Adaptation and Psychometric Properties of a Romanian Version of Metacognitive Awareness Inventory for Teachers (RoMAIT)

Abstract

(1) Background: Metacognition plays a crucial role in education, with teachers’ metacognitive awareness being a critical determinant of effective teaching and learning. Despite its importance, validated instruments for assessing teachers’ metacognition are scarce, especially in non-English-speaking contexts. This study aims to adapt and evaluate the psychometric properties of the Metacognitive Awareness Inventory for Teachers (MAIT) to the Romanian cultural and educational contexts. (2) Methods: After rigorous cross-cultural adaptation procedures, including translation, back-translation, and expert review, the MAIT was administered to 444 Romanian teachers (86.5% female, mean age = 47.18 years, SD = 8.92). Exploratory and confirmatory factor analyses were conducted to assess the structural validity of the instrument. (3) Results: A 21 item second-order two-factor model including knowledge of cognition (KC) and regulation of cognition (RC) showed good fit indices (χ²/df = 3.5, CFI = 0.965, TLI = 0.96, SRMR = 0.023, RMSEA = 0.075). Limitations of the study are discussed.

1. Introduction

1.1. Background of This Study

The interest in what is now called metacognition (MC) can be traced back to the commandment “Know thyself!” inscribed on the frontispiece of the temple of Apollo at Delphi in ancient Greece, and early signs of this interest can be found in the writings of Plato and Aristotle (Georghiades, 2004). However, the concept of metacognition was formally coined only in 1976 by John H. Flavell, who defined it as “thinking about thinking” or “cognition about cognitive phenomena” (Flavell, 1976).

Subsequent definitions of MC, e.g., “awareness and management of one’s own thought” (Kuhn & Dean, 2004), “the monitoring and control of thought” (Martinez, 2006), “knowledge or beliefs about what factors or variables act and interact in what ways to affect the course and outcome of cognitive enterprises” (Flavell, 1979), and “the knowledge of cognition and strategies to regulate it” (Muijs & Bokhove, 2020), emphasize the existence of two dimensions of MC, namely the knowledge (awareness, monitoring) of cognition (KC) and the regulation (management, control) of cognition (RC).

According to Schraw and Moshman (1995), knowledge of cognition (KC) refers to “what individuals know about their own cognition or about cognition in general”, while regulation of cognition (RC) refers to “metacognitive activities that help control one’s thinking or learning”. Other authors, e.g., Güner and Erbay (2021) and Tuononen et al. (2023), also use the terms metacognitive awareness for KC and metacognitive skills for RC.

The knowledge of cognition has three main components (Cross & Paris, 1988; Schraw & Moshman, 1995; Schraw et al., 2006) as follows:

Declarative knowledge (DK) is broadly defined as epistemological understanding of the processes associated with thinking.

Procedural knowledge (PK) refers to the management of cognition, including knowledge about certain strategies and heuristics applicable for solving the cognitive tasks.

Conditional knowledge (CK) is the knowledge about when and why to use specific strategies.

RC includes planning, monitoring, and evaluation of the thinking process.

Planning (P) involves goal setting, selection of the appropriate strategies, and defining the temporal sequence of the execution of tasks.

Monitoring (M) is the appraisal of the degree of congruence between the cognitive goal and the current status of the execution of cognitive tasks, and—in case of miscongruence—the adjustment of strategies.

Finally, the evaluation (E) is the global appraisal of the outcome and of the regulatory process and may include revisiting the cognitive goals and strategies.

According to Schraw and Dennison (1994), RC has two more components, namely Information Management Strategies (IM), consisting of “organizing, summarizing, and transforming information to facilitate learning and comprehension”, and Debugging Strategies (DS), which refer to “identifying and correcting errors or obstacles in one’s cognitive process”, but IM and DS are often integrated into the broader components of planning and monitoring.

This model of metacognition—graphically synthesized in Figure 1—is now widely accepted (Craig et al., 2020; Fleur et al., 2021; Georghiades, 2004) and laid the foundation for the development of many of the available instruments used to measure metacognition (Ellis et al., 2012).

The study of metacognition was initially approached from a purely theoretical perspective—see the comprehensive literature reviews (Craig et al., 2020; Fleur et al., 2021; Georghiades, 2004). However, as the research advanced, it became evident that MC plays an important practical role in education as a determinant of self-regulated learning (Schuster et al., 2020) and of learning outcomes (Guo, 2022).

MC is a crucial component in the instruction for the development of higher-order thinking skills (Hamzah et al., 2022), and it is closely linked with some of the so-called 21st-century skills (He et al., 2024; Karatas & Arpaci, 2021), including critical thinking (Ellerton, 2015; Mitsea et al., 2021), creativity (McMillan et al., 2022; Preiss, 2022), and efficient communication (Chaisriya et al., 2023; Maor et al., 2023).

MC is particularly important for teachers too. In order to help students improve their metacognitive awareness and skills, teachers should have a good understanding of how students learn, along with the awareness of their own metacognitive abilities (Prytula, 2012). Teachers act as metacognitive models for their students (Lin et al., 2005), and their metacognitive awareness is directly linked with students’ learning outcomes (Tuononen et al., 2023).

1.2. Measuring Metacognition in Education

The recognition of the importance of MC in education fueled the interest of the researchers in how metacognition can be taught and learned and in how to measure it (Swanson et al., 2024; Veenman, 2016; Zohar & Barzilai, 2013).

The methods of assessment of MC were roughly classified as online (e.g., think-aloud protocols) and offline methods (e.g., self-report questionnaires), depending on when data is collected relative to the actual execution of the cognitive tasks (Saraç & Karakelle, 2012).

For practical reasons, the self-report questionnaires are largely used. Among these instruments, one of the most widely used is the Metacognition Awareness Inventory (MAI) proposed in (Schraw & Dennison, 1994), a scale with 52 items and 8 factors (DK, PK, CK, P, M, E, IM, and DS).

MAI was translated, adapted, and validated in several languages, including Spanish (Gutierrez De Blume et al., 2024), Portuguese (Bártolo-Ribeiro et al., 2020), Turkish (Akin et al., 2007), Chinese (Li et al., 2024), and Vietnamese (Nguyen & Phung, 2021). Two abridged versions of MAI, with 12 and 18 items respectively, were adapted for children (Junior MAI, or JrMAI) (Sperling et al., 2002). For a comprehensive review of the available instruments aimed at measuring metacognition of children aged 4–16 years, see (Gascoine et al., 2017).

In what concerns the assessment of teachers’ metacognition, a relatively popular instrument is the Metacognitive Awareness Inventory for Teachers (MAIT)—a variant of MAI with 24 items, proposed in (Balcikanli, 2011). An 18 item version of MAIT (Kallio et al., 2017) was translated and adapted in Finland. Another carefully designed instrument worth mentioning is the Teacher Metacognition Inventory (TMI), a 28 item scale described in (Jiang et al., 2016).

Self-report measures of metacognition have been criticized (Craig et al., 2020) for a variety of reasons, among which are:

• Subjectivity and social desirability bias;
• Poor relationship with the actual performance of the subjects;
• Limited replicability;
• Insufficient theoretical explanation of the factor choice and lack of comparison between different factor models;
• Inconsistencies in correlations with other measures of the same constructs and with related constructs.

Despite these drawbacks, self-report measures of MC remain useful. After a comprehensive review and meta-analysis of the existing studies, Craig et al. (2020) conclude that “self-reports provide a useful overview of two factors—metacognitive knowledge and metacognitive regulation. However, metacognitive processes as measured by self-report subscales are unclear” (p. 155). Further to this, Craig et al. (2020) argue that self-report analysis “can be used to evaluate general skills of two factors distinguishing knowledge from regulation but cannot adequately measure distinct subcomponents within the two factors” (p. 174). Finally, Craig et al. (2020) concludes that “there is not a single self-report that can be recommended as the industry standard (i.e., reliable and replicable)” (p. 173).

The above-mentioned limitations of the self-report measures of MC are partly compensated by several practical advantages, namely, they can be easily and efficiently administered (with or without support from digital technology), and they are suitable for large sample sizes, allowing the assessment of MC across entire schools with minimum logistical effort. In addition, they require relatively fewer resources and less time to analyze compared to other assessment methods, while of course taking into account and trying to address the limitations mentioned above.

1.3. Objectives

Though the interest of the Romanian researchers in the study of metacognition in education, as measured by the number of publications on this topic, seems to be growing in the past years, we found very few articles describing Romanian versions of the instruments commonly used to measure metacognition. In rare cases, when the participants were proficient in English, the original scales were used (Susnea et al., 2023), but many other reports, e.g., (Ciascai & Haiduc, 2011; Șuteu et al., 2021), do not specify whether the instruments used were translated and validated for the Romanian population.

In this context, the objective of the research described in this work is to fill this gap and provide the Romanian researchers and educators with a simple and reliable tool for measuring teachers’ metacognition. We chose MAIT as the starting point of the adaptation process for its theoretical soundness and relative popularity.

2. Method

2.1. Participants

A total of 444 Romanian teachers volunteered to participate in the study, following an invitation sent by email to over 5000 public schools in rural and urban areas from all provinces of Romania. The survey was anonymous. Only basic demographic information (age, gender, years of professional experience, didactic degree) was asked besides questions of the actual psychometric scale.

Data were collected online, between April and July 2024, by means of the Google Docs platform.

The sample comprised 384 (86.5%) females and 60 (13.5%) males, with ages spanning between 18 and 74 (M = 47.18, SD = 9.97), having between 1 and 51 years of didactic experience (M = 21.38, SD = 10.88). The majority of them (65.3%) had the highest didactic degree, and 75% live and work in urban areas. See Figure 2 and Figure 3 for demographic information on the participants.

Since the present study is part of a more comprehensive research aimed at exploring the connections between metacognition and the occupational well-being of the teachers, several other scales were included in the questionnaire, among which were the scales for self-efficacy and subjective well-being recommended in the OECD framework for data collection and analysis in the research of teachers’ wellbeing (Viac & Fraser, 2020). The description of these scales falls beyond the scope of the present report.

2.2. Procedures

Following the general guidelines of Briskin’s translation model (Ali, 2016), the original MAIT scale in English was translated into Romanian (with permission from the author) by two independent, experienced English teachers. The translations were compared, and the incongruities were discussed and eliminated. The resulting Romanian version was then back-translated into English by a third English teacher and compared with the original English version. Several corrections were made in the panel by the three translators. The researchers then asked five other teachers to read and fill out the final version of the questionnaire and report any issues concerning possible ambiguities and difficulties in understanding the content.

The Romanian version of the scale (presented in

Table A1. The initial Romanian version of MAIT (RO-MAIT).

Factor	Item	Item Text in Romanian for MAIT
DK	MAIT1	Sunt constient(ă) de punctele mele tari si slabe ca profesor.
	MAIT7	Știu de ce competențe ai nevoie pentru a fi un bun profesor.
	MAIT13	Dețin controlul asupra modului în care predau.
	MAIT19	Stiu ce ar trebui să fac la clasă.
PK	MAIT2	Incerc să folosesc la clasă strategii despre care stiu că au funcționat în trecut.
	MAIT8	Am o motivație pentru fiecare din strategiile pe care le folosesc la clasă.
	MAIT14	Sunt constient(ă) de strategiile pe care le folosesc atunci când sunt la clasă.
	MAIT20	Aleg în mod automat strategiile care mă ajută la clasă.
CK	MAIT3	La clasă, îmi folosesc punctele tari ca să le compensez pe cele slabe.
	MAIT9	Mă pot motiva să continui munca la clasă atunci când e nevoie.
	MAIT15	Folosesc strategii diverse în functie de situația cu care mă confrunt la clasă.
	MAIT21	Știu când fiecare din strategiile pe care obișnuiesc să le folosesc va fi cea mai eficienta
P	MAIT4	Imi organizez timpul la clasă cat mai eficient, în așa fel încât să reușesc să mă încadrez în timp.
	MAIT10	Înainte de a începe să predau îmi fixez obiective specifice.
	MAIT16	Obișnuiesc să mă întreb ce materiale voi folosi la clasă în ziua respectivă.
	MAIT22	Îmi organizez timpul în așa fel încât să-mi ating obiectivele.
M	MAIT5	Periodic, mă întreb în timpul orei dacă îmi ating obiectivele fixate.
	MAIT11	Analizez cât de eficiente sunt strategiile pe care le folosesc chiar în timp ce predau la clasă.
	MAIT17	Verific în mod regulat dacă elevii mei înțeleg ce predau.
	MAIT23	În timp ce predau, evaluez cât de bine o fac.
E	MAIT6	Odată ce mi-am terminat ora, mă întreb dacă mi-am îndeplinit obiectivele fixate la început.
	MAIT12	După fiecare oră mă întreb dacă aș fi putut folosi și alte strategii diferite de predare sau evaluare.
	MAIT18	După ce am predat un/o subiect/lecție, mă întreb dacă aș putea fi mai eficient(ă) data viitoare.
	MAIT24	Mă întreb dacă am luat în considerare toate aspectele posibile după predarea unui subiect/unei lecții.

in Appendix A) was reviewed item by item from the perspective of content validity by two experts in education, holding a PhD degree and having relevant experience in advanced educational projects. Both experts indicated the items MAIT7 (I know what skills are more important in order to be a good teacher), MAIT13 (I have control over how well I teach), and MAIT19 (I know what I am expected to teach) as problematic since it may attract responses biased by social conformity (If I am a good teacher, which I certainly am, then I obviously know what skills are needed to be a good teacher).

We marked the items MAIT7, MAIT13, and MAIT19 as potentially problematic, but we included them in the initial version of the questionnaire. The scale was then compiled into a Google Docs questionnaire, and the link was distributed via email in over 5000 schools across rural and urban areas from 35 counties of Romania (out of the total of 41) along with a letter explaining the purposes of the study. A number of 444 teachers volunteered to anonymously participate in the study and filled out the questionnaire. The answers were collected by means of a 5-point Likert scale. Raw data were further processed using Microsoft Excel, SPSS version 20, and Amos Graphics version 23.

The model fit indices for Confirmatory Factor Analysis (CFA) were evaluated according to the guidelines proposed in a previous study (Byrne, 2013; Hu & Bentler, 1999) by comparing the values of the ratio Chi-square/degrees of freedom (χ²/df), Standardized Root Mean Square Residual (SRMR), Root Mean Square Error of Approximation (RMSEA), Tucker–Lewis index (TLI), and Comparative fit index (CFI) with a series of fixed thresholds listed in

Table 1. Threshold values of the main indices in CFA.

Index	Description	Threshold
χ2/df	Chi-square divided by degrees of freedom	≤3
TLI	Tucker–Lewis index	≥0.90
CFI	Comparative fit index	≥0.90
RMSEA	Root Mean Square Error of Approximation	≤0.06
CR	Composite Reliability score	≥0.7
AVE	Average Variance Extracted	≥0.5

.

2.3. Preliminary Data Processing

The analysis of raw data revealed that a number of 43 out of the total 444 (9.68%) responses had missing data. Though relatively high, this proportion is still below the acceptable threshold of 10% suggested in a previous study (Dong & Peng, 2013). Therefore, we preferred to preserve the sample size and used mean substitution for the missing values. Further analysis was performed using the imputed dataset.

In the next step, we compared our data with the available statistics published by the Romanian Ministry of Education for 2023 (Romanian Ministry of Education, 2023) regarding teachers’ age distribution, gender, and work environment (urban/rural). The results are shown in Figure 4 and Figure 5.

The age groups are defined as follows: AG1: under 25, AG2: between 25 and 29, AG3: between 30 and 34, AG4: between 35 and 39, AG5: between 40 and 44, AG6: between 45 and 49, AG7: between 50 and 54, AG8: between 55 and 59, AG9: between 60 and 65, and AG10: over 65 years old.

A χ² goodness-of-fit test was conducted to compare the distributions of age, gender, and work environment in the sample with their corresponding distributions in the population. The results, shown in

Table 2. Results of the χ² goodness-of-fit tests.

Variable	Chi-Square	df	p
Age	63.184	9	0.000
Gender	4.541	1	0.033
Work environment	23.525	1	0.000

, indicate that the null hypothesis must be rejected: the distributions are not identical; thus, the sample is not representative of the population of Romanian teachers.

To improve the representativeness of our sample, we implemented a post-stratification weighting procedure (Cochran, 1977) based on three stratification variables (SV): age (with 10 categories: AG1…AG10), gender (with two categories: MALE, FEMALE), and work environment (with two categories: URBAN, RURAL). This leads to a total number of strata S = 10 × 2 × 2 = 40. For each stratum, we computed weighting factors:

$$w_S={\displaystyle \frac{p_S^{\left(population\right)}}{p_S^{\left(sample\right)}}}$$

(1)

based on the proportions of the respective stratum in the population and in the sample.

Since each individual respondent can be characterized by a unique combination of SVs, we assigned combined weights to each respondent $j\in [1,N]$:

$$w_j=w_{AGi}\times w_{Gi}\times w_{WEi}$$

(2)

$$\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e}AGiϵ\left[{AG}_1\dots .{AG}_{10}\right],G_iϵ\left\{MALE,FEMALE\right\},{WE}_iϵ\{URBAN,RURAL\}$$

In practice, using the full product of these separate ratio weights may lead to excessively large variation in weights, which can inflate the variance of estimators and reduce statistical precision (Gelman, 2007). To moderate this effect, we applied a square-root transformation to the combined weight coefficients. Consequently, each individual’s data in the dataset was multiplied by $\sqrt{w_j}$. The results of the post-stratification process are presented in Figure 6 and Figure 7.

The results of the χ² goodness-of-fit test comparing the distributions of age, gender, and work environment in the sample after post-stratification with their corresponding distributions in the population, shown in

Table 3. Results of the χ² goodness-of-fit tests after post-stratification.

Variable	Chi-Square	df	p
Age	2.611	9	0.978
Gender	0.566	1	0.452
Work environment	0.353	1	0.553

, indicate that the weighted sample is representative of the population of Romanian teachers with respect to the considered SVs. However, we should mention that our sample is not fully representative, because 65% of the respondents have the highest didactic degree. Since there are no official statistics about the proportions of teachers according to the didactic degree in the population, we could not consider this variable for post-stratification.

3. Results

3.1. Item Analysis

The descriptive statistics along with inter-item correlations for the Romanian version of MAIT are presented in

Table 4. Descriptive statistics for the items of the Romanian version of MAIT (RO-MAIT).

	M	SD	Skewness	Kurtosis	Correlations
	Declarative knowledge				MAIT1	MAIT7	MAIT13	MAIT19
MAIT1	4.32	1.27	0.88	1.56	1
MAIT7	4.34	1.25	0.87	1.50	0.834 **	1
MAIT13	4.21	1.23	1.09	2.36	0.819 **	0.871 **	1
MAIT19	4.35	1.29	1.07	1.63	0.740 **	0.784 **	0.800 **	1
	Procedural knowledge				MAIT2	MAIT8	MAIT14	MAIT20
MAIT2	4.10	1.34	0.75	0.88	1
MAIT8	4.26	1.28	0.93	1.55	0.808 **	1
MAIT14	4.27	1.24	0.91	1.38	0.803 **	0.880 **	1
MAIT20	3.99	1.35	0.68	0.84	0.715 **	0.762 **	0.792 **	1
	Conditional knowledge				MAIT3	MAIT9	MAIT15	MAIT21
MAIT3	4.09	1.31	0.86	1.35	1
MAIT9	4.21	1.25	0.96	1.49	0.765 **	1
MAIT15	4.29	1.25	0.78	0.94	0.794 **	0.857 **	1
MAIT21	3.96	1.31	1.03	1.64	0.728 **	0.814 **	0.812 **	1
	Planning				MAIT4	MAIT10	MAIT16	MAIT22
MAIT4	4.23	1.26	0.91	1.48	1
MAIT10	4.18	1.28	0.83	0.87	0.857 **	1
MAIT16	4.04	1.38	0.84	1.19	0.684 **	0.728 **	1
MAIT22	4.15	1.22	0.85	1.03	0.872**	0.872 **	0.715 **	1
	Monitoring				MAIT5	MAIT11	MAIT17	MAIT23
MAIT5	3.73	1.32	0.65	0.78	1
MAIT11	4.08	1.24	0.63	0.64	0.670 **	1
MAIT17	4.34	1.26	0.76	0.87	0.617 **	0.779 **	1
MAIT23	4.05	1.26	0.88	1.05	0.660 **	0.787 **	0.829 **	1
	Evaluating				MAIT6	MAIT12	MAIT18	MAIT24
MAIT6	3.94	1.30	0.90	1.42	1
MAIT12	3.87	1.32	0.90	1.46	0.758 **	1
MAIT18	4.09	1.30	0.82	1.15	0.761 **	0.825 **	1
MAIT24	4.01	1.28	0.80	1.06	0.781 **	0.799 **	0.842 **	1
Valid N (listwise)	444

. Mean values for all items were M = 4.13 and SD = 1.28. These values are higher than those expected for a 5-point Likert scale, indicating a ceiling effect. We found that 22 out of the total of 24 items received the highest possible score (5) from more than 30% of participants, and 8 items received the highest score from more than 50% of the participants. However, the values of skewness and kurtosis for all items remain within the acceptable limits.

Initially, we interpreted the ceiling effect visible in our data as the result of the large proportion of experienced teachers in our sample. However, by computing the correlation between the measure of professional experience of the participants and the items of the scale, we found negative and significant values (r < −0.3, p < 0.01) for all items. This is surprising, because in any profession, most of the indicators of performance tend to improve with experience, as many actions and evaluations gradually shift to system 1 of thinking—a fast and automatic mode of cognition described by Daniel Kahneman (2011).

Our data suggest that metacognition follows an opposite pattern—at least in the case of teachers. The decline of the measure of metacognition may be because actions and evaluations that are performed automatically are less likely to be the subject of metacognitive analysis. This hypothesis is partly supported by the findings in (Khonamri et al., 2024), but additional research is obviously needed to clarify this topic.

The ceiling effect is probably due to a special type of social conformity: Romanian teachers have been, for many years, chronically underpaid. Therefore, most of them enroll in a perpetual competition for a merit-based salary increase awarded to teachers with outstanding professional performances. As a result, whenever they have an opportunity, they tend to overestimate or place extra emphasis on any aspect of their activity that might be seen as a marker of performance. This interpretation remains hypothetical, and further empirical research is needed to confirm it.

In particular, the items MAIT7, MAIT13, and MAIT19, reported as potentially problematic, are indeed among the 8 items with the most pronounced ceiling effect. An even greater issue with these items is that—at least in the Romanian version of the scale—they seem to reflect a construct different from metacognition—namely the respondents’ confidence in their own expertise as teachers. Therefore, we marked these items for deletion.

The ceiling effect remains a significant limitation of the present study, but the value of SD (1.27) suggests that, despite the high value of M and the ceiling effect, there is still a reasonable level of variability in the data allowing for meaningful analysis.

3.2. Factor Analysis

CFA of the original 6-factor model of MAIT (see Figure 8) proposed in (Balcikanli, 2011) indicated a poor fit with our data: CMIN/df = 4.796, CFI = 0.944, TLI = 0.960, SRMR = 0.032, RMSEA = 0.093.

The extreme high values of the correlations between latent factors indicate serious discriminant validity issues, suggesting that the theoretical constructs associated with the factors substantially overlap. Moreover, by removing the items MAIT7, MAIT13, and MAIT19, the factor DK remains with a single item, which is insufficient for adequately capturing or representing the associated construct.

Therefore, we decided to reconsider the factorial structure of the model and conducted an Exploratory Factor Analysis (EFA) using Principal Axis Factoring for extraction and Oblimin rotation.

The dataset was tested for sampling adequacy using the Kaiser–Meyer–Olkin (KMO) test, which produced a value above 0.94. The Bartlett test of sphericity returned χ² = 13123, df = 210, p < 0.001.

Based on eigenvalues greater than 1 (see the scree plot in Figure 9), the EFA extracted two factors that explain 80.12% of the variance in the data. Items MAIT7, MAIT13, and MAIT19 were not considered in EFA.

The resulting model is shown in Figure 10.

Based on the modification indices, we allowed covariances between the residual variables associated with the pairs (MAIT5-MAIT6), (MAIT14-MAIT15), and (MAIT1-MAIT2) because they share unexplained common variance beyond what is accounted for by respective latent factors. This overlap is most likely due to similar wording. (e.g., MAIT5: I ask myself periodically if I meet my teaching goals while I am teaching vs. MAIT6: I ask myself how well I have accomplished my teaching goals once I am finished).

This produced acceptable goodness-of-fit indices for the model: CMIN/df = 3.5, CFI = 0.965, TLI = 0.960, SRMR = 0.023, and RMSEA = 0.075. The reliability indices CR(KC) = 0.984 and CR(RC) = 0.950, and the AVE values are well above the threshold values: AVE(KC) = 0.806, AVE(RC) = 0.759. The values of Cronbach’s alpha were 0.984 for KC and 0.952 for RC.

However, the square root of AVE(RC) is less than the correlation of RC with KC; thus, the discriminant validity is not satisfactory. This is obviously due to the high correlation between KC and RC (r = 0.893).

To solve this issue, we implemented a second-order factor model, wherein a higher-order factor of General Metacognition (MC) accounts for the shared variance between KC and RC, as shown in Figure 11.

The problem of the high correlation between KC and RC could also be solved by considering a single-factor model or a bifactor model. However, the solution based on a second-order model offers a couple of significant advantages. First, it provides a theoretically meaningful explanation for the high interfactor correlation while preserving the distinct nature of the two components. Second, it effectively addresses potential multicollinearity issues by modeling the shared variance through the higher-order MC factor rather than through direct factor correlation, providing a more nuanced understanding of the measurement structure. While KC and RC are distinct in content, they also share a substantial general metacognitive core.

The final form of the instrument for measuring teachers’ metacognition in the Romanian educational context, which has a satisfactory statistical model fit based on our analysis and the implicit modeling decisions previously detailed, is presented in

Table A2. The final Romanian version of MAIT (RO-MAIT).

Factor	Item	Item Text in Romanian for MAIT
KC	MAIT1	Sunt constient(ă) de punctele mele tari si slabe ca profesor.
	MAIT2	Incerc să folosesc la clasă strategii despre care stiu că au funcționat în trecut.
	MAIT3	La clasă, îmi folosesc punctele tari ca să le compensez pe cele slabe.
	MAIT4	Imi organizez timpul la clasă cat mai eficient, în așa fel încât să reușesc să mă încadrez în timp.
	MAIT8	Am o motivație pentru fiecare din strategiile pe care le folosesc la clasă.
	MAIT9	Mă pot motiva să continui munca la clasă atunci când e nevoie.
	MAIT10	Înainte de a începe să predau îmi fixez obiective specifice.
	MAIT11	Analizez cât de eficiente sunt strategiile pe care le folosesc chiar în timp ce predau la clasă.
	MAIT14	Sunt constient(ă) de strategiile pe care le folosesc atunci când sunt la clasă.
	MAIT15	Folosesc strategii diverse în functie de situația cu care mă confrunt la clasă.
	MAIT17	Verific în mod regulat dacă elevii mei înțeleg ce predau.
	MAIT20	Aleg în mod automat strategiile care mă ajută la clasă.
	MAIT21	Știu când fiecare din strategiile pe care obișnuiesc să le folosesc va fi cea mai eficienta
	MAIT22	Îmi organizez timpul în așa fel încât să-mi ating obiectivele.
	MAIT23	În timp ce predau, evaluez cât de bine o fac.
RC	MAIT5	Periodic, mă întreb în timpul orei dacă îmi ating obiectivele fixate.
	MAIT6	Odată ce mi-am terminat ora, mă întreb dacă mi-am îndeplinit obiectivele fixate la început.
	MAIT12	După fiecare oră mă întreb dacă aș fi putut folosi și alte strategii diferite de predare sau evaluare.
	MAIT16	Obișnuiesc să mă întreb ce materiale voi folosi la clasă în ziua respectivă.
	MAIT18	După ce am predat un/o subiect/lecție, mă întreb dacă aș putea fi mai eficient(ă) data viitoare.
	MAIT24	Mă întreb dacă am luat în considerare toate aspectele posibile după predarea unui subiect/unei lecții.

in Appendix B.

4. Discussion

4.1. Related Work

To the best of our knowledge, there are only three previous attempts to translate and/or adapt MAIT in other languages or cultural contexts.

A brief summary of these studies is shown in

Table 5. Previous adaptations of MAIT.

Study	Country	Sample Size	Model Structure	Number of Items
(Kallio et al., 2017)	Finland	N = 208	6-factor, first order	18
(Parsons, 2019)	U.S.A.	N = 282	2-factor, bifactorial	13
(Gutierrez De Blume & Montoya Londoño, 2020)	Colombia	N = 755	6-factor, third order	21

.

Note that none of the existing adaptations of MAIT preserved the original structure of the scale with 6 factors and 24 items.

4.2. Limitations

It is important to highlight the limitations of this study beyond the intrinsic drawbacks of self-report measures of metacognition outlined in a previous study (Craig et al., 2020) such as subjectivity and social desirability bias. First, our sample is not fully representative of the Romanian population of teachers, as it comprises mostly experienced teachers. Further research should replicate this study with more diverse samples. Second, the cultural context that creates pressure for positive self-evaluations among teachers resulted in a visible ceiling effect that creates artificial similarity between items and reduces the generality of the results. The hypothesis that the clustering of the responses towards the upper limit of the scale is due to cultural factors is supported by the fact that it is not visible in the Finnish sample (Kallio et al., 2017), where the mean value of the scores for all items is M = 3.86 (SD = 0.74), but it is even more significant in the Colombian sample (Gutierrez De Blume & Montoya Londoño, 2020) (M = 4.24, SD = 0.73).

Because RoMAIT represents the first attempt to validate a metacognition scale for Romanian teachers, future research should conduct a bilingual pilot—having participants complete both RoMAIT and, for example, the original MAIT or the TMI in English—to test measurement invariance and formally compare factor structures across languages.

The negative correlation between the professional experience and the scores of MAIT is interesting and worth additional attention from researchers. It is clearly not accidental. An independent sample t-test conducted on our sample using experience as a grouping variable revealed that novice teachers have significantly higher scores than experienced teachers for all items, with mean differences ranging from 64% to 95% of the standard deviation (p < 0.001). Further research should investigate this aspect using a different measure of metacognition to determine whether it is due to a measurement artifact or captures a meaningful aspect of how metacognitive awareness evolves with teaching experience.

5. Conclusions

Considering the importance of metacognition in education, the scarcity of the available instruments for measuring teachers’ metacognition is surprising. In this context, the present study contributes to filling this gap and makes a first step towards the development of a reliable instrument adapted for the Romanian teachers.

The Romanian adaptation of the Metacognitive Awareness Inventory for Teachers (RoMAIT), based on a theoretically sound two-factor model of metacognition, demonstrates in the end good psychometric properties and strong reliability indices. This instrument provides Romanian educators and researchers with a culturally adapted, reliable tool for assessing metacognitive awareness in teaching.

Continuing this process of testing and validating measures in many cultural contexts will allow for valid cross-cultural research and comparative analyses across educational systems in the future. We encourage other researchers to continue this process to allow for an in-depth comparative analysis of teachers’ metacognitive awareness and of its relationship to other teacher and student variables.

Despite the aforementioned limitations, RoMAIT can still be used—alongside other validated assessment methods, such as online classroom observation platforms or computer-based tasks capturing real-time metacognitive processes—for a variety of research purposes, including investigating links to educational outcomes, evaluating professional development interventions, or probing how metacognitive awareness shapes instructional decision-making.

Future research should address the limitations of this study and determine whether the remaining imperfections of the scale stem from structural issues or measurement biases.