Lean but strained: a socio-technical explanation of the safety paradox in just-in-time modular construction

Modular construction has transformed the global construction industry by relocating labour-intensive activities to structured factory environments, enabling standardisation, reduced material variability, and tighter supply chain coordination (Wu et al., 2022; Vargas et al., 2025). Although modular settings are widely regarded as safer than traditional sites due to improved ergonomic conditions and structured workflows (Vithanage et al., 2022), this presumed safety advantage rests on an assumption that warrants scrutiny. Just-In-Time (JIT) production principles have been extensively integrated into modular construction to synchronise material flows and enforce sequencing discipline across factory, transport, and site operations (Bortolotti et al., 2013; Ivanov and Dolgui, 2021), creating tightly coupled, low-slack systems that introduce important operational challenges alongside their efficiency gains.

Tightly synchronised delivery systems leave limited room for recovery when disruptions occur, resulting in compressed work schedules, heightened coordination pressure, and increased reliance on rapid, real-time adjustments. Under such conditions, workers may be required to absorb system variability through accelerated task execution, multitasking, and intensified coordination. These dynamics raise concerns that efficiency gains may be accompanied by increased behavioural demands, particularly in environments characterised by unreliable supply flows and logistical uncertainty.

This tension reflects a broader unresolved debate in lean research. While JIT enhances coordination and eliminates waste, it simultaneously reduces operational slack and increases system sensitivity to variability (Ivanov and Dolgui, 2021). Prior studies offer mixed perspectives: tightly coupled systems may improve safety through standardisation, or undermine it by intensifying work pace and reducing recovery capacity (Shah and Ward, 2007; Hedvall et al., 2025). This debate remains insufficiently theorised within modular construction, where safety is often treated as a presumed co-benefit of process discipline rather than as an outcome governed by distinct causal pathways (Gao et al., 2023).

In tightly synchronised systems, disruptions propagate across interdependent activities, transferring operational pressure onto workers through accelerated work and coordination demands (Bode and Wagner, 2015; Ivanov and Dolgui, 2021). Limited attention has been paid to the behavioural mechanisms through which JIT-induced system tightening affects safety, or the conditions under which these effects are amplified or mitigated.

These gaps converge on a single question: through what mechanism does JIT influence employee safety, and under what upstream conditions does this mechanism strengthen or attenuate? Four research questions address this progressively, from mechanism identification (RQ1, RQ2) through pathway testing (RQ3) to boundary condition analysis (RQ4).

To address this gap, this study advances the notion of a “lean–strained paradox,” whereby efficiency-oriented production systems may generate increased behavioural demands while maintaining or even improving observed safety outcomes. Drawing on Socio-Technical Systems (STS) theory, JIT is conceptualised not merely as an inventory control technique but as a work-structuring mechanism that reshapes both technical flows and human behaviour (Pasmore et al., 2019; Clegg et al., 2017).

Building on this perspective, the study introduces operational strain as the key mediating mechanism linking JIT to employee safety. Operational strain captures the workload intensity, coordination burden, and time compression associated with tightly synchronised systems. By reducing buffers and increasing interdependencies, JIT does not eliminate variability but redistributes it within the system, shifting the burden of adjustment from structural buffers to human actors. In this sense, operational strain represents the behavioural pathway through which efficiency-oriented system design translates into workforce outcomes.

Furthermore, supply reliability is conceptualised as the critical boundary condition determining whether JIT stabilises coordination or amplifies operational strain (Bode and Wagner, 2015; Ivanov and Dolgui, 2021).

Accordingly, this study develops and tests a moderated mediation model explaining how JIT is associated with employee safety through operational strain, and how this relationship is conditioned by supply reliability in modular construction systems. The study addresses the following research questions:

The study makes the following contributions to research on production and operations management. First, it reconceptualises strain induced by just-in-time (JIT) systems by theorising operational strain as a short-term compensatory coordination mechanism that can temporarily support safety in tightly coupled production systems, but only under stable upstream supply conditions. Second, it advances lean theory by reframing JIT as a production planning and control (PPC) regime that restructures material flow, work organisation, and real-time coordination, extending lean beyond efficiency outcomes to worker-level safety implications. Third, it introduces operational strain as a mediating construct, providing a mechanism-based explanation of how tightly coupled PPC designs transfer variability and coordination pressure onto workers. Fourth, it identifies system alignment, particularly supply reliability, as a boundary condition for JIT’s safety effects and, using evidence from modular construction in a developing economy, extends POM theory on coupling and coordination to contexts of infrastructural constraint and high supply variability. In doing so, the study extends recent operations management evidence on JIT contingency and variability redistribution (Alemsan et al., 2025; Salman et al., 2024; Yeni et al., 2025) into the behavioural safety domain.

Beyond these theoretical contributions, the study offers practical guidance for construction managers, lean practitioners, and safety regulators. It shows that monitoring operational strain provides more diagnostic value than lagging incident metrics, and that supply reliability is a safety boundary condition with direct implications for JIT system design in environments with unreliable upstream flows.

Section 2 develops the theoretical framework; Sections 3 and 4 present the methodology and results; Section 5 discusses the findings; Section 7 concludes.

Socio-Technical Systems (STS) theory explains organisational outcomes as the result of alignment between technical system design and social adaptation processes (Trist and Emery, 2015; Clegg et al., 2017). Rather than treating performance as a function of technical optimisation alone, STS emphasises that changes in workflows, scheduling, and production structures inevitably reshape coordination demands, work pacing, and behavioural expectations. System effectiveness, therefore, depends on the social subsystem's capacity to adjust to these technical configurations.

This perspective is particularly relevant in tightly coupled production environments such as modular construction, where Just-In-Time (JIT) practices reduce buffers and synchronise interdependent activities across factory, logistics, and site operations (Shou et al., 2023; Ivanov and Dolgui, 2021). While such configurations enhance workflow discipline and coordination, they also increase the system's sensitivity to disruptions by limiting its ability to absorb variability. As a result, deviations are more likely to be managed through behavioural adjustment rather than structural slack.

Under stable conditions, JIT can improve safety by reducing congestion and enhancing workflow clarity. However, when variability exceeds system tolerance, tightly synchronised systems generate increased coordination pressure (Bode and Wagner, 2015; Hedvall et al., 2025), positioning JIT as a socio-technical intervention whose safety effects depend critically on operating conditions.

This framing aligns with Perrow's (1984) Normal Accidents Theory, which argues that tightly coupled systems are structurally predisposed to cascading disruptions when local deviations propagate through interdependent components. In modular construction, such cascades manifest as routine workflow compressions: delayed deliveries generating sequencing mismatches that intensify coordination demands. Weick and Sutcliffe (2007) extend this logic within High Reliability Organisation (HRO) theory, demonstrating that safety in tightly coupled systems depends on collective mindfulness: heightened sensitivity to operational cues, deliberate task sequencing, and adaptive response capacity. Critically, HRO theory suggests that under short-term strain, workers may exhibit increased vigilance as a compensatory response, providing prospective theoretical grounding for the unexpected strain-safety relationship explored empirically below. Together, Perrow and Weick contextualise the STS framework within the modular production setting: the technical-social interdependence of factory-transport-site supply chains makes this context productive for STS analysis, and supply reliability the critical condition determining whether the coupling enables or overwhelms worker safety-sustaining capacity.

JIT synchronises deliveries with production schedules, reducing inventory and handling waste (Montazeri et al., 2024; Li et al., 2026). In modular construction, where factory, transport, and site assembly are tightly coupled, disciplined sequencing is central to performance (Babalola and Aliu, 2025), but reduced buffers convert upstream deviations into downstream instability (James and Ebin, 2022; Dzreke and Dzreke, 2025). JIT must therefore be theorised as an interdependent system in which reliability, pacing, and contingency capacity jointly shape outcomes (Mweshi et al., 2025).

These dynamics are particularly pronounced in developing economy contexts, where JIT encounters structural conditions absent from advanced industrial settings. Supply chains in such environments exhibit higher baseline variability, weaker supplier reliability, limited logistics infrastructure, and fragmented governance (Bandoophanit and Pumprasert, 2022; Mweshi et al., 2025). These conditions amplify the fragility of tightly coupled systems: the buffers JIT eliminates may be the precise reserves that compensate for structural supply instability, suggesting that JIT’s safety implications are conditioned by upstream stability and institutional context. In addition, Salman et al. (2024) demonstrate that synchronisation failures between lean planning and upstream supply conditions are the primary barrier to JIT effectiveness in developing-economy manufacturing settings, confirming that the challenges identified here are not sector-specific.

The relationship between JIT practices and employee safety is theoretically ambiguous. On one hand, JIT enhances workflow organisation by reducing clutter, improving sequencing, and creating more predictable task environments, which can support safer work execution (Shah and Ward, 2007; Love et al., 2018). In modular construction, these effects are particularly relevant due to spatial constraints and high task interdependence. However, some studies suggest that lean discipline can reduce strain by eliminating unnecessary motion and clarifying task sequences (Shah and Ward, 2007; Love et al., 2018), underscoring that the JIT-strain relationship is not deterministic and warrants empirical testing.

On the other hand, JIT reduces buffers and increases system sensitivity to variability, requiring disruptions to be absorbed through real-time coordination and intensified work effort (Ivanov and Dolgui, 2021; Bandoophanit and Pumprasert, 2022). This variability redistribution mechanism has been documented in lean supply chain settings by Yeni et al. (2025), who show that lean system efficiency is systematically undermined when upstream supplier reliability is insufficient to absorb schedule variation.

Two established frameworks illuminate why these demands raise safety concerns. Karasek's (1979) Job Demand-Control (JDC) model predicts that high-demand, low-control environments (precisely those generated by tightly synchronised JIT systems) elevate strain and increase the risk of safety error. The Job Demands-Resources (JD-R) model (Bakker and Demerouti, 2007) extends this by distinguishing demands that impose costs from resources that buffer them; in JIT-intensive modular construction, compressed schedules and multitasking demands constitute significant job demands that may impair safety vigilance. However, the challenge-hindrance stressor framework (Cavanaugh et al., 2000; Prem et al., 2017) cautions that time pressure and coordination demands may function as challenge stressors, stimulating heightened attentiveness, consistent with Małysa et al.'s (2024) finding that lean workflow discipline can simultaneously increase workload and improve safety adherence. This empirical ambiguity motivates H3, which reflects the dominant JDC prediction:

STS theory provides a unifying explanation by conceptualising safety as an outcome of workers' responses to system-induced pressures, while Factory Physics highlights that buffer reduction redistributes variability within the system rather than eliminating it (Clegg et al., 2017; Bode and Wagner, 2015). In tightly coupled modular construction environments, this redistribution often shifts the burden of adjustment to workers, linking system design directly to behavioural outcomes. (Hopp and Spearman, 2011)

This redistribution manifests as operational strain, defined as the behavioural and cognitive pressure arising from time compression, workload intensification, and coordination demands in tightly synchronised systems (Prem et al., 2017; Fireman et al., 2023). Operational strain represents the mechanism through which JIT translates system-level efficiency into worker-level consequences. Accordingly, JIT is expected to influence safety both directly, through improved coordination, and indirectly, through increased strain.

Based on this reasoning, the following hypotheses are proposed:

Supply reliability is a critical condition shaping how JIT operates in tightly coupled production systems. Prior research shows that JIT performance depends heavily on the consistency and predictability of inbound flows, particularly in environments characterised by high interdependence and limited buffering (Bode and Wagner, 2015; Ivanov and Dolgui, 2021). In modular construction, where factory production, logistics, and site assembly are synchronised, delivery disruptions can propagate rapidly across activities, increasing coordination demands and destabilising workflow execution.

This framing is consistent with supply chain resilience theory, which conceptualises resilience as the ability to return to a desired state after disruption (Christopher and Peck, 2004). Alemsan et al. (2025) corroborate this notion, demonstrating that JIT's operational effectiveness is contingent on supply reliability, with JIT failures consistently linked to unstable upstream conditions across all disruption scenarios examined.

In tightly coupled JIT systems, low supply reliability reduces resilience by eliminating the redundancy that enables recovery (Sheffi, 2005). Critically, what resilience theory identifies at the supply chain level, STS theory identifies at the socio-technical level: variability unabsorbed by system buffers must be absorbed by human actors. Supply reliability, therefore, links upstream resilience to downstream worker safety, providing theoretically integrated justification for its moderating role.

From a Socio-Technical Systems perspective, supply reliability reflects the degree of alignment between technical scheduling systems and social coordination processes (Clegg et al., 2017; Delikhoon et al., 2024). When supply is reliable, synchronised workflows can be executed as planned, enabling workers to operate within manageable coordination demands. In contrast, unreliable supply disrupts this alignment, requiring workers to compensate by making reactive adjustments, intensifying coordination, and compressing execution timelines.

These dynamics suggest that supply reliability conditions the extent to which JIT-induced variability is absorbed within the technical system or transferred to workers (Bode and Wagner, 2015; Xiao and Qi, 2016). Under high reliability, variability is largely contained, reducing the behavioural burden associated with JIT. Under low reliability, variability is amplified and redistributed to workers, increasing operational strain and exposing the limits of system stability.

Accordingly, supply reliability is expected to moderate the relationship between JIT practices and operational strain such that the strain-inducing effect of JIT is weakened under stable supply conditions (Ivanov and Dolgui, 2021; Roehrich et al., 2021).

Figure 1 integrates the hypothesised relationships, showing the direct and indirect effects of Just-in-Time practices on employee safety through operational strain. It also illustrates the moderating role of supply reliability in the relationship between Just-in-Time practices and operational strain.

This study adopts a two-wave, multi-informant survey design to examine the relationship between Just-In-Time (JIT) practices and employee safety in modular construction, incorporating the mediating role of operational strain and the moderating effect of supply reliability. The design separates data collection across time and respondent roles to improve measurement validity and reduce common method bias.

Data were collected at two points in time. At Time 1 (T1), managerial respondents with responsibility for planning, logistics, and production coordination provided assessments of JIT practices and supply reliability. At Time 2 (T2), approximately four weeks later, frontline respondents directly involved in execution activities reported on operational strain and employee safety. This temporal and role-based separation helps reduce consistency bias while preserving the stability of operational conditions.

The empirical context comprises modular construction projects operating in logistically constrained urban environments characterised by limited storage space, high interdependence, and variability in inbound deliveries. These conditions create tightly coupled production systems in which JIT practices are both necessary and potentially destabilising, making the context appropriate for examining the interactions among system design, behavioural strain, and safety outcomes.

A multi-stage sampling approach was employed. First, a sampling frame of 230 modular and off-site construction firms/projects was identified through industry directories, professional networks, and publicly available records. Of these, 103 firms agreed to participate, representing a 44.8% firm-level participation rate.

Within participating firms, role-based purposive sampling was used to select knowledgeable respondents. At Time 1, surveys were administered to managerial personnel responsible for production planning, logistics coordination, or scheduling. At Time 2, surveys were administered to frontline personnel, including site supervisors and operational staff directly exposed to workflow and safety conditions. In total, 792 professionals were invited across both waves. A total of 371 responses were received, and after screening for completeness, matching, and response quality, 346 valid matched responses were retained, yielding a usable response rate of 43.7%.

Responses were matched across waves using anonymised project-level identification codes. Where multiple frontline responses were obtained within a project, aggregation was conducted in accordance with agreement assessment procedures to ensure reliability. Non-response bias was assessed using early–late response comparisons, with no significant differences observed.

Sample size adequacy was assessed using the inverse-square-root method for PLS-SEM (Kock and Hadaya, 2018). Given the smallest path coefficient (β = 0.109, H5), the recommended minimum is approximately 295 respondents; the retained 346 matched responses exceed this threshold. G*Power post-hoc analysis confirmed adequate power (1 − β > 0.80) at α = 0.05 for f² = 0.15. Where multiple frontline responses were obtained within a project, aggregation was conducted following Bliese (2000). Intraclass correlation coefficients (ICC1 = 0.16, ICC2 = 0.71) and within-group agreement (rwg mean = 0.78) confirmed adequate aggregation justification (LeBreton and Senter, 2008).

Measurement instruments were developed using established scale adaptation procedures. Items were drawn from prior literature in lean operations, production management, and safety research, and refined through expert review and pilot testing to ensure clarity and contextual relevance.

All constructs were modelled as reflective latent variables and measured using a 5-point Likert scale ranging from strongly disagree to strongly agree. Measurement items were adapted from established empirical scales and contextualised to modular construction settings. Where standardised measures were not directly transferable, items were developed from established conceptual definitions and refined through expert review and pilot testing, following recommended scale-adaptation procedures (Lambert and Newman, 2023).

JIT practices (four items; Time 1, managerial) capture schedule alignment, buffer minimisation, delivery synchronisation, and installation flow continuity. Operational strain (four items; Time 2, frontline) measures time pressure, work pace, coordination demand, and disruption-induced workload. Employee safety (four items; Time 2, frontline) covers hazard minimisation, incident reduction, adherence to procedures, and control of site conditions. Supply reliability (three items; Time 1, managerial) assesses delivery consistency, disruption frequency, and schedule adherence. The full list of measurement items is provided in Appendix I.

Three control variables were included: firm size (number of employees), firm age (years in operation), and modular production experience (years), all measured at Time 1 via single-item scales consistent with prior lean and construction operations research.

The hypothesised model was estimated using Partial Least Squares Structural Equation Modelling (PLS-SEM) in SmartPLS. PLS-SEM is appropriate given the study’s predictive orientation, its inclusion of mediation and moderation effects, and its suitability for complex models involving latent constructs.

Prior to model estimation, key PLS-SEM distributional assumptions were examined. Skewness and kurtosis statistics indicated acceptable levels of non-normality (skewness < |2|, kurtosis < |7|), supporting bootstrapping-based inference. Non-linearity was assessed following Sarstedt et al. (2022) by sequentially adding squared predictor terms to the model; none reached statistical significance (all p > 0.10), confirming the adequacy of the linear specification.

The analysis followed a two-stage approach. First, the measurement model was assessed for indicator reliability, internal consistency, convergent validity, and discriminant validity. Indicator loadings, composite reliability, and average variance extracted (AVE) were examined, while discriminant validity was assessed using the heterotrait–monotrait ratio (HTMT).

Second, the structural model was evaluated by examining path coefficients, t-values, and p-values obtained via bootstrapping. The model’s explanatory power was assessed using R² values, while predictive relevance was evaluated using Q² statistics.

Mediation was assessed using bootstrapped indirect effects, with significance determined by confidence intervals. Moderation was tested using the two-stage interaction approach, which reduces multicollinearity and is appropriate for reflective constructs. Significant interactions were further examined using simple slope analysis. Model fit and predictive performance were assessed using SRMR, d_ULS, d_G, and PLSpredict, consistent with current best practices in PLS-SEM (Hair et al., 2022).

Common method bias was addressed through both design-based and statistical procedures. The two-wave, multi-informant design serves as the primary procedural remedy by separating predictor and outcome variables across both time and respondent source, thereby reducing consistency bias and method-driven covariance (Podsakoff et al., 2003).

As an additional diagnostic, Harman’s single-factor test was conducted, with results indicating that no single factor accounted for more than 40% of the variance, suggesting that CMV is unlikely to be a serious concern (Podsakoff et al., 2003). Full collinearity variance inflation factors (VIFs) were also examined (Table 1), with all values below the recommended threshold of 3.3, suggesting that common method bias is unlikely to affect the results (Kock, 2015) materially.

To mitigate potential endogeneity concerns, the study combines temporal separation with statistical diagnostics. Specifically, the Gaussian copula (GC) approach was implemented within the PLS-SEM framework, following the approach of Hill et al. (2021). This method transforms non-normally distributed regressors into copula terms, which are subsequently incorporated into the structural model as control functions. The absence of statistically significant copula terms suggests that reverse causality and omitted variable bias are unlikely to contaminate the estimated relationships.

The GC results indicate that the copula-adjusted paths from JIT practices to operational strain (β = 0.479 vs adjusted β = 0.471, Δβ = 0.008) and from JIT practices to employee safety (β = 0.401 vs adjusted β = 0.396, Δβ = 0.005) remain substantively unchanged after copula correction. The Gaussian Copula terms for both endogenous constructs are non-significant (p > 0.10), indicating no meaningful endogeneity bias (Hill et al., 2021).

Overall, the GC analysis provides no evidence of endogeneity across the examined paths. The robustness of the findings was further verified by PLSpredict cross-validated predictive relevance (Q² > 0 for all endogenous constructs), confirming adequate predictive model performance.

One residual bias warrants clear acknowledgement. Both operational strain and employee safety were reported by frontline respondents at Time 2, creating a same-source dependency between these constructs that temporal separation of waves cannot eliminate. The Gaussian Copula results provide no evidence of endogeneity in the strain-safety path (β = 0.031, t = 0.324, p > 0.05), and full collinearity VIFs remain below the CMV contamination threshold (Kock, 2015). Future studies should nonetheless collect strain data from supervisors or objective workload records to enable fully source-independent testing.

The measurement model was assessed for indicator reliability, construct reliability, convergent validity, and discriminant validity (Table 1). All indicator loadings exceeded 0.70 (range: 0.760–0.912), confirming adequate indicator reliability. Internal consistency was confirmed via Cronbach's alpha (range: 0.813–0.909) and composite reliability (range: 0.878–0.936), both exceeding the 0.70 threshold. Average variance extracted (AVE) values ranged from 0.644 to 0.786, all exceeding the 0.50 benchmark, confirming convergent validity (Hair et al., 2022). Discriminant validity is reported in Table 2 (Fornell-Larcker) and Table 3 (HTMT).

Discriminant validity was evaluated using the Fornell-Larcker criterion and heterotrait-monotrait (HTMT) ratios (Tables 2 and 3). The square roots of AVE for each construct exceed inter-construct correlations, satisfying the Fornell-Larcker criterion. While most HTMT values fall below the conservative 0.85 threshold, two pairs (JIT-Safety, 0.857; Operational Strain-Safety, 0.843) approach this boundary. Three considerations support their theoretical distinctiveness: (1) constructs differ in referent (system design vs worker outcome); (2) measurement items differ lexically and theoretically; (3) Voorhees et al. (2016) note that values below 0.90 are generally acceptable for conceptually proximate constructs. This proximity is acknowledged as a limitation, and future studies may examine these construct boundaries using confirmatory composite analysis. Importantly, the two construct pairs with elevated HTMT values (JIT Practices–Employee Safety and Operational Strain–Employee Safety) are the same pairs most exposed to the Time 2 same-source dependency acknowledged in Section 4.1. These two limitations are not fully independent and should be read in conjunction by readers assessing the construct boundaries of these pairs.

The structural model was assessed using path coefficients, bootstrapping significance tests, and bias-corrected 95% confidence intervals (Figure 2; Table 4). JIT has a significant positive effect on employee safety (β = 0.401, t = 5.647, p < 0.001; CI [0.262, 0.540]), supporting H1, and on operational strain (β = 0.479, t = 8.547, p < 0.001; CI [0.369, 0.589]), supporting H2. Operational strain also has a significant positive effect on employee safety (β = 0.316, t = 4.227, p < 0.001; CI [0.169, 0.463]). However, the positive direction contradicts the hypothesised negative relationship; hence, H3 is not supported. This unexpected finding suggests a more complex relationship between strain and safety than initially theorised.

The JIT × supply reliability interaction (moderation) has a significant negative effect on operational strain (β = −0.109, t = 3.309, p = 0.001; CI [−0.174, −0.044]), supporting H5. Thus, higher supply reliability attenuates the strain-inducing effect of JIT, confirming its buffering role in tightly coupled production systems.

The mediating role of operational strain was assessed using bootstrapping (5,000 subsamples, bias-corrected and accelerated). As shown in Table 4, the indirect effect of JIT practices on employee safety via operational strain was significant (β = 0.151, t = 4.183, p < 0.001, 95% BCa CI [0.080, 0.222]), confirming partial mediation (H4 supported). The confidence interval entirely excludes zero, meeting the bias-corrected bootstrap standard for establishing mediation significance in PLS-SEM (Hair et al., 2022). Since the direct effect of JIT on employee safety remains significant (β = 0.401, p < 0.001), the results indicate partial mediation.

A simple slope analysis (Aiken and West, 1991) confirmed the moderation pattern with supply reliability at ±1 SD from the mean. At low supply reliability (−1 SD), the JIT-strain relationship was positive and significant (β = 0.587, p < 0.001); at high supply reliability (+1 SD), this relationship remained significant but was materially attenuated (β = 0.370, p < 0.001). The significant slope difference (Δβ = 0.217) confirms that supply reliability meaningfully conditions the strain-inducing consequences of JIT, functioning as a buffer rather than an eliminator.

Similarly, the conditional indirect effects reveal that the mediating effect of operational strain is stronger when supply reliability is low and weaker when reliability is high. This indicates that the strain-based pathway linking JIT to safety outcomes is contingent on upstream stability conditions.

The model demonstrates substantial explanatory power (R² = 66.8% for employee safety; R² = 69.1% for operational strain), warranting critical examination. Three considerations address potential construct overlap concerns: first, predictors and outcomes were measured across different time points and by different respondent groups, reducing common-source explanations; second, full collinearity VIFs remain below 3.3 (Table 1), indicating predictor redundancy is not inflating estimates; third, Q² values (0.677 and 0.632) confirm genuine out-of-sample predictive relevance not susceptible to overfitting bias. Nonetheless, the theoretical proximity between JIT-induced demands and safety-related conditions is acknowledged as a potential limitation.

Effect size analysis (Table 5) further shows that JIT has a large effect on operational strain (f² = 0.400) and a moderate effect on employee safety (f² = 0.190), while supply reliability exerts a moderate effect on operational strain (f² = 0.295). These findings reinforce the central role of JIT and supply reliability in shaping system dynamics (see Table 6).

Control variables (firm size, firm age, and modular production experience) were included to account for potential firm-level and experiential effects. The results (Table 7) indicate that firm size has a small but significant positive effect on employee safety (β = 0.12, p < 0.05), while its effect on operational strain is not significant. Firm age and modular production experience do not exhibit significant effects on either operational strain or employee safety. Overall, the inclusion of control variables does not materially alter the hypothesised relationships, supporting the robustness of the model.

The findings provide a nuanced and theoretically rich account of how Just-In-Time (JIT) practices influence employee safety in tightly coupled modular construction systems. Consistent with H1, JIT exhibits a significant positive association with employee safety, reinforcing prior research linking lean practices with improved workplace organisation, reduced congestion, and enhanced workflow discipline (Shah and Ward, 2007; Love et al., 2018; Liu et al., 2020). In modular construction, synchronised deliveries and minimal on-site inventory contribute to clearer work zones and more predictable execution patterns (Ahn et al., 2020; Vithanage et al., 2022; Efiok et al., 2015). These findings support the view that JIT, as a coordination mechanism, can enhance safety by structuring flow and reducing physical hazards.

At the same time, the results confirm H2, showing that JIT significantly increases operational strain. This aligns with both Socio-Technical Systems (STS) theory and Factory Physics, which suggest that tightly coupled systems with reduced buffers are inherently more sensitive to variability (Bode and Wagner, 2015; Ivanov and Dolgui, 2021). By compressing slack and enforcing synchronisation, JIT shifts the burden of disruption from the system to workers, requiring them to absorb variability through intensified coordination, accelerated task execution, and continuous adjustment. This reinforces the view that efficiency gains in lean systems are not costless but are often achieved through heightened behavioural demands. This is consistent with Hedvall et al. (2025), who identify reduced slack as a primary driver of operational pressure in lean manufacturing. Unlike prior studies that measure strain indirectly through incident records, the present study operationalises strain as a behavioural construct, providing a more proximal account of how system tightening affects workers. These findings reinforce Salman et al. (2024) and Yeni et al. (2025) on lean system fragility, demonstrating that the burden of lean-induced variability redistribution is measurable at the individual worker level.

Contrary to H3, operational strain demonstrates a significant positive association with employee safety (β = 0.316, p < 0.001). As theorised prospectively in Section 2, both HRO theory (Weick and Sutcliffe, 2007) and the challenge-hindrance stressor framework (Cavanaugh et al., 2000; Prem et al., 2017) provide a basis for this pattern: time pressure and coordination demands, core components of operational strain, may function as challenge stressors that elicit heightened vigilance, more deliberate task sequencing, and stronger adherence to safety-critical routines rather than impairing performance. Consistent with this, Małysa et al. (2024) found that lean workflow discipline can simultaneously increase coordination demands and improve safety adherence in construction settings. Three cautions apply: the cross-sectional design limits temporal inference; compensatory behaviours likely reflect short-term adaptive capacity that may collapse under sustained escalation (Weick and Sutcliffe, 2007); and the modest effect size (β = 0.316) suggests that individual variability in adaptive capacity exists. Strain should therefore be understood as a dual-threshold construct: moderate levels may trigger safety-sustaining vigilance; sustained elevated strain is likely to exceed adaptive capacity and generate adverse outcomes consistent with Perrow's (1984) account of cascade failure in tightly coupled systems.

The mediation results (H4) further clarify this mechanism. The significant indirect effect indicates that JIT influences employee safety partly through operational strain, supporting a partial mediation structure. This demonstrates that JIT operates through both structural pathways, such as improved workflow organisation, and behavioural pathways, such as increased coordination effort.

The moderating role of supply reliability (H5) provides important insight into the conditions under which these dynamics unfold. Consistent with expectations, supply reliability weakens the positive relationship between JIT and operational strain, indicating that stable upstream flows reduce the behavioural burden placed on workers. This finding aligns with STS theory, which emphasises that system outcomes depend on alignment between technical configurations and social coordination processes (Clegg et al., 2017; Delikhoon et al., 2024). When supply reliability is high, synchronised workflows can be executed as planned, enabling JIT to function as a disciplined coordination mechanism with manageable coordination demands. This extends Christopher and Peck’s (2004) supply chain resilience framework to the domain of worker safety and reinforces Alemsan et al.’s (2025) finding that JIT effectiveness is contingent on supply reliability, demonstrating that this contingency extends from logistical outcomes to worker-level safety consequences, as Sheffi (2005) predicts for systems with unabsorbed variability.

In contrast, low supply reliability disrupts this alignment and amplifies the consequences of tight system coupling. Consistent with Factory Physics, variability that buffers cannot absorb propagates through the system, manifesting as delivery delays, sequencing mismatches, and reactive rescheduling (Bode and Wagner, 2015). These disruptions compress execution windows and intensify coordination demands, increasing operational strain as workers compensate through rapid adjustment and increased effort. The conditional effects further confirm that the JIT–strain relationship is stronger under low-reliability conditions and weaker under high-reliability conditions.

Importantly, the moderation results also extend the mediation logic by showing that the strain pathway itself is contingent on upstream stability. When supply reliability is high, variability is largely contained within the technical system, reducing the behavioural burden associated with JIT. When reliability is low, variability is increasingly transferred to workers, strengthening the strain-based pathway through which JIT influences safety outcomes.

This study examined how JIT practices influence employee safety in modular construction, introducing operational strain as a mediating mechanism and supply reliability as a moderating boundary condition. JIT simultaneously enhances safety through improved workflow coordination and increases operational strain through system tightening, a dual effect captured by the partial mediation structure. Contrary to H3, operational strain associates positively with safety, theorised here as a short-term compensatory effect consistent with HRO theory and the challenge-hindrance stressor framework. Supply reliability attenuates the strain-inducing pathway, confirming that tightly coupled systems generate manageable behavioural demands only when upstream flows are stable.

Theoretically, the study advances three claims. First, JIT is reframed as a mechanism for redistributing variability that shifts operational pressure from structural buffers to human actors, extending lean theory beyond its efficiency orientation to encompass worker-level safety implications. Second, operational strain is established as a mediating construct providing a mechanism-based explanation of how system-level design produces workforce-level outcomes, addressing gaps identified by Gao et al. (2023) and Małysa et al. (2024). Third, the supply reliability boundary condition integrates STS theory, Factory Physics, and supply chain resilience theory into a unified explanatory framework for tightly coupled production environments.

In practice, JIT implementation should be accompanied by a supply reliability assessment, since the strain-amplifying effects of JIT are substantially stronger under low-reliability conditions. Operational strain should be monitored as a leading safety indicator, using workload intensity and coordination pressure measures, rather than relying on lagging incident-based metrics that may mask compensatory worker effort. Regulators should expand safety frameworks beyond incident reporting to incorporate system-level indicators of workload intensity and schedule compression, particularly in developing economies where infrastructure constraints amplify the fragility of tightly coupled systems. A critical caution is warranted: practitioners should not interpret rising operational strain as inherently safety-positive. The dual-threshold framing in Section 5 indicates that moderate strain may elicit short-term compensatory vigilance, but sustained elevated strain is likely to exceed workers' adaptive capacity and generate adverse safety outcomes consistent with Karasek's (1979) Job Demand-Control prediction. Strain indicators should therefore be treated as early warning signals requiring managerial intervention through workload redistribution, schedule buffering, or supply reliability improvement, not as evidence that the system is self-regulating.

Limitations include: the cross-sectional design within each wave restricts temporal inference; perceptual measures may introduce response bias; the modular construction context in a developing economy constrains direct generalisability; and the same-source dependency between strain and safety at Time 2 remains a methodological limitation despite the CMV controls employed. Future research should employ longitudinal designs to examine whether the positive strain-safety relationship holds over extended exposure or reverses under sustained strain accumulation. Incorporating objective strain indicators and extending the framework to healthcare, logistics, and aerospace manufacturing would test the generalisability of the variability-redistribution explanation of the lean-safety relationship.

The supplementary material for this article can be found online.