The Bangla Big Five Inventory-2: a comprehensive psychometric validation

Abstract

The Big Five Inventory-2 (BFI-2) is a widely recognized tool for assessing personality traits across five domains and fifteen facets. However, its psychometric properties in non-Western cultures like Bangladesh remain unexplored. This study aimed to validate the Bangla BFI-2 (BFI-2-B) within a Bangladeshi community sample to provide a culturally adapted personality assessment tool. A cross-sectional survey was conducted on 1,095 participants, where 646 participants (59%; 358 female; Age_mean = 24.25 years, SD = 4.47) passed all attention checks. Participants responded to a demographic questionnaire, Bangla Big Five-2 (BFI-2-B), and Bangla NEO Five-Factor Inventory (NEO FFI). The domain-level structural validity was analyzed using Exploratory Structural Equation Modeling (ESEM). A series of five different latent models were tested by Confirmatory Factor Analysis (CFA) for facet-level structural validity. Full Measurement invariance across gender, education level, and language were tested. The item quality was assessed using Item Response Theory analysis (IRT). Convergent and discriminant validity were assessed by correlating the BFI-2-B domains with the NEO-FFI. The BFI-2-B demonstrated high internal consistency across domains (> 0.70) and facets (> 0.60; except energy level, intellectual curiosity, and respectfulness). ESEM confirmed structural validity at the domain level (CFI & TLI = 0.96). CFA analysis revealed that at the facet level, a three-facet structure with an acquiescence factor yielded the most acceptable fit (CFI& TLI 0.95; RMSEA 0.06; SRMR 0.08). Full measurement invariance was established across gender and educational levels, but only weak invariance was found across languages, indicating linguistic challenges. The similar domains of BFI-2-B and NEO-FFI had strong correlations (r 0.59), and distinct domains exhibited low correlations, indicating strong convergent and discriminant validity. IRT analysis showed that most items had high to moderate discrimination. The BFI-2-B is a reliable and valid tool for assessing personality in the Bangladeshi context, with robust psychometric properties across domains and most facets. Addressing linguistic nuances and testing in more diverse samples can further enhance its cross-cultural applicability.

The stage 1 protocol for this Registered Report was accepted in principle on 24/09/24. The protocol, as accepted by the journal, can be found at: 10.17605/OSF.IO/7DTQG.

Introduction

Personality traits are relatively enduring patterns of thoughts, emotions, and behaviors guiding an individual’s tendency to respond in a certain way under a given circumstance¹. There is a consensus in the literature that personality traits robustly influence our health and wellness², job performance³, and academic achievement⁴. Hence, it is critically important to measure personality traits precisely. In that vein, a major psychometric theory in the personality field- Big Five (BF) theory^5–7 was brought to light, which categorized personality traits into five independent bipolar dimensions: Extraversion, Neuroticism or Negative Emotionality, Agreeableness, Consciousness, and Open mindedness. In addition to this five-domain model of the BF, literature has extended its support to the hierarchical nature of the BF domains with several facets nested under each domain^8–10. The main difference between domain and facet is the degree of conceptual breadth they cover. Each personality trait domain of the BF (e.g., BF-Extraversion) enjoys a higher bandwidth of conceptual breadth and represents a large amount of behavioral information⁷. In contrast, personality trait facets are narrowly defined (e.g., the “sociable” facet under the “BF-Extraversion” domain) and provide a specific description of behaviors.

Rooted in the BF theory, several personality assessment tools were developed, including the NEO Personality Inventory-revised¹¹, International Personality Item Pool-NEO⁸, and Big Five Aspect Scale¹². These long, in-depth measures provide a clear picture of BF domains with incremental information regarding the facets within each domain. However, over the years, researchers faced a major challenge of balancing the content adequacy of the personality measures with the procedure’s length and cost-effectiveness. The Big Five Inventory-1 (BFI-1) was introduced nearly 20 years ago^7,13 to address this. BFI-1 incorporates 44- brief, easy-to-understand trait descriptive phrases representing the BF domains to facilitate coherence, clarity, and efficiency with a substantial reduction in completion time (5–10 min depending on the reader’s ability) compared to other personality measures¹⁴. However, the focus of BFI-1 was limited solely to the domain level of traits.

Recently, BFI-1 received a major revision¹⁴, in which the authors incorporated the hierarchical model of personality traits by including three prominent facet traits under each BF domain. In the new BF Inventory 2 (BFI-2), the item number increased to 60 (15 facets with four items each, nested under five domains) with an equal number of positively and negatively keyed items (both at the domain and facet level) to address the consenting responding style—participants’ tendency of consistently providing agree or disagree responses in the survey items, regardless of the item content. With a more balanced item formulation, BFI-2 offers greater predictive precision¹⁴. Soto and John¹⁴, in their validation study, reported satisfactory psychometric properties of the BFI-2 among a U.S. student sample (n = 470). Despite the increase in length, BFI-2 remains economical compared to other scales¹⁵ and has been widely used in research and applied fields. As of 1 November 2023, the initial BFI-2 paper¹⁴ has been cited 2259 times (Google Scholar). BFI-2 has been translated into more than 30 languages and psychometrically validated in Chinese¹⁵, Turkish¹⁶, Danish¹⁷, Dutch¹⁸, Russian¹⁹, and German²⁰.

It is important to acknowledge that the initial psychometric properties of the BFI-2 were obtained from Western, Educated, Industrialized, Rich, and Democratic (WEIRD) populations. Given the observed differences in personality domain structures and other correlates of personality traits among the personality measures across cultures, using them without proper psychometric validations on non-Western cultures would lead to erroneous interpretation^21,22. To what extent these findings could be generalized to other countries, such as Bangladesh – an Asian country – remains less explored. Zhang, et al.¹⁵, in their Chinese BFI-2 validation with four diverse samples, successfully reproduced the BF domains however, two facets—trust (BF-Agreeableness domain) and intellectual curiosity (BF-Open Mindedness)—exhibited relatively low reliability (n = 765). Ahya and Siaputra²³, in their validation work of Indonesian BFI-2, indicated that only 50 items exhibited satisfactory factor loadings in an exploratory factor analysis. Additionally, the five-domain structure did not yield adequate fit through confirmatory factor analysis. However, the facet-level model was successfully reproduced. Altogether, the findings of the adaptation work of BFI-2 in Asian countries indicate the necessity of psychometrically validating the BFI-2 in the Bangladeshi context to increase the generalizability and precision of this measurement. Moreover, there is a lack of published psychometrically valid scales in Bangla to measure personality. Only a Bangla ten-item Personality Inventory²⁴ is available in the published literature, developed based on BFI-1 and employing only two items per domain. Though short measures are sometimes handy, a full-length measure of personality, such as BFI-2, which accounts for both domains and facets, would facilitate an in-depth understanding of the personality of the Bangladeshi community people. Thus, in this paper, we took the initiative to adapt and psychometrically validate the BFI-2 in Bangla.

The BFI-1, which has been widely used globally for the last 20 years in various cultural and linguistic settings, is reported to be susceptible to gender differences, where women tend to score higher than men in some trait domains^25–27. Further, Dahmann and Anger²⁸ found that low education level led to more extraversion and less emotional stability among young German adults, indicating the susceptibility of personality domain to education level. These measurement invariance (MI) issues across gender and education levels in BF-1 have been studied robustly^25,28. However, for the BFI-2, measurement invariance (MI) has been investigated mostly cross-culturally (English vs Native language): U.S. student and internet validation sample (n = 1470) vs. Chinese sample (n = 1718)¹⁵, and U.S. internet validation sample (n = 1000) vs. German sample (n = 1338)²⁰. Research on measurement invariance (MI) focusing on gender, education level, and language (Bangla vs. English) invariance for BFI-2 is currently lacking but highly recommended.

Validation studies of BFI-2 following factor analytic strategies (exploratory and confirmatory) conducted over the years could now be considered suboptimal methods. Many studies used Principal Component Analysis (PCA) to report the domain-level factor structure of BFI-2^14,19,29. Though commonly used, PCA does not differentiate between common variance (variance due to the hypothesized factors) and unique variance (variance attributed to the respondents’ characteristics). Hence, PCA is less suited to examine the factor structure of personality measures. Furthermore, conventional reporting of Exploratory Factor Analysis (EFA) lacks specific goodness-of-fit criteria and follows some rules of thumb to investigate the factor structure. Factor structures obtained by these conventional EFA are often susceptible to sample size and do not replicate easily³⁰. Hence, it became a common practice in the personality domain to employ Confirmatory Factor Analysis (CFA), which focuses on validating a hypothesized factor structure³⁰. Yet reproducing the factor structure of personality inventories, including BF, using the CFA method remains controversial^30,31. The invariance of factor structure across different populations could be attributed to the tenuous simple structure assumption of the CFA analysis strategy, which typically requires items of BF measures to be trait-specific. This issue could be avoided by employing the Exploratory Structure Equation Modeling (ESEM)³².

ESEM accounts for non-trivial cross-loading consistently found in personality measurements³³. It’s encouraging to observe that recent adaptation studies utilize the ESEM technique to investigate the hypothesized factor structure, especially the domain-level structure of BFI-2^15,20,34,35. However, individual item quality assessment of BFI-2 has scarcely been conducted using Item Response Theory (IRT). In contrast to Classical Test Theory (CTT) approaches (ESEM, factor analysis) that assume all items’ contributions are equal, IRT acknowledges that some items could be more difficult than others and considers the interaction of item difficulty and personal characteristics (latent traits) of the respondents while measuring psychological constructs, i.e., personality. By assessing item difficulty, item discrimination, item bias, and item information carried across the latent trait, IRT-based analysis supplements the CTT-based analyses to increase the precision and accuracy of a measure. Soto and John³⁶ employed IRT-based analysis to develop short and extra-short forms of the BFI-2. However, a full-scale examination of the BFI-2 using IRT principles is highly required.

This study aims to validate the Big Five Inventory-2 among Bangladeshi community samples. In that pursuit, we seek answers to four questions: (Q1) What is the latent structure of the Bangla BFI-2 (BFI-2-B)? (Q2) What is the extent of measurement invariance of the BFI-2-B across gender, education level, and language (English vs Bangla language)? (Q3) What is the quality of items in the BFI-2-B when assessed using Item Response Theory (IRT)? (Q4) What is the convergent and discriminant validity of the BFI-2-B? The Table 1 summarizes the research questions addressed in this study. To answer these four questions, we set four specific objectives for this study. First, we will investigate the structural validity of the domain-level and facet-level structure of the Bangla BFI-2 (BFI-2-B; Q1). We hypothesize that the domain level and facet level structure of BFI-2 would be reproducible in the Bangladeshi community sample (H1). Second, we will investigate the measurement invariance of the BFI-2-B across gender, education level, and language (Q2). We hypothesize that full invariance of the BFI-2-B would be established for gender, education, and language (H2). Third, we will employ item response theory (IRT) based analysis to assess the item quality of the BFI-2-B (Q3). Lastly (Q4), we will gather convergent and discriminant validity of the BFI-2-B by calculating correlations between the big five domains of personality measured using the BFI-2-B and the Bangla-translated NEO Five-Factor Inventory (NEO-FFI)³⁷. Previous studies^14,38 while establishing the measurement properties of BFI-2, reported strong correlations ( between the similar domains of BFI-2 and NEO-FFI (convergent validity, e.g., BFI-2 Extraversion with NEO-FFI Extraversion) and very weak to moderate correlations between distinct domains (discriminant validity, e.g., BFI-2 Extraversion with NEO-FFI Conscientiousness). Hence, we hypothesize strong (≥ 0.50) correlations between similar domains (showing convergent validity) and discriminant validity (correlations between distinct domains) would be lower than the convergent validity of BFI-2-B and Bangla-translated NEO-FFI (H3).

Table 1.

Design table.

Question	Hypothesis (if applicable)	Sampling plan	Analysis plan	Interpretation given to different outcomes
Q.1. What is the latent structure of Bangla BFI-2?	H1: The domain level and facet level structure of BFI-2 would be reproducible in the Bangladeshi community sample	For structural equation modeling framework-based analysis it is to have 10 participants per item. BFI-2-B has 60 items, at least 600 participants are required. Monte Carlo Simulation studies suggested a sample size of 200–500 to evaluate the obtained result safely when analyzing ordinal data with the “Weighted Least Square with mean and variance” (WLSMV) estimator. Hence, we target to attain a sample size of 600. Data collection will be stopped once we have 600 complete data	We would investigate the structural validity of the BFI-2-B both at the domain and facet levels. The domain-level structural validity of BFI-2-B would be investigated using Exploratory Structural Analysis Modeling (ESEM). Facet-level structural validity of each domain, we would employ confirmatory factor analysis (CFA) with the “Weighted Least Square with mean and variance” (WSMV) estimator	To assess the model fit in ESEM and CFA we would follow the popular suggestions of Hu and Bentler51: Comparative fit index (CFI) and the Tucker Lewis index (TLI): good fit 0.95, acceptable fit 0.90); the root means square error of approximation (RMSEA): good fit < .06, acceptable fit < .08; and the standardized root mean square (SRMR) good fit < 0.08, acceptable fit < 0.10
Q.2. What is the extent of measurement invariance of the BFI-2-B across gender, education level and language (English Vs. Bangla)?	H2: A full measurement invariance across gender, education level, and language would be observed	We would follow the sampling plan of Q1 given that the analyses strategies fall under same family	ESEM-based five-domain model would be used investigate the full measurement invariance using the 13-model taxonomy of Marsh, et al.52	To assess measurement invariance, we would follow the suggestions of Cheung and Rensvold55, where invariance between two models would be indicated by ΔRMSEA 0.01 and ΔCFA =-0.01
Q.3. What is the quality of items in BFI-2-B when assessed using Item Response Theory (IRT)?	Not applicable	Sampling adequacy for IRT analysis was analyzed using a Monte Carlo Simulation using the “SimDesign” package (Chalmers & Adkins, 2020) with a sample size varying from 100–600. IRT models at the facet level with 4 items was considered the average root mean squared error (RMSE) was calculated to estimate the optimal sample size for the graded response model. The RMSE became stable with a range of 1.34–1.39 for N = 500 to 600. Our planned sample size exceeds the requirement	We would subject each facet to item response theory-based item analysis. For each facet, we would fit a graded response-based IRT model57 with a marginal maximum likelihood estimation method with the MHRM algorithm	We would report the item difficulty and item discrimination and categorize the items based on discriminatory power following the guidelines of Baker59: none = 0, very low = 0.01–0.34, low = 0.35–0.64, moderate = 0.65–1.34, high = 1.35–1.69, very high > 1.70. The test information curve will be inspected to identify the range of latent continuum () with the highest information and least standard error of measurement. We will estimate and report the marginal reliability coefficients for the fitted IRT models
Q.4. What is the convergent and discriminant validity of the BFI-2-B	H3: We expect a strong (≥ 0.50) correlation for convergent validity. Convergent validity is expected to be higher than discriminant validity (H3)	A power analysis with 80% power with alpha = 0.05, at least 82 participants is required for a person-product moment correlation analysis. Our planned sample size exceeds the requirement	Convergent validity: Correlation of similar domains of BFI-2 and NEO-FFI, e.g., BFI-2 Extroversion and Neo Extroversion Discriminant validity: Correlation of distinct domains of BFI-2 and NEO-FFI, e.g., BFI-2 Extraversion with NEO-FFI Conscientiousness	We would check the statistical significance (p < 0.05), directionality and strength of the correlations. We expected that the correlation of similar domains would be strong () and higher than the correlations of distinct domains

Methods

Ethics information

The project has obtained ethics clearance from the Monash University Human Research Ethics Committee (Project ID: 41934). All methods were performed in accordance with the relevant guidelines and regulations. Digital informed consent from all participants were recorded at the beginning of the online survey. Participation was voluntary, and participants could withdraw from participation at any time without being impacted negatively. Participants received a gift worth 3 USD and feedback on their personality scores for their contribution.

Design

Materials

Demographic questionnaire. The demographic questionnaire asked about participants’ age, gender, socio-economic status, religion, marital status, occupation, and education level.

Big Five Inventory- 2. The Big Five Inventory-2 (BFI-2)¹⁴ is a 60-item scale incorporating the hierarchical structure with 15 facets nested under 5 major personality domains: BF-Extraversion, BF-Negative Emotionality, BF-Agreeableness, BF-Conscientiousness, and BF-Open Mindedness. These 60 items are short, easily understood (e.g., “Is outgoing, sociable”, “is relaxed, handles stress well”) and use a 5-point Likert type response format (1 = Disagree strongly; 2 = Disagree a little; 3 = Neutral; no opinion; 4 = Agree with a little; 5 = Agree strongly). Soto and John¹⁴ reported satisfactory reliability (0.81–0.90 for domains and 0.59–0.86 for facets across the internet validation sample-1000 adults and student validation sample-470 students).

Big Five Inventory-2-Bangla. The BFI-2-B was developed by translating and adapting BFI-2 items using the International Test Commission (ITC) guidelines³⁹. Three bilingual scholars, natives of Bangladesh, conducted the forward translation of the BFI-2 to Bangla. The translated versions were judged and synthesized by the authors. Subsequently, three bilingual scholars back-translated the Bangla scale into English. The authors again synthesized the three back-translations, compared them with the original scale, and made necessary amendments. Twelve mental health professionals assessed the content validity of the BFI-2-B by independently evaluating the relevance of the 60 items using a 4-point Likert type scale (1: not at all relevant, 2: slightly relevant, 3: quite Relevant, 4: Highly Relevant). The item-level content validity (I-CVI) and scale-level content validity index (S-CVI) were estimated as content validity indicators. Items with an I-CVI score > 0.83 and S-CVI > 0.90 indicate adequate content validity^40,41. Eight items were below the cut-off value. These eight items were readjusted and re-evaluated by the same experts. After adjustment, I-CVI scores of all items were acceptable. The S-CVI for the total scale was 0.94, estimated using the average method, and indicated satisfactory content validity^40,41.

NEO Five-Factor Inventory. The NEO Five-Factor Inventory (NEO FFI)³⁷ is a 60-item scale that measures personality using the same five domains as BFI-2: NEO-Extraversion, NEO-Negative Emotionality, NEO-Agreeableness, NEO-Conscientiousness and NEO-Open Mindedness. NEO-FFI measures each of the five domains using 12 items with a 5-point Likert-type scale (1 = Disagree strongly; 2 = Disagree a little; 3 = Neutral; no opinion; 4 = Agree a little; 5 = Agree strongly). We translated the NEO-FFI into Bangla using the forward–backward translation method and used it in the current study.

Procedure

We first ran a pilot (N = 50) study to double-check the questionnaire translation and debug the R scripts for analysis. After this, we launched a quantitative, cross-sectional, anonymous online survey. Participants were recruited using social media platforms. On the online survey’s landing page, explanatory statements contained information on inclusion and exclusion criteria and data confidentiality statements. It was mentioned in the explanatory statement that their participation would be voluntary and that they could withdraw from participation at any time without being impacted negatively. If happy with the explanatory statements, participants agreed to participate (offered consent online). Participants first responded to a demographic questionnaire. Second, they completed the 60-item BFI-2 and, finally, responded to the 60-item NEO-FFI-B. The survey required around 20–25 min to complete. Participants received a gift worth 3 USD and feedback on their personality scores for their contribution.

Sampling

As a rule of thumb, 10 participants per item were required for analysis based on the structural equation modeling framework^42,43. Since BFI-2-B has 60 items, at least 600 participants were required. Monte Carlo Simulation studies suggested a sample size of 200–500 to evaluate the obtained result safely when analyzing ordinal data with the Weighted Least Square with Mean and Variance (WLSMV) estimator. Further, we assessed the sampling adequacy for IRT analysis using a Monte Carlo Simulation using the “SimDesign” package⁴⁴ with a sample size varying from 100–600. We simulated the IRT models at the facet level with 4 items. We calculated the average root mean squared error (RMSE) to estimate the optimal sample size for the graded response model. The RMSE became stable with a range of 1.38–1.40 for N = 500 to 600. Lastly, a power analysis using G*Power⁴⁵ indicated that 82 participants are required for a significant correlation of medium effect (r = 0.3;α = 0.05) with 80% statistical power. Our planned sample size well exceeded the requirement. We launched an online survey based on these guidelines to collect data from 600 participants at least. Four items for attention checks were included in the survey to ensure that participants are attentive while they respond to the survey (e.g., This is an attention check. Please select “Strongly disagree”). Once we launched the survey, we checked weekly for completeness (100% completeness with all attention checks correctly answered). We planned to stop data collection once we had at least 600 participants and a 100% completion rate. In case of excess data, we incorporated them in our analysis.

Exclusion and inclusion criteria

Any Bangladeshi national living in Bangladesh, aged over 18, who could read and write the Bangla language were eligible to participate in the survey. Bangladeshi nationals who did not permanently reside in Bangladesh (settled overseas) were excluded due to their exposure to multiple social norms and cultures. Participants who failed the attention check were excluded. Table 1 summarizes the design of the study.

Analysis plan

We used R⁴⁶ with RStudio for all the analysis. Several statistical packages, including psych⁴⁷, esemComp⁴⁸, and lavaan⁴⁹ were used. Figure 1 summarizes the data analysis steps.

Fig. 1.

Data analysis steps.

Descriptive statistics and reliability analysis

We reported the descriptive statistics and estimated the reliability of each domain and facet of the BFI-2-B in our sample using the psych⁴⁷ package. We reported Cronbach’s α and McDonald’s ω_t coefficient for reliability estimates.

Domain and facet level structural validity

We investigated the structural validity of the BFI-2-B at both the domain and facet levels. The domain-level structural validity of the BFI-2-B was investigated using Exploratory Structural Equation Modeling (ESEM) using esemComp⁴⁸ package. ESEM is a substantive-methodological synergy that integrates the computational power of confirmatory factor analysis and exploratory factor analysis^32,50. We examined whether the five BFI-2-B domains could be recovered in our sample at the item level (each item serves as an indicator of the Big Five domain). We allowed all items to load on all five domains using a rotated target matrix (60 × 5) where the primary loading’s positions were left unspecified, and the cross-loadings were set to zero. To assess the model fit, followed the popular suggestions of Hu and Bentler⁵¹: Comparative fit index (CFI) and the Tucker Lewis index (TLI): good fit 0.95, acceptable fit 0.90); the root means square error of approximation (RMSEA): good fit < 0.06, acceptable fit < 0.08; and the standardized root mean square (SRMR) good fit < 0.08, acceptable fit < 0.10.

To investigate the facet-level structural validity of each domain, we employed confirmatory factor analysis (CFA) with the Weighted Least Square with Mean and Variance (WSMV) estimator. lavaan⁴⁹ was used to run the CFA. Here, we tested the five models that Soto and John¹⁴ used in their study to investigate the facet level structure of BFI-2: (M1) Single Domain Structure- all 12 items of a domain were allowed to load on a single factor; (M2) Single Domain with acquiescence- in addition to the single domain factor an acquiescence method factor were added. In line with their study, we constrained the factor loadings of all items on the acquiescence factor equal to 1, making both true-keyed and false-keyed items load in the same direction. To ensure the proper representation of individual differences that were caused by response style but not due to personality content, the acquiescence factor was not allowed to correlate with domain factors; (M3) positive and negative items—two correlated factors where true keyed items were loaded on one factor, and false-keyed were loaded on another; (M4) the three facets model- representing the three facets within each Big Five Domain; (M5) three facets plus acquiescence model- An acquiescence factor were added to the three facets model. In this model, each item was allowed to load on its corresponding facet and the acquiescence method factor. All factor loadings in the acquiescence factor were constrained to 1, and no correlation will be allowed between the acquiescence factor and the three facets. To assess the model fit, we followed the aforementioned guidelines of Hu and Bentler⁵¹.

Measurement invariance

We extended the ESEM-based five-domain model to investigate the full measurement invariance using the 13-model taxonomy of Marsh, et al.⁵² across gender, educational level, and language (English vs. Bangla). lavaan⁴⁹ was used to run the measurement invariance analyses. To assess the invariance across Bangla and English versions of the test, we used the internet sample (N = 1000) collected by the third author and used to finalize BFI-2 items, their domain, and facet structures¹⁴. We compared that data with the data accumulated from Bangladeshi participants using the Bangla version of the test. For gender, language, and education level full measurement invariance testing, we used the ESEM paradigm, which allowed us to compare the invariance using raw data, thus accounting for individual item functioning. Typically, measurement invariance investigation incorporates a sequence of model fit, starting with the least restricted model—configural invariance with no parameter invariance constraints⁵³. Next, the weak measurement invariance model constrains factor loadings to be equal across the groups. Strong measurement invariance assumes the indicator means and factor loadings to be invariant over groups. Strict measurement invariance constrains the factor loadings, indicator means, and item uniqueness to be equal over the groups. Marsh, et al.⁵⁴ proposed a 13-model taxonomy that expands this four-model-based invariance testing by integrating both factorial and measurement invariance techniques where the sequence of testing propagates from configural invariance to a model of full invariance (Complete factorial invariance) that posits strict invariance as well as the invariance of latent mean and factor variance–covariance matrix. At least, strict measurement invariance would be required to compare the BFI-2-B scale’s scores across different attributes such as gender, education levels, and language (For a more extended discussion of the 13 model taxonomy, please see Marsh, et al.⁵²). To assess measurement invariance, we followed the suggestions of Cheung and Rensvold⁵⁵, where invariance between two models would be indicated by ΔRMSEA 0.01 and ΔCFA .

Item response theory-based analysis

We subjected each facet to item response theory (IRT)-based item analysis. mirt⁵⁶ package was used to run the IRT-based item analysis. We fit a graded response-based IRT model⁵⁷ for each facet using the marginal maximum likelihood estimation method with the MHRM algorithm. We gathered information on item difficulty, discrimination, and test information. We assessed the local fit of the items using infit, outfit, S- and S- associated RMSEA statistics. Person fit was evaluated using statistics⁵⁸. < -2 was considered a misfit for each facet. We reported the item difficulty and item discrimination and categorized the items based on discriminatory power following the guidelines of Baker⁵⁹ none = 0, very low = 0.01–0.34, low = 0.35–0.64, moderate = 0.65–1.34, high = 1.35–1.69, very high > 1.70. The overall test information curves were inspected to identify the range of latent continuum () with the highest information and least standard error of measurement. We estimated marginal reliability coefficients for the fitted IRT models.

Convergent and discriminant validity

Lastly, we investigated the convergent and discriminant validity of the BFI-2-B by using Pearson’s partial product-moment correlation between the scores of BFI-2-B and the Bangla-translated NEO-FFI, controlling for the gender and age of the participants. We expected strong (≥ 0.50) correlations between similar domains for convergent validity. Discriminant validity (correlations between distinct domains) was expected to be lower than convergent validity.

Deviations from the preregistered plan

We acknowledge two deviations from the preregistered plan. First, due to technological limitations, we aggregated item scores for each facet and treated the facet scores as continuous indicators while fitting the ESEM model for full invariance testing, instead of analyzing item-level data as planned. Second, the pilot data, collected before Stage 1 Acceptance of the manuscript, to evaluate the feasibility of the plan and the accuracy of the analysis code as requested by reviewers, were included in the full study sample rather than reported separately as initially planned.

Results

Pilot

We conducted a pilot study with 50 participants to double-check the questionnaire translation and debug the R scripts for analysis. The entire survey took approximately 20 min to complete. To collect data from 50 participants with a 100% completion rate (attention check items included), we had to record data from 57 participants (87.72% success rate). Among the 50 participants, 28 were male (Age_mean = 22.68; Age_SD = 1.83), and 22 were female (Age: Mean = 23.59; SD = 4.24). The majority were undergraduate students (98%), Muslim (98%) and unmarried (96%). In the pilot, the reliability of the five domains of BF were extraversion: Cronbach’s α = 0.76, McDonald’s ω_t = 0.78; agreeableness: Cronbach’s α = 0.8, McDonald’s ω_t = 0.80; conscientiousness: Cronbach’s α = 0.76, McDonald’s ω_t = 0.78; negative emotionality: Cronbach’s α = 0.83, McDonald’s ω_t = 0.84; open mindedness: Cronbach’s α = 0.80, McDonald’s ω_t = 0.81. All analysis codes were executable. The pilot data indicated that the proposed study is realistic and feasible.

Characteristics of study sample

We have collected 1095 participants’ responses (including the pilot sample), resulting in complete responses (100% completion) of 646 participants who passed all four attention checks. Table 2 summarizes the characteristics of the current sample. Among the participants, 358 were females (Age_mean = 24.59 years; Age_SD = 4.69) and 288 males (Age_mean = 23.33 years; Age_SD = 4.14). Most participants were Muslim (91%), with similar proportions across genders (92% for females and 91% for males); 82% were unmarried, and 84% were university students. Participants with government jobs represented 13% of the sample, while those in business or private sector jobs were less common (1.7% and 1.2%, respectively). The mean score of participants’ social stance (an individual’s perceived social position in a social hierarchy measured by a single-item 10-point Likert scale: 1 for the lowest and 10 for the highest social position) was 6.65 (SD = 1.63), with females showing a slightly higher mean (6.72, SD = 1.65) than males (6.56, SD = 1.61).

Table 2.

Characteristics of study sample (N = 646).

Variable	Overall, N = 6461	Female, N = 3581	Male, N = 2881
Age	24.25 (4.47)	24.59 (4.69)	23.83 (4.14)
Religion
Islam	590 (91%)	329 (92%)	261 (91%)
Hindu	47 (7.3%)	23 (6.4%)	24 (8.3%)
Christian	5 (0.8%)	5 (1.4%)	0 (0%)
Others	3 (0.5%)	0 (0%)	3 (1.0%)
Budda	1 (0.2%)	1 (0.3%)	0 (0%)
Marital status
Unmarried	530 (82%)	272 (76%)	258 (90%)
Married	113 (17%)	84 (23%)	29 (10%)
Divorced	3 (0.5%)	2 (0.6%)	1 (0.3%)
Profession
Student	544 (84%)	285 (80%)	259 (90%)
Government job	83 (13%)	63 (18%)	20 (6.9%)
Business	11 (1.7%)	6 (1.7%)	5 (1.7%)
Private job	8 (1.2%)	4 (1.1%)	4 (1.4%)
Education
Undergraduate	555 (86%)	299 (84%)	256 (89%)
Post-graduate	91 (14%)	59 (16%)	32 (11%)
Perceived social status	6.65 (1.63)	6.72 (1.65)	6.56 (1.61)

¹Mean (SD); n (%)

Characteristics of internet sample collected earlier

The internet validation sample (N = 1000) gathered for the development of BFI-2¹⁴ earlier (English version) was compared with our sample (completed the Bengali version) to investigate the measurement invariance for language. 50% of that sample was female. Their age ranged from 18 to 74 (Mean = 28.73 years, SD = 11.68), with most (65%) under 30. They were mostly White/Caucasian (66%) and residing in the United States (82%). A detailed description of the sample is published elsewhere¹⁴.

Descriptive statistics and reliability analysis

Tables 3 and 4 present the descriptive statistics of BFI-2-B. All items violated the normality assumption, due to most participants agreeing with more desirable items and disagreeing with undesirable items. The corrected item-total correlations ranged between 0.12–0.57 (Mean = 0.35; SD = 0.10). Table 4 provides the descriptives of BFI-2-B at the domain and facet level with mean-level gender differences within each sample. Women tend to describe themselves as somewhat more agreeable, conscientious, and emotional (for all instances, p < 0.05). At the facet level, women described themselves as more compassionate, trustworthy, organized, anxious, depressed, and emotionally volatile than men (for all instances, p < 0.05). Men described themselves as more intellectual and creative (for all instances, p < 0.05).

Table 3.

Descriptives of BFI-2-B items.

Items	Mean	SD	Skew	Kurtosis	Normality	Corrected item total correlation
BFI01	3.52	1.01	− 0.71	− 0.13	0.85*	0.3
BFI06	3.63	0.89	− 0.69	0.56	0.86*	0.45
RBFI11	3.17	1.1	− 0.11	− 1.01	0.89*	0.15
RBFI16	2.64	1.25	0.37	− 1.01	0.88*	0.17
BFI21	2.7	1.13	0.26	− 0.93	0.89*	0.17
RBFI26	3.27	1.12	− 0.19	− 1.05	0.88*	0.37
RBFI31	2.02	0.87	1.22	1.71	0.76*	0.22
RBFI36	2.85	1.07	0.14	− 0.93	0.89*	0.35
BFI41	3.28	1.01	− 0.18	− 0.77	0.90*	0.54
BFI46	3.47	1.12	− 0.51	− 0.59	0.88*	0.25
RBFI51	2.91	1.01	0.18	− 0.66	0.90*	0.31
BFI56	3.72	0.81	− 0.79	0.78	0.82*	0.48
BFI02	4.17	0.75	− 1.11	2.42	0.77*	0.22
BFI07	4.44	0.64	− 1.51	5.24	0.68*	0.31
RBFI12	3.48	1.13	− 0.3	− 1.01	0.88*	0.31
RBFI17	3.99	1.09	− 1.26	0.94	0.78*	0.31
RBFI22	4.1	0.95	− 0.98	0.3	0.80*	0.35
BFI27	4.02	0.9	− 1.05	1.17	0.81*	0.28
BFI32	4.02	0.76	− 0.77	1.03	0.81*	0.38
RBFI37	3.16	1.14	− 0.01	− 1.2	0.87*	0.38
RBFI42	2.64	1.08	0.41	− 0.74	0.88*	0.33
RBFI47	3.99	1.02	− 1.02	0.5	0.82*	0.4
BFI52	4.29	0.58	− 0.47	1.56	0.72*	0.38
BFI57	3.64	0.98	− 0.69	− 0.06	0.85*	0.34
RBFI03	3.34	1.21	− 0.33	− 1.02	0.88*	0.43
RBFI08	2.82	1.22	0.15	− 1.09	0.90*	0.57
BFI13	3.67	0.97	− 0.79	0.26	0.85*	0.19
BFI18	3.59	1.05	− 0.7	− 0.15	0.86*	0.36
RBFI23	2.98	1.16	− 0.06	− 1.1	0.89*	0.5
RBFI28	2.28	0.93	1.00	0.69	0.79*	0.29
BFI33	3.55	1.07	− 0.52	− 0.51	0.88*	0.46
BFI38	3.71	0.77	− 0.65	0.63	0.82*	0.49
BFI43	4.2	0.75	− 0.97	1.42	0.78*	0.34
RBFI48	3.84	1.1	− 0.86	− 0.05	0.84*	0.4
BFI53	3.66	0.99	− 0.71	0.03	0.86*	0.42
RBFI58	2.83	1.11	0.28	− 1.05	0.87*	0.39
RBFI04	2.75	1.11	0.27	− 0.73	0.91*	0.46
RBFI09	2.36	0.99	0.73	0.08	0.86*	0.46
BFI14	2.98	1.21	0.09	− 1.13	0.89*	0.35
BFI19	2.95	1.11	0.07	− 0.99	0.90*	0.44
RBFI24	2.41	1.09	0.67	− 0.24	0.87*	0.41
RBFI29	2.74	1.09	0.51	− 0.65	0.87*	0.49
BFI34	3.62	1.14	− 0.53	− 0.72	0.87*	0.45
BFI39	3.69	1.05	− 0.64	− 0.38	0.86*	0.34
RBFI44	2.45	1.06	0.62	− 0.35	0.87*	0.42
RBFI49	3.24	1.15	− 0.18	− 1.04	0.89*	0.26
BFI54	3.53	1.12	− 0.59	− 0.57	0.87*	0.41
BFI59	2.87	1.19	0.21	− 1.08	0.89*	0.52
RBFI05	3.26	1.23	− 0.19	− 1.15	0.89*	0.25
BFI10	4.03	0.8	− 0.79	0.7	0.82*	0.29
BFI15	3.71	0.82	− 0.72	0.56	0.83*	0.5
BFI20	3.93	0.98	− 1.1	1.02	0.81*	0.22
RBFI25	3.8	1.02	− 0.78	− 0.01	0.85*	0.35
RBFI30	3.42	1.06	− 0.49	− 0.63	0.87*	0.46
BFI35	4.13	0.75	− 0.98	1.88	0.78*	0.33
BFI40	3.73	0.91	− 0.65	0.07	0.85*	0.25
RBFI45	3.93	0.97	− 0.81	0.02	0.83*	0.21
RBFI50	3.69	1.19	− 0.79	− 0.27	0.85*	0.26
RBFI55	3.37	0.91	− 0.08	− 0.6	0.89*	0.12
BFI60	3.42	0.97	− 0.42	− 0.3	0.89*	0.42

*p < 0.001.

Table 4.

Descriptive statistics for BFI-2-B domains and facets.

Domain	Combined1, N = 646	Female1, N = 358	Male1, N = 288	p
BF-extraversion	3.52 (1.01)	3.45 (1.01)	3.61 (1.02)	0.047
Sociability	3.52 (1.01)	3.45 (1.01)	3.61 (1.02)	0.047
Assertiveness	3.63 (0.89)	3.64 (0.90)	3.62 (0.87)	0.7
Energy level	3.17 (1.10)	3.24 (1.10)	3.09 (1.10)	0.070
BF-agreeableness	4.17 (0.75)	4.22 (0.72)	4.10 (0.79)	0.037
Compassion	4.17 (0.75)	4.22 (0.72)	4.10 (0.79)	0.037
Respectfulness	4.44 (0.64)	4.45 (0.64)	4.42 (0.65)	0.6
Trust	3.48 (1.13)	3.56 (1.11)	3.39 (1.16)	0.062
BF-conscientiousness	3.34 (1.21)	3.45 (1.20)	3.21 (1.21)	0.009
Organization	3.34 (1.21)	3.45 (1.20)	3.21 (1.21)	0.009
Productiveness	2.82 (1.22)	2.84 (1.25)	2.80 (1.18)	0.6
Responsibility	3.67 (0.97)	3.64 (0.99)	3.72 (0.94)	0.4
BF-negative emotionality	2.75 (1.11)	2.98 (1.10)	2.45 (1.04)	< 0.001
Anxiety	2.75 (1.11)	2.98 (1.10)	2.45 (1.04)	< 0.001
Depression	2.36 (0.99)	2.47 (1.01)	2.23 (0.96)	< 0.001
Emotional volatility	2.98 (1.21)	3.14 (1.22)	2.77 (1.18)	< 0.001
BF-open-mindedness	3.26 (1.23)	3.33 (1.19)	3.16 (1.28)	0.11
Intellectual curiosity	4.03 (0.80)	3.92 (0.82)	4.16 (0.76)	< 0.001
Aesthetic sensitivity	3.26 (1.23)	3.33 (1.19)	3.16 (1.28)	0.11
Creative imagination	3.71 (0.82)	3.60 (0.85)	3.85 (0.76)	< 0.001

¹Mean (SD); p value is reported for independent sample t-tests.

Table 5 presents the reliability coefficients of BFI-2-B and NEO-FFI. The values of Cronbach’s α, and McDonald’s Omega for all five domains of BFI-2-B and NEO-FFI (except NEO- conscientiousness & NEO- Open mindedness) were > 0.70. Supplementary Table S1 presents the reliability coefficients of the 15 facets of BFI-2-B. The reliability indices for the 15 facets ranged between 0.38 to 0.89 for Cronbach’s α and between 0.40 to 0.89 for McDonald’s ω_t. All facets except energy level, responsibility, and intellectual curiosity had reliability values > 0.60.

Table 5.

Reliability of the BFI-2-B and Bangla NEO-FFI.

	Cronbach’s α		McDonald’s ωt
	BFI-2-B	NEO-FFI	BFI-2-B	NEO-FFI
Extraversion	0.76	0.84	0.83	0.84
Agreeableness	0.78	0.72	0.77	0.78
Conscientiousness	0.81	0.61	0.79	0.62
Negative emotionality	0.86	0.87	0.86	0.87
Open mindedness	0.76	0.58	0.77	0.64

Domain and facet level structural validity

Table 6 presents the factor loading of domain-level validity analysis. The fitted ESEM model with five domains exhibited good model fit [χ² = 3774.55, df = 1480, p < 0.001 CFI = 0.96, TLI = 0.96, RMSEA = 0.06 (90% CI:0.05–0.05), SRMR = 05]. As shown in Table 6, all items under BF-Negative Emotionality exhibited the highest loadings under the corresponding domain. Among the 12 items of the BF-Extraversion domain, 7 items had the highest factor loading on the targeted Extraversion domain, one item (RBFI11) had the highest loading on Negative Emotionality, and four items had the highest loading on Open Mindedness (BFI06, BFI21, BFI41, BFI56). Results showed that the items of BF-Agreeableness equally spread over two domains-Agreeableness (RBFI12, RBFI17, RBFI22, RBFI37, RBFI42, RBFI47) and Open Mindedness (BFI02, BFI07, BFI27, BFI32, BFI52, BFI57). Among the 12 items of BF-Conscientiousness, 9 were loaded highest on the corresponding factor, and three were loaded highest on Open Mindedness (BFI13, BFI38, BFI43). Items of BF-Open Mindedness were spread over Agreeableness ( RBFI5, RBFI20, BFI35, RBFI50, RBFI55), Open Mindedness (BFI10, BFI15, BFI40, RBFI45, BFI60) and Extraversion (RBFI25, RBFI30) domains. These results indicate that the BFI-2’s intended multidimensional structure can be recovered from the BFI-2-B, but with cross-cultural differences in the precise location of some items.

Table 6.

ESEM factor loadings of the 60 items of BFI-2-B.

	Negative Emotionality	Conscientiousness	Extraversion	Open mindedness	Agreeableness
BF-extraversion
BFI01	0.022	0.048	0.412	0.227	− 0.045
BFI06	− 0.228	0.135	0.116	0.314	− 0.006
RBFI11	0.369	− 0.044	0.190	0.151	− 0.090
RBFI16	0.151	− 0.065	0.802	− 0.199	0.075
BFI21	− 0.121	0.059	0.304	0.312	− 0.482
RBFI26	− 0.122	0.188	0.363	− 0.004	0.060
RBFI31	− 0.112	− 0.049	0.635	− 0.253	0.033
RBFI36	− 0.211	− 0.042	0.395	0.201	− 0.156
BFI41	− 0.243	0.300	0.218	0.375	− 0.191
BFI46	0.163	0.027	0.568	0.093	0.022
RBFI51	− 0.081	0.137	0.413	0.079	− 0.137
BFI56	− 0.072	0.095	0.336	0.468	− 0.056
BF-agreeableness
BFI02	0.080	0.168	− 0.265	0.454	0.130
BFI07	− 0.154	0.176	− 0.298	0.506	0.055
RBFI12	− 0.183	0.058	− 0.034	− 0.121	0.460
RBFI17	0.099	0.026	0.015	0.172	0.526
RBFI22	− 0.230	0.087	− 0.262	0.096	0.417
BFI27	− 0.142	0.191	− 0.323	0.368	0.050
BFI32	0.011	0.275	− 0.199	0.467	0.128
RBFI37	− 0.156	0.047	− 0.027	− 0.032	0.517
RBFI42	− 0.195	0.097	0.118	− 0.302	0.420
RBFI47	0.023	0.135	0.017	0.070	0.559
BFI52	− 0.111	0.221	− 0.313	0.539	0.137
BFI57	− 0.205	0.174	− 0.199	0.261	0.087
BF-conscientiousness
RBFI03	0.028	0.841	0.058	− 0.221	0.049
RBFI08	− 0.156	0.551	0.283	− 0.234	0.221
BFI13	0.013	0.083	− 0.004	0.338	− 0.003
BFI18	0.105	0.875	− 0.057	− 0.034	− 0.076
RBFI23	− 0.259	0.300	0.252	− 0.114	0.149
RBFI28	− 0.191	0.408	0.076	− 0.162	− 0.027
BFI33	0.114	0.960	− 0.023	− 0.074	0.001
BFI38	− 0.200	0.266	0.107	0.426	− 0.126
BFI43	− 0.041	0.297	− 0.173	0.495	− 0.053
RBFI48	0.138	0.806	0.015	− 0.157	0.163
BFI53	− 0.046	0.489	0.024	0.263	− 0.098
RBFI58	− 0.186	0.237	0.146	− 0.146	0.232
BF-negative emotionality
RBFI04	0.646	− 0.067	0.186	− 0.340	0.174
RBFI09	0.595	− 0.116	0.127	− 0.277	0.143
BFI14	0.434	− 0.023	− 0.074	0.178	− 0.198
BFI19	0.534	− 0.030	0.042	0.232	− 0.383
RBFI24	0.416	− 0.221	− 0.045	− 0.177	0.188
RBFI29	0.812	0.004	0.182	− 0.235	0.117
BFI34	0.747	0.095	− 0.126	0.150	− 0.083
BFI39	0.659	0.061	− 0.232	0.261	− 0.005
RBFI44	0.608	− 0.153	0.342	− 0.296	0.131
RBFI49	0.520	− 0.039	0.104	− 0.088	0.172
BFI54	0.665	0.018	− 0.223	0.170	0.033
BFI59	0.777	0.051	0.045	− 0.066	− 0.069
BF-open mindedness
RBFI05	0.107	− 0.198	0.301	0.002	0.517
BFI10	0.050	− 0.206	0.254	0.473	0.131
BFI15	− 0.137	0.081	0.180	0.541	− 0.006
BFI20	0.482	− 0.267	0.517	0.047	0.790
RBFI25	− 0.066	− 0.171	0.267	0.259	0.249
RBFI30	− 0.061	− 0.064	0.318	0.288	0.271
BFI35	0.347	− 0.197	0.416	0.164	0.752
BFI40	− 0.174	0.205	− 0.256	0.416	− 0.169
RBFI45	0.061	− 0.202	0.171	0.354	0.158
RBFI50	0.300	− 0.206	0.392	− 0.117	0.780
RBFI55	0.068	− 0.276	0.242	0.065	0.306
BFI60	− 0.120	0.014	0.195	0.468	− 0.019

The highest factor loading for each row is in bold.

Table 7 presents the model fit statistics for the series of five-facet models within each domain. For all fitted models, the CFIs were > 0.90 and SRMR < 0.10 (except single domain open-mindedness). For most cases, a better fit was observed for three facets plus acquiescence models (extraversion, agreeableness, conscientiousness, open-mindedness). For negative emotionality, the RMSEA values were over the upper threshold (RMSEA > 0.08) for all fitted models, but all other fit statistics were acceptable for the three facets plus acquiescence model. Taken together, these results suggest that, within each Big Five domain, BFI-2-B item responses can be modeled using a combination of three substantive facet factors and an acquiescence method factor.

Table 7.

Fit statistics for confirmatory factor analysis of the BFI-2-B items.

Model	χ2	df	CFI	TLI	RMSEA	RMSEA 90% CI		SRMR
						Lower	Upper
Extraversion
M1: single domain	483.24	54	0.906	0.885	0.111	0.102	0.120	0.084
M2:single domain plus acquiescence	481.66	53	0.908	0.883	0.112	0.103	0.121	0.084
M3: positive and negative items	470.55	53	0.908	0.886	0.111	0.101	0.120	0.081
M4: three facets	226.96	51	0.961	0.950	0.073	0.064	0.083	0.062
M5:three facets plus acquiescence	204.16	50	0.966	0.955	0.069	0.059	0.079	0.059
Agreeableness
M1: single domain	338.72	54	0.944	0.932	0.091	0.082	0.100	0.077
M2: single domain plus acquiescence	208.56	53	0.970	0.962	0.068	0.058	0.077	0.058
M3: positive and negative items	219.95	53	0.967	0.959	0.070	0.061	0.080	0.062
M4: three facets	282.47	51	0.955	0.941	0.084	0.075	0.094	0.071
M5: three facets plus acquiescence	196.57	50	0.971	0.962	0.068	0.058	0.078	0.057
Conscientiousness
M1:single domain	474.28	54	0.972	0.966	0.110	0.101	0.119	0.086
M2: single domain plus acquiescence	273.01	53	0.985	0.982	0.080	0.071	0.090	0.067
M3: positive and negative items	460.28	53	0.973	0.966	0.109	0.100	0.118	0.085
M4: three facets	227.67	51	0.988	0.985	0.073	0.064	0.083	0.064
M5: three facets plus acquiescence	119.34	50	0.995	0.994	0.047	0.036	0.058	0.045
Negative emotionality
M1: single domain	501.60	54	0.962	0.954	0.113	0.104	0.123	0.079
M2: single domain plus acquiescence	501.33	53	0.962	0.953	0.115	0.105	0.124	0.079
M3: positive and negative items	360.64	53	0.974	0.968	0.095	0.086	0.104	0.064
M4:three facets	398.11	51	0.971	0.962	0.103	0.093	0.112	0.074
M5: three facets plus acquiescence	386.65	50	0.972	0.962	0.102	0.093	0.112	0.074
Open-mindedness
M1: single domain	672.83	54	0.898	0.875	0.133	0.124	0.142	0.101
M2: single domain plus acquiescence	475.31	53	0.930	0.913	0.111	0.102	0.120	0.087
M3: positive and negative items	648.49	53	0.901	0.877	0.132	0.123	0.141	0.099
M4: three facets	163.74	51	0.981	0.976	0.059	0.049	0.069	0.053
M5: three facets plus acquiescence	158.76	50	0.982	0.982	0.058	0.048	0.068	0.053

df = degrees of freedom; CFI = Comparative fit index; TLI = Tucker-Lewis Index; RMSEA = Root means square error of approximation, CFI and TLI values 0.900, RMSEA values , and SRMR values are bolded. All Chi-Square values are significant at p < 0.001.

Measurement invariance

Due to technological limitations, while fitting the ESEM model for full invariance testing, we aggregated the item scores for each facet and treated the facet scores as continuous indicators. Parallelling the domain structure, the facets were rotated towards a 15 × 5 target matrix with unspecified primary loadings and secondary loadings set to zero. Such a technique to reduce the computational power required to compare complex latent models is common in the literature^15,20 and does not hinder our ability to test the hypothesis that the full invariance of the BFI-2-B would be established for gender, education, and language (H2).

Supplementary Table S2 presents the results of the full invariance analysis of BFI-2-B data. For gender-focused measurement invariance (male vs. female), all ΔCFAs were = − 0.01 except M5 (strong factorial invariance), M7 (strict factorial invariance), and all ΔRMSEAs were 0.01 except M10. For education-focused measurement invariance analysis, all ΔCFAs were = − 0.01 except M3, M5 (strong factorial invariance), M7 (strict factorial invariance), and M9, and all ΔRMSEAs were =0.01. In all cases, CFIs were 0.90, and RMSEAs were 0.0.08. No fitted models were rejected by both the cutoff values (ΔRMSEA 0.01 and ΔCFA ) simultaneously. These results suggest the highest level of measurement invariance—complete factorial invariance—for gender and education levels. By contrast, for language-focused measurement invariance analysis, only M1, M2, and M5 models exhibited acceptable fit indices (CFI > 0.90). Thus, only weak factorial invariance was observed across language.

Item response theory-based analysis

Supplementary Table S3 presents the IRT parameters, including item discrimination (a), item difficulty, and item-fit indices for each of the 15 facets. All items exhibited acceptable item fit (infit & outfit < 1.4, S-χ² associated RMSEA statistics . Supplementary Fig. S1 indicates that for all 15 models, Z_h statistics are larger than − 2 for most participants, suggesting a good person fit. As shown in Supplementary Table S3, among the 60 items, 3 items were very low, 2 items were low, 21 items were moderate, 12 were high, and 22 had very high discrimination scores. Test information curves (TIC) for the fitted models (Supplementary Fig. S2 & Supplementary Table S4) indicated that each facet had a good range of coverage across the underlying traits with some facets (Compassion: − 4.5 to 1.8, Respectfulness: − 5.3 to 1.7, Intellectual Curiosity: − 4.5 to 2) skewed toward lower trait levels. Marginal reliability ranged between 0.57 to 0.88. All facets had marginal reliability > 0.60 except the responsibility facet (Supplementary Table 3S). Taken together, these results indicate that items of BFI-2-B are well-calibrated to measure the intended traits across a wide range of the latent continuum, demonstrating adequate psychometric properties for assessing personality facets within the Bangladeshi population.

Convergent and discriminant validity

Table 8 presents domain-level correlations of the BFI-2-B with NEO-FFI. Results indicated that BFI-2-B converged strongly with NEO-FFI while controlling for gender and age. The correlations between the similar domains were ≥ 0.59 for all instances. Correlations between distinct domains were ≤ 0.39 for all instances. These results indicate good convergent and discriminant validity for the BFI-2-B.

Table 8.

Correlations of the BFI-2-B domains with NEO-FFI.

	NEO-agreeableness	NEO-conscientiousness	NEO-extraversion	NEO-NE	NEO-open mindedness
Correlation of BFI-2 and NEO-FFI (controlled for gender)
BF-agreeableness	0.59	0.35	0.25	− 0.29	0.09
BF-conscientiousness	0.19	0.79	0.32	− 0.35	− 0.01
BF-Extraversion	− 0.05	0.38	0.73	− 0.35	0.10
BF-NE	− 0.26	− 0.38	− 0.26	0.79	− 0.01
BF-Open mindedness	0.13	0.30	0.23	− 0.21	0.64
Correlation of BFI-2 and NEO-FFI (controlled for age)
BF-Agreeableness	0.59	0.32	− .23	− 0.24	0.10
BF-Conscientiousness	0.17	0.77	− .29	− 0.01	− 0.22
BF-Extraversion	− 0.07	− .38	− 0.73	− 0.35	0.09
BF-NE	− 0.18	− 0.33	− 0.24	0.78	0.01
BF-Open mindedness	0.13	0.30	0.23	− 0.21	0.64

All correlations are significant (p < 0.001). The highest correlation in each row is in bold.

Discussion

The current study aimed to validate the Bangla Big Five Inventory-2 (BFI-2-B) by examining important psychometric properties, including reliability and validity (structural, convergent, and discriminant), measurement invariance (across gender, education level and language), and item quality (item discrimination and test information). The study was designed to test three specific hypotheses: (H1) that the domain and facet-level structures of the BFI-2 would be reproducible in the Bangladeshi culture, (H2) that full measurement invariance would be established across gender (male vs female), education levels (Undergrads vs Postgrads) and language (Bangla vs English), (H3) that the BFI-2-B would show strong convergent validity (r for similar domains measured using the Bangla NEO-FFI, as well as discriminant validity between distinct domains.

The BFI-2-B demonstrated satisfactory internal consistency for the five major domains, with Cronbach’s α and McDonald’s ω_t values surpassing 0.76 for all domains. These findings align with the reliability indices reported for the original English BFI-2¹⁴ (α = 0.83–0.90) and adaptations in other languages, such as Chinese¹⁵ and Dutch¹⁸. Most facets also demonstrated adequate reliability. However, intellectual curiosity under BF-Open Mindedness, energy level under BF-Extraversion, and responsibility under BF-Conscientiousness exhibited lower reliability. These discrepancies may be due to cultural differences in how personality traits are valued and expressed in Asian versus Western contexts. Intellectual curiosity, which entails a desire for new ideas and intellectual engagement, is a highly valued trait in many Western societies, where individualism and self-exploration are encouraged⁶⁰. Studies show that Western cultures emphasize intellectual pursuits for personal growth and self-expression⁶¹. In contrast, intellectual curiosity may hold a different position within Asian cultures, where collectivist values may prioritize conformity, respect for tradition, and group harmony over individual exploration⁶². This cultural difference could affect how respondents interpret items related to intellectual curiosity, leading to inconsistent responses and lower reliability scores. The energy level facet within BF-Extraversion reflects enthusiasm, vitality, and social engagement, traits that are prominently valued in Western societies where extraversion is often seen as an asset in social and professional settings⁶³. In Asian contexts, however, social restraint and modesty are often more valued, and high-energy behaviors may be interpreted differently or even discouraged, especially in social interactions⁶⁴. Consequently, items measuring energy level might not resonate as strongly with respondents in Asian cultures⁶⁴. The lower reliability of the responsibility (BF-Conscientiousness) facet may also reflect cultural nuances in interpreting this trait. In Western cultures, responsibility is typically associated with individual dependability and personal accountability⁶⁵, traits that are seen as stable and integral to one’s personality. However, in many Asian societies, responsibility may be viewed more contextually and may be tied closely to roles within family and community structures⁶¹. Rather than being a stable, individual characteristic, responsibility may be perceived as a situational obligation that one upholds in specific contexts (e.g., family responsibilities) rather than as a broad, consistent trait​. This variability in interpretation can lead to inconsistent responses and lower reliability for responsibility-related items in Asian contexts.

The results supported our first hypothesis that the domain and facet-level structures of the BFI-2¹⁴ would be reproducible in the Bangladeshi culture. It is reassuring that at the domain level, our results supported the existence of the Big Five traits mirroring the results found in other international adaptations^15,16,29. However, some items were observed to have cross-loadings, especially for the BF-Open Mindedness and BF-Agreeableness domains. This suggests that though the overall domain structure of BFI-2 is recognizable in its Bangla adaptation, item-level variations were observed, which may be due to the cultural and linguistic contexts. For example, certain items related to interpersonal behavior or emotional expressions might carry distinct connotations in the Bangladeshi context, thus altering their interpretations. The translation process, while rigorous, may not fully capture the subtleties embedded in the original items. At the facet level, consistent with the original development work of the BFI-2¹⁴, the best model fit for the BFI-2-B was achieved when each of the five major personality domains was modeled with three distinct facets per domain and an additional acquiescence factor. This approach aligns with the BFI-2’s design philosophy. Including an acquiescence factor is particularly beneficial as it accounts for response bias related to individuals’ tendency to agree with statements regardless of content. This factor enhances the model’s robustness by adjusting for variance arising from acquiescent response styles rather than true differences in personality traits. Thus, structuring each domain with three specific facets alongside an acquiescence factor mirrors the theoretical underpinnings and methodological rigor of the BFI-2-B. This structure has proven effective across various languages and contexts, indicating that this facet-based, multidimensional personality measurement approach is versatile and adaptable.

Supporting our second hypothesis, results showed that BFI-2-B demonstrated full measurement invariance across gender (male vs. female) and education levels (undergraduates vs. postgraduates). Similar findings were reported in the Japanese adaptation²⁹ of BFI-2, where the highest level (strict) of measurement invariance was established across genders. The establishment of MI across gender and education level suggests that the BFI-B items are interpreted similarly among male and female participants and individuals with different educational levels within the Bangladeshi sample. This cross-group comparability highlights the scale’s suitability for diverse applications, making it a valuable tool for research and practice in varied contexts. However, one important pattern to be highlighted is that for some domains and facets, females tended to score higher than males (at the mean level). In the current study, females described themselves as more agreeable, conscientious, and emotional. At the facet level, women described themselves as more compassionate, trustworthy, organized, anxious, depressed, and emotionally volatile than men. By contrast, men described themselves as more intellectually creative. Such findings are aligned with the existing literature^14,15. Soto and John¹⁴ reported that females tend to describe themselves as more extroverted, agreeable, conscientious, and emotional than males. Yoshino, et al.²⁹ while adapting BFI-2 in the Japanese language reported that within the BF-Extraversion domain, males had a higher assertiveness score than females, while females had a higher sociability score. Future research is warranted to investigate these gender-based differences in personality traits.

In contrast to our expectation that full measurement invariance across language (English vs Bangla) would be observed for BFI-2-B, our results supported only weak measurement invariance for language. Similar challenges in language-based measurement invariance were also observed in the German adaptation²⁰, where exact measurement invariance was estimated using CFA, configural invariance was reported for agreeableness and open-mindedness, and weak measurement invariance was reported for extraversion and conscientiousness. These findings collectively point to potential challenges related to cultural and linguistic interpretations of specific constructs and items. In Western cultures, traits like openness and extraversion are often positively associated with self-expression and individual achievement, whereas in Bangladesh and other South Asian cultures, these traits may be viewed through a lens that emphasizes social harmony, modesty, and family obligations^66,67. Furthermore, words associated with traits like responsibility or intellectual curiosity may hold slightly different connotations in Bangla than in English, potentially affecting how respondents interpret these items. The lack of MI between Bangla and English BFI-2 may also be partly attributed to the differences in response styles between the two cultures. Research indicates that people from collectivist cultures, such as Bangladesh, may respond to personality assessments in ways that reflect social desirability or normative behaviors, prioritizing community-oriented values over individual traits. In contrast, respondents in individualistic Western cultures might prioritize self-expression, leading to different response patterns^66,68. These variations can affect item functioning, as some BFI-2 items may resonate differently in a collectivist versus an individualist cultural context, even if the underlying construct is similar.

The Item Response Theory (IRT) analysis of the BFI-2-B provided insights into item functioning and measurement precision across the fifteen personality facets. Overall, most items displayed acceptable fit indices, indicating their adequacy in capturing the underlying traits within each facet. Items demonstrated a range of discrimination values, with 22 items showing very high discrimination, 12 items high, 21 items moderate, and a few items with lower discrimination scores. The test information curves (TICs) further indicated that most facets covered a wide range of the latent trait continuum, particularly around the mean levels of traits. However, certain facets, such as compassion and respectfulness within BF-Agreeableness and aesthetic sensitivity within BF-Open Mindedness, showed coverage skewed towards lower trait levels, suggesting that the BFI-2-B items within these facets provide more reliable information for individuals with lower trait expressions. Marginal reliability estimates were satisfactory for most facets, with values above 0.60 across facets, except for responsibility (BF-Conscientiousness), which fell slightly below the threshold. The lower reliability of this facet may reflect that cultural interpretation of responsibility is context-dependent rather than a stable, individual trait⁶¹. This variability suggests that further refinement of items within these facets could enhance reliability in the Bangladeshi setting.

The results of the current study supported our third hypothesis that the BFI-2-B would show strong convergent validity (> 0.50) for the similar domains measured using the Bangla NEO-FFI, and discriminant correlations between distinct domains would be lower than the convergent correlations. This reinforces the robustness of the BFI-2-B as a valid personality assessment tool. The BFI-2-B demonstrated strong convergent validity, with correlations for similar domains exceeding r = 0.59. These high correlations suggest that the BFI-2-B effectively measures personality constructs comparable to those captured by the NEO-FFI, aligning with expected patterns based on the theoretical overlap between these two well-established personality models. Discriminant validity was confirmed since the correlations between distinct domains were substantially lower than the correlations between similar domains. This differentiation among domains suggests that the BFI-2-B effectively distinguishes between personality traits. These findings are consistent with results from the development work of English BFI-2¹⁴ that reported high convergent and discriminant validity with established personality measures. In both Western and non-Western adaptations, high correlations between the BFI-2 and NEO-FFI domains underscore the reliability of the BFI-2 structure across different cultural and linguistic settings, supporting the broad applicability of the Big Five model.

Limitations and future directions

While the BFI-2-B has demonstrated promising psychometric properties in this study, several limitations must be acknowledged. First, although the sample size was adequate, the study relied on an internet-based sample, which may introduce sampling bias. Internet samples may not fully represent the diversity of the Bangladeshi population, particularly regarding socioeconomic, geographical, and educational backgrounds. Additionally, this study retained data from 646 participants out of the initial 1,095. This exclusion of 41% of the data due to participants failing attention checks, may impact the representativeness and generalizability of the results. Second, measurement invariance across languages showed only weak invariance, indicating potential cultural and linguistic differences in item interpretation between Bangla and English versions. This limitation suggests that the translation process may not capture subtle cultural nuances in certain personality traits, potentially impacting the cross-linguistic validity of the BFI-2-B. Furthermore, certain facets within domains, such as energy level (BF-Extraversion), responsibility (BF- Conscientiousness), and intellectual curiosity (BF-Open-mindedness), exhibited lower reliability. This could indicate cultural differences in trait interpretation, which may affect the reliability of these facets in non-Western samples.

To address these limitations and enhance the robustness of the BFI-2-B, future studies should consider employing mixed-method approaches, including cognitive interviews, to gain deeper insights into how Bangladeshi respondents interpret specific items. This could help identify cultural nuances influencing responses and improve translation accuracy. Additionally, future research should seek to replicate these findings across more diverse samples, including rural populations and those from varied socioeconomic backgrounds, to enhance generalizability. Finally, to improve cross-linguistic measurement invariance, researchers might develop culture-specific items or adapt existing items to better align with local norms and values​.

Conclusion

This study represents a comprehensive psychometric validation of the Bangla Big Five Inventory-2 (BFI-2-B), providing a valuable tool for assessing personality traits in the Bangladeshi population. The BFI-2-B demonstrated strong internal consistency (especially at the domain levels) and structural validity (at both the domain and facet levels), reinforcing its alignment with the theoretical framework of the Big Five personality model. The study successfully established full measurement invariance across gender and education levels, highlighting the BFI-2-B’s suitability for diverse applications within these demographic groups in Bangladesh. Additionally, the strong convergent and discriminant validity demonstrated by the BFI-2-B supports its capability to measure distinct personality constructs effectively. However, challenges emerged in achieving full measurement invariance across languages, suggesting potential cultural and linguistic nuances in item interpretation that warrant further exploration. Lower reliability scores for specific facets, such as energy level and intellectual curiosity, underscore the need for caution when interpreting these facets in Bangladeshi contexts. Overall, the BFI-2-B provides a psychometrically robust, culturally adapted measure of personality that holds promise for research and applied settings in Bangladesh. With continued research and adaptation, the BFI-2-B can serve as a foundational tool for advancing personality assessment in Bangladesh, contributing to more nuanced understandings of personality traits within this cultural context.

Author contributions

M.A.S., F.K., C.J.S, and S.H. designed the study. M.A.S. and F.K. curated the data. M.A.S. conducted the formal analysis and prepared all figures and tables. M.A.S., F.K., C.J.S, and S.H. wrote and reviewed the main manuscript text. All authors reviewed the manuscript.

Data availability

All raw data and materials are publicly available on Open Science Framework (https://osf.io/7dtqg/) and GitHub (https://github.com/mind-psychometry/BanglaBigFive2).

Code availability

Data analysis codes are publicly available on Open Science Framework (https://osf.io/7dtqg/) and GitHub (https://github.com/mind-psychometry/BanglaBigFive2).

Declarations

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-90264-0.