for the domain specificity of epistemic beliefs about school subject domains, which raises the demand for a classification system to specify major fields of personal epistemology.

HIGHLIGHTS

• We examined pre-service teachers’ epistemic beliefs about school subject domains.

• We compared beliefs about the natural, mathematical, social, and linguistic sciences.

• A questionnaire with four dimensions reliably measured beliefs across domains.

• All four domains largely differed from each other in all four dimensions.

• Results provide strong evidence for the domain-specificity of epistemic beliefs.

Keywords:

• Epistemic beliefs
• personal epistemology
• domain specificity/domain generality
• school subjects
• pre-service teachers

Introduction

Epistemic beliefs are personal or implicit views about the source, certainty, simplicity, and justification of knowledge (Hofer, 2004). Their research is grounded in epistemology—a branch of theoretical philosophy dealing with questions about principles and methods of knowledge and knowing. Epistemic beliefs are active when persons are dealing with information from their environment and trying to construct personal knowledge (Schommer, 1990). Epistemic beliefs are said to act like a filter that restricts the perception and processing of information (Hofer, 2004), or like a metacognitive tool that regulates and controls mental processes (Barzilai & Zohar, 2014). Empirical research has shown that epistemic beliefs were found to be linked to motivation (Bråten & Strømsø, 2004; Winberg et al., 2019), critical thinking (Feinkohl et al., 2016; Rott, 2021), reasoning (Kremer et al., 2014; Krist et al., 2019; Yang et al., 2019), problem solving (Lindfors et al., 2019; Shin & Song, 2016), strategy use (Madjar et al., 2017; Urhahne, 2006), self-regulated learning (Muis & Franco, 2009; Pieschl et al., 2008), and academic achievement (Greene et al., 2018; Kampa et al., 2016).

For a long time, research was guided by the question of whether epistemic beliefs should be considered as domain-general or domain-specific. While in the early days of research, a domain-general perspective on personal epistemology was prevalent (Perry, 1970; Schommer, 1990), this view has changed considerably over time. Muis et al. (2006), in a review of 19 empirical studies, concluded that epistemic beliefs should be regarded as both domain-general and domain-specific. Evidence of domain specificity was not only found when students from different disciplines indicated their beliefs about knowledge and knowing but also when individuals specified their epistemic views about different academic disciplines. Consistently, all of the studies reviewed by Muis et al. (2006) provided evidence for some degree of domain specificity in personal epistemology.

Muis et al. (2006) hypothesised in their theory of integrated domains in epistemology (TIDE) that domain-specific epistemic beliefs begin to develop when learners enter the educational system. Through experience with different subjects, epistemic beliefs are getting more specific and evolve along certain dimensions (Bendixen & Rule, 2004; King et al., 1994). After some belief-shaping years in school, students learn about the similarities and differences between school subjects, so that by the end of schooling, distinctive epistemic views should be found across school subject domains.

However, previous research has largely focussed on the differences in epistemic beliefs across academic disciplines represented at universities (Donald, 1990; Paulsen & Wells, 1998; Rosman et al., 2017, 2020; Schommer-Aikins et al., 2003). According to the theoretical assumptions made by Muis et al. (2006), we intended to examine the epistemic beliefs about different areas of school subjects in more detail. The guiding question of our research was: Can we find evidence for the specificity of epistemic beliefs about school subject domains?

Theoretical background

In the following, we will elaborate on the different dimensions of epistemic beliefs that underlie our research approach. We report which domains have been compared in previous studies and point out that research has been limited by a lack of systematicity in the selection of domains. To categorise domains in the school setting, we follow Biglan’s (1973) classification, more specifically the distinction between so-called ‘hard’ and ‘soft’ subjects.

Multidimensional structure of epistemic beliefs

In a synthesis of various epistemological theories, Hofer and Pintrich (1997) identified four dimensions that, in their view, represent the core structure of personal epistemology. Epistemic beliefs can be differentiated at a superordinate level into two general areas: the nature of knowing and the nature of knowledge, each of which comprises two dimensions. The nature of knowing is composed of the dimensions source of knowledge and justification for knowing. Some researchers, however, combined these two dimensions in later works (Bråten et al., 2019; Cheng et al., 2021). To map the nature of knowledge, Conley et al. (2004) defined the dimensions certainty of knowledge and development of knowledge in accordance with the stability of knowledge in the multidimensional approach by Schommer (1998; Schommer-Aikins et al., 2000).

The source of knowledge refers to the belief of whether school knowledge is handed down to individuals by omniscient authorities, such as teachers, scientists, and experts, or whether individuals themselves can become active constructors of meaning (King & Kitchener, 1994). Individuals often view their own role in gaining knowledge in a subject as passive. They rely on sources, such as teachers, books, or family members to provide them with the knowledge they need, rather than constructing it on their own (Elder, 2002). Individuals take a more nuanced position when knowing is based on intentional activity that involves goal-directed investigation, exploration, and questioning (Weinstock et al., 2020).

The justification for knowing stands for beliefs dealing with evidence and evaluating knowledge claims. A naïve view that justification is based on appeals to authority or based on what feels right is contrasted with an advanced view that justification is based on quality standards of inquiry and evaluation of sources (Greene et al., 2018; Hofer, 2004). In domains, such as natural sciences and mathematics, scientific experiments and mathematical proofs can often be used to justify knowledge. In the humanities and social sciences, knowledge is not considered provable and individuals have to rely on multiple sources of knowledge by making use of the expertise of different authorities (Marra & Palmer, 2008).

The certainty of knowledge is reflected in the belief that theories and results are considered certain and absolutely true. In case of doubt, experts are able to provide the one correct answer to a problem (Trautwein & Lüdtke, 2007). This dualistic view of labelling knowledge as either right or wrong contrasts with a relativistic view. Knowledge is considered as a human construction that is subject to uncertainty and may prove to be wrong (Perry, 1970). For complex problems, there may be more than one correct answer and the context must be considered to assess the appropriateness of the proposed solution.

The development of knowledge represents the temporal perspective, individuals’ beliefs about the scientific progress in a field of research. Knowledge can be viewed as tentative and changeable or as fixed and absolute (Schommer, 1990). Individuals may differ in their views of whether new ideas and discoveries can change existing concepts and theories, and whether long-held truths can be modified to better suit evidence.

Domain specificity of epistemic beliefs

Once epistemic beliefs had been established as a multidimensional construct (Schommer, 1990), researchers quickly raised questions about the generality or specificity of personal epistemologies (Hofer, 2000; Schommer & Walker, 1995). With indications that individuals’ self-concept (Shavelson & Marsh, 1986) or experts’ knowledge (Chi et al., 1981) are strongly domain-specific, assumptions were made that epistemic beliefs might also vary across domains. However, characterising the structure of epistemic beliefs is not the only intention to examine their specificity. Domain-specific epistemic beliefs are essential prerequisites to understand the knowledge and practices in academic areas, such as science (Elby et al., 2016), mathematics (Depaepe et al., 2016), or history (Stoel et al., 2022; VanSledright & Maggioni, 2016). To promote the development of domain-specific epistemic beliefs in secondary schools and tertiary institutions can thus be understood as an important objective of education.

The philosopher of science Kuhn (1962) advocated the idea that an academic domain is characterised by a unified paradigm. It refers to a consensus about what questions are appropriate for research and what methods can be used to address them. Biglan (1973) found, by applying multidimensional scaling, that members of a university ranked different disciplines according to their dominant paradigm. On a dimension from hard to soft, natural sciences, such as physics, chemistry, and biology were considered hard disciplines, closely followed by mathematical sciences, such as mathematics and computer science. Social sciences, such as psychology, education science, economics, sociology, and political science counted more towards the soft disciplines. At the end of the spectrum and belonging to the soft disciplines were the linguistic sciences, such as English, Russian, and German linguistics. Interestingly, recent research could provide impressive support for Biglan’s highly-cited classification of disciplines (Simpson, 2017; Stoecker, 1993). A correspondence analysis by Simpson (2017) yielded the same ranking of mentioned disciplines on a spectrum from hard to soft disciplines. Moreover, Rosman et al. (2017) perceived this first dimension in Biglan’s three-dimensional classification scheme compared to the other dimensions ‘pure-applied’ and ‘life-nonlife’ as particularly appropriate for exploring epistemic beliefs.

Numerous comparisons have been made of how epistemic beliefs differ across domains (Muis et al., 2006). Looking more closely at those studies, it is striking that often a hard discipline has been contrasted with a soft discipline (Buehl et al., 2002; Hofer, 2000; Schommer-Aikins et al., 2003; Stodolsky et al., 1991). This trend can also be detected in more recent studies when science was compared with history (Iordanou et al., 2019), biology with history (Barzilai & Weinstock, 2015; Thomm et al., 2017), mathematics with education (Löfström & Pursiainen, 2015), mathematics with psychology (Muis et al., 2016), biology with psychology (Rosman et al., 2020), or computer science with psychology (Rosman et al., 2017). Less frequently was tested whether personal epistemologies differ between two similar disciplines, such as physics and biology (Lee & Tsai, 2012), or physics and mathematics (Liu & Liu, 2011). Researchers assumed that people conceive scientific knowledge in the hard disciplines as objective truths and having an absolute character, whereas knowledge in the soft disciplines should be based more on subjective opinions, supporting relativistic beliefs of scientific knowledge as tentative and evolving (e.g. Rosman et al., 2017).

These comparative studies also show that there is not enough regularity in selecting the domains. While there are some studies that explicitly follow Biglan’s classification of academic disciplines (Jehng et al., 1993; Karimi, 2014; Paulsen & Wells, 1998; Rosman et al., 2017, 2020; Schommer-Aikins et al., 2003), there is still a long way to go towards domain-specific epistemic beliefs being systematically understood. Comparative studies that make use of a unified overarching framework, such as TIDE (Muis et al., 2016) would be desirable, so that knowledge about epistemological understandings of different domains and subjects can become more cumulative.

The present study

In this study, pre-service teachers are asked about their epistemic beliefs across four school subject domains. Hofer (2006) has explicitly pointed out the importance of domain-specific epistemic beliefs in teacher education. Pre-service teachers are characterised by a high level of subject interest (Glutsch & König, 2019) and a high level of interest in educational studies (Rösler et al., 2013) and should therefore be motivated to disclose their beliefs about different domains of knowledge. In school and at university, they have gained many years of experience with different subject cultures, which should enable them to express differentiated beliefs about the nature of knowledge and knowing. In line and partially derived from the TIDE framework (Muis et al., 2006), the following questions and hypotheses guided this investigation:

1. Can epistemic beliefs in four different school subject domains be measured reliably? We expected that a domain-specific questionnaire can reliably measure epistemic beliefs through the contextualisation of the items.

2. How specific are epistemic beliefs across four school subject domains? We expected significant differences in epistemic beliefs to emerge across domains.

3. What structure best characterises the relationships among domain-specific epistemic beliefs? We excepted to find a theoretical structure that can map the empirical relationships among dimensions and domains of epistemic beliefs.

Methods

Sample

The sample consisted of 196 pre-service teachers from two German universities. Among the study sample were 45 men and 151 women with an average age of 21.20 years (SD = 2.63). Participants studied a variety of disciplines (biology, chemistry, mathematics, computer science, geography, history, politics, economics, social studies, English, German, French, art, music, physical education, and Catholic religious education) for M = 2.82 semesters (SD = 1.50). In addition to pedagogy and psychology, every teacher training program includes at least two teaching disciplines and didactics in each of these subjects. To get a better idea of their major field of study, we classified participants’ first-named disciplines. Hence, 2.1% of the participants studied natural sciences, 24.4% mathematical sciences, 23.3% social sciences, and 50.3% linguistic sciences.

Instrument

The instrument was based on the self-report items used to measure epistemic beliefs by Conley et al. (2004). This questionnaire is considered the most widely used instrument in science education for measuring epistemic beliefs (Lee et al., 2021; Schiefer et al., 2022). It has been tested in various target samples, including pre-service teachers (Lee et al., 2021), typically shows satisfying reliability (e.g. Cheng, 2018; Cheng et al., 2021; Yang et al., 2019), replicable factor structure (Lin et al., 2012; Tsai et al., 2011), and correlates positively with academic achievement (Greene et al., 2018).

The questionnaire by Conley et al. (2004) was slightly modified to compare beliefs about the natural, mathematical, social, and linguistic sciences (see also Appendix). We regarded as important that all school subject domains were on the same level of abstraction (Schommer-Aikins et al., 2003). For the purpose of comparison, we accepted that the items did not do full justice to the uniqueness of each domain and discipline (Hofer, 2006).

To establish the different domains of the survey, the various subjects offered at all types of schools were examined on the web pages of the relevant Ministry of Education. Then, the attempt was made to cluster the school subjects on a continuum from hard to soft subjects using Biglan’s classification scheme. The four resulting domains almost completely cover the canon of subjects taught at secondary school. In the instruction of the questionnaire, the four domains were described for the participants as follows: The natural sciences encompass the subjects of biology, chemistry, and physics. The mathematical sciences relate to the subjects of mathematics and computer science. The social sciences include the subjects of history, social studies, and economics. The linguistic sciences contain the subjects of native language and foreign languages.

The domain specificity of the questionnaire was made clear by giving each item an addendum to highlight the domain of interest. The underlined gaps in the items were filled with the terms ‘natural’, ‘mathematical’, ‘social’, or ‘linguistic’. The questionnaire scales consisted of six items about the justification for knowing (e.g. ‘In ___ science, there can be more than one way for ___ scientists to test their ideas’), with higher scores indicating that knowledge is justified through systematic inquiry and integration of multiple sources of information. Five items were used to measure the source of knowledge (e.g. ‘If you read something in a ___ science book, you can be sure that it is true’), with higher scores indicating that knowledge is passively acquired and handed down by authority. Six items asked about the development of knowledge (e.g. ‘Ideas in ___ science sometimes change’), with higher scores representing that knowledge can change over time as a result of new discoveries. Six items measured the certainty of knowledge (e.g. ‘All questions in ___ science have one right answer’), with higher scores indicating that knowledge is considered certain and unchanging. Thus, the questionnaire was composed of two positively poled scales of justification and development and two negatively poled scales of source and certainty. Three items from the original questionnaire that related to justifying knowledge through experimentation were removed as they could hardly be applied to other domains than science. All items were rated on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree).

Procedure

Via a circular email, the pre-service teachers were invited to participate in the study. By calling up a link, they could view the questionnaire on a university website and worked on it for about 15–20 min. The school subject domains were presented in random order. The items of the four different scales were also randomly intermixed but their sequence was identical in each domain. Detailed descriptions of the school subject domains were included in the questionnaire instruction and were continuously visible while the participants were answering the items. Participants were asked to think about their personal experiences from school as well as their studies at university. All participants were assured that the questionnaire is anonymous, answers would be treated confidentially, only be used for research purposes, and not be passed on to third parties. As a reward, the participants could receive twenty euros in a raffle for conscientious and complete processing of the questionnaire. Therefore, they had to provide an email address at the end of the questionnaire. The addresses were saved in a separate file that was not linked to the other data and being deleted after the raffle ended. The winners were promptly informed to pick up their winnings at the secretary’s office of the professorship.

Statistical analyses

One hundred ninety-six pre-service teachers completed the questionnaire in full. Only this data was used for further statistical analyses so that no missing values had to be replaced. Data from 30 other teacher trainees were excluded from the analyses as they only responded to the items on the first domain, which means that 75% or more of their answers were missing. As these participants were not interested to continue, it is likely that their data in the other three domains were not missing at random.

To answer the first research question on the reliability of the applied instrument, we report item-total correlations and Cronbach’s Alpha coefficient. Since epistemic belief questionnaires frequently lack sufficient reliability (Clarebout et al., 2001; DeBacker et al., 2008), this question deserves particular attention. Otherwise, significant differences across domains could also be due to measurement error.

To answer the second and core question of our research about the domain specificity of epistemic beliefs, we applied a two-way repeated measures ANOVA and a test on correlation differences. From the multitrait-multimethod matrix by Campbell and Fiske (1959) can be deduced: Evidence for the domain specificity of epistemic beliefs is given when different constructs within domains are more highly correlated than the same constructs between domains. This would require heteroconstruct-monodomain correlations to be higher than monoconstruct-heterodomain correlations (Bong & Hong, 2010).

To answer the third research question, we made use of structural equation modelling to map the structure of domain-specific epistemic beliefs. To this end, we tested four models in which either (1) the dimensions or (2) the school subject domains determined the structure. In addition, we tested two higher-order models in which either (3) beliefs about the nature of knowledge and knowing (Hofer & Pintrich, 1997) or (4) beliefs about hard and soft subjects (Schommer-Aikins et al., 2003) determined the structure.

To judge the model structures, the four structural models were confirmatory tested by using AMOS 25 (Arbuckle, 2017). For each latent variable, the factor loading of one observed variable was fixed at one, while the other loadings from that factor were freely estimated. To take repeated measures into account, error terms were permitted to covary between scales of the same domain in dimensional models and between scales of the same dimension in domain models. All endogenous variables within the models had an error term with path coefficients fixed at one. Normal theory-based maximum-likelihood (ML) was used as a method for parameter estimation (McDonald & Ho, 2002).

To evaluate the goodness of fit of the four structural equation models, we used the Chi-square value with degrees of freedom, the Tucker-Lewis-Index (TLI), the Comparative Fit Index (CFI), and the Root Mean-Square Error of Approximation (RMSEA). For the TLI and CFI, values >.90 are acceptable and values >.95 are excellent. For the RMSEA, values <.08 are acceptable and values <.05 are excellent (Hu & Bentler, 1999).

Results

Item and scale statistics

The first research question addresses the reliability of the employed measuring instrument. Table 1 shows the descriptive statistics, item-total correlations, and reliabilities of the domain-specific epistemic belief questionnaire. Means and standard deviations suggest neither floor nor ceiling effects. The vast majority of the item-total correlations shows desired values above .30 (Cortina, 1993). The reliabilities of all scales are higher than .60 and only in three cases below the desired level of .70 (Cortina, 1993). These are favourable values for a questionnaire in the area of epistemic beliefs. They support the assumption that the employed questionnaire can capture epistemic beliefs in four domains of school learning with sufficient reliability.

Table 1. Item and scale statistics of the domain-specific epistemic belief questionnaire.

	No. of items	Min	Max	M	SD	Corri-t	Alpha
Natural sciences
Justification	6	2.17	5.00	4.01	0.49	.31–.49	.60
Source	5	1.00	4.60	2.82	0.61	.39–.56	.72
Development	6	2.33	5.00	3.90	0.54	.29–.62	.74
Certainty	6	1.00	4.00	2.25	0.55	.31–.43	.66
Mathematical sciences
Justification	6	1.83	5.00	3.53	0.63	.22–.53	.66
Source	5	1.20	5.00	3.32	0.72	.37–.70	.81
Development	6	1.50	5.00	3.05	0.66	.17–.65	.76
Certainty	6	1.17	5.00	3.14	0.70	.41–.61	.77
Social sciences
Justification	6	1.83	5.00	4.10	0.56	.26–.56	.71
Source	5	1.00	4.20	2.53	0.67	.47–.68	.81
Development	6	2.33	5.00	4.04	0.61	.31–.73	.79
Certainty	6	1.00	3.83	2.08	0.60	.43–.58	.76
Linguistic sciences
Justification	6	1.00	5.00	3.42	0.66	.37–.58	.72
Source	5	1.00	5.00	3.16	0.72	.33–.74	.80
Development	6	1.50	5.00	3.23	0.72	.46–.71	.83
Certainty	6	1.00	4.67	2.77	0.72	.32–.65	.81

Corr_i-t: range of corrected item-total correlations.

Mean differences

To answer the second research question on domain specificity of epistemic beliefs, a two-way repeated measures ANOVA was computed with the repeated-measures factors domain and dimension as independent variables (Field, 2013). Table 1 shows the associated means and standard deviations for this analysis. A significant main effect was found for the factor domain, F(3, 193) = 7.79, p < .001, η_p² = .108, as well as for the factor dimension, F(3, 193) = 256.05, p < .001, η_p² = .799. The interaction of domain and dimension, F(9, 187) = 54.79, p < .001, η_p² = .725, also became significant. Bonferroni-adjusted post-hoc analysis [adjusted p-value: .05 ÷ (6 × 4) = .00208] was conducted to compare the different domains on each dimension. Pairwise comparisons revealed significant differences between nearly all domains of personal epistemology in all four dimensions. The natural sciences did not differ significantly from the social sciences on the dimensions of justification (p = .025) and development (p = .005). In addition, mathematical sciences did not differ significantly from linguistic sciences on the dimension justification (p = .039).

Figure 1 illustrates furthermore that the profiles of natural and social sciences and the profiles of mathematical and linguistic sciences show some similarities. The justification for knowing through own studies is seen as most likely in the social and natural sciences. Moreover, participants assume the strongest development of knowledge in these two school subject domains. In the mathematical and linguistic sciences, on the other hand, great reliance is placed on authorities as a source of knowledge. The pre-service teachers perceived the mathematical sciences offering the greatest certainty of knowledge, again followed by the linguistic sciences.

Figure 1. Means and standard errors of epistemic beliefs across domains.

Correlational analysis

Manifest intercorrelations were used to further investigate the second research question about the domain specificity of epistemic beliefs. Triangles in Table 2 highlight the correlations of different constructs within domains, which can also be referred to as heteroconstruct-monodomain correlations. It gets obvious that all correlations between different dimensions within domains are significant at a moderate to high level. The average correlation of the 24 absolute values in the triangles of Table 2, determined by the use of Fisher Z-transformation, amounts to r = .50. For the purpose of comparison, Table 2 equally shows the correlations between domains, which can also be referred to as monoconstruct-heterodomain correlations, marked in bold. The same hypothetical constructs correlate on a low to moderate level across domains. The average correlation of these 24 bolded values, determined by means of Fisher Z-transformation, amounts to r = .36. A statistical test on correlation differences between the mean correlations of epistemic beliefs within and across domains becomes significant, Z = 2.383, p < .01. Since the heteroconstruct-monodomain correlations are significantly higher than the monoconstruct-heterodomain correlations, the outcome can be regarded as evidence for the domain specificity of epistemic beliefs in school subject domains (Bong & Hong, 2010).

Table 2. Manifest intercorrelations of epistemic beliefs.

Note. p < .05 if r > |.14|, p < .01 if r > |.18|, p < .001 if r > |.23|. Letters J, S, D, and C denote justification, source, development, and certainty. Triangles indicate heteroconstruct-monodomain correlations, while numbers in bold highlight monoconstruct-heterodomain correlations.

Structural equation modeling

To answer the third research question and more fully assess the relationships between domain-specific epistemic beliefs, structural equation modelling was conducted. As previously mentioned, four theoretical models were constructed to specify the relations between the different domain-specific epistemic beliefs. The first model in Figure 2 is based on the assumption that the four dimensions structure the latent relationships between the epistemic beliefs best. In the second model, the possibility of two higher-order dimensions is taken into consideration. As suggested by Hofer and Pintrich (1997), these are ‘beliefs about the nature of knowing’ and ‘beliefs about the nature of knowledge’. The third model in Figure 2 is based on the assumption that the four domains structure the latent relationships between the epistemic beliefs best. The fourth model is an extension of the third model. The model is based on the hypothesis that participants distinguish in their epistemic beliefs between hard and soft sciences as higher-order domains.

Figure 2. SEM models for the relationships between domain-specific epistemic beliefs. Note. Upper left part: Four-dimensions model. Lower left part: Two-higher-dimensions model. Upper right part: Four-domains model. Lower right part: Two-higher-domains model. First letters N, M, S, and L denote natural, mathematical, social, and linguistic sciences. Second letters J, S, D, and C denote justification, source, development, and certainty.

Table 3 reveals that the four-dimensional model is fitting the empirical data best. The four-dimensions model has a reasonable fit with TLI over .90, CFI over .95, and RMSEA under .08 (Hu & Bentler, 1999). Figure 3 depicts factor correlations and factor loadings of this best fitting model. Extending the model by two dimensions of higher order yields to a significantly poorer model fit as CFI is reduced by .01 (Cheung & Rensvold, 2002). The fit values for models in which domains are in the foreground are even weaker. The four-domains model and the two-higher-domains model do not show a reasonable fit to the data and must be regarded as not acceptable.

Figure 3. Factor correlations and factor loadings of the four-dimensions model. Note. First letters N, M, S, and L denote natural, mathematical, social, and linguistic sciences. Second letters J, S, D, and C denote justification, source, development, and certainty. All factor correlations and factor loadings are significant at the p < .001 level. Residual covariances between scales within domains are not shown for the sake of clarity.

Table 3. Fit indices of confirmatory factor analyses for the relationships among domain-specific epistemic beliefs.

Model	χ^2	df	TLI	CFI	RMSEA
Four-dimensions	149.632***	74	.924	.953	.072 [.056–.089]
Two-higher-dimensions	166.269***	75	.909	.943	.079 [.063–.095]
Four-domains	261.666***	74	.810	.883	.114 [.099–.129]
Two-higher-domains	269.424***	75	.806	.879	.115 [.101–.130]

TLI: normed fit index; CFI: comparative fit index; RMSEA: root mean-square error of approximation (90% confidence intervals are presented in brackets).

***p < .001.

Discussion

Based on the theory of integrated domains in epistemology (Muis et al., 2006), the question of domain specificity of epistemic beliefs was investigated in more detail. We chose pre-service teachers that already gained experiences with the subject domains in school and are now specialising in these domains at the university. An attempt was made to investigate more domains than in previous studies. Instead of comparing only two or three disciplines, epistemic beliefs about four overarching school subject domains were examined based on Biglan’s classification of hard and soft subjects. The natural, mathematical, social, and linguistic sciences represented almost the entire spectrum of subjects taught in secondary schools.

A questionnaire based on Conley et al. (2004) was able to capture four dimensions of personal epistemology in all four school subject domains with sufficient reliability. Regarding the core research question of domain specificity of epistemic beliefs, all four school subject domains were found to largely differ from each other on all four dimensions of personal epistemology. These findings were supported by correlations that also indicate domain specificity of epistemic beliefs. A model with the four dimensions as latent variables and the domain-specific beliefs of these dimensions as indicators best reflected the structures and relationships in the data set.

Specificity of school subject domains

The modified epistemic belief questionnaire by Conley et al. (2004) had sufficient measurement reliability for all four school subject domains. Comparing the four domains to each other, the reliabilities in the natural sciences were somewhat lower than in the other three domains. It is unlikely that this problem is due to the measuring instrument. The questionnaire by Conley et al. (2004) has already been used quite frequently and has proven its good measurement properties, especially in the natural sciences (Cheng et al., 2021; Lee et al., 2021). We assume the reason for the lower reliability in the fact that hardly any of our pre-service teachers was studying biology, chemistry, or physics. Conversely, the highest reliabilities were obtained in the linguistic sciences⸺an area that about half of the pre-service teachers studied.

The main question of the study was directed at the specificity of epistemic beliefs in four domains of school learning. We made inquiries about beliefs on school subject domains, which represent, according to Biglan’s classification of hard and soft disciplines, clearly distinguishable areas (Biglan, 1973; Simpson, 2017). Each domain comprised several subjects that share a common underlying paradigm (Kuhn, 1962). Thus, there was homogeneity of school subjects within domains but heterogeneity between the four school subject domains.

The significant mean differences between the natural, mathematical, social, and linguistic sciences on all four dimensions of personal epistemology demonstrate that pre-service teachers view each domain individually. During their school years, they might have learned that each of these four domains has a specific epistemic position (Muis et al., 2006). Similarly, the Programme for International Student Assessment (OECD, 2019), for example, distinguishes between scientific, mathematical, and language literacy. In addition to general competencies, such as logical thinking and problem solving, each domain requires specific competencies unique to this domain. Thus, students learn that each domain of knowledge requires a particular perspective contributing to the development of their own domain-specific epistemic beliefs. In this respect, it seems pretty obvious that participants’ beliefs about knowledge and knowing differed significantly across the four school subject domains.

Despite all the differences, however, there were also commonalities between school subject domains. Thus, the profiles of the natural sciences and the social sciences were somehow similar, as were the profiles of the mathematical sciences and the linguistic sciences. It may be speculated that the similarities of the natural sciences and the social sciences are due to the empirical approach, whereby the knowledge generated is perceived as less static and more evolving. The mathematical sciences and linguistics, on the other hand, are both based on strong laws and rules, whereby knowledge is perceived as more certain and more handed down by authority. Please note that this outcome is in contrast to Biglan’s classification of hard and soft disciplines. Perhaps it would be useful to add another dimension ‘empirical vs. theoretical/hermeneutical’ to the classification scheme.

Further evidence for the domain specificity of epistemic beliefs stems from correlational findings. The same constructs between different domains produced significantly smaller correlations than different constructs within the same domains (Bong & Hong, 2010; Campbell & Fiske, 1959). This suggests that the domain has a strong influence on individuals’ rating of the constructs, which is another argument for the domain specificity of epistemic beliefs. These results are noteworthy in that linguistically almost the same items were used to examine the same constructs across domains. They differed only in a domain-specific addendum. In contrast, linguistically very different items have been used to study different constructs within domains. They resembled each other only in a domain-specific addendum. Nevertheless, the heteroconstruct-monodomain correlations turned out to be significantly higher than the monoconstruct-heterodomain correlations. We thus aimed to show that correlations can be useful in identifying differences in epistemic beliefs between domains.

In a final exploratory step, four structural equation models were tested to map the relationships between dimensions and domains of epistemic beliefs. The model with the four dimensions as latent variables and the four domains as indicators showed the best fit to the data. It is logically derived from the correlational findings as the relationships between dimensions in the data are closer than the relationships between domains. The model can be interpreted as showing that the different dimensions of personal epistemology are closely related but each dimension has a domain-specific component. This is another way to demonstrate the importance of viewing epistemic beliefs in the school subject domains as domain-specific rather than domain-general.

Limitations

This study is subject to some limitations. First, a questionnaire measuring epistemic beliefs in science (Conley et al., 2004) was used and adapted to other contexts. This approach ensures comparability across domains but does not do justice to the specificity of each scientific domain (Hofer, 2006). However, the study results indicate domain specificity of epistemic beliefs, which are even more likely when domain specificities are considered as well. Second, the data were collected in a repeated-measures design, which means that the pre-service teachers involved might have lacked sufficient expertise to answer all questions appropriately. Although the participants were capable of indicating their epistemological understanding due to many years of experience with the domains in school, participants in a between-groups design can process the questionnaire with considerably more expertise. While we surveyed a convenience sample, a targeted study with students of the four domains might yield more accurate findings. Groups from the four epistemic domains would have to assess only their own domain, but similar differences should emerge as in the within-group design. For reasons of validity, it would be worthwhile trying to confirm the study outcomes in a between-groups design (Muis et al., 2006; Rosman et al., 2020). Students with different expertise (bachelor vs. master or primary vs. secondary education) could also be compared to highlight possible developments from naïve to nuanced positions of personal epistemology (Miguel-Revilla et al., 2021; Stoel et al., 2017). Third, each school subject domain included several subjects that were similar to each other but not identical in their scientific paradigm (Kuhn, 1962). For example, some researchers also determined differences in epistemic beliefs between biology, chemistry, and physics (Lee & Tsai, 2012; Topcu, 2013) that we treated all the same as the natural sciences. Our goal was to make comparisons on a higher level across a wide range of subjects. Nonetheless, it seems worthwhile to identify commonalities and differences in epistemic beliefs at the subject or topic level as well (e.g. Strømsø & Bråten, 2009).

Conclusions

The issue of the domain specificity of epistemic beliefs needs to be examined with great rigour and even greater breadth. Therefore, this study covered a broader range of subjects by specifically selecting higher-level school subject domains. We provided evidence that a single set of measures can be used to assess the nature of epistemic beliefs in a wider range of domains. A comparison was made between the natural, mathematical, social, and linguistic sciences that clearly differed on a continuum from hard to soft subjects. The results show that each of the four domains comes up with a different profile of epistemic beliefs. They provide strong evidence for the domain specificity of epistemic beliefs about school subject domains, which raises the demand for a classification system to specify major fields of personal epistemology.

While previous research has made clear that epistemic beliefs are better studied in a domain-specific rather than a general way (Muis et al., 2006), the question of which domains exist and how many can be empirically separated from each other is currently completely unresolved. Our research demonstrates a way to compare a larger number of domains based on a well-established measurement tool that helps elucidate the structure of domain-specific epistemic beliefs. We hope that similar to the research on academic self-concept, this can be an impetus to further investigate and clarify the question of the number and interconnections of domains of epistemic beliefs in more detail in the future.

Disclosure statement

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

Appendix.

Self-report items to measure epistemic beliefs across domains

The questionnaire is a slightly modified version of the Epistemological Beliefs Questionnaire by Conley et al. (2004). Three items from the original questionnaire that related to justifying knowledge through experimentation were removed as they could hardly be applied to other domains than science. Further changes in item wording are highlighted in italics. The original wording follows in square brackets. To measure epistemic beliefs about domains, the underlined gaps were filled with the terms ‘natural’, ‘mathematical’, ‘social’, or ‘linguistic’.

Justification

Good ideas in ___ science can come from anybody, not just from ___ scientists.

In ___ science, there can be more than one way for ___ scientists to test their ideas.

Ideas in ___ science can come from your own questions and research [experiments].

Good answers in ___ science are based on evidence from many different studies [experiments].

One important part of ___ science is getting to the bottom [doing experiments] to come up with new ideas how things work.

Ideas about ___ scientific research [science experiments] come from being curious and thinking about how things are connected [work].

Source

In ___ science, you have to believe what the ___ science books say about stuff.

Only ___ scientists know for sure what is true in ___ science.

Everybody has to believe what ___ scientists say.

Whatever the teacher says in ___ science class is true.

If you read something in a ___ science book, you can be sure that it is true.

Development

The ideas in ___ science books sometimes change.

Some ideas in ___ science today are different than what ___ scientists used to think.

Ideas in ___ science sometimes change.

Sometimes ___ scientists change their minds about what is true in ___ science.

There are some scientific questions that even ___ scientists cannot answer.

New findings [discoveries] can change what ___ scientists think is true.

Certainty

All questions in ___ science have one right answer.

Once ___ scientists have a research result [result from an experiment], that is the only answer.

The most important part of doing ___ science is coming up with the right answer.

___ scientists pretty much know everything about ___ science; there is not much more to know.

___ scientists always agree about what is true in ___ science.

___ scientific knowledge is always true.