Validation of the Workplace COVID-19 Knowledge and Stigma Scale (WoCKSS) in Malaysia: a cross-sectional study using item response theory and factor analyses

Abstract

Objective

To validate the Workplace COVID-19 Knowledge and Stigma Scale (WoCKSS) using item response theory (IRT), exploratory factor analysis (EFA) and confirmatory factor analysis (CFA).

Design

Cross-sectional psychometric validation.

Setting

Manufacturing companies registered in Malaysia.

Participants

A total of 137 factory workers participated in the exploratory phase and 300 in the confirmatory phase. Inclusion criteria were Malaysian nationality and ability to read Malay.

Methods

The knowledge domain was examined using the two-parameter logistic IRT model in two stages: an exploratory IRT analysis in phase 1 to screen items and a confirmatory IRT analysis in phase 2 to evaluate the final item set. The stigma domain was analysed using EFA followed by CFA. Reliability was assessed using Cronbach’s alpha and McDonald’s omega (ω). The development process and content and face validity results were previously published.

Results

14 knowledge items were retained after exploratory IRT and formed the final knowledge scale evaluated in confirmatory IRT. For these final items, discrimination parameters ranged from 0.77 to 3.17 and difficulty values from −4.47 to 0.23, with unidimensionality supported (p=0.644). EFA supported a three-factor stigma structure (stereotype, fear, prejudice), and CFA confirmed excellent model fit (χ²=8.91, df=11, p=0.630; root mean square error of approximation=0.00; Comparative Fit Index=1.00; Tucker-Lewis Index=1.00; standardised root mean square residual=0.021). Composite reliability by McDonald’s omega ranged from 0.691 to 0.893.

Conclusion

WoCKSS is a reliable and valid instrument for assessing workplace COVID-19 knowledge and stigma in industrial sectors in Malaysia.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study uses robust psychometric methods including item response theory (IRT), exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) with separate exploratory and confirmatory datasets.

• The instrument is validated in the Malay language and tailored to the workplace context, where COVID-19 stigma was highly prevalent.

• The tool addresses a gap in occupational health research by focusing on factory workers, an understudied population.

• Cross-sectional data limit causal interpretation.

• Findings are based on convenient sampling and may result in answer bias.

Introduction

The COVID-19 pandemic created significant health, social and economic disruption worldwide, accompanied by widespread fear and misinformation. Such uncertainty contributed to stigmatising attitudes towards individuals infected with COVID-19, mirroring patterns observed during previous infectious disease outbreaks such as Severe Acute Respiratory Syndrome (SARS), Ebola and H1N1, where fear and blame reinforced social exclusion.^{1 2} Stigma is known to influence health-seeking behaviour, psychological well-being and social cohesion and is shaped by broader social norms and social processes, as described in foundational stigma theories.^{3 4} In Malaysia, the manufacturing sector recorded a high concentration of workplace clusters, where close physical proximity and shared facilities heightened both transmission anxiety and stigma among workers.^{5 6} Although COVID-19 has transitioned to endemic management, its social consequences, including stigma, have not resolved in parallel. In Malaysia, continued public health monitoring of COVID-19 beyond the acute pandemic phase, particularly in workplace settings, underscores the need for context-specific and psychometrically robust measurement tools.

Multiple instruments have been developed internationally to assess COVID-19 knowledge and stigma. Knowledge questionnaires typically measure factual understanding of transmission, symptoms and prevention behaviours, yet many were designed for the general population, healthcare workers or students.7,9 Several COVID-19 stigma instruments also exist, including the COVID-19 Public Stigma Scale (COVID-PSS) which assess domains such as fear, stereotyping and prejudice.¹⁰ However, validated Malay-language tools remain limited, and few existing instruments have undergone comprehensive validation using both exploratory factor analysis (EFA) and confirmatory factor analyses (CFA) or item response theory (IRT).

A validated instrument tailored to workplace settings is particularly important, as factory workers differ from the general population in terms of educational background, work environment and access to health information. Existing tools are not designed for this context and may not capture the specific forms of COVID-19 knowledge and stigma experienced in industrial workplaces. To address this gap, the Workplace COVID-19 Knowledge and Stigma Scale (WoCKSS) was developed, with its content and face validation previously reported by the authors.¹¹ This study extends that work by evaluating the scale through IRT, EFA and CFA, providing comprehensive evidence of its validity and reliability for use among Malaysian factory workers.

Methods

Study design and participants

This study employed a two-phase cross-sectional design comprising an exploratory validation (phase 1) and a confirmatory validation (phase 2). Data collection for phase 1 was conducted from late April 2023 to November 2023, and phase 2 from January 2024 to April 2024. Six registered manufacturing companies were randomly selected for phase 1 (n=137), and five additional companies were included in phase 2 (n=300), which helped reduce selection bias at the company level. Eligible participants were Malaysian factory workers able to read Malay. Workers who participated in prior content or face validation were excluded.

Data were collected using an online questionnaire administered through Google Forms. Company representatives disseminated the survey link to workers at each participating company, and all eligible workers who received the link were invited to participate. A total of 137 workers completed the questionnaire in phase 1 and 300 in phase 2. Because the survey link was disseminated by company representatives through internal workplace communication channels, the total number of eligible workers who received the invitation and the number who actively declined participation could not be quantified; non-participation therefore reflected non-response. Participation was voluntary and anonymous to reduce response and social desirability bias, and informed consent was obtained electronically.

Instrument

The WoCKSS was developed as a workplace-specific, Malay-language instrument to capture both factual understanding and attitudinal responses related to COVID-19. It consists of two domains: a knowledge domain comprising 17 items with ‘true’, ‘false’ or ‘I do not know’ response options, and a stigma domain comprising 22 items rated on a Likert scale. The stigma domain was conceptualised as a multidimensional construct comprising three latent factors: fear, stereotype and prejudice, informed by stigma theory. The knowledge domain was designed to assess factual understanding of COVID-19 relevant to workplace settings, while the stigma domain captures attitudinal components. The development process and content and face validity findings have been published previously.¹¹ In the present study, the knowledge domain was evaluated using IRT as a unidimensional construct, while the stigma domain was evaluated as a multidimensional construct using EFA and CFA.

Sample size

For the EFA of the stigma domain, the minimum recommended sample size was initially based on five participants per item.¹² With 22 stigma items retained after face validation, a minimum of 110 respondents was required. To enhance robustness, a target sample size of approximately 150 participants was planned, in line with methodological recommendations for EFA.13,15 EFA was conducted using phase 1 data (N=137), meeting the lower-bound heuristic for item-to-participant ratios. Beyond heuristic rules, methodological and simulation studies indicate that stable factor solutions for multidimensional constructs can be achieved with sample sizes in the range of 100–150 when communalities are moderate to high and factor loadings are substantial.¹⁵ The adequacy of the phase 1 sample was further supported by satisfactory Kaiser-Meyer-Olkin (KMO) values and a significant Bartlett’s test of sphericity, indicating that the stigma domain data were suitable for factor analysis.

For reliability analysis using Cronbach’s alpha, the required sample size was determined using the following parameters: α=0.05, power=0.80, number of items=22, required Cronbach’s alpha=0.70 and reference value=0.50. The minimum required sample size was 66 respondents, based on Statstodo’s sample size calculator.¹⁶

Following EFA, CFA was conducted exclusively on the stigma domain, as specified in the Instrument section. The CFA tested the three-factor structure (fear, stereotype and prejudice) identified in the EFA of the stigma items. The final CFA model comprised nine observed items loading onto three latent factors, with 12 freely estimated parameters. The required sample size was based on a ratio of at least 10 respondents per freely estimated parameter, yielding a minimum of 120 participants. A larger target sample of 300 respondents was selected to ensure stable parameter estimation and robust model fit in accordance with recommendations for CFA in complex measurement models.^{13 15}

For IRT analyses, only the knowledge domain of the WoCKSS was analysed. Although no universal sample size guideline exists for IRT, recommended ranges typically fall between 100 and 500 respondents.¹⁷ The sample sizes obtained for both phases were within this range.

Phase 1: exploratory validation

Exploratory item response theory for knowledge domain

IRT analyses were conducted only for the knowledge domain, which was treated as a unidimensional construct. Local independence was assessed through residual correlations, and item characteristic curves were examined to evaluate item discrimination and difficulty parameters. The exploratory IRT analysis applied the two-parameter logistic (2-PL) model to all 17 knowledge items to estimate item difficulty (b) and discrimination (a). Item fit was evaluated using χ² statistics and inspection of item characteristic curves (ICCs). Modified parallel analysis was used to assess the assumption of unidimensionality, which compares the observed data structure with that expected under a unidimensional IRT model. Analyses were conducted in RStudio using the ‘ltm’, ‘psych’ and ‘irtoys’ packages.

Exploratory factor analysis for stigma domain

EFA was conducted using principal axis factoring (PAF) with promax rotation in the ‘psych’ and ‘GPArotation’ packages. Sampling adequacy was assessed using the KMO statistic, and factorability was evaluated using Bartlett’s test of sphericity. Factor retention was determined using multiple complementary criteria: eigenvalues greater than 1, visual inspection of the scree plot, Horn’s parallel analysis (parallel analysis scree plot comparing observed eigenvalues with those generated from randomly simulated datasets of identical size), Very Simple Structure (VSS) and Velicer’s Minimum Average Partial (MAP) test.18,20 Factors were retained when observed eigenvalues exceeded simulated eigenvalues in the parallel analysis. Items were removed if they demonstrated loadings below 0.50, communalities below 0.30 or cross-loadings defined as secondary loadings ≥0.30 on more than one factor. Internal consistency for the resulting factors was assessed using Cronbach’s alpha as a conventional internal consistency index, with acceptable thresholds defined as ≥0.70.

Phase 2: confirmatory validation

Confirmatory item response theory for knowledge domain

The 14 knowledge items retained from phase 1 were reanalysed using the 2-PL model to confirm the stability of item parameters. Unidimensionality was reassessed using modified parallel analysis. Model fit was evaluated using χ² statistics, and measurement precision was examined through the test response function (TRF).

Confirmatory factor analysis for stigma domain

CFA was conducted using the ‘lavaan’, ‘semTools’, ‘semPlot’ and ‘psych’ packages. Multivariate normality was assessed, and due to violations, robust maximum likelihood estimation was applied. Model fit was evaluated using the χ² statistic, root mean square error of approximation (RMSEA), Comparative Fit Index (CFI), Tucker-Lewis Index (TLI) and standardised root mean square residual (SRMR).²¹ Competing models were compared using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). Composite reliability was assessed using McDonald’s omega (ω) as the primary model-based reliability estimate aligned with the confirmatory measurement model.²²

Data handling and software

All analyses were performed in RStudio (V.2025.09.2).

Patient and public involvement

Patients or members of the public were not involved in the design, conduct, reporting or dissemination plans of this research.

Results

Participant characteristics

Table 1 presents the sociodemographic characteristics of participants in the exploratory (n=137) and confirmatory (n=300) samples. The mean age of participants was similar across both phases, ranging from 31.6 to 32.6 years old. The exploratory sample had a higher proportion of males (65.7%), whereas the confirmatory sample had more females (56.7%). A substantial proportion reported a previous COVID-19 infection (64.2% in the exploratory sample; 68.0% in the confirmatory sample). Most participants reported no family members who died due to COVID-19, and the majority were non-smokers.

Table 1. Sociodemographic characteristics of participants in the exploratory and confirmatory samples.

Variables	Exploratory (n=137)	Confirmatory (n=300)
	n (%)	n (%)
Age	Mean (SD): 31.64 (9.0)	Mean (SD): 32.6 (8.6)
Gender
Female	47 (34.3)	170 (56.7)
Male	90 (65.7)	130 (43.3)
Education
SPM	27 (19.7)	38 (12.7)
Pre-university/diploma	48 (35.0)	62 (20.7)
At least bachelor’s degree	62 (45.2)	200 (66.7)
Monthly income
<RM 2500	52 (38.0)	101 (33.7)
RM 2501 to RM 4849	54 (39.4)	149 (49.7)
≥RM 4850	28 (20.4)	48 (16.0)
COVID-19 infection history
No	49 (35.8)	96 (32.0)
Yes	88 (64.2)	204 (68.0)
Family members died due to COVID-19
No	123 (89.8)	264 (88.0)
Yes	12 (8.8)	36 (12.0)
Smoking status
Not smoking	112 (81.8)	271 (90.3)
Smoking	25 (18.2)	29 (9.7)

RM, Malaysian ringgit; SPM, Sijil Pelajaran Malaysia/Malaysian Certificate of Education.

Item response theory findings

The exploratory IRT analysis evaluated all initial knowledge items. Three items (K5, K11, K15) were removed due to poor discrimination, extreme difficulty or significant misfit. Among the retained items, discrimination values ranged from 0.41 to 2.18, and difficulty parameters ranged from −5.40 to 0.46. Items such as K7, K13 and K17 displayed strong discrimination. However, unidimensionality was not supported in this phase (p<0.001). This may be attributable to the smaller sample size, which can yield unstable parameter estimates.

Following exploratory IRT, the confirmatory IRT analysis re-evaluated the 14 retained items. Item parameters showed improved stability, with discrimination values ranging from 0.77 to 3.17 and difficulty values from −4.47 to 0.23. Table 2 presents a comprehensive summary of discrimination, difficulty and item fit statistics for the 14 retained items across both exploratory and confirmatory analyses.

Table 2. Combined item response theory (IRT) results for the knowledge domain (validation parts 1 and 2).

Item	Exploratory				Confirmatory
	Discrimination (a)	Difficulty (b)	χ2 (df=8)	P values	Discrimination (a)	Difficulty (b)	χ2 (df=8)	P values
K1	0.75	−5.40	8.17	0.417	0.97	−4.47	17.55	0.025
K2	0.79	−2.93	13.76	0.088	1.94	−2.13	28.81	<0.001
K3	0.82	−3.28	12.50	0.130	1.03	−3.04	15.99	0.043
K4	1.13	−2.16	17.31	0.027	1.43	−2.04	7.15	0.520
K6	1.33	−0.33	51.30	<0.001	1.42	−0.64	51.58	<0.001
K7	2.11	−2.72	21.96	0.005	2.02	−2.90	23.03	0.003
K8	1.38	−2.56	16.45	0.036	2.79	−2.21	7.94	0.440
K9	1.19	−2.96	13.50	0.096	1.92	−2.51	28.83	<0.001
K10	1.43	−3.34	9.72	0.286	1.21	−3.95	10.92	0.206
K12	0.41	−3.55	10.94	0.205	0.77	−2.60	60.04	<0.001
K13	2.18	−0.96	19.67	0.012	1.78	−1.48	42.28	<0.001
K14	2.02	0.46	35.07	<0.001	1.60	0.23	16.41	0.037
K16	0.82	−3.42	8.92	0.350	1.59	−2.52	19.51	0.012
K17	18.63	−0.73	2.00	0.981	3.17	−1.33	19.51	0.012

The ICCs for the confirmatory dataset (figure 1) illustrate the performance of each item across the latent ability continuum. Items K1, K3, K4 and K10 were relatively easy and appeared left-shifted, indicating high probability of correct response at lower ability levels, while K14 and K17 appeared to be more difficult. Strong discrimination was observed for K7, K8, K13 and K17 as indicated by steeper slopes. Figure 2 shows the TRF, demonstrating a clear monotonic increase in expected total score with increasing ability. Approximately 76.34% of the total test information was captured within the ability range of −3 to +3, indicating reliable performance across a wide range of knowledge levels. The assumption of unidimensionality was supported in the confirmatory sample, as evidenced by the non-significant result from the modified parallel analysis (p=0.644).

Figure 1. Item characteristic curves for confirmatory IRT analysis. IRT, item response theory.

Figure 2. Test response function for the knowledge domain (confirmatory IRT). IRT, item response theory.

Exploratory factor analysis

EFA supported a three-factor solution representing stereotype, fear and prejudice (KMO=0.72). Following iterative item removal based on low factor loadings (<0.50), cross-loadings, low communalities (<0.30) and high inter-factor correlations, a final structure of nine items was retained. The factors were labelled as ‘stereotype’, ‘fear’ and ‘prejudice’, consistent with the theoretical framework of stigma. Factor loadings ranged from 0.558 to 0.968, and all communalities exceeded 0.40. Cronbach’s alpha coefficients for the three subscales were 0.84 (stereotype), 0.73 (fear) and 0.85 (prejudice), indicating satisfactory internal consistency (table 3).

Table 3. Construct validity and internal consistency of WoCKSS stigma domain by EFA and Cronbach’s alpha.

Factor	Item	Factor loading	Communalities	Cronbach’s alpha
1 (stereotype)	S4	0.796	0.652	0.84
	S5	0.951	0.866
	S6	0.668	0.446
2 (fear)	S7*	0.777	0.606	0.73
	S8*	0.730	0.547
3 (prejudice)	S15	0.558	0.436	0.85
	S16	0.607	0.426
	S17	0.968	0.850
	S18	0.929	0.781

^*Inverse item.

EFA, exploratory factor analysis; WoCKSS, Workplace COVID-19 Knowledge and Stigma Scale.

Confirmatory factor analysis

CFA was conducted to validate the factorial structure of the stigma domain identified in EFA. A total of three competing models were evaluated sequentially to assess model fit, optimise parsimony and enhance theoretical coherence, as shown in table 4.

Table 4. Fit indices for three CFA models.

Fit indices	Model
	Model 1	Model 2	Model 3
χ2 (df)	57.12 (24)	40.37 (17)	8.91 (11)
χ2 p value	<0.001	0.001	0.630
SRMR	0.054	0.049	0.021
RMSEA	0.070	0.071	0.00
90% CI	0.047 to 0.094	0.043 to 0.100	0.000 to 0.055
CFit p value	0.075	0.102	0.927
CFI	0.97	0.98	1.00
TLI	0.95	0.96	1.00
AIC	6431.64	5591.33	4778.92
BIC	6509.42	5661.70	4841.89

AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; CFA, confirmatory factor analysis; CFI, Comparative Fit Index; CFit, close fit; RMSEA, root mean square error of approximation; SRMR, standardised root mean square residual; TLI, Tucker-Lewis Index.

Model 1 comprised all nine items from the initial EFA structure. Although it demonstrated acceptable fit (χ² (24) = 57.12, p<0.001; RMSEA=0.070; CFI=0.97; TLI=0.95; SRMR=0.054), high modification indices (MI) were observed between S17 and S18 (MI=17.40) and S16 and S18 (MI=10.83), indicating potential local misfit. Additionally, items S15 and S16 demonstrated low factor loading and substantial residuals. These findings suggested model refinement was necessary. Reallocating S16 and S17 to a different factor, as suggested by the MI, was not theoretically justifiable. Thus, S15 was removed in the next model.

Model 2 excluded item S15, resulting in a minor improvement in model fit (χ² (17) = 40.37, p=0.001; RMSEA=0.071; CFI=0.98; TLI=0.96; SRMR=0.049). However, item S16 continued to exhibit poor performance, evidenced by high MI values and substantial residuals warranting its removal in the next model.

Model 3, the final model, excluded both S15 and S16. This model achieved excellent fit: χ² (11) = 8.91, p=0.630; RMSEA=0.000 (90% CI 0.000 to 0.055); CFI=1.00; TLI=1.00; SRMR=0.021. Model parsimony was also supported by the lowest AIC (4778.92) and BIC (4841.89) among the three models, confirming model 3 as the best-fitting solution.

Factor loadings and composite reliability (ω) for model 3 are shown in table 5. All standardised loadings were significant and ranged from 0.594 to 0.999. Composite reliability was acceptable across all three factors: stereotype (ω=0.691), fear (ω=0.867) and prejudice (ω=0.893). The final CFA model structure, illustrating the three-factor solution of the stigma domain, is presented in figure 3.

Table 5. Standardised factor loadings and composite reliability for final CFA model (model 3).

Factor	Item	Factor loading	ω
F1 (stereotype)	S4	0.699	0.691
	S5	0.594
	S6	0.860
F2 (fear)	S7*	0.863	0.867
	S8*	0.931
F3 (prejudice)	S17	0.789	0.893
	S18	0.999

^*Inverse item.

CFA, confirmatory factor analysis.

Figure 3. Path diagram of final three-factor CFA model. F1 = stereotype, F2 = fear, F3 = prejudice. CFA, confirmatory factor analysis.

Discussion

Item response theory analysis

The IRT analysis of the knowledge domain provided a detailed assessment of item functioning and measurement precision. Both phases used the 2-PL model, which allows difficulty and discrimination parameters to vary across items and is appropriate for knowledge-based constructs.¹⁷ In the exploratory phase, most items fell within the recommended difficulty range of −3 to +3,²³ although K1 and K17 were at the extremes. These items were retained because they assess essential aspects of COVID-19 knowledge that are highly relevant in workplace settings: K1 captures workers’ understanding of key symptoms, which is fundamental for early recognition and reporting, while K17 addresses misconceptions about reinfection risk, a prevalent belief that has substantial implications for workplace prevention practices. Discrimination values were generally moderate to high, with K7 and K13 showing strong discriminatory capacity.²³ A few items demonstrated weaker discrimination, suggesting limited ability to differentiate respondents across knowledge levels.

Unidimensionality was not supported in the exploratory phase, likely reflecting sampling variability or the multifaceted nature of COVID-19 knowledge.^{17 24} The confirmatory phase addressed these limitations, as larger samples generally improve IRT parameter stability and model assumptions.^{24 25} Item parameters remained consistent, and unidimensionality was confirmed through modified parallel analysis, indicating that the retained items measured a single latent construct.

The ICCs and TRF further supported the strength of the scale. Easier items (K1, K3) were located at lower ability levels, while more difficult items (K17) showed steeper slopes, indicating higher discrimination across ability levels.²⁶ Approximately three-quarters of total information fell within the −3 to +3 range, suggesting optimal precision for low to moderate knowledge levels.²³ These findings are consistent with previous IRT applications to COVID-19 knowledge scales, which also reported mid-range information concentration and varied discrimination.^{27 28} Overall, the knowledge domain of WoCKSS demonstrates strong measurement properties and stable item functioning.

Construct validity and reliability of the stigma domain

Construct validity for the stigma domain was examined using EFA followed by CFA. Sampling adequacy was acceptable (KMO=0.7),²⁹ and Bartlett’s test supported factorability.¹³ PAF was used due to its robustness under non-normality.^{30 31} Several retention procedures including the scree plot, VSS and MAP tests strengthened confidence in the extracted structure and reduced reliance on eigenvalues alone.^{18 19} Promax rotation was applied given the expectation of correlated stigma dimensions.³¹

The EFA results supported three distinct but related factors (stereotype, fear, prejudice), with all retained items showing loadings ≥0.50 and communalities ≥0.400.²¹ Inter-factor correlations were below 0.70, indicating adequate separation of constructs. This pattern is consistent with the structure reported in the COVID-PSS.¹⁰ CFA further confirmed and refined this solution. Two items (S15 and S16) were removed due to weak loadings and high MI, a refinement approach commonly applied in stigma scale development.^{32 33} The final model demonstrated excellent fit, exceeding the recommended thresholds.³⁴

Although the fear and prejudice factors each included only two items, high loadings and conceptual coherence justified their retention, reflecting psychometric recommendations that permit two-item factors under such conditions.^{35 36} Similar decisions have been reported in recent COVID-19 stigma validations.^{32 33}

Reliability analysis

Reliability was supported by both Cronbach’s alpha in the exploratory phase and ω in the confirmatory phase, the latter offering a more accurate estimate of internal consistency when item loadings vary.^{37 38} Although the stereotype subscale fell slightly below the conventional 0.70 threshold, it was retained based on its theoretical importance and acceptable loadings, a practice also observed in previous COVID-19 stigma validations.^{32 33} The high reliability of the two-item prejudice subscale is noteworthy; despite recommendations for three items per factor, two-item factors are acceptable when items are strongly correlated and conceptually coherent.^{35 36} Taken together, these indices demonstrate that the WoCKSS stigma domain has satisfactory internal consistency for use in workplace settings.

Strengths, limitations and implications

This study has several methodological strengths. The dual-phase design with separate exploratory and confirmatory samples enabled rigorous testing of item functioning, dimensionality and reliability across independent datasets. Scale development was preceded by systematic content and face validation, ensuring that items were conceptually sound and culturally appropriate for Malaysian factory workers.^{11 39} The use of IRT for the knowledge domain and EFA and CFA for the stigma domain allowed detailed evaluation of item performance and factorial structure, supporting the development of a coherent and contextually grounded instrument. Reliability assessment using ω further strengthened the internal consistency evidence for the final model.²¹

Some limitations should be acknowledged. Although factories were randomly selected, convenience sampling at the worker level may limit generalisability beyond the participating companies. The cross-sectional design restricts the ability to assess test-retest reliability or responsiveness to change. The fear and prejudice factors each comprise only two items; however, both demonstrated strong factor loadings and acceptable reliability, supporting their suitability within the final measurement model. While the EFA met established adequacy criteria, the phase 1 sample size represents the lower bound of recommended ranges for multidimensional scales. Although methodological and simulation-based evidence suggests that stable factor recovery is achievable under these conditions when communalities and factor loadings are adequate, some degree of sampling variability cannot be entirely excluded. To mitigate this limitation, the factor structure identified in the exploratory phase was subsequently evaluated using confirmatory factor analysis in an independent and larger sample, thereby strengthening confidence in the stability and generalisability of the final measurement model. These limitations should be considered when applying WoCKSS in other occupational settings.

Conclusion

This study provides evidence that the WoCKSS is a valid and reliable instrument for assessing COVID-19-related knowledge and stigma among factory workers in Malaysia. The knowledge domain showed stable item parameters and supported unidimensionality under a 2-PL IRT model, while the stigma domain demonstrated a coherent three-factor structure with acceptable internal consistency. WoCKSS therefore offers a concise, contextually grounded measure that can support occupational health research and practice focused on understanding and addressing workplace COVID-19 stigma.

Data availability statement

Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.