Measuring individual differences in adults’ mindreading

Developmental research into mindreading has been dominated by questions about when and how children acquire mindreading concepts. Given that older children and adults already have a conceptual understanding of what thoughts, beliefs and desires are, that they may differ between people, and that they can impact behaviour, researchers have devised a wide range of tasks that are more subtle and complex in nature to assess mindreading in middle childhood and adolescence [10] and in adulthood [11]. However, mindreading tasks devised to be complex enough to capture individual differences in adults’ mindreading are often characterised by a lack of convergent validity, i.e., performance on one measure of mindreading does not correlate with performance on another measure [12, 13] or construct validity (i.e., the same task can be used to measure different constructs) [14, 15]. Some tasks used to assess mindreading such as the Animations Task [11, 16], which requires participants to explain the movements of triangles that exhibit either a simple interaction (e.g., two triangles appearing to be dancing), or a more complex interaction (e.g., one triangle appearing to surprise the other triangle), are sometimes also utilised to measure other constructs, such as anthropomorphism [17, 18]. Similarly, the Reading the Mind in the Eyes test (RMET) [19], originally developed to measure mindreading, may be more suitable as a measure of emotion perception given that the task requires participants to match suitable adjectives with pictures of faces expressing emotions [14, 15]. It is therefore unclear what “advanced” mindreading tasks really measure [14, 15, 20, 21]. Furthermore, mindreading research in adults has predominantly focused on group-level comparisons between clinical and non-clinical groups [12, 22, 23] and does not provide satisfactory answers as to whether and why there are individual differences in adults’ mindreading.

Notwithstanding these limitations, some mindreading tasks have also been successfully administered to typical adults [12, 24–26]. The Movie for the Assessment of Social Cognition [MASC, 26] requires participants to reason about the mental states of four characters who are shown in a realistic social interaction based on a multiple-choice question regarding the characters’ mental states. Neurotypical adults have been found to select correct responses approximately 60% of the time [23, 26], which indicates that the MASC is suitable for measuring individual differences in non-clinical adult samples. It should, however, be noted that the multiple-choice format of the MASC does not allow for the differentiation of frequency of mental state terms; a quantitative marker of an individual’s tendency to mindread based on the frequency with which they refer to other people’s mental states, and the appropriateness of participants’ mindreading responses; contextually justified mindreading.

The Silent Films Task [SFT; 10] has specifically been developed to detect individual differences in mindreading beyond the pre-school years. The task requires participants to answer six open-ended questions about five video clips that portray a main character interacting with a range of peripheral characters. Although this task exhibits convergent validity (e.g., scores on the SFT are strongly correlated to scores on another–strange stories–task) [10], criterion validity [27], and test-retest reliability [28], it has yet to be formally evaluated beyond early adolescence. Furthermore, the original questions used in the SFT prompt participants to reason about mental states and do not differentiate between frequency of MST and response appropriateness. Thus, despite these advances in the creation of “advanced” mindreading tasks, challenges persist in the measurement of how mindreading varies in adults [3, 12, 14, 15, 20, 29].

Finally, there are grounds to think that frequency of MST and response appropriateness may be separable constructs, although they are often confounded in open-ended response coding schemes. The frequency of spontaneously produced MST is distinct from performance on mindreading tasks in middle childhood [30], in adults [24], and in older adults [31]. Furthermore, there is evidence that the use of mental state terms can vary with different interaction partners [31], perhaps because some social interactions are more motivating than others.

The role of motivation in individual differences in adults’ mindreading

A person’s motivation fundamentally shapes their behaviour [32]. Indeed, making mentalistic attributions is an effortful process that places high demands on memory and cognitive control [33–36]. Individual differences in motivation might therefore also influence performance on mindreading tasks [4, 5, 6, 32].

Mindreading motivation has been defined as a stable orientation to engage effortfully with others’ minds and mental states [4], and it has been suggested that autistic people sometimes perform less well on mindreading tasks because they have lower social motivation rather than lower social ability [37–39]. Likewise, there is evidence that neurotypical adults’ mindreading motivation is positively related to their performance on mindreading tasks [4]. Specifically, Carpenter et al. (2016) found small positive correlations between mindreading motivation, assessed via the 13-item Mind Reading Motivation Scale and performance on the interpersonal perception test 15 [IPT-15; 4], which requires participants to make judgements videotaped real-world social situation. It has been suggested that mindreading motivation and mindreading ability are conceptually and empirically separable [4–6], and there is evidence that social motivation and mindreading ability make distinctive contributions to explaining social outcomes [5]. The present study combined this observation with a revised coding scheme for mindreading that distinguishes between the quantity of mental state terms and the appropriateness of mindreading responses.

Overall, the above-described concerns and previous research findings motivate the first objective of introducing a scoring system that distinguishes between (a) mindreading quantity, a potential indicator of motivation, operationalised as frequency of MST in responses, and (b) mindreading quality, operationalised as the “appropriateness” of participants’ answers when responding to open-ended versions of the Silent Films Task [SFT; 10] and the Movie of the Assessment of Social Cognition (MASC; 26).

Criterion validity of MST and response appropriateness

One way to distinguish between two closely related constructs is to investigate differential relations between the constructs and other variables [4]. Although previous results have not always been consistent and may have been subject to some of the psychometric shortcomings of mindreading tasks [12, 21], they provide potential criteria for evaluating the different factor structures used to conceptualise mindreading in the present study. The distinctiveness of individual differences in mindreading quality and mindreading quantity was therefore examined by investigating whether these constructs might show unique associations with variables that have been linked with mindreading performance, such as measures of social competence [10, 7, 40] and personality tendencies such as religiosity [41–43], authoritarianism [44], and anthropomorphism [17, 45].

The present study

In summary, we sought to shed light on the nature of individual differences in mindreading in adults, by creating a novel scoring system that leveraged existing open-ended mindreading tasks. To this end, whether mindreading quantity can be distinguished from mindreading quality was assessed. More specifically, the present study examined whether individual differences in neuro-typical adults’ (N = 128) and frequencies of explicit references to others’ mental states and response appropriateness on two open-ended response mindreading tasks were (a) separable constructs, (b) accounted for by differences in individuals’ mindreading motivation, and (c) related differentially to a variety of outcome measures that had previously been empirically or theoretically linked with mindreading.

It was hypothesised that if frequencies of MST and response appropriateness are separable constructs, then response appropriateness and frequencies of MST will load onto separate latent factors (i.e., one representing response appropriateness and one frequencies of MST, a potential indicator of motivation). Alternatively, if frequencies of MST and response appropriateness test the same underlying construct (i.e., “mindreading”), response appropriateness and frequencies of MST will load onto one latent factor. Finally, it is possible that individual differences in performance will be driven by characteristics of the two different tasks, rather than the psychological constructs of response appropriateness, frequencies of MST or mindreading. This effect would lead indicators (i.e., MST and appropriateness) coded from the SFT to load onto one latent factor (i.e., representing performance on the SFT) and both indicators coded from the MASC to load onto a separate factor. The possibility of shared method variance will be accounted for in our analyses. Furthermore, it was hypothesised that mindreading motivation will be specifically related to the quantity of mental state terms produced in participants mindreading responses.

Participants

This study was pre-registered prior to the start of data collection. This is a link to the preregistration: https://osf.io/uerkx/). A total of 133 participants were recruited between the 24^th of January and 28^th of March 2021 (the link to the study was accessible between these dates) via word of mouth and via the University of Birmingham School of Psychology undergraduate database.

Power-analysis based on an a-priori sample size calculator for structural equation models [46] indicated that a minimum sample size of 119 participants was needed to obtain a level of acceptable statistical power of .80, and for a model that has a maximum of three latent variables and 15 indicator and observed variables. Our minimum target sample size was therefore 119 participants. All participants took part in the study in exchange for course credit and were psychology students. To be eligible participants had to be at least 18 years old, report no current or past neurological or psychiatric problems, and be English-speaking (including native and non-native English speakers). Based on these pre-registered exclusion criteria, 5 participants were excluded due to reporting a past neurological or psychiatric problem, leaving a final sample of 128 participants (113 females, 15 males, Mage = 19.47 years, SD = 1.47 years, range = 18–28). The participation rate was 96.24%. One hundred and seven participants (83.6%) reported English as their native language, with 71.9% of participants (N = 92) being monolingual, 25% (N = 32) bilingual and 3.1% (N = 4) multilingual (speaking more than two languages). Most participants (95.3%, N = 122) had completed secondary school or equivalent (e.g., A levels), 2.3% (N = 3) held a bachelor’s degree, and 1.6% a master’s degree (N = 2). Participants were invited to identify their ethnicity according to the descriptions recommended by the United Kingdom Office for National Statistics (ONS) that are comprised of five broad categories and 18 sub-categories. In order of size, the five broad categories were presented as follows: 70.4% (N = 90) of participants identified as White, 16.5% (N = 21) identified as Asian, 5.5% (N = 7) as Black, 4.7% (N = 6) as Mixed, and one participant (0.8%) as an ethnicity not listed. Data on religion was also recorded according to the questions proposed by the ONS: 56.3% (N = 72) of participants reported no religion, 21.1% (N = 27) identified as Christian, 2.3% (N = 3) of participants as Hindu, 7% (N = 9) as Jewish, 10.2% (N = 13) as Muslim, 1.6% (N = 2) as Sikh, and 1.6% (N = 2) preferred not to answer the question.

Design and procedure

As this was an individual differences study, questionnaires and measures were administered in a fixed order, following Goodhew and Edwards (2019) [47]. The rationale is that a fixed order of presentation makes it more likely that any effects of task order are equivalent for all participants, which minimizes the unexplained variance when the focus of analysis is on the pattern of individual differences between participants. A fixed order of presentation of course raises the possibility that fatigue or practice, or order effects, will result in unreliable measurement. For example, highly fatigued participants might show floor-level performance, or answer at random. All measures were therefore checked for reliability. All questions were administered via the Qualtrics survey platform. Testing was completed in one session and no feedback was provided in relation to response content. The time participants had to complete the tasks was not limited and the total duration of the study was approximately one hour. Informed written consent was obtained from all participants prior to participating and the study was approved by the Ethical Review Committee at the (Masked) (protocol code: ERN_09–719 2020). The recruitment system and software for online testing ensured that participants’ identities were never known to the researchers. Participants’ data was stored confidentially and only available to the researchers involved in this study at the University of Birmingham.

Response appropriateness and frequencies of Mental State Terms (MST)

Ten Item Personality Measure (TIPI) [49]

The 10-item measure of the Big Five (or Five-Factor Model) dimensions of personality was administered. Participants indicated the degree to which they ascribed different personality traits to themselves, using a 7-point Likert scale ranging from 1(Disagree strongly) to 7(Agree strongly). Items 2, 4, 6, 8 and 10 were reversely coded (e.g., for these items 7 was recoded with a 1, a 5 with a 3 etc.). The average of two items (standard item and reverse-scored item) made up a scale (e.g., extraverted, enthusiastic & reserved, quiet, a = .77, Critical, quarrelsome & Sympathetic, warm, a = .29, Dependable, self-disciplined & Disorganised, careless, a = .64, Anxious, easily upset & Calm, emotionally stable, a = .65, Open to new experiences, complex & Conventional, uncreative, a = .37).

Wechsler abbreviated scale of intelligence [50]

Current intellectual functioning (verbal ability) was assessed using the Similarities Test of the Wechsler Adult Scale of Intelligence. This vocabulary subtest has been recognized as among the most widely used measure of general mental ability. Participants completed 20 trials (with 20 items), explaining why two words were alike. For example, participants were asked: “In what way are grapes and strawberries alone?”, to which the correct answer would be “both fruit”. Responses were scored using the WASI manual and items were scored on a 3-point scale from 0 to 2. Scores were added up, with higher scores indicating greater expressive language ability, α = .33, range = 22–35.

Analysis plan

First, measurement models for mindreading appropriateness (i.e., response appropriateness for the MASC and SFT) and frequencies of mental state terms (from responses to the MASC and SFT) were tested. Following this, whether mindreading motivation as well as a range of outcome measures were associated with individual differences in mindreading using structural equation modelling was examined. The latent factors were regressed onto age, gender (0 = male, 1 = female), education, whether participants spoke English as their native language (0 = no, 1 = yes) and verbal ability. The data were analysed using SPSS and R, the Lavaan package [51].

A robust maximum likelihood estimator was used to account for the potential non-normal distribution of each variable where variables were continuous. For ordinal variables, diagonally weighted least squares estimation (WLSMV) was used. To evaluate whether the models fit the data, three recommended methods were used: the root mean square of approximation (RMSEA) with < 0.05 indicative of good model fit and values between 0.05 and 0.08 as adequate model fit [52]; the comparative fit index (CFI); and the Tucker-Lewis index (TLI), for which values above .95 are considered as a good fit [53]. Where two or more models met all three criteria, the simplest model that was most strongly supported by theory for further analyses was selected. Covariates (e.g., gender, age, education, English as native language and verbal ability) were treated as separate variables. In line with recommendations for individual differences researchers, correlations of .10 were considered as a small effect, .20 a typical (or moderate) effect and > .3 as a relatively large effect [54].

Missing data

The percentage of missing values ranged from 0.8% (N = 1) for demographic variables such as education up to 7.8% for religious upbringing (N = 10). The only reason for missing data was participants’ nonresponse. A full information maximum likelihood (FIML) approach to missing data was adopted. This method has been shown to produce unbiased parameter estimates, standard errors and test statistics that are consistent and efficient when data is either missing at random (MAR) or missing completely at random (MCAR) [55]. To test for patterns within the missing data, Little’s Completely at Random test (MCAR) was conducted and showed that data was not systematically missing, χ² (1205) = 1133.893, p = .928.

Preliminary and descriptive statistics

Table A in the S1 Table shows the bivariate correlations between the item-level data of the SFT and MASC as well as the descriptive statistics for each item of the measures. For correlations between the original coding of the MASC and SFT with the new coding applied in the present study, see Table B in the S2 Table. Descriptive statistics of all summed study variables can be seen in Table 1. The SFT mental state terms data were normally distributed, with one participant using no mental state terms at all (.8%) and two participants (1.6%) using 15 mental state terms to describe the actions in the video clips. There were no ceiling or floor effects for response appropriateness data, with no participant scoring 0 and only 4 participants achieving full scores (3.1%). For the MASC mental state terms data, 3 participants made only two mental state references (2.3%), and one participant made 75 mental state references (.8%) whilst the second highest number of MST was 51 (N = 1, .8%). The data were normally distributed. In terms of appropriateness, there were no ceiling or floor effects, with no participant scoring 0, one participant scoring 1 (.8%), and three participants scoring 2 (2.3%). No participant achieved full scores, 3 participants scored 17, (2.3%), and a further 3 participants achieved 16 across all items of the mask. Skewness and kurtosis of all study measures, except for the social network size measure, were within the acceptable ranges of -3 and +3 for skewness and -10 to +10 for kurtosis. The social network size measure was positively skewed due to an outlier (one participant indicated 110 social contacts, whilst the second highest number was 43).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics for each study measure.

https://doi.org/10.1371/journal.pone.0305270.t001

Main analysis

The latent factor structure of mindreading (i.e., whether response appropriateness and frequencies of MST are separable or overlapping) was examined. A schematic overview of the different models that were tested is shown in Fig 1 and the summary statistics of the different models are shown in Table 1. Both a one-factor (see model A.123) and two-factor (see model C.23) model provided an acceptable fit for the data (One factor model: RMSEA = .032, CFI = .937, TLI = .927, Factor loadings = .115 - .679; Two factor model: RMSEA = .031, CFI = .943, TLI = .933, Factor loadings = .132 - .733). Additionally, the latent factors of response appropriateness and frequencies of MST were significantly positively correlated, r = .75 (see Table 2). Therefore, in terms of our first criterion for judging the value of coding MST and appropriateness separately versus in a single construct, although indices were marginally in favour of a two-factor solution, there is no clear evidence in favour of a one- or two-factor model. The second criterion was therefore considered. Specifically, the associations between the resulting factors from these models were compared with the outcome variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Schematic representation of the models that were tested to investigate the best fit for the item-level data of the SFT (5 items) and MASC (10 items), with each item being associated with two unique scores; a mindreading response appropriateness score and a frequencies of MST score (making 32 individual scores).

A = One-Factor model where frequencies of MST and response appropriateness load onto one factor, regardless of task; 1 = Method factors due to the different tasks are accounted for (MASC); B = Two-Factor / Two-Task Model where both constructs (i.e., MST and response appropriateness) coded from the SFT load onto one factor (representing performance on the SFT) and both constructs coded from the MASC load onto a separate factor to test if shared task characteristics drove individual differences in response appropriateness and MST; C = Two-Factor model where response appropriateness and MST load onto separate factors.

https://doi.org/10.1371/journal.pone.0305270.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Bivariate correlations between all variables of interest.

https://doi.org/10.1371/journal.pone.0305270.t002

Next, associations between the resulting factors from these models and the social outcome variables were compared. Table 2 shows the bivariate correlations between all study variables. The latent factors for response appropriateness and frequencies of MST were strongly positively correlated, indicating that there was a large amount of shared variance between both constructs. Although there was a small significant association between response appropriateness and mindreading motivation, frequencies of MST and mindreading motivation were not significantly correlated. Anthropomorphism was weakly positively associated with response appropriateness and negatively with loneliness. Authoritarianism was weakly negatively associated with mindreading motivation.

Table 3 displays the summary statistics for the structural equation models that were estimated to test how either response appropriateness and frequencies of MST, separately as part of a two-factor model, or jointly, as part of a one-factor model, related to other constructs of interest. Each outcome variable was separately regressed onto these two latent factors. To control for potential confounding variables, the outcome variables were regressed onto each latent factor and education, gender, age, whether they spoke English as their native language, verbal ability and personality traits. Covariates were permitted to correlate with each other in each model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary statistics of different models to assess whether response appropriateness and frequencies of MST are separable or overlapping.

https://doi.org/10.1371/journal.pone.0305270.t003

An assessment of the unique associations between mindreading motivation, frequencies of MST and response appropriateness showed that MRM was not predicted by the amount of MST people used in their responses or by the appropriateness of their responses. Likewise, a one-factor solution where frequencies of MST and response appropriateness loaded onto one factor was not a significant predictor of MRM. A one-factor solution did not predict any of the outcome variables significantly. However, for a two-factor solution there were significant, differential associations between the latent factors of frequencies of MST and response appropriateness with religiosity. MST predicted religiosity positively while greater response appropriateness was a significant negative predictor of religiosity (see Fig 2 below for a simplified path diagram of the standardized estimates for significant regression paths). This finding indicates that there was valid non-shared variance between both constructs, which might be obscured by confounding or not differentiating between the degree to which participants quantitatively engage with others’ mental states and the degree to which their responses are appropriate. In terms of our second criterion for judging the value of coding MST and appropriateness separately versus in a single construct, evidence of greater explanatory power for a two-factor solution for predicting outcomes related to mindreading was found.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. ** p < .001, *p < .05.

Simplified path diagram showing standardized estimates for statistically significant regression paths for the two-factor religiosity latent factor model (see Table 4). Covariances between predictor variables are not displayed. Non-significant variables controlled for in this model are age, gender, verbal ability, English as native language, and personality (e.g., extraversion, agreeableness, conscientiousness). Rel = Religiosity; APP = Response appropriateness; MST = Mental State Terms.

https://doi.org/10.1371/journal.pone.0305270.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Regression summary statistics.

Separate one- and two-factor models were run for each listed outcome variable individually.

https://doi.org/10.1371/journal.pone.0305270.t004

Evidence for a two-factor structure of mindreading

A significant current challenge for the field is to advance theoretical understanding of how mindreading abilities continue to vary in adults and to devise psychometrically robust ways to measure these individual differences [29]. Previous research has proposed that mindreading may comprise multiple component processes such as cognitive versus affective [12, 56], social-perceptual versus social-cognitive [57] or effortful versus automatic [58]. Whereas these proposals focus on the cognitive architecture of mindreading, we sought to characterise sources of variation in performance. This is relevant for any given component process, for “advanced” tasks like the MASC and SFT that combine multiple components, and for everyday use of mindreading outside of scientific settings.

Unlike many “laboratory” tasks [13, 29] the MASC and SFT are known to generate robust individual differences and to predict social outcomes, but their coding schemes combine the quality and quantity of mindreading responses into a single scale reflecting the degree to which a response is “correct”. It might be that adults’ mindreading truly varies on a single dimension. However, it is not possible to test this hypothesis when mindreading quality and quantity are baked into a unidimensional coding scheme. Based upon growing evidence that motivation contributes to individual differences in mindreading [4–6], it was hypothesised that some people might be disposed to mindread more than others, even if they were not especially successful in doing so. Therefore, mindreading quality and quantity were coded separately, and whether these factors were indeed reflected in the covariance structure of the tasks was tested.

To do so, we tested (a) whether both constructs loaded onto one or two latent factors; and (b) whether a one- or two-factor model was superior at predicting outcome variables which had previously been linked with mindreading. Evidence on the first criterion was equivocal, as both a one- and two-factor solution provided an acceptable fit for the data. Whilst response appropriateness and frequencies of MST were from the same two tasks, there was a risk that either the one-factor solution, or the large amount of shared variance between the constructs in the two-factor solution, were due to shared methods rather than similarity in the underlying constructs. However, no evidence was found for this because shared method variance was accounted for by running a two-task model, which did not provide a good fit for the data.

In relation to our second criterion, a one-factor model (representing the shared variance between MST and response appropriateness) was found to not predict any of the outcomes measured in this study. In contrast, a two-factor solution revealed significant, differential associations between the MST latent factor and response appropriateness latent factor (two-factor model) with religiosity. This finding suggests that despite substantial shared variance, response quality and quantity have explanatory power as distinct constructs. It remains important for future work to determine whether the non-shared or shared variance between MST and response appropriateness predicts other criterion variables above and beyond the ones employed here.

In summary, given the support found for both a one-factor and two-factor solution, consideration of parsimony might favour the one-factor solution. However, since the two-factor solution has superior power at predicting outcomes, we believe there are good grounds to favour the two-factor solution.

Mindreading motivation

Given that the MASC and SFT involve mental concepts that participants should have mastered by adulthood, variation in performance could indicate differences in motivation to reason about others [5, 29, 33–36]. However, the present results indicated that self-reported motivation was not associated with either the amount of MST people used in their responses or with the appropriateness of their responses. It should be noted that minimally 13 and maximally 91 points could be achieved on the MRM scale but in the current sample, the lowest MRM score was 47, and the highest 89 (with data normally distributed between this range). This suggests that the variance of the MRM variable may have been restricted in the current study, perhaps because most of the participants were undergraduate psychology students. Likewise, a one-factor solution where frequencies of MST and response appropriateness loaded onto one factor was not associated with MRM. The present study therefore provided no evidence that frequencies of MST, and / or response appropriateness captured meaningful individual differences in self-reported mindreading motivation. However, it is important to note that the present study used a relatively brief self-report measure of mindreading motivation. It remains possible that different effects might be observed within a more diverse sample, or with more direct measures of motivation, or on tasks that assess a broader range of social motivations, including but not limited to motivation for mindreading. The origin of variation in MST remains currently unclear and future research will need to examine whether the observed lack of association between frequency of MST and mindreading motivation reflected a genuine lack of association between the two constructs.

Social outcome variables

A one-factor mindreading factor was not uniquely associated with any of the outcome variables, but there were significant, differential associations between the latent factors of frequencies of MST and mindreading response appropriateness with religiosity. Specifically, a higher degree of MST predicted religiosity (composed of the sub-components of religious belief, affiliation, practice, and upbringing) positively, whereas greater response appropriateness was a significant negative predictor of religiosity. This means that religious participants were less likely to give appropriate (or context-sensitive) answers on two mindreading tasks and more likely to use mental state language than non-religious participants. Previous research is in line with the notion that religiosity may be negatively (or not at all) related to appropriate mentalising about other human minds, which differ qualitatively from supernatural agents’ minds [43, 59]. The converse positive association between frequencies of MST and religiosity is likewise in line with theoretical accounts arguing that religiosity may have emerged because of the human propensity to attribute mental states [60, 61].

The present findings highlight the potential distinctiveness of mindreading quantity (i.e., frequency of MST) and mindreading quality (i.e., response appropriateness) in the structure of individual differences in mindreading.

Limitations and future directions

First, the measures available to investigate the constructs in the present study are limited. For instance, while a validated self-report questionnaire to measure mindreading motivation was used, participants’ scores on this measure were restricted in range. In addition, some of the covariate measures (i.e., personality, verbal ability) exhibited low internal reliability, which may have prevented the detection of relationships. Researchers often distinguish between religiosity and spirituality [62], and this distinction might be relevant to the issue of mindreading. Religion is commonly perceived as more institutionalised and dogmatic than spirituality [63, 64]. Therefore, if spirituality is less tied to religious concepts (e.g., a personal God who is concerned with each individual’s life), it may not share with religiosity its positive relation with making frequent explicit mentalistic attributions to others. Finally, it should also be noted that a large proportion of participants in the present study were female psychology students. While caution must be used when generalising the present findings to a broader population, the possibility of reduced variance in the current relatively homogeneous sample if anything reduced the likelihood that reliable factor structure for the present measures would have been found. Further research should use a non-university sample, as well as include a range of educational backgrounds and ages to be more inclusive and generalisable to a larger number of individuals.