Reading comprehension and metacomprehension skills in women with Autism Spectrum Disorder in Higher Education: An exploratory correlational study

Introduction

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects multiple areas of human functioning, such as communication, social interaction, behaviour and certain aspects of cognitive processing; it is characterised by challenges in social interaction and communication, and by restricted or repetitive patterns of behaviour (DSM-5-TR, APA, 2022). In Chile, recent estimates suggest that one in every 51 children has this diagnosis (Yáñez et al., 2021). This picture becomes even more complex when we consider the gender disparity in detection rates, with a significantly higher prevalence among men than among women (López-Espejo, 2025).

Several hypotheses have been put forward to explain this gender gap; one of them is that it may be due to the different ways in which ASD symptoms manifest in women, which tend to go unnoticed in clinical and educational settings (Lockwood-Estrin et al., 2021). Historically, studies on ASD have also predominantly focused on samples comprising mainly men, which has resulted in theoretical models and assessment tools that are not particularly sensitive to the female expression of the spectrum (Hull et al., 2021; Begeer et al., 2013). In many cases, girls with ASD are not identified at an early stage because their behaviour is perceived as being more well-adjusted, which delays the establishment of an accurate diagnosis (Hiller et al., 2014). Added to this is the frequent use of social camouflage strategies by women diagnosed with ASD, such as imitating behaviours and superficially conforming to social norms. This makes it difficult to detect their condition and can have significant consequences for their emotional wellbeing and academic performance (Hull et al., 2020; Dean et al., 2017).

In turn, women diagnosed with ASD have specific cognitive and emotional profiles, which may influence the way in which they approach reading comprehension and the activation of metacognitive skills (Weiner et al., 2023). Beckerson et al. (2024) point out that these cognitive profiles may involve differences in processes such as cognitive flexibility, inferential processing and the attribution of mental states – skills that play a significant role in text comprehension and in the construction of meaning whilst reading. This situation takes on particular significance in the context of higher education, where inclusion has begun to establish itself as a priority that still faces challenges.

Reading comprehension is a complex, multimodal skill in which various factors relating to the reader, the texts and the context interact. In this process, readers draw not only on their knowledge of language, but also on their understanding of the world and a range of cognitive and metacognitive skills that enable them to construct a coherent representation of the textual content. This process culminates in the construction of a situation model (Kintsch, 1998).

The use of metacognitive strategies enables the reader to maintain the coherence of their mental model of the text and to adapt their reading process in line with the demands of the task (McNamara, 2004). Strategic readers do not only construct meaning, but also actively assess and regulate their own comprehension process by engaging higher-order cognitive processes, which enable them to construct and integrate a model of the situation (McNamara, 2004). It should be noted that, in readers diagnosed with ASD, this skill may be influenced by their ability to infer the mental states of others and by patterns of cognitive rigidity that may hinder the flexible updating of interpretations and inferences whilst reading (Dindar et al., 2023).

Metacognitive accuracy refers to the correspondence between the readers’ judgements about their performance and their actual performance in reading comprehension tasks (Soto et al., 2018). A high level of accuracy means that the reader can accurately assess their level of understanding, enabling them to make appropriate strategic decisions. Conversely, low accuracy indicates difficulties with self-assessment and the regulation of learning. Research in typically developing populations has shown that metacognitive accuracy is positively correlated with reading performance, prior knowledge, study time and the feedback received (Gutiérrez-de-Blume, 2017).

Micai et al. (2019) used eye-tracking technology to analyse how people with ASD process sentences containing semantic and spelling errors. The authors found that explicit instructions improved their performance, and that sentences containing semantic errors generated a greater number of fixations. These findings reinforce the idea that, whilst they may use reading monitoring strategies, their effectiveness is enhanced in structured contexts with explicit support. In the same spirit, Rebolledo-Luna et al. (2021) compared the metacognitive accuracy of students with ASD with that of their neurotypical peers following a comprehension activity. Although both groups improved their accuracy after reading, students diagnosed with ASD showed a lower degree of correspondence between their judgements and their actual performance. This finding suggests that people diagnosed with ASD face specific challenges in monitoring and calibrating their reading comprehension.

To date, there remains a considerable gap in our understanding of how women diagnosed with ASD approach reading comprehension and to what extent metacognitive variables influence this process, which limits our understanding of both their potential in reading comprehension and the specific challenges and barriers they face throughout their educational journey. In particular, few studies have adopted an integrated approach to examining reading comprehension, the accuracy of metacognitive judgements, and the relationship between metacognitive skills and reading performance in female university students diagnosed with ASD.

In response to this need, this study aims to provide up-to-date empirical evidence on reading comprehension performance and its relationship with metacognitive skills in female university students diagnosed with ASD, and the following specific research objectives were proposed: (1) To describe the levels of reading comprehension, knowledge of cognition and regulation of cognition in female university students diagnosed with ASD; (2) To analyse the accuracy of metacognitive performance judgements, both prospective (before the task) and retrospective (after the task), using measures of absolute calibration; (3) To explore the linear association between the metacognitive dimensions – knowledge and regulation of cognition – and performance in reading comprehension.

METHOD

The study was quantitative in nature (Sánchez-Flores, 2019) and was a non-experimental cross-sectional study (Rodríguez & Mendivelso, 2018). The study had an exploratory-correlational scope. First, the links between variables were analysed; subsequently, simple linear regression models were estimated for exploratory purposes, with a view to examining the strength of the linear association between each metacognitive dimension and reading comprehension performance.

RESULTS

Analysis strategy

The analysis strategy was defined taking into account the exploratory nature of the study and the available sample size. Firstly, descriptive statistics were calculated to characterise the participants’ reading and metacognitive performance. Spearman’s correlations were then used to analyse the associations between variables, as the assumption of normality was not met for some metacognitive dimensions. Finally, independent simple linear regression models were estimated, incorporating one metacognitive dimension per model. Multiple regression models were not estimated due to the small sample size and the risk of overfitting. Consequently, the regression results were interpreted as exploratory evidence of a linear association rather than as confirmatory predictive evidence.

It should be noted that, in order to address the first research objective – which concerned the description of the participants’ performance in reading comprehension and metacognitive skills – the scores obtained from the various instruments were standardised. This was necessary because of the variety of scales used, which made direct comparisons difficult. The results were therefore converted into achievement percentages, with the maximum possible score in each assessed dimension – including absolute calibration – set at 100%; descriptive statistics such as the mean, standard deviation and the minimum and maximum values (range) were then calculated.

Table 1 shows the descriptive statistics for the results achieved in the reading comprehension test. The overall percentile and the percentile achieved in each of the instrument’s dimensions are shown. The results revealed that the dimension which performed best was the textual dimension, whilst the critical dimension performed the worst. Furthermore, the overall test percentile shows a general trend towards normal performance. However, it can be seen that they achieve relatively lower scores on the implicit questions compared with the explicit ones.

Table 1Descriptive results on reading comprehension and metacomprehension 

Statistics	Overall percentile	Textual	Review	Pragmatics	Implicit	Explicit
Mean	73,478	72,565	40,652	65,348	69,261	72,348
Standard deviation	24,445	25,743	25,723	30,394	26,273	28,394
Minimum	6	9	1	1	3	14
Maximum	99	99	99	99	99	99

The results of the self-report on metacomprehension skills are presented in Table 2. The total scores show variation in the level of knowledge and metacognitive regulation. When analysing the sub-dimensions of the instrument, the highest score was observed in the planning dimension, followed by the cognitive knowledge dimension. By contrast, the lowest mean score was observed for regulation of comprehension. The dimensions ‘process evaluation’, ‘outcome evaluation’ and ‘regulation of non-comprehension’ showed intermediate values.

Table 2Results of the self-report on metacomprehension skills 

	KC	PL	ERP	ERO	RC	RNC	Total MI
Mean	16,478	17,522	10,478	11,391	9,652	11,565	77,087
SD	2,372	3,591	1,275	1,500	1,824	1,121	7,211
Min.	10	7	8	7	6	10	62
Max.	19	23	13	13	13	13	92

Note. KC = Knowledge of Cognition; PL = Planning; ERP = Evaluation of the Reading Process; ERO = Evaluation of Reading Outcomes; RC = Regulation of Comprehension; RNC = Regulation of Non-comprehension; Total MI = Total Metacognitive Inventory; SD = Standard Deviation; Min. = Minimum; Max. = Maximum.

To address the second objective, prospective (pre-task) and retrospective (post-task) performance judgements were analysed, using absolute calibration as an indicator. In this study, a simple difference approach was used to estimate the calibration error, which makes it possible to identify biases that overestimate or underestimate performance. In this context, positive values indicate that performance has been overestimated, whilst negative values reflect an underestimation. These results show that performance was underestimated to a greater extent in the judgements made after completing the reading comprehension task.

To assess whether these differences were statistically significant, a Student’s t-test for paired samples was applied, comparing the prospective and retrospective calibration errors. The results showed a statistically significant difference between the two types of judgement, t(22) = 6.423, p < .001. The effect size, estimated using Cohen’s d for paired samples, was large (d = 1.34). However, this result should be interpreted with caution due to the small sample size (n = 23), which could affect the stability of the effect estimate.

Table 3Results for prospective and retrospective errors 

Type of error	Mean	SD	Min.	Max.	d (Cohen)
Prospective	-3,478	34,395	-49,000	94,000	1.34
Retrospective	-16,522	37,396	-69,000	94,000	—

N.B. SD = Standard Deviation; Min. = Minimum; Max. = Maximum; d (Cohen) = effect size for the difference between prospective and retrospective errors (related sample).

For the third research objective: To investigate the linear relationship between metacognitive dimensions – knowledge of cognitive and regulation of cognition – and reading comprehension performance, a Spearman’s rank correlation analysis was first conducted. This type of analysis was chosen because the Shapiro-Wilk test for each variable (which is more suitable for small samples) indicated that the metacognitive variables did not meet the assumption of normality.

Table 4 shows the analysis, which revealed moderate positive associations between various metacognitive dimensions. The relationship between planning and regulation of comprehension (p < .01) is highlighted, as is the relationship between process assessment and outcome assessment in reading (p < .01). Furthermore, several dimensions showed significant associations with the total score on the metacognitive inventory, particularly comprehension regulation (p < .001), suggesting that this skill constitutes a central component within the metacomprehension system. These correlations were not used as a criterion for establishing the absence of multicollinearity in a multivariate model, as no multiple regression with simultaneous predictors was estimated.

Table 4Correlation between variables 

Variable	KC	PL	ERP	ERO	RC	RNC	Total MI
1. CsC	—
2. Pl	-.036	—
3. EPL	.251	.166	—
4. ERL	.460*	-.195	.616**	—
5. RC	.308	.553**	.178	-.022	—
6. RI	-.016	.147	.280	.254	.287	—
7. Total IN	.415*	.623**	.596**	.440*	.742**	.398	—

N.B. * p < .05, ** p < .01, *** p < .001

Note. KC = Knowledge of Cognition; PL = Planning; ERP = Evaluation of the Reading Process; ERO = Evaluation of Reading Outcomes; RC = Regulation of Comprehension; RNC = Regulation of Non-comprehension; Total MI = Total Metacognitive Inventory.

Secondly, given the small sample size of the study (n = 23), no multiple regression models were estimated, nor were any predictor selection procedures carried out. Instead, separate simple linear regression models were estimated, incorporating a single metacognitive dimension per model. This decision was in keeping with the exploratory nature of the study and was intended to estimate the magnitude of the linear association between each metacognitive dimension and reading comprehension performance, whilst avoiding the simultaneous inclusion of multiple predictors, which could increase the risk of overfitting and parameter instability. Consequently, the coefficients obtained should not be interpreted as confirmatory predictive evidence, but rather as preliminary indicators of association that need to be tested in subsequent studies with larger samples.

As no multiple regression model with simultaneous predictors was estimated, no diagnostics for multicollinearity, such as the variance inflation factor or tolerance, were reported. These indicators are relevant when two or more predictors are included together in the same model. In this study, the decision not to estimate multivariate models was due to the small sample size and the exploratory nature of the analysis. Consequently, the simple linear regression models were interpreted as independent estimates of the linear association between each metacognitive dimension and reading performance. Therefore, the interpretation of the models focused on the magnitude of the observed association, the coefficient of determination and statistical significance, whilst maintaining a cautious approach due to the sample size (Tabachnick & Fidell, 2013).

The sample size was not used to support a confirmatory predictive inference, but rather to define the scope of the analyses. As the sample comprised 23 participants, the models were restricted to simple independent linear regressions, thereby avoiding the simultaneous inclusion of multiple predictors. This decision helped to reduce the risk of overfitting and unstable estimates, although it does not eliminate it entirely. Consequently, the coefficients of determination obtained should be interpreted as indicators of the shared variance observed in this specific sample, and not as stable or generalisable estimates of the predictive power of the metacognitive variables. In this study, the coefficients of determination obtained for the various models (see Tables 5 and 6) range from R² = .77 to R² = .85, which corresponds to very large effect sizes according to conventional criteria (f² > 3). However, the results obtained should be interpreted with caution due to the sample size. It must be acknowledged that, due to the small sample size, the parameter estimates may be unstable and the coefficients of determination (R²) may be overestimated. Whilst this phenomenon has been extensively documented in studies with small sample sizes, where models tend to capture variability specific to the sample under analysis (Button et al., 2013), caution should be exercised when interpreting the effect sizes found in this study; these should only be regarded as preliminary evidence that needs to be confirmed by future research involving larger samples.

The results obtained from the simple linear regression models are shown in Tables 5 and 6. In this sample, the total score on the metacognitive inventory showed a strong linear link with reading comprehension performance (R² = .825). This figure indicates that both variables accounted for a significant proportion of the variance; however, it should be interpreted with caution given the small sample size and the exploratory nature of the analysis. Subsequently, the knowledge of cognition dimension was analysed. The model was found to be significant, with a common variance with reading comprehension of 79%.

For the planning dimension, the model was once again significant, with a common variance with the reading comprehension performance of 79%. With regard to the variables used to assess the reading process and evaluate reading outcomes, the models were significant in both cases and share a common variance of 84% with the predicted variable. With regard to the last two proposed predictor variables – regulation of comprehension and regulation of incomprehension – the models were significant. The first predictor variable identified shares 77% of the common variance with the predicted variable, whilst for the regulation of incomprehension, the common variance is 85%. Whilst the dimensions above showed strong linear associations with reading performance, the observed R² values indicate a significant proportion of variance shared between each metacognitive dimension and reading comprehension. However, given the small sample size and the exploratory nature of the study, these results should not be interpreted as evidence of independent or generalisable predictive effects.

Although the estimated models show a statistically significant relationship between metacognitive variables and reading comprehension performance, these results should be interpreted with caution – as noted previously – given that the small sample size could lead to overfitting and a possible inflation of the coefficients of determination (R²), which could result in an overestimation of the magnitude of the observed effects. In order to mitigate this risk, the analyses were carried out using separate simple regression models for each predictor, thereby avoiding the simultaneous inclusion of multiple variables, which could lead to unstable estimates. Consequently, the findings should be interpreted as exploratory evidence regarding the relationship between metacognitive skills and reading comprehension in female university students diagnosed with ASD; the robustness of these findings will need to be tested in future research using larger samples and confirmatory designs.

Table 5Summary of exploratory simple linear regression models for reading comprehension 

Model	Variable	P	R²	Adjusted R²	RMSE
H₁	Total MI	0,908	0,825	0,817	12,264
H₁	KC	0,891	0,794	0,784	13,306
H₁	PL	0,891	0,793	0,784	13,318
H₁	ERP	0,917	0,841	0,833	11,700
H₁	ERO	0,914	0,836	0,828	11,873
H₁	RC	0,879	0,773	0,762	13,912
H₁	RNC	0,922	0,850	0,843	11,355

Note. Total MI = Total Metacognitive Inventory; KC = Knowledge of Cognition; PL = Planning; ERP = Evaluation of the Reading Process; ERO = Evaluation of Reading Outcomes; RC = Regulation of Comprehension; RNC = Regulation of Non-comprehension.

N.B. Each model was estimated independently, incorporating a single metacomprehension dimension as a predictor. The coefficients should be interpreted as exploratory associations observed in this sample, rather than as confirmatory predictive effects.

Table 6Coefficients of the exploratory linear regression models for reading comprehension 

Model	Variable	Unstandardized	Standard error	Standardized	t	p
H₁	Total MI	0,291	0,029	0,225	10,180	8.717×10-10
H₁	KC	1,342	0,146	0,339	9,204	5.336×10-9
H₁	PL	1,239	0,135	0,475	9,194	5.445×10-9
H₁	ERP	2,161	0,201	0,334	10,772	3.069×10-10
H₁	ERO	1,960	0,185	0,311	10,585	4.248×10-10
H₁	RC	2,174	0,251	0,428	8,649	1.580×10-8
H₁	RNC	1,958	0,175	0,236	11,161	1.579×10-10

Note. Total MI = Total Metacognitive Inventory; KC = Knowledge of Cognition; PL = Planning; ERP = Evaluation of the Reading Process; ERO = Evaluation of Reading Outcomes; RC = Regulation of Comprehension; RNC = Regulation of Non-comprehension.

N.B. The coefficients correspond to independent simple linear regression models. Due to the small sample size, the results should be interpreted with caution and require replication in larger samples.

In order to summarise the results obtained from the various regression models estimated, Figure 1 presents a comprehensive overview of the relationship between each of the metacognitive variables analysed and reading comprehension performance. This figure provides a graphical summary of the standardised coefficients obtained from the independent simple linear regression models estimated for each metacognitive dimension.

Figure 1Summary of simple regression models between metacognitive skills and reading comprehension 

DISCUSSION AND CONCLUSIONS

The findings of this research enable us to characterise the reading and metacognitive profiles of female university students diagnosed with ASD. Overall performance in reading comprehension was at a normal level, which contrasts with previous reports; for example, Solari et al. (2019) report that only 31.2 per cent of people on the spectrum achieve an average level in this area. This difference should be interpreted with caution, as the sample in this study consists of female university students with a diagnosis consistent with low levels of support. Therefore, the results may be linked to characteristics of the sample, rather than exclusively to gender. Consequently, future research should examine these variables in greater depth in order to gain a better understanding of the reading profile of women with ASD.

When analysing Lectum’s dimensions, lower performance was identified in the critical dimension. This dimension assesses higher-order skills such as the transfer of information to new contexts, problem-solving based on textual content, making value judgements, engaging in reflective dialogue with the text, and understanding non-conventional meanings, such as metaphors, irony or figurative language (Riffo et al., 2018). Furthermore, lower scores were observed on the implicit questions, which specifically measure inferential skills, the transfer of information and the understanding of metaphors or irony (Riffo et al., 2018). These difficulties may be linked to cognitive characteristics associated with theory of mind (Atkinson et al., 2017; Baron-Cohen et al., 1985) and to a tendency towards a more literal interpretation of language (McIntyre et al., 2017). In this regard, they may face challenges in determining the author’s intentions, recognising and discussing different perspectives, employing interpretative processes such as transferring information gleaned from the text to a different context in order to complete a task, and carrying out complex inferential processes that require cognitive flexibility, sensitivity to the rhetorical use of language, and sound social interpretation.

With regard to metacognitive skills, we follow the guidelines set out by Pintrich et al. (2000), in which various instruments were used to assess aspects of metacomprehension. The results of the Metacomprehension Inventory (MI) revealed a high level of awareness in the moments leading up to reading tasks, suggesting that students with ASD begin reading in an organised manner, demonstrating awareness and use of control mechanisms prior to starting a reading task. It is worth noting that they score lower in the ‘comprehension regulation’ sub-dimension, which assesses the ability to actively monitor their understanding whilst reading – a skill that is fundamental to self-regulation (Schraw & Dennison, 1994). This finding suggests that they may face greater challenges in identifying when they are not understanding what they are reading, which could compromise their ability to apply comprehension strategies used during reading, such as rereading, paraphrasing, elaborative inferences and bridging inferences, amongst others. This is consistent with the findings of Rebolledo-Luna (2023), who notes that people with ASD make limited use of reading comprehension strategies whilst reading.

As for metacognitive calibration skills, whilst it has been suggested in the literature that retrospective judgements tend to be more accurate than prospective ones – given that the reader forms them after completing a reading task, which allows them to adjust their judgement based on direct feedback (Dunlosky et al., 2021) – the findings of this study suggest a different pattern in women diagnosed with ASD. The participants’ judgements showed that they underestimated their performance both before and after; however, in their retrospective judgements, the underestimation is significantly greater. This may reflect lower confidence or a more critical appraisal of their performance. This finding is consistent with previous research which has documented significantly lower levels of overall self-esteem and perceived power in people diagnosed with ASD (Nguyen et al., 2020). Similarly, Strang et al. (2024) reported higher levels of cognitive alexithymia – a difficulty experienced by some people in identifying, understanding and expressing their own emotions (Sifneos, 1973) – and low self-esteem, which could contribute to a less accurate perception of their own abilities.

Exploratory simple linear regression models revealed strong associations between metacognitive skills and reading comprehension performance. In particular, the total score on the metacognitive inventory was associated with a high proportion of shared variance in this sample (R² = .82). However, this result should be interpreted with caution, as the small sample size (n = 23) could lead to unstable estimates and a possible overestimation of the coefficient of determination. For this reason, the results should be regarded as exploratory evidence that needs to be confirmed in larger samples.

The results of the exploratory models suggest that metacognitive skills are significantly associated with the reading performance of female university students diagnosed with ASD. This pattern is consistent with the existing literature on metacomprehension and reading comprehension, although it should not be interpreted as confirmatory evidence due to the sample size and the cross-sectional design of the study. This suggests that a better understanding of the various reading strategies available and when to use them is associated with better performance, which could compensate for some core characteristics of ASD, such as a tendency to interpret language literally, cognitive rigidity and difficulty in making inferences (Kalandadze et al., 2022; McIntyre et al., 2017). Furthermore, the ability to set reading goals and anticipate the content of a text forms part of an essential process that enhances comprehension. Similarly, skills such as pausing, rereading or adjusting one’s strategy when faced with a difficulty are positively linked to reading performance; and, finally, the ability to actively recognise when one has not understood a passage and to take steps to rectify this is essential for reading performance.

In theoretical terms, these findings are consistent with the idea that metacomprehension complements the cognitive component of reading comprehension in women diagnosed with ASD. It is proposed that this skill plays a crucial role in the comprehension integration phase, as it enables readers to select, organise, evaluate and regulate the quality of the mental representation constructed during reading, thereby allowing them to discard irrelevant information and construct a macrostructural representation and a situational model in accordance with a standard of coherence specific to reading.

In particular, the skills needed to monitor comprehension are key to effective reading. According to Carpenter and Williams (2023), people with ASD display reduced and variable metacognitive accuracy, which means that some people may underestimate their level of understanding. This is particularly relevant for women with ASD, who may have reduced automaticity in the use of such metacognitive strategies and require explicit instruction that promotes both an understanding of these strategies and their application in real-life reading situations.

A key limitation of the study relates to the sample size. Although the number of participants is understandable given the specific nature of the population studied – female university students with a formal diagnosis of ASD – the sample size limits the stability of the regression models and the generalisability of the coefficients obtained. For this reason, the regression analyses were restricted to simple linear models, with only one predictor at a time, and the results were interpreted solely as exploratory associations. Future research should replicate these findings using larger samples, include comparison groups, and estimate multivariate models that enable the relative weight of each metacognitive dimension to be assessed. Furthermore, the recruitment process may have been subject to self-selection bias, which could mean that students with a greater interest or higher academic achievement were more likely to participate. Furthermore, the use of self-reported measures to assess metacognitive skills could introduce bias into the information reported by the participants. Furthermore, the absence of a comparison group comprising female university students without a diagnosis of ASD makes it impossible to determine to what extent the observed results are specifically associated with the diagnosis.

The sample excluded certain comorbidities frequently associated with ASD, which could limit its representativeness in terms of the diversity of profiles found across the spectrum. Furthermore, the late diagnosis of the participants could influence variables such as academic self-esteem or perceived competence, which in turn could affect metacognitive calibration. Furthermore, the cross-sectional, non-experimental design of the study prevents the establishment of causal relationships between the variables analysed; the results should therefore be interpreted in terms of association.

The findings suggest a need to promote teaching practices that foster the development of metacognitive skills. Among the strategies that could be implemented are explicit guidance in the university classroom; for example, teaching students to identify gaps in comprehension whilst reading and encouraging the use of strategies that facilitate the construction of local and global coherence within the text, such as rereading, paraphrasing and drawing inferences.

In conclusion, the results suggest a significant association between metacognitive skills – particularly those related to monitoring and adjustment during reading – and reading comprehension performance in female university students diagnosed with ASD. Given the exploratory nature of the study and the small sample size, these findings should be interpreted as preliminary evidence, not as proof of generalisable predictive power.