The Effect of Total Quality Management on Organisational Performance: A Systematic Review and Meta-Analysis of Structural Equation Modelling (SEM) Studies

Abstract

This systematic review and meta-analysis, conducted in accordance with PRISMA 2020 guidelines, synthesises SEM-based evidence on the relationship between Total Quality Management (TQM) and organisational performance. The Web of Science, Scopus, and Semantic Scholar were searched for peer-reviewed studies, published between 2015 and 2026, that reported direct TQM-to-performance structural paths. Of 255 full-text records assessed, 148 were coded, and 92 contributed to the primary random-effects model. The pooled standardised path coefficient indicates a moderate positive effect. Between-study heterogeneity was substantial. Composite TQM operationalisation produced stronger effects than individual dimensions. Aggregate and sustainability performance responded more strongly than other performance types. Meta-regression revealed no statistically significant contextual moderators. Key limitations include geographic concentration in South and Southeast Asia and the MENA region, reliance on cross-sectional survey designs, and a single-coder approach. No risk-of-bias tool was applied, as no validated instrument exists for this study type in management research. TQM consistently improves organisational performance across contexts. Future research should prioritise longitudinal designs, broader sectoral coverage, and non-English literature.

1. Introduction

There is an extensive body of literature supporting the positive effects of TQM on organisational performance, spanning traditional, manufacturing and quality-related measures to environmental, social, and sustainable ones. The positive effects are quite stable, as studies across industries that used different methodological approaches conducted in different national contexts have yielded similar results [1,2,3]. However, the strength of this relationship may vary across TQM dimensions, performance outcomes, and organisational settings. This variation, rather than uncertainty, continues to attract scholarly interest. The differences in measurement may be due to how TQM constructs have been operationalised in different studies. TQM has been measured both as a single variable and as a composite construct. Unlike the first, which measures the overall level of effects, the latter is often defined as a multidimensional set of specific measures, such as leadership, people management, process management, quality data, supplier management, and customer focus [2,4]. It can be argued that both methods examine the same phenomenon. However, different levels of measurement can lead to different results. Composite scores may combine effects that impact relationships differently, while separate constructs can highlight which specific practices are responsible for particular outcomes in a given context [3,5].

The measurement frameworks for the dependent variable have evolved considerably over time. Early TQM studies focused on operational and quality-based outcomes, like manufacturing efficiency, defect rates, and process reliability, reflecting the origins of quality management in production contexts [6]. Financial performance measures followed as researchers sought to establish the business case for TQM investment, linking quality practices to profitability, market share, and return on assets [7,8]. Innovation performance emerged subsequently, capturing TQM’s potential to stimulate new product development, process development and organisational learning [9]. By raising awareness to achieve sustainability, organisations have linked TQM with the capacity to pursue environmental, social, and economic objectives simultaneously, thereby gaining considerable attention as a distinct performance dimension [10,11,12,13]. Connecting these differences across measurement frameworks may shed light on theoretically meaningful variations that are only partially addressed in previous studies [1,2,3,14,15].

Previous meta-analyses rely most on correlations, leaving unresolved debates about the effects of path coefficients, which are the main source of effect sizes in structural equation modelling (SEM) studies. In addition, SEM has become a dominant methodology in TQM-related studies since 2015 [16]. Apart from the growing application of SEM in TQM studies, there is a shift in the research landscape. Most recent studies originate from South and Southeast Asia and the MENA region [3,14,15]. Furthermore, a growing number of studies have shifted attention to linking TQM with sustainability performance, examining outcomes such as corporate sustainability, green performance, and environmental governance across manufacturing and service contexts [11,17,18,19,20]. However, there is a lack of meta-analyses that have systematically synthesised path-coefficient evidence for the TQM–performance relationship, leaving its average effect size and boundary conditions empirically unresolved. This study addresses the aforementioned gap by analysing SEM-derived path coefficients from 255 studies published between 2015 and 2026 through five research questions:

• RQ1: What is the overall magnitude of the direct TQM–performance relationship when estimated from SEM path coefficients?
• RQ2: Do different performance outcomes (financial, operational, quality, innovation, overall, and sustainable) affect study results?
• RQ3: Does TQM operationalisation (composite construct vs. individual dimensions) moderate the effect?
• RQ4: To what extent do industry, geographic, and firm size characteristics explain between-study heterogeneity?
• RQ5: Do CB-SEM and PLS-SEM studies consistently produce different path coefficient estimates?

The study is positioned within the fields of industrial engineering and engineering management and organisation and management. Quality management systems represent a core area of industrial engineering, concerned with the design, implementation, and evaluation of organisational practices that improve performance outcomes. How organisations adopt, operationalise, and sustain these practices across different industries, firm sizes, and national contexts is a central question in organisation and management research. This meta-analysis bridges both fields by synthesising quantitative evidence on the TQM–performance relationship from SEM-based studies spanning manufacturing, service, and mixed-sector organisations.

This study makes four methodological and substantive contributions. First, it provides the meta-analytic synthesis of standardised path coefficients derived from SEM-based TQM studies, moving beyond the correlation-based effect sizes that characterised prior meta-analyses [1,2,3]. Second, it tests TQM operationalisation as a composite construct versus its individual dimensions, a distinction that prior reviews have mainly described qualitatively rather than quantitatively. Third, it examines the role of contextual and methodological factors, including industry sector, firm size, geographic region, and SEM type, as potential sources of between-study variance, providing an assessment of these boundary conditions within a path-coefficient framework. Fourth, it extends previous findings of the TQM—organisational performance relationship, aggregating a body of SEM-based evidence that has grown substantially since 2015.

The rest of this paper is organised as follows: Section 2 reviews the theoretical foundations and construct boundaries; Section 3 describes the methodology; Section 4 presents the results; Section 5 interprets the findings; and Section 6 concludes the study.

2. Theoretical Background and Literature Review

2.1. TQM Dimensional Structure

Total Quality Management (TQM) is often viewed as a broad collection of various organisational concepts [3]. The diversity in scholarly traditions, whether emphasising it as a philosophy, organisational improvement, or a collection of practices and tools, reflects a range of methodological implications [21,22,23]. Studies that operationalise TQM as a single composite construct or decompose it into a multidimensional framework may measure the same phenomenon. However, the measurement outcomes can differ substantially. This is not a methodological weakness but a sign of the construct’s true multidimensionality. For instance, Saraph et al. [24] proposed a multi-item scale for TQM success factors. Flynn et al. [4] later developed a seven-dimensional instrument to assess TQM essentials. Anderson et al. [25] and Powell [26] created refined measurement tools for these dimensions. In addition, business excellence frameworks have defined quality management concepts with operationalisation approaches that are similar yet distinct, influencing a large proportion of empirical studies on TQM practices.

A distinction in the academic literature also exists between “soft” and “hard” TQM concepts. Soft dimensions are mostly related to innovative outcomes, employees, customer-related performance, and their relationships, while hard dimensions are mostly linked to operational efficiency improvements and to mechanistic, tool-based practices such as process management and quality data systems [3,4]. Variation in the application of soft and hard concepts may produce similar dispersion while measuring performance metrics. This distinction is particularly relevant in the context of sustainability-related outcomes. Khalil & Muneenam [18] found that specific TQM dimensions vary considerably in their contribution to corporate green performance, with people management emerging as the dominant driver. Similarly, Akanmu et al. [13] demonstrated that not all TQM dimensions contribute equally to sustainability outcomes, with process-oriented practices showing stronger effects than leadership-related ones. These findings suggest that the soft–hard distinction has implications beyond traditional performance outcomes, and that the importance of TQM dimensions may also vary across organisational sustainability contexts. This variety of operationalisation measures across the literature is a major source of variation between studies that this research aims to explore.

2.2. The TQM–Performance Relationship (RQ1)

Two theoretical frameworks underpin the expectation of a positive relationship between TQM and performance. First, the resource-based view (RBV) treats TQM as an organisational capability that is difficult to imitate, embedded in routines, shared values, and cross-functional practices, rather than in a single tool or technique [27,28]. From this perspective, TQM generates a sustained performance advantage, because competitors cannot easily replicate the tacit knowledge and organisational commitment it requires. Second, contingency theory complements the RBV by predicting that the strength of the TQM–performance effect is not uniform, but depends on the alignment between quality practices and the organisation’s structural, sectoral, and cultural context [28,29]. These two frameworks imply that TQM should have a positive effect on organisational performance. However, the effect might vary across industries, firm sizes, and geographies. Therefore, the fundamental question addressed in this study is whether TQM has a positive effect on organisational performance, while assessing its consistency across different research contexts.

This positive relationship is evident in previous studies across manufacturing and service sectors [27,30,31,32]. This study is based on similar assumptions, including various contextual variables in the study design. For example, Jiménez-Jiménez et al. [33] found a strong positive impact of TQM on organisational performance within the multinational context of European manufacturing firms. Valmohammadi & Roshanzamir [34] verified this relationship among pharmaceutical companies in Iran. Also, Ahmad et al. [35] showed similar findings to those in the Malaysian automotive sector, and Homaid et al. [36] repeated these findings in the Yemeni microfinance sector. More recently, this positive relationship has been confirmed across sustainability dimensions. Abbas [10] found that TQM has a significant positive impact on corporate sustainability across environmental, social, and economic dimensions in Pakistani manufacturing and service organisations. Hassis et al. [37] highlighted this finding in Palestinian manufacturing, reporting that customer focus and human resource management were the most effective TQM practices in driving corporate sustainability, with corporate social responsibility serving as a partial mediator. Hudnurkar et al. [38] confirmed a direct relationship between TQM and corporate sustainability across all three sustainability dimensions in Indian SME manufacturing, noting that innovation capability mediates this link at the dimensional level but not at the aggregate level. Also, Nazarian et al. [17] found that TQM positively influences sustainable development in the Iranian sports goods industry. Jermsittiparsert et al. [39] showed that TQM positively affects sustainable performance in Thai electronics manufacturing. Khan & Naeem [9] demonstrated that strategic quality orientation drives sustainable business growth both directly and through innovation capabilities in Pakistani service industries. Similar positive effects were confirmed by other authors, who linked TQM to sustainable outcomes [18,20,40].

Although the relationship is generally positive, results may vary across different contextual factors. Specifically, Yücel [40] found that the TQM effect on sustainability performance is significantly stronger in medium-sized firms than in small or large firms. In contrast, Saeidi et al. [11] found that organisational age does not mediate the TQM—green performance relationship, while organisational size affects TQM. Khalil & Muneenam [18] found that TQM practices significantly enhance corporate green performance, with organisational culture serving as a mediator. Loedphacharakamon & Worakittikul [20] showed that TQM strongly cultivates a green organisational culture, which in turn mediates its impact on green performance, whereas the direct path did not reach significance.

Since the results of these relationships might differ significantly across studies, it can be argued that this variation may not be merely random noise. Results from Nair [1], Xu et al. [2], Prashar [3], Ahmad et al. [35], and Ahmad et al. [41] suggest that heterogeneity across studies warrants further evaluation. Reviewing 28 studies published between 1995 and 2015, Xu et al. [2] found that most individual QM practices were positively correlated with organisational performance. However, even within specific TQM practices, results varied. Analysing 135 studies from 31 countries, Prashar [3] observed significant residual heterogeneity that was rarely explained by performance and national culture dimensions. It should be noted that the mentioned studies mostly rely on correlations as a measure of effect size.

Beyond the corpus of correlation-based results, Magno et al. [16] documented the increasing dominance of Structural Equation Modelling (SEM) as a key methodology in TQM studies since 2015. Therefore, a substantial amount of empirical evidence based on standardised path coefficients remains to be incorporated into the TQM literature. Additionally, scholars have concluded that disparities in results are not solely due to methodological differences. Such differences may also arise from geographical characteristics, industry-specific factors, and other contextual variables, highlighting that contextual factors may drive between-study variance, an issue that previous studies have poorly addressed. For example, Ahmad et al. [35] and Ahmad et al. [41] discovered that the region of origin influences the average TQM–performance relationship. Ahmad et al. [35] emphasised that studies from Asia tend to report higher estimates than those from Europe or America. Also, manufacturing industries are among the most studied sectors, in contrast to other research contexts [15]. These study characteristics may not be random limitations. They might be a source of variation that should be included in the study design and modelled alongside the pooled effect sizes.

Three prior meta-analyses are particularly relevant for comparison. Nair [1] synthesised correlational evidence across 23 studies establishing a positive TQM–performance relationship, but did not examine path coefficients or test contextual factors. Xu et al. [2] extended this work across 28 studies, yet likewise relied on correlations between individual QM practices and performance outcomes, without addressing SEM-derived estimates. Prashar [3] is the most recent, analysing 135 studies, within the context of national culture, but again on a correlational basis. The present study differs from these three. It restricts its effect-size metric to standardised path coefficients from SEM models, which are not interchangeable with correlations. Furthermore, it focuses exclusively on studies published since 2015, when SEM became the dominant methodology in TQM research [16], emphasising operationalisation type, performance dimensionality, geographic and firm-size characteristics as simultaneous key variables within a single meta-analytic framework.

2.3. Performance Dimensionality (RQ2)

Although TQM consistently shows positive effects on organisational performance across studies, the organisational performance measures used in measurement frameworks vary widely across outcome variables.

Table 1. Synthesis of performance outcomes.

Performance Type	Key Performance Constructs	Mediating/Moderating Mechanisms
AGG—Aggregate	Organisational/firm/business performance, business success, strategic performance, supply chain performance, hotel performance, SME performance, balanced scorecard, organisational excellence, corporate reputation, non-financial performance	Innovation capability, organisational culture, IT capability, knowledge management, lean practices, operational performance, employee satisfaction, competitiveness, process management, innovation
SP—Sustainability	Corporate sustainability, green performance, environmental sustainability, corporate sustainable development, social sustainability, sustainable business growth, sustainability performance, organisational green performance	Green innovation, knowledge management, CSR, organisational culture, talent management, innovation capability, environmental management practices, green organisational culture, employee green behaviour
OP—Operational	Operational performance, efficiency, internal business process, process performance, organisational effectiveness, inventory management, operational quality performance	Agile manufacturing, organisational learning, knowledge management, process management, supply chain practices, innovation speed, product innovation, process innovation
FP—Financial	Financial performance, market performance, financial results, external quality results, and financial and market performance	Knowledge management, non-financial performance, process management, customer focus, leadership, agile manufacturing, supplier quality management
IP—Innovation	Innovation performance, exploitative/explorative innovation, radical/incremental product and process innovation, organisational innovation, marketing innovation	Organisational learning, knowledge transfer, business innovation capability, process management, supply chain integration, proactive behaviour, and environmental management
QP—Quality	Quality performance, product performance, internal quality results	Process management, lean six sigma, Industry 4.0, supply chain integration, customer focus, supplier quality management
CS—Customer Satisfaction	Customer satisfaction, customer results, customer behaviour, customer perspective	Industry 4.0, innovation, continuous improvement
JOBSAT—Job Satisfaction	Job satisfaction, affective commitment, job involvement, turnover intention, employee satisfaction	—
EP—Employee Performance	Employee performance	Customer focus, leadership, process management, supplier quality management

presents the synthesis of performance outcomes used in TQM studies, along with moderating/mediating variables.

The synthesis shown in Table 1 reveals considerable variability across performance outcomes. Although positive effects of TQM on performance lead to heterogeneity in relationship pathways, studies measuring aggregate and sustainability performance show the highest proportions of TQM-positive effects, suggesting that TQM’s impact is most consistently observed when performance is assessed broadly [6,42] or in sustainability terms [10,39,43].

Studies examining operational, quality, and customer-related outcomes follow a similar pattern, with a notably higher proportion of mixed results; however, with fewer contested paths than in financial performance, suggesting that TQM effects vary considerably across studies and that contextual factors may temper TQM effects [44,45,46,47,48,49].

Financial performance is the most contested outcome. Nearly half of all TQM—performance paths are either non-significant or negative, consistent with prior meta-analytic evidence that financial returns from TQM investment might be delayed and sensitive to measurement design [3,8,50,51,52].

Studies on innovation performance are also a heterogeneous category, highlighting the relevance of organisational learning, innovation capability, and knowledge transfer as mediating factors in the TQM–innovation link [53,54]. Similarly, heterogeneity in TQM’s effects on employee-related outcomes suggests that its impact on employee-level outcomes may depend more on intermediate organisational mechanisms [52].

The diversity of mediating and moderating mechanisms identified across categories further underscores that the TQM–performance relationship is not a single, uniform effect, but a context-dependent concept that may vary depending on the outcome being measured.

2.4. TQM Construct Operationalisation: Composite Versus Dimensional (RQ3)

Choosing to operationalise TQM as a composite or as separate dimensions has implications for measurement. The composite approach to operationalisation captures the combined effects among TQM dimensions. In contrast, dimensional operationalisation treats individual practices as separate levers with distinct performance effects [4,24]. Comparing the application of composite measures with individual ones reveals a significant pattern. Composite measures produce significantly greater impact on organisational performance. Measuring TQM as a whole allows the path coefficient to reflect the combined influence of all practices, including their interdependencies. In contrast, modelling each dimension separately yields a path coefficient that captures only the marginal contribution of each. Whether this pattern remains statistically significant when path coefficients are used as the effect-size measure is one of the specific questions this study aims to explore.

This distinction is also reflected in the temporal evolution of measurement practices across the study corpus. Earlier studies, particularly those published between 2015 and 2016, more frequently operationalised TQM through individual dimensions, consistent with the tradition established by Flynn et al. [4] and Saraph et al. [24], who developed multi-item instruments to capture distinct TQM practices separately. Over time, the composite approach has become increasingly prevalent. Recent sustainability-oriented studies illustrate this shift clearly. Abbas [10] operationalised TQM as a composite of six derived practices, applying a single latent construct in the path model. Similarly, Jermsittiparsert et al. [39], Bouzaabia & Ben Salem [43] and Saeidi et al. [11] treated TQM as a single composite factor when examining its effect on sustainable and green performance outcomes. This progression from dimensional to composite operationalisation reflects a broader tendency in the literature to capture the synergistic rather than the additive effects of TQM practices.

2.5. Contextual Moderators: Industry, Geography, and Firm Size (RQ4)

According to contingency theory, organisational performance and management practices should be aligned with specific situational and organisational factors [15,28]. However, such contextual factors may lead to variations in effect sizes, and findings may differ accordingly. For instance, Sila [29] found that contextual factors, such as industry type, firm size, and organisational learning, do not provide support for the argument that TQM and TQM–performance relationships are context-dependent. However, Sila [32] confirmed that country and sector might serve as key moderators of the TQM–performance relationship. He concluded that such control variables should be an integral part of every study design. Alateyyat et al. [15] also reached a similar conclusion, noting that manufacturing organisations are most represented in the literature, while the service and public sectors fall behind.

Geographical context might be an important moderator [3,35,41]. For example, Prashar [3] identified that national culture dimensions, such as uncertainty avoidance and institutional collectivism, are important moderators of the TQM—performance relationship. This study includes papers from South and Southeast Asia and the MENA region, with India, Pakistan, Thailand, Iran, and the UAE being the most frequently studied countries. Consequently, it is essential to investigate whether geographic moderators produce larger effects and whether differences in cultural or broader contextual factors are significant.

Firm size is a third contingency moderator. TQM implementations often rely on dedicated organisational functions, such as quality, training infrastructure, and supplier development programmes. Small and medium-sized organisations may lack the resources needed to support these organisational functions. Abbas [10] found a fairly robust positive effect of TQM on corporate sustainability across environmental, social, and economic dimensions. Yücel [40] highlighted differences in the effect of TQM on sustainability performance between medium-sized firms and small or large firms. Similar conclusions are drawn by Saeidi et al. [11]. They found that organisational age does not mediate the relationship between TQM and green performance, whereas organisational size affects TQM. Consequently, contextual factors, along with firm-size categories, may moderate path coefficients that differ significantly between individual studies.

2.6. Methodological Moderators: CB-SEM Versus PLS-SEM (RQ5)

Covariance-based Structural Equation Modelling (CB-SEM) and Partial Least Squares SEM (PLS-SEM) differ in their treatment of latent constructs, and these differences make SEM type a candidate methodological moderator of the TQM–performance path coefficient. CB-SEM constructs assume reflectiveness, common factors, with an emphasis on overall model fit. In contrast, PLS-SEM treats constructs as weighted composites, optimising them for prediction [16,55]. Such differences may affect path coefficient estimates. Specifically, PLS-SEM’s composite specification tends to maximise explained variance in the dependent construct, which may inflate path coefficients relative to CB-SEM under certain model configurations. Hence, the methodological approach may be a potential source of variance in effect sizes. Magno et al. [16] noted that PLS-SEM applications are subject to fewer established methodological guidelines. This study encompasses both methodological approaches, such as those of Turkyilmaz et al. [42], Iqbal et al. [47], and ElMelegy et al. [48].

3. Materials and Methods

3.1. Search Strategy and Study Selection

A systematic literature search was conducted across Web of Science (WoS), Scopus, and Semantic Scholar databases. WoS and Scopus were accessed between 12 and 15 February 2026, using the following search string:

(“total quality management” OR “TQM practices” OR “TQM”) AND (“structural equation” OR “PLS” OR “AMOS” OR “LISREL”) AND (“organizational performance” OR “organisational performance” OR “firm performance” OR “business performance”).

In WoS, the search used the topic field (TS) and was limited to the Article document type and English language. In Scopus, the equivalent TITLE-ABS-KEY field was applied with the same terms and filters. Semantic Scholar was searched on 16 February 2026 using the same keyword combination, without any publication-year filter.

WoS returned 326 records, Scopus 178, and Semantic Scholar 271. After removing duplicates, WoS and Scopus yielded 398 unique records. The inclusion of Semantic Scholar results yielded 613 unique records. The review was limited to studies published from 2015 onwards. This boundary reflects three related considerations. First, SEM became the dominant methodology in TQM research from the mid-2010s onward [16]. Second, standardised path coefficients are not directly comparable with the correlations on which earlier meta-analyses were built. Combining both metrics would introduce heterogeneity in effect sizes, undermining the validity of the pooled estimates. Furthermore, SEM studies that report correlation matrices primarily assess discriminant validity among constructs rather than isolate the TQM–performance association, making it impractical to extract a consistent correlational effect size from this corpus.

Title and abstract screening identified 80 papers for exclusion because their titles and abstracts did not contain any of the target keywords. Moreover, 238 records were excluded because their publication years were older than 2015. Of the remaining 295 records, 40 were excluded because the full text could not be retrieved. This left 255 records for full-text assessment. Study selection followed the PRISMA 2020 guidelines [56].

Studies were included if they met all of the following criteria: (1) the study was an empirical investigation using CB-SEM or PLS-SEM; (2) the structural model included a direct path from TQM or one of its dimensions to organisational performance, with TQM as the independent variable; (3) standardised path coefficients (β) were reported or could be calculated from available test statistics; and (4) the study was published in a peer-reviewed journal. Studies were excluded if they: (1) employed a non-SEM methodology, Hayes PROCESS macro, or interpretive structural modelling; (2) used a primary independent variable conceptually distinct from TQM (e.g., ISO 9001, EFQM, balanced scorecard, environmental management system); (3) reported only indirect TQM—performance paths without direct ones; (4) were conference papers, book chapters, conceptual works; or (5) had a poor model fit. Figure 1 presents the study selection process in the PRISMA 2020 flow diagram. This review was not registered, and no protocol was prepared prior to data collection. The PRISMA 2020 checklist is provided in Supplementary Material SA.

Figure 1. PRISMA 2020 flow diagram illustrating the selection of 148 coded studies from an initial pool of 613 unique records. The 460 IV paths extracted from these 148 studies form the basis of the vote-counting synthesis reported in

Table 2. Characteristics of included studies (all 148 coded studies).

Characteristic	Value	Note
Studies coded	148	Following full-text eligibility assessment
Years covered	2015–2026	Pre-2015 excluded by design
CB-SEM	79 (53.4%)	Of 148 coded studies
PLS-SEM	69 (46.6%)	Of 148 coded studies
Mean sample size	283 respondents/firms	Median: 244, min 28, max 1069
Top country	India (k = 20)	13.5% of the sample
Total D3 = IV rows	460	Including 76 with E1 = NP
Studies with numeric β	132	Primary analytical sample
Studies in REM (k)	92	With β and t-statistic

. Of the 148 coded studies, 92 met the additional requirement of a reported t- or z-statistic and contributed to the primary random-effects model. A list of included studies with references is given in Supplementary Material SC.

Figure 1. PRISMA 2020 flow diagram illustrating the selection of 148 coded studies from an initial pool of 613 unique records. The 460 IV paths extracted from these 148 studies form the basis of the vote-counting synthesis reported in Table 2. Of the 148 coded studies, 92 met the additional requirement of a reported t- or z-statistic and contributed to the primary random-effects model. A list of included studies with references is given in Supplementary Material SC.

3.2. Coding Procedure

Of the 255 records retrieved for full-text assessment, 148 met the eligibility criteria. They were coded into the database. A database was formed in Microsoft Excel. Each row represented one direct path from a TQM construct to a performance outcome, for each study. Studies reporting multiple direct paths within the same model, for example, TQM—financial performance and TQM—operational performance, were assigned one row per path, with identical bibliographic data but distinct entries for the performance variable and effect sizes.

The database consisted of five blocks. The bibliographic block (Block A) recorded the Zotero reference key, publication year, authors, journal title, and digital object identifier. The contextual block (Block B) included: country, industry sector, firm size (SME, large, or mixed), sample size, and unit of analysis (respondent on organisational or individual level). Block C coded information about study methods (CB-SEM or PLS-SEM), the software used, fit indices, reliability estimates (Cronbach’s α, composite reliability, or both), average variance extracted, and whether a correlation matrix was provided. The construct block (Block D) recorded modelling details, including the name and theoretical mapping of the TQM construct, its role in the model (independent variable, mediator, or moderator), and the name and type of the performance outcome. The results block (Block E) was reserved for the standardised path coefficient (β), its associated t- or z-statistic, p-value, and a directional vote-counting category.

TQM constructs were mapped according to Xu et al. [2]. Therefore, coding included: F1 (Leadership), F2 (People Management), F3 (Process Management), F4 (Product/Service Design), F5 (Quality Data and Information), F6 (Supplier Management), and F7 (Customer Focus). Studies applying TQM as a single composite were coded as “COMPOSITE”. This classification follows the practice established in prior meta-analyses [2,3] to preserve the distinction between composite and dimension-specific TQM measurements. Performance outcomes were classified as follows: aggregate or mixed performance (AGG), sustainability performance (SP), operational performance (OP), financial performance (FP), innovation performance (IP), quality performance (QP), customer satisfaction (CS), job satisfaction (JOBSAT), and employee performance (EP).

The standardised path (β) coefficient extracted from the TQM—performance paths served as the primary effect-size measure. Reported β values alongside a t or z-statistic were used to compute the standard error with SE = |β|/|t| [57]. Studies that reported β without a test statistic were excluded from random-effects calculations, but retained for vote-counting. Studies reporting only directional significance, non-standardised coefficients, or bivariate correlations, without a structural model, were also excluded from the random-effects analysis.

3.3. Statistical Analysis

For statistical analyses, Python v3.14.3 (NumPy v2.4.3, SciPy v1.17.1) was used with the DerSimonian—Laird (DL) random-effects estimator [58]. It is equivalent to the restricted maximum likelihood (REML) method in the R metafor package [59]. A random-effects model was selected because studies do not estimate a single universal effect. Rather, they draw from a distribution of effects that varies across contexts.

3.3.1. Effect Size Aggregation

In the case of multiple direct TQM—performance paths, the within-study mean β and within-study mean SE were computed. Given that the values of individual paths were also recorded, it was possible to aggregate multiple non-independent effect sizes from the same study as a single observation, while preserving individual paths for subgroup analyses [57].

3.3.2. Heterogeneity

Between-study heterogeneity was tested by Cochran’s Q statistic, I², H², τ², and τ [60]. Following academic recommendations, I² values of 25%, 50%, and 75% are classified as low, moderate, and high heterogeneity, respectively. Given the diversity of sectors, countries, and TQM operationalisations, substantial heterogeneity was anticipated.

3.3.3. Sub-Group Analyses

Random-effects models were estimated for each level of the following categorical variables: type of SEM (CB-SEM vs. PLS-SEM), performance outcome (D7: AGG, SP, OP, FP, QP, IP, CS, JOBSAT, EP), TQM factor (D2: “COMPOSITE” vs. F1–F7), industry sector (B2), firm size (B3), geographic region (B1), and unit of analysis (B5). A minimum of 5 studies per subgroup was required for a random-effects estimate. Subgroups with fewer than 5 were used solely for vote counting.

3.3.4. Meta-Regression

Meta-regression complements subgroup analyses by treating predictors as continuous rather than categorical variables. This allows for a more direct test of whether specific study characteristics predict variation in effect size across the full sample. Two predictors were examined, publication year (A4) and sample size (B4). Publication year matters because the corpus spans twelve years, during which measurement frameworks and SEM adoption practices have evolved considerably. If more recently published studies tend to report smaller path coefficients, this would suggest that methodological rigour has improved over time. In contrast, if larger, it might indicate growing enthusiasm for TQM as a research topic, thereby inflating estimates. Sample size was included because larger samples produce more stable estimates. A positive coefficient would indicate that better-powered studies report stronger effects, while a negative one would suggest small-sample inflation in parts of the literature. A multivariate model was also estimated, replacing sample size with a geographic variable (South Asia vs. Europe), to test whether regional context predicts effect size after accounting for publication year.

3.3.5. Sensitivity Analyses

Two sensitivity analyses were conducted. First, a leave-one-out (LOO) analysis was used to estimate the pooled β by excluding each study in turn. The results were assessed using Cook’s distance 4/k. Second, an extended model with 40 studies without t-statistics was used to approximate standard errors via the inverse-normal transformation of reported p-values: SE = |β|/Φ⁻¹ (1 − p/2) [57]. Studies reporting only a non-significant result, without a p-value, were assigned a p-value of 0.10.

3.3.6. Publication Bias

Funnel plot asymmetry was assessed using Egger’s regression [61]. The standard formulation regressing the z-statistic on precision was not used as the primary test, because a single study with an extreme standard error (SE = 1.985) produced an artefactual intercept. The trim-and-fill procedure [62] was applied to estimate the number of potentially missing studies and to assess whether the pooled estimate would change accordingly. Robustness against publication bias was further evaluated using the Rosenthal fail-safe N and the Orwin fail-safe N, computed with a threshold of β = 0.05.

3.4. Distinction Between Q-Between and R²

Two complementary analytical approaches were used to examine between-study variation. In the subgroup analyses, separate random-effects models were estimated for each level of a categorical variable, yielding distinct pooled β values for each subgroup. These analyses address whether the TQM–performance effect holds across different performance types, TQM operationalisations, and how large it is within each category. The moderator analysis takes a different approach. Rather than estimating separate models, it uses two statistics, the Q-between (Q_b) and R², to evaluate whether a categorical variable systematically shifts effect sizes across the full sample. Q_b tests whether the pooled βs differ across subgroups beyond what sampling error would predict. R² quantifies how much of the total between-study variance (τ²) is reduced when that variable is used to partition studies. Crucially, these two statistics can point in different directions. A moderator may account for a large share of τ² by reducing within-group heterogeneity, while Q_b remains non-significant when subgroup means are close to one another.

4. Results

4.1. Study Flow and Sample Characteristics

The database search identified 613 unique records across three sources. Following title and abstract screening, 80 records were excluded because the target keywords were absent from the title or the abstract. A further 238 records were excluded because they predated the 2015 boundary. Of the remaining 295 records, 40 could not be retrieved in full text and were excluded, leaving 255 records for full-text eligibility assessment.

Out of these 255 records, 107 were excluded in full-text review. The key reasons for exclusion were: non-SEM methodology, including only exploratory/confirmatory factor analysis, and/or regression, Hayes PROCESS macro (k = 30); inappropriate independent variable construct, such as ISO 9001, EFQM, or balanced scorecard (k = 24); study containing only indirect TQM—performance paths (k = 17); other eligibility violations, including conceptual works and poor model fit (k = 22); pilot or validation studies without SEM (k = 4); TQM construct(s) only as the dependent variable (k = 4); confirmed duplicates (k = 3); and file retrieval errors (k = 3). The remaining 148 studies were fully coded and entered into the database.

Of the 148 coded studies, 11 were assigned only mediator or moderator roles for TQM, with no direct TQM-to-performance paths, coded as D3 = MEDIATOR or MODERATOR. These were retained, but excluded from the quantitative synthesis, leaving 137 studies with at least one direct path. Further, five studies were excluded because all their IV rows reported no effect size, yielding an analytical sample of 132 studies with at least one numeric β. Of these, 40 studies did not report a t- or z-statistic alongside β, making it impossible to compute the standard error as SE = |β|/|t|. These studies contributed to the vote-counting synthesis and the extended sensitivity model. However, they were excluded from the primary random-effects model. The primary model consisted of k = 92 studies. The complete study flow is presented in Figure 1, which serves as the entry point for the vote-counting synthesis in

Table 3. Vote-counting results by type of performance outcome (D7, N = 460 IV rows).

D7	k Rows	Pos. Sig.	Pos. ns.	Neg.	Pos. Sig. (%)
AGG	155	127	25	3	81.9%
SP	37	30	6	1	81.1%
OP	66	40	25	1	60.6%
FP	61	31	22	8	50.8%
IP	63	36	20	7	57.1%
QP	33	22	9	2	66.7%
CS	24	16	7	1	66.7%
JOBSAT	15	6	8	1	40.0%
EP	6	3	3	0	50.0%
Total	460	311 (67.6%)	125 (27.2%)	24 (5.2%)	—

. Excluded studies are given in Supplementary Material SB, and included ones in SC.

Table 2 presents the characteristics of the 132 studies with numeric β. CB-SEM and PLS-SEM were represented in near-equal proportions. The sample spanned twelve publication years, with a concentration of studies from South and Southeast Asia and the MENA region. India was the most frequently studied country. The mean sample size was 283 respondents or firms, with considerable variation across studies. Of the 460 D3 = IV rows coded across all 148 studies, 384 carried a numeric β, and 76 were coded as E1 = NP; these 460 rows formed the basis of the vote-counting.

4.2. Vote Counting

Vote counting was applied to all 460 D3 = IV rows, across the 148 coded studies, including the 76 rows where β was not reported (E1 = NP). For each row, the directional vote was recorded in column E6 of the coding table as one of four categories: positive and statistically significant; positive and non-significant; negative and non-significant; or negative and statistically significant. Vote counting treats each path as a single observation, regardless of sample size. Results are presented in Table 3.

Across all 460 paths, positive effects predominated overwhelmingly. The combined proportion of positive paths, significant and non-significant, was 94.8%, with only 24 paths (5.2%) coded as negative. Of these, a single path (0.2%) reached statistical significance in the negative direction for the innovation performance category. The distribution of significant positive paths varied across performance types. The 155 AGG paths and the 37 SP paths both showed high proportions of statistically significant positive effects (81.9% and 81.1%, respectively). Job satisfaction (JOBSAT), employee performance (EP), and financial performance (FP) showed greater variability. Customer satisfaction (CS), job satisfaction (JOBSAT), and employee performance (EP) had small sub-samples. These results should be interpreted with caution.

4.3. Primary Random-Effects Model

The random-effects model (DerSimonian—Laird estimator, k = 92) yielded a pooled standardised path coefficient of β = 0.367 (95% CI, 0.283, 0.452, z = 8.56, p < 0.001), indicating a moderate positive effect of TQM on organisational performance. The 95% confidence interval does not include zero, meaning that the pooled estimate is statistically significant at all levels. The model output is presented in

Table 4. Primary random-effects model results (DerSimonian—Laird estimator, k = 92).

Parameter	Value	Interpretation
k (studies in model)	92	From 132 with numeric β; 40 without t-statistic excluded
Pooled β (REM, DL)	0.367	Moderate positive effect
95% CI	[0.283, 0.452]	Does not include zero
z-statistic	8.56	—
p-value	<0.001	Highly significant
Q-statistic	5883 (df = 91)	p < 0.001
I2	98.5%	Extreme heterogeneity
H2	65.12	—
τ2	0.157	Between-study variance
τ	0.396	SD of true effect distribution

. The distribution of study-level effect sizes and 95% confidence intervals is presented in Figure A1, Appendix A.

Between-study heterogeneity was substantial. The Q statistic was highly significant (Q = 5.883, df = 91, p < 0.001), indicating rejection of the null hypothesis of homogeneity. I² indicates that 98.5% of the observed variance in effect sizes reflects true differences in effects across studies rather than sampling error. The 95% prediction interval for the effect of the new study ranges from β = −0.41 to β = 1.14 [57]. This indicates that, while the average effect is positive and moderate, individual study effects vary substantially across contexts. The upper bound exceeds 1.0, which is a known issue in extremely heterogeneous distributions [57].

4.4. Sub-Group Analyses

The subgroup analyses in this section address two research questions.

Table 5. Sub-group random-effects model for performance types.

D7	k	Pooled β	95% CI	p	I2	τ
AGG	54	0.430	[0.318, 0.541]	<0.001	98.9%	0.407
SP	8	0.324	[0.213, 0.435]	<0.001	72.6%	0.125
QP	5	0.419	[0.182, 0.656]	0.001	92.9%	0.258
OP	15	0.350	[0.172, 0.529]	<0.001	95.6%	0.336
IP	13	0.266	[0.142, 0.390]	<0.001	84.1%	0.207
FP	11	0.253	[0.058, 0.448]	0.011	94.3%	0.303
CS	2	—	—	—	—	Vote count only
JOBSAT	1	—	—	—	—	Vote count only
EP	0	—	—	—	—	Vote count only
Qb (6 groups)	—	—	Qb = 5.13, df = 5	p = 0.400 ns	—	—

examines whether the pooled TQM–performance effect holds equally across different outcome categories or is stronger for some performance types than others.

Table 6. Sub-group random-effects model for TQM operationalisation (D2).

D2	k	Pooled β	95% CI	p	I2	τ
“COMPOSITE”	67	0.408	[0.309, 0.508]	<0.001	98.5%	0.398
F1—Leadership	11	0.115	[0.059, 0.170]	<0.001	42.5%	0.058
F2—People Management	11	0.200	[0.115, 0.284]	<0.001	70.4%	0.116
F3—Process Management	11	0.219	[0.131, 0.308]	<0.001	78.8%	0.128
F4—Product Design	4	—	—	—	—	Vote count only
F5—Quality Data	8	0.127	[0.074, 0.180]	<0.001	39.8%	0.048
F6—Supplier Management	5	0.107	[0.038, 0.175]	0.002	25.6%	0.040
F7—Customer Focus	9	0.124	[0.017, 0.231]	0.024	81.6%	0.144
Qb (7 groups)	—	—	Qb = 33.44, df = 7	p < 0.001	—	—

presents results for a similar question regarding TQM operationalisation: Does it matter whether a study measures TQM as a composite or through its individual dimensions? Both tables estimate separate random-effects models for each subgroup with k ≥ 5 studies, so each subgroup has its own pooled β, confidence interval, and heterogeneity estimate. A subgroup with a moderate pooled β and low τ tells a different story than one with a similarly sized β but high within-group dispersion. The first suggests a consistent effect; the second means that something else is still driving variation within that category.

4.4.1. Performance Types

Sub-group random-effects models were estimated separately for each performance type with k ≥ 5 studies. Results are given in Table 5. All six sub-groups yielded statistically significant pooled effects. The largest effects were observed for aggregate performance (AGG, β = 0.430) and quality performance (QP, β = 0.419), followed by operational performance (OP, β = 0.350) and sustainability performance (SP, β = 0.324). Financial performance (FP, β = 0.253) and innovation performance (IP, β = 0.266) showed smaller but still significant effects. Customer satisfaction (CS, k = 2), job satisfaction (JOBSAT, k = 1) and employee performance (EP, k = 0) fell below the minimum threshold for a random-effects estimate. Therefore, they are used for vote counting (Table 3).

The Q-between statistic across the six subgroups was non-significant (Q_b = 5.13, df = 5, p = 0.400), indicating that the pooled β does not differ statistically across performance types. Within-group heterogeneity was substantial across all sub-groups (I² = 72.6–98.9%). Notably, sustainability performance showed considerably lower heterogeneity (I² = 72.6%, τ = 0.125) than other categories, suggesting that TQM–sustainability studies share more consistent contextual conditions. Overall, performance type alone does not account for the observed between-study variance.

4.4.2. TQM Factors

Sub-group models were estimated for seven levels of the TQM operationalisation variable (D2), with k ≥ 5 studies per level. Product/Service Design (F4) with k = 4 fell below the threshold. Therefore, it was included only in the vote counting. Results are presented in Table 6.

The Q-between statistic was highly significant (Q_b = 33.44, df = 7, p < 0.001), indicating that TQM operationalisation produces the highest statistically significant differences in pooled β across subgroups. Studies measuring TQM as a composite construct (“COMPOSITE”) yielded a substantially larger pooled effect (β = 0.408) than those measuring TQM on an individual level (β = 0.107–0.219). It could be argued that a composite measure captures the combined variance of all TQM dimensions. In contrast, a single-dimension measure isolates only one component of what is, in fact, a multifactor construct [2].

Individual dimensions produce lower within-group heterogeneity than “COMPOSITE” studies. F6 (Supplier Management, τ = 0.040) and F5 (Quality Data and Information, τ = 0.048) showed the most consistent effects across contexts, while F7 (Customer Focus, τ = 0.144) and F3 (Process Management, τ = 0.128) showed moderate within-group variability. “COMPOSITE” studies retained the same extreme heterogeneity as the overall model (I² = 98.5%, τ = 0.398), indicating that factors beyond TQM operationalisation drive the variance within subgroups.

4.5. Moderator Analysis

The results of the moderator analysis across categories are provided in

Table 7. Moderator analysis.

Moderator	k *	Qb p	R2	Findings
TQM factor D2	92	< 0.001	—	COMPOSITE β = 0.408 vs. dims. 0.107–0.219
Industry B2	92	0.665 ns	42.2%	mfg. τ = 0.219 vs. service τ = 0.455
Firm size B3	85	0.774 ns	45.1%	SME τ = 0.180 vs. large τ = 0.450
Unit of analysis B5	92	0.700 ns	46.2%	org. τ = 0.194 vs. ind. τ = 0.489
Performance type D7	91	0.400 ns	23.7%	AGG β = 0.430, SP β = 0.324; IP/FP β ≈ 0.253–0.266
SEM type C1	92	0.514 ns	18.7%	CB-SEM β = 0.396 vs. PLS-SEM β = 0.348
Geographic region B1	92	0.304 ns	48.9%	South Asia β = 0.425, Europe β = 0.278

* k per moderator may differ from the model (k = 92) due to missing values in the coding column.

. For each moderator, the Q-between statistic and the R² are reported, in accordance with the methodological guidelines.

As shown, the TQM operationalisation was the only categorical variable to reach statistical significance, indicating that the pooled βs did not differ significantly across industry sector, firm size, unit of analysis, performance level, SEM type, or geographical region.

Nevertheless, several moderators substantially reduced within-group heterogeneity, as reflected in their R² values. The industry sector (B2) accounted for some share of between-study variance (R² = 42.2%), primarily through differences in within-group τ rather than in pooled β. Therefore, manufacturing studies showed lower heterogeneity (τ = 0.219) than service sectors (τ = 0.455), while their pooled β values were similar. Firm size (B3, R² = 45.1%) and unit of analysis (B5, R² = 46.2%) showed comparable patterns, with SME and organisation-level studies providing substantially lower within-group τ than their counterparts. Geographic region (B1) explains the largest share of between-study variance (R² = 48.9%). South Asian studies have yielded a higher pooled effect (β = 0.425) than European ones (β = 0.278). In this regard, geographic region (B1) illustrates a clear distinction between Q-between and R². It produces the largest R² of all moderators examined (48.9%), reducing a notable portion of within-group heterogeneity. However, Q_b does not reach significance (Q_b p = 0.304), because the pooled βs for South Asian (β = 0.425) and European (β = 0.278) studies, although differing in magnitude, fall within the range of sampling variability given the number of studies per region.

4.6. Meta-Regression

Univariate meta-regression models were estimated for publication year and sample size as continuous predictors (k = 92, DL estimator). Neither predictor was statistically significant. Publication year produced a regression coefficient of β = −0.015 (p = 0.306), and sample size yielded a coefficient of β ≈ 0.000 (p = 0.978). These results indicate that the pooled effect of β = 0.367 is stable across the study period 2015–2026 and does not depend on sample size.

A multivariate meta-regression model was also estimated. Results are similar to the univariate ones. Neither predictor achieved statistical significance. Publication year yielded β = −0.014 (p = 0.363), and South Asia versus Europe yielded β = 0.066 (p = 0.546). These results are consistent with the univariate findings and indicate that publication year or geographic region does not predict effect-size variation. Results are presented in

Table 8. Meta-regression results.

Model	K	Predictor	β Coefficient	SE	z	p
Univariate	92	Publication year	−0.015	0.015	−1.023	0.306 ns
Univariate	92	Sample size	≈0.000	0.000	−0.028	0.978 ns
Multivariate	92	Publication year	−0.014	0.015	−0.91	0.363 ns
Multivariate	92	South Asia vs. Europe	0.066	0.110	0.60	0.546 ns

ns—statistically non-significant.

.

4.7. Sensitivity Analyses

The leave-one-out (LOO) analysis re-estimated the pooled β after removing each of the 92 studies in turn. The pooled estimate ranged from β = 0.356 to β = 0.372 across all 92 iterations, representing a maximum deviation of Δβ = 0.011 from the primary estimate of β = 0.367. The most influential study was 2I26TUIZ (β = 0.926), whose exclusion produced the largest fall in value. No study exceeded the Cook’s distance threshold of 4/k = 0.044, confirming that no single study affected the pooled estimate. A change of Δβ = 0.011 across 92 iterations is, by any practical standard, negligible. The pooled path coefficient is stable regardless of which study is excluded. No single observation drives the result, and the corpus as a whole produces a consistent estimate.

As a second sensitivity check, standard errors for the 40 studies lacking a t-statistic were approximated using the inverse normal transformation of the reported p-values (SE = |β|/Φ⁻¹ (1 − p/2)), following Borenstein et al. [57]. Studies reporting only a non-significant result, without a p-value, were assigned a conservative p-value of 0.10. Four studies without a t-statistic or a p-value were excluded. The extended model (k = 130) yielded β = 0.378 (95% CI, 0.308, 0.449, I² = 97.9%, τ = 0.389), which is very close to the primary model (Δβ = 0.011). This confirms that excluding 40 studies from the model due to missing t-statistics did not affect the pooled estimate. Results are given in

Table 9. Sensitivity analysis.

Parameter	Primary Model (k = 92)	Extended Model (k = 130)	Comment
Pooled β	0.367	0.378	Δ = 0.011
95% CI low	0.283	0.308	-
95% CI High	0.452	0.449	-
I2	98.5%	97.9%	Δ = −0.6 pp
τ	0.396	0.389	Δ = −0.007
p-value	<0.001	<0.001	No significant difference

.

4.8. Publication Bias

Funnel plot asymmetry was assessed using Egger’s regression [61]. The intercept was statistically significant (0.376, t = 13.85, p < 0.001), indicating a small-study effect. Therefore, less precise studies tend to report larger (β) path coefficients (SE ≈ 1/√n).

The standard Egger’s test formulation (z ~ precision) yielded an intercept of −6.404, but this could be interpreted as an artefact. One study (Q6NVI3HZ) had an unusually large standard error (SE = 1.985). The study was not excluded because its large standard error reflects sampling variability rather than a coding error. Not removing it was an arbitrary decision. Instead, the β ~ SE formulation was used as the primary test.

The trim-and-fill [62] estimated L0 = 0 missing studies, indicating that the observed funnel asymmetry does not reflect omission of studies in a specific direction [63]. The adjusted pooled estimate was identical to the original (β = 0.367, Δβ = 0.000).

Robustness of the pooled effect against publication bias was further evaluated using two fail-safe N statistics. The Rosenthal fail-safe N of 89,978 indicates that approximately 90,000 unpublished null-result studies would be required to reduce the pooled effect to non-significance, showing the robustness of the pooled estimate against publication bias. The Orwin fail-safe N of 578 indicates that 578 studies with a mean effect of zero would be required to reduce the pooled β to below the 0.05 threshold. Therefore, the Rosenthal threshold of 89,978 is more than 350 times that number, meaning that for publication bias alone to render the pooled effect non-significant, the literature would need to contain a hidden body of null results far exceeding anything plausible in this field. The Orwin threshold of 578 tells a similar story at a more conservative benchmark. Taken together, these figures confirm that the pooled estimate of β = 0.367 is not a statistical artefact of selective publication. Results are summarised in

Table 10. Publication bias assessment summary.

Test	Result	Interpretation
Egger’s test (β ~ SE)	intercept = 0.376, t = 13.85, p < 0.001	Significantly small-study effect
Egger’s test (z ~ precision)	intercept = −6.404	Artefact—not used as primary test
Trim-and-fill L0	0 missing studies	No systematic omission of studies
Adjusted pooled β	0.367	Δβ = 0.000
Rosenthal fail-safe N	89,978	No practical effect
Orwin fail-safe N (β = 0.05)	578	No practical reduction of the pool effect

. The funnel plot is presented in Figure A2, Appendix B.

5. Discussion

Based on 92 SEM-based studies published between 2015 and 2026, a meta-analysis of TQM—performance relationships was conducted. Previous meta-analytic studies were mostly based on correlations [2,3]. In contrast, this study focuses on examining standardised path coefficients, including the effects of contextual factors.

5.1. Overall Effect (RQ1)

The pooled path coefficient, β = 0.367 (95% CI, 0.283, 0.452; p < 0.001), confirms that TQM has a positive, statistically significant effect on organisational performance. This influence can be categorised as moderate in strength [64]. Vote counting across 460 structural paths provides the same conclusion (94.8%). The results are consistent with previous studies. For example, Ahmad et al. [41] reported a positive relationship between TQM practices and business performance, emphasising larger effect sizes in Asian studies than in European ones. Xu et al. [2], examining 28 studies published between 1995 and 2015, reported positive correlations between QM practices and organisational performance across all individual TQM dimensions. On a sample of 135 studies from 31 countries, Prashar [3] underlined positive associations between TQM and all performance types. However, it should be noted that path coefficient sensitivity might depend on the predictors included in the structural model.

The overall estimate incorporates a growing body of evidence linking TQM to sustainability performance. All sustainability-oriented studies in the research corpus show predominantly positive effects of TQM on sustainability-related performance, spanning manufacturing and services, across South Asia, Southeast Asia, the MENA region, and beyond. These positive effects are documented across diverse contexts, including Palestinian manufacturing [37], Tunisian manufacturing [43], UAE manufacturing SMEs [12], Jordanian manufacturing [19], Pakistani services [9,18], Indian MSMEs [38], and Thai service sectors [20]. This convergence of evidence across diverse contexts reinforces the conclusion that TQM’s positive performance effect extends to environmental, sustainable, and social dimensions, not only to conventional operational or financial outcomes [10,11,39,43,65].

Such results are consistent with the broader relationship between TQM and sustainability, which has received growing systematic attention. Silvestri et al. [66] conducted a systematic literature review of 139 studies published between 1996 and 2023. They examined the links between TQM and sustainability through the Triple Bottom Line framework, concluding that TQM catalyses sustainability. Abdellaoui Fethallah et al. [67] demonstrated that TQM produces significant positive effects across all three sustainable development dimensions, economic, social, and environmental. Ronalter et al. [68], analysing 4292 companies, found that firms operating quality management systems achieve statistically significantly higher ESG scores than companies without such systems. The present meta-analytic findings, based on SEM-derived path coefficients from 92 studies, are consistent with this body of evidence. The pooled effect of β = 0.367 reflects not only conventional performance gains but also the cumulative influence of TQM on sustainability-oriented outcomes, as evidenced by the SP subgroup (β = 0.324, k = 8).

5.2. Performance Dimensions (RQ2)

Organisational performance outcomes show a consistent pattern across sub-groups, even though the differences between them do not reach statistical significance (Q_b = 5.13, df = 5, p = 0.400). Aggregate performance shows the strongest pooled effect (β = 0.430), with quality performance close behind (β = 0.419), followed by operational performance (β = 0.350), sustainability (β = 0.324), innovation (β = 0.266) and financial performance (β = 0.253). The strength of the aggregate performance effect is consistent across the corpus.

These results are similar to previous findings. For example, Anil & Satish [6] documented positive TQM effects across all seven performance indicators. Additionally, composite measures might yield higher values due to cumulative effects across multiple domains. Similarly, Sahoo [69] found that TQM practices have strong direct effects on aggregate manufacturing performance. Abbas [10] documented a moderate positive effect of TQM on corporate sustainability. Hamdoun et al. [65] linked quality management to sustainability-oriented innovation outcomes.

The sustainability performance showed the lowest within-group heterogeneity of all performance categories (I² = 72.6%, τ = 0.125). This suggests that TQM produces a relatively consistent effect on sustainability outcomes across diverse contexts, more so than financial or innovation performance. This consistency is noteworthy because SP constructs across the corpus are not uniform. Some studies operationalise sustainability as a composite outcome [10,12,37], others target specifically green or environmental performance [18,19,43], while others measure individual Triple Bottom Line dimensions separately [38]. Despite this conceptual diversity, the pooled effect remains stable, suggesting that TQM’s positive influence on sustainability is robust across different operationalisations. Silvestri et al. [66] similarly concluded, across 139 studies, that TQM catalyses sustainability across TBL dimensions.

In studies that decompose the sustainability concept into individual dimensions, environmental and social outcomes consistently show stronger TQM effects than economic outcomes. Abdellaoui Fethallah et al. [67] reported stronger path coefficients for the environmental and social dimensions than for the economic one. Ronalter et al. [68] confirmed this at a larger scale, finding that firms with quality management systems achieve significantly higher, more consistent ESG scores across environmental and social dimensions. At the same time, the governance pillar showed comparatively weaker gains. This dimensional pattern suggests that TQM’s sustainability effect operates through operational and human practices, such as process management, employee involvement, and waste reduction. These practices more directly affect environmental and social outcomes than economic or governance ones. Future research should systematically disaggregate sustainability performance into its constituent dimensions to better understand where TQM’s sustainability effect is strongest.

Financial performance is the weakest of the performance dimensions. This is consistent with Prashar’s [3] finding that financial performance yields the smallest average effect across TQM dimensions. Kulenović et al. [51] reported near-zero or negative direct effects of TQM on financial performance, attributing this to the mediating role of innovation performance. Augustyn & Akamavi [8] found that the specifics of the structural model strongly influence the TQM–financial performance relationship. Similarly, Zehir & Zehir [49] found near-zero direct paths from several TQM dimensions to financial performance. Thus, it could be argued that the benefits of TQM implementation lag behind in financial gains.

The lower effect for innovation performance might imply that process standardisation constrains the creativity required for innovation outcomes. Escrig-Tena et al. [53] documented mixed innovation performance across TQM dimensions, arguing that the relationship depends on the balance between hard and soft TQM practices. Mushtaq & Peng [5] found that TQM’s effect on innovation performance is predominantly indirect, with business innovation capability as the mediating factor. Honarpour et al. [70] found a positive relationship between TQM and process innovation, emphasising the significant role of knowledge management. They suggest that TQM alone is insufficient to drive innovation without the support of knowledge management practices.

5.3. TQM Operationalisation (RQ3)

The most notable finding of the subgroup analysis concerns the operationalisation of TQM. The between-group Q-statistic is highly significant (Q_b = 33.44, df = 7, p < 0.001), underscoring the TQM factor’s statistical significance. Studies operationalising TQM as a composite yielded a pooled effect of β = 0.408, compared to studies that operationalise TQM as individual dimensions. Composite operationalisation combines the variance of all dimensions simultaneously. The path coefficient reflects the cumulative effect of the TQM composite, rather than models in which TQM is operationalised through individual dimensions. This relates to previous findings. Anil & Satish [6] found that composite TQM operationalisation produces strong effects on organisational performance. Carmona-Márquez et al. [71] and Prashar [3] reached similar conclusions.

However, with higher effects comes greater heterogeneity for the composite measures, in contrast to smaller but more consistent effects for individual dimensions. This pattern is informative when examined in the context of sustainability performance. Corpus studies that examine sustainability performance operationalise TQM mostly as a composite construct. Therefore, it would be expected that the SP subgroup would show similarly high within-group heterogeneity. However, the SP subgroup yields a considerably lower value (I² = 72.6%). This suggests that sustainability as a performance outcome introduces a certain consistency. In other words, when the outcome is sustainability, composite TQM studies converge more tightly than they do in the case of other performance dimensions. Dimensional operationalisation in sustainability studies likely yields smaller but more consistent effect sizes [67].

5.4. Contextual Moderators and Heterogeneity (RQ4)

The between-study heterogeneity is substantial (I² = 98.5%, τ = 0.396). The 95% prediction interval for the true effect in a new study encloses β in the range (−0.41, 1.14). This level of dispersion indicates that the effect size of TQM—performance, while robust, varies substantially across contexts. The moderator analyses provide partial explanations for this variation.

TQM was the only moderator with statistical significance. This implies TQM operationalisation has a meaningful effect on measurement outcomes. The industry sector explained 42.2% of the between-study variance in R², with manufacturing studies showing considerably lower heterogeneity (τ = 0.219) than service studies (τ = 0.455). Firm size accounted for 45.1%, with SME studies clustering more tightly (τ = 0.180) than large-firm studies (τ = 0.450). Unit of analysis explained 46.2%. None of these moderators had a significant effect on shifting β. This aligns with findings that the TQM—performance relationship is robust against contextual factors [29]. Although not statistically significant, contextual factors may account for some of the variance in the measurement framework [32].

Geographic region does explain some variance. However, it is not a significant categorical moderator (Q_b p = 0.304, R² = 48.9%). South Asian studies yielded a higher pooled effect (β = 0.425) than European ones (β = 0.278), though this difference does not reach statistical significance. The results of Ahmad et al. [41], Carmona-Márquez et al. [71], Sila [32], Anil & Satish [6], and Sahoo [69] provide grounds for concluding that European studies show considerably more variability and that Asian studies yield higher TQM–performance associations. This might be due to factors which were not directly measured in this study. However, they seem to be embedded in the measurement framework and the national research context. For example, Prashar [3] identified national culture dimensions as moderators of the TQM—performance relationship. In addition, the multivariate model did not yield a statistically significant trend in publication year (β = −0.014, p = 0.363). The direction of the coefficient suggests that more recently published studies may report slightly smaller path coefficients, though this pattern does not reach significance.

The sustainability-oriented studies in the corpus span a broad range of industries, including manufacturing [12,19,37,43], services [9,18,20], and firm sizes, ranging from SMEs [12,13,38] to large organisations [17,18]. Despite this contextual diversity, the SP studies show the lowest within-group heterogeneity of all performance categories (I² = 72.6%). This suggests that studies targeting sustainability outcomes are more homogeneous than those targeting other performance dimensions. This finding is broadly consistent with Ronalter et al. [68], who found that the positive relationship between quality management systems and ESG performance holds across Europe, East Asia, and North America, and across different company sizes.

It should be noted that non-significant Q_b statistics do not rule out the presence of moderation. In subgroup analyses, the statistical power of the between-group test depends on the number of studies per subgroup and the precision of their estimates. Where subgroups are small or heterogeneous within themselves, even between-group differences may not reach statistical significance. This is a recognised limitation of moderator analyses in meta-analysis more broadly, and applies to the findings reported in Table 5, Table 6 and Table 7. The null results for industry sector, firm size, unit of analysis, geographic region, and SEM type should therefore be interpreted as inconclusive rather than as evidence of the absence of moderation. Larger, more balanced corpora with greater representation across sectors, regions, and firm sizes would provide a more definitive test of these potential boundary conditions.

5.5. SEM Method Bias (RQ5)

CB-SEM and PLS-SEM studies produced similar pooled estimates (β = 0.396 and β = 0.348, respectively). The difference is small and statistically non-significant (Q_b p = 0.514). This finding is relevant to ongoing debates over the comparability of the two approaches. Magno et al. [16] conducted a systematic review of 107 PLS-SEM applications, identifying systematic deficiencies in its application. Researchers frequently do not follow established guidelines for structural model evaluation, such as using insufficient bootstrap samples, conducting inadequate predictive model assessment, and failing to evaluate higher-order constructs adequately. Despite these application deficiencies, the present data do not support the assumption that PLS-SEM produces systematically larger path coefficients than CB-SEM in the TQM literature. The SEM type does not drive the pooled estimate of β = 0.367.

The nature of research itself might explain the predominance of PLS-SEM in sustainability-oriented TQM studies. Sustainability constructs are inherently complex and multidimensional, and are often measured in emerging economy contexts with smaller sample sizes, which provide recognised justifications for selecting PLS-SEM over CB-SEM. However, the concerns identified by Magno et al. [16] may also apply to sustainability studies. If PLS-SEM applications in sustainability-oriented TQM research do not follow established guidelines, the path coefficients may not be directly comparable across studies.

Future research should address this methodological gap directly. A systematic comparison of CB-SEM and PLS-SEM estimates within TQM studies, conducted under controlled conditions, might clarify whether the SEM-type equivalence observed in the overall model holds specifically for performance outcomes.

5.6. Publication Bias

Egger’s test detected a significantly small-study effect (intercept = 0.376, t = 13.85, p < 0.001), indicating that less precise studies tend to report larger path coefficients. However, the trim-and-fill procedure estimated zero missing studies, leaving the pooled estimate unchanged at β = 0.367. This apparent contradiction is caused by the substantial between-study heterogeneity. When true effects vary substantially, funnel plot asymmetry reflects variation across contexts rather than suppressed null results [63]. The fail-safe N statistics strengthen this conclusion. The Rosenthal fail-safe N of 89,978 and the Orwin fail-safe N of 578 indicate that the pooled estimate is quite robust. Thus, publication bias does not appear to be an issue for these findings.

5.7. Limitations

This review has several limitations worth noting.

The pooled estimate is a standardised path coefficient (β) from structural equation models, not a Pearson correlation. Direct numerical comparison with prior correlation-based meta-analyses is not appropriate.

Forty studies were excluded from the primary model because they did not report a t- or z-statistic alongside β. A sensitivity analysis including these studies (k = 130) produced a nearly identical estimate (Δβ = 0.011). Studies without t-statistics may slightly skew toward the positive, as non-significant results are less likely to report test statistics.

The corpus skews heavily toward South and Southeast Asia and the MENA region, and toward private manufacturing firms. Service, public, and third-sector organisations are underrepresented. Almost all primary studies use cross-sectional survey designs, which limit causal inference and might introduce common-method bias. The search was also restricted to English, which likely misses relevant work given the field’s geographic profile.

A single coder carried out all screening and data extraction. For numerical values, this is a manageable limitation. Beta and t statistics were read directly from reported tables and cross-checked using SE = |β|/|t|. However, categorical decisions are a different matter. Classifying TQM operationalisation type, performance category, and study eligibility all involve decisions that a second independent coder would have challenged and refined. Future studies conducting similar meta-analyses should treat dual coding of categorical variables, with reported Cohen’s kappa or percentage agreement, as a methodological standard rather than an optional step.

No formal risk-of-bias tool was applied. No validated instrument exists for cross-sectional SEM studies in management research. Heterogeneity was handled statistically through I², τ², subgroup analyses, and meta-regression.

These limitations should be weighed against several strengths. To the authors’ knowledge, this study constitutes the largest SEM-specific synthesis of TQM–performance evidence to date, with 148 coded studies, 460 individual structural paths, and a verified primary model of k = 92 studies. Using standardised path coefficients rather than correlations avoids the scale-dependency problems that affect prior meta-analyses and keeps the effect-size metric directly interpretable within the SEM framework that produced it. All effect sizes were cross-verified using SE = |β|/|t|, which substantially reduces the risk of transcription error. The pooled estimate was then stress-tested from multiple angles, including leave-one-out analysis, an extended sensitivity analysis (k = 130), and three complementary publication bias assessments, yielding consistent results across all. The subgroup analyses cover six performance types and seven TQM operationalisation categories, each with its own pooled estimate and heterogeneity profile. Finally, by treating sustainability performance as a distinct outcome category, this study extends TQM–performance synthesis into a domain that previous correlation-based meta-analyses have not reached.

5.8. Practical Implications

The findings of this meta-analysis carry several implications for practitioners and policymakers. Organisations should consider investing in TQM, given that a positive TQM—performance relationship holds regardless of sector, geographic context, or firm size.

Also, managers should be aware that organisations deploying TQM as an integrated management system, where TQM dimensions are pursued simultaneously, report substantially stronger performance gains than those that target individual practices. Selective adoption of single TQM dimensions, such as supplier management or leadership development alone, captures only a fraction of the performance improvement available through coordinated, organisation-wide implementation.

In particular, TQM’s effect on sustainability outcomes is not only statistically significant but also more consistent across contexts than its effect on financial or innovation performance. Such a conclusion holds across studies of different contexts, including UAE manufacturing SMEs, Palestinian manufacturing, Indian MSMEs, and Thai service sectors. For organisations that strive to strengthen ESG performance, this convergence of evidence across diverse industrial and geographic contexts suggests that TQM represents a well-evidenced and practically viable instrument for sustainability improvement, extending beyond its traditional role as an operational management tool.

6. Conclusions

Across 92 SEM-based studies published between 2015 and 2026, TQM consistently improves organisational performance. Vote counting across 460 independent paths points in the same direction. Positive effects dominate across all performance types.

TQM operationalisation also matters. Composite implementations, which capture the effects of multiple practices, produce stronger effects than those testing individual dimensions. Geography and industry shape the variance, though neither emerged as a statistically significant moderator in meta-regression. This may be because the corpus is too concentrated in the South and Southeast Asian manufacturing sectors.

Future studies should prioritise longitudinal designs, broader sectoral coverage beyond manufacturing, greater representation of underrepresented regions and non-English literature. The growing body of evidence on TQM–sustainability suggests that TQM practices enhance ESG and sustainability performance.