A panel data analysis of labour productivity determinants in Vietnam’s textile and garment industry

The textile and garment industry has long been one of Vietnam’s most important manufacturing sectors, playing a central role in employment creation, export earnings and industrialisation. Since the Doi Moi reforms, the industry has experienced rapid expansion, driven by trade liberalisation, integration into global value chains and relatively abundant labour supply. Despite its sustained growth and strong export performance, labour productivity in the sector remains modest compared with regional competitors, posing a key challenge for Vietnam’s efforts to move up the value chain and achieve more sustainable industrial development. Improving labour productivity is particularly critical for labour-intensive industries such as textiles and garments, where competitiveness increasingly depends not only on low labour costs but also on efficiency, quality upgrading and institutional support. In recent years, rising wages, tighter labour markets and growing international competition have intensified pressure on Vietnamese firms to enhance productivity through technological upgrading, better workforce management and improved access to finance. However, productivity performance across firms remains highly heterogeneous, reflecting differences in capital intensity, scale, market positioning and local business environments. A growing body of international literature has examined the determinants of labour productivity in manufacturing industries, highlighting the roles of capital deepening, labour quality, technological capability, agglomeration effects and institutional conditions (Lin et al., 2011; Islam and Syed Shazali, 2011; Fu, 2015). These studies provide valuable insights but are often based on specific country contexts or rely on cross-sectional data, limiting their ability to capture productivity dynamics over time. Moreover, empirical evidence for developing and transition economies – particularly at the firm level – remains relatively limited. In the Vietnamese context, existing studies on labour productivity in the textile and garment industry are still fragmented. Several studies rely on survey-based or cross-sectional data and focus on specific localities or narrow sets of explanatory factors (Viet, 2018; Ha et al., 2023). While these studies contribute to understanding firm behaviour and workforce characteristics, they do not adequately address unobserved firm heterogeneity or productivity changes over time. Other studies examine broader institutional or trade-related factors but do not directly link firm-level productivity outcomes with sub-national institutional quality (Le, 2020; Tran et al., 2021). As a result, there remains a lack of comprehensive panel-data-based evidence on the determinants of labour productivity in Vietnam’s textile and garment sector at the national level.

This study seeks to fill this gap by examining labour productivity determinants using a large firm-level panel data set covering Vietnamese textile and garment enterprises over the period 2017–2023. By using panel data econometric techniques, the analysis is able to control for time-invariant unobserved firm characteristics – such as managerial practices or organisational structure – and to capture within-firm productivity variations over time. This approach provides more reliable estimates than cross-sectional methods and allows for a clearer identification of key productivity drivers. Specifically, the study investigates the effects of capital intensity, labour scale, labour quality, financial leverage, market share and provincial institutional quality on enterprise-level labour productivity. By integrating firm-level characteristics with sub-national institutional indicators, the analysis offers a more comprehensive perspective on productivity dynamics in a labour-intensive export-oriented industry. In addition, the study explores heterogeneity across sub-sectors, ownership types and firm size, thereby shedding light on how productivity determinants vary across different segments of the industry. The contribution of this study is threefold. Firstly, it provides new empirical evidence on labour productivity determinants in Vietnam’s textile and garment industry using large-scale panel data. Secondly, it extends the productivity literature by jointly considering firm-level inputs and institutional quality in a developing-country context. Thirdly, it offers policy-relevant insights for industrial upgrading, workforce development and institutional reform aimed at enhancing productivity and sustaining long-term competitiveness in Vietnam’s manufacturing sector.

Research on labour productivity in manufacturing industries highlights the importance of capital accumulation, labour attributes, firm structure and institutional conditions. In labour-intensive sectors such as textiles and garments, productivity performance is shaped by the interaction between firm-level production inputs and broader structural and institutional conditions, particularly in developing and transition economies where technological upgrading remains uneven.

Capital intensity and technological capability are widely recognised as key determinants of labour productivity. Empirical studies consistently show that higher capital–labour ratios enhance productivity by enabling firms to adopt more efficient production processes and improve resource utilisation (Islam and Syed Shazali, 2011; Fu, 2015). Firm-level evidence from Asian textile industries suggests that investment in machinery and equipment contributes positively to productivity, although the magnitude of the effect is often constrained by the labour-intensive nature of production and limited complementary skills (Lin et al., 2011). Using data envelopment analysis combined with the Malmquist productivity index, Chia-Nan Wang et al. (2022) demonstrated that capital-related inputs such as total assets and production facilities play a significant role in explaining productivity differences among Vietnamese textile and garment enterprises, highlighting the importance of capital deepening even in labour-intensive industries.

Labour-related factors constitute another central strand of the productivity literature. Labour quality – commonly proxied by wages, skills or training – has been found to improve productivity by enhancing workers’ efficiency, adaptability and motivation (Islam and Syed Shazali, 2011). Case-based evidence from the textile industry further underscores the importance of human capital and technical competence. Ephrem (2019), in an in-depth study of a textile enterprise in Ethiopia, identifies inadequate technical training, weak maintenance skills and poor organisational practices as major constraints on productivity, suggesting that labour quality and management capacity are critical complements to capital investment. In contrast, the relationship between labour scale and productivity is less clear-cut. While firm expansion may generate economies of scale, empirical studies frequently report diminishing or negative returns to labour in labour-intensive manufacturing when workforce growth is not matched by improvements in technology and management (Tekleselassie et al., 2018). Evidence from Vietnam similarly indicates that rapid expansion in labour size can reduce average productivity due to co-ordination problems and inefficient task allocation (Viet, 2018; Ha et al., 2023).

Financial structure and market positioning further influence productivity outcomes at the firm level. Access to external finance and the use of financial leverage can facilitate investment in capital, technology and working capital, thereby supporting productivity growth (Njagi et al., 2017). However, excessive reliance on debt may expose firms to financial vulnerability, particularly in export-oriented industries subject to demand shocks. Market share, by contrast, is often associated with higher productivity through economies of scale, learning-by-doing and stronger bargaining power within value chains. Empirical studies indicate that firms with larger market shares tend to exhibit superior productivity performance, especially in highly competitive international markets (Fu, 2015).

Beyond firm-level characteristics, a growing literature emphasises the role of the institutional environment in shaping productivity and performance. Institutional quality – capturing regulatory efficiency, transparency and administrative capacity – affects transaction costs and firms’ incentives to invest and innovate. In the Vietnamese context, studies show that improvements in provincial-level governance and business-support institutions are positively associated with firm productivity (Le, 2020). Related research on the garment sector highlights the importance of trade-related institutions and policy frameworks. Tran et al. (2021), using a stochastic frontier approach, find that institutional factors such as trade facilitation, business freedom and infrastructure quality significantly enhance export efficiency in Vietnam’s garment industry, indirectly reinforcing firms’ productivity and competitiveness.

Overall, the existing literature highlights multiple channels through which labour productivity is determined but also reveals important limitations. Many studies rely on cross-sectional or survey-based data, focus on isolated determinants or examine productivity indirectly through performance or export outcomes. Moreover, few studies integrate firm-level production characteristics with sub-national institutional conditions using panel data. By addressing these gaps, the present study contributes new empirical evidence on labour productivity determinants in Vietnam’s textile and garment industry, offering a more comprehensive understanding of how capital, labour, market structure and institutional environments jointly shape productivity in a developing-country context.

To estimate the factors affecting labour productivity, this study uses the estimation model presented in the study of Bhaumik et al. (2012) and establishes the research model from the Cobb–Douglas production function as follows:

where:

• $Y$ is the output;

• $K$ and $L$ are, respectively, the capital and labour levels used by the enterprise in the production process;

• $A>0$ is a parameter representing the production technology; and

• $0<α,β<1$ are parameters representing the individual elasticity of output with respect to production inputs.

Dividing both sides of the production function by, we get the equation representing labour productivity $(Y/L)$ in terms of total factor productivity (⁠$A$⁠), capital stock per worker (⁠$K/L$⁠)and firm size measured by labour size (⁠$L$⁠):

Total factor productivity can be described by the equation:

With $Z$ being a vector of control factors affecting technological progress such as firm characteristics and institutional environment quality and finally random error representing other unobserved factors. In addition, the study extends the model by incorporating additional determinants of firm labour productivity, namely labour quality (Liu et al., 2001; Kien, 2008), firm market share (Le and Ho, 2020), and other firm-specific factors (Njagi et al., 2017).

In combination, the study proposes to use the following model to study the factors affecting labour productivity of textile and garment enterprises:

With $i$ is the index showing the business $(i=1,N-$⁠), $t$ is the time index (⁠$t=1,T-$⁠), $j$ is an index representing the province (⁠$j=1,J-$⁠). $LP$ is labour productivity, $LQ$ is a variable representing labour quality, $DR$ is the debt ratio, $TP$ is the business market share, $PCI$ is a proxy for institutional quality, $ui$ represents the fixed, time-specific characteristics of each firm and $εit$ is the random error of the model.

The equations above are based on three regression models with panel data, including pooled ordinary least square (OLS), fixed effects (FE) or random effects (RE). These models are based on different assumptions and will be selected through appropriate tests when research data is available. The pooled OLS model is based on the assumption that individual characteristics of enterprises do not affect labour productivity (⁠$ui=0$⁠). If $ui≠0$⁠, the FE and RE models would be more appropriate. The FE model would be used in the context of firm-specific characteristics that are correlated with other explanatory variables in the model. Then, by eliminating the cross-sectional mean over time for each unit, the endogeneity effect caused by $ui$ will be eliminated and the coefficient estimates of the productivity equation will be consistent estimates. In contrast, the RE model with the assumption $ui$ uncorrelated with the independent variables and establishes a two-component composite random error structure $ui$ and $εit$⁠. This error structure has a time-invariant autocorrelation phenomenon, so generalized least squares estimation is performed to solve that problem and obtain efficient estimates for the slope coefficients in the above equation.

This study uses an unbalanced panel data set of Vietnamese textile and garment enterprises over the period 2017–2023, constructed from the Annual Enterprise Survey conducted by the General Statistics Office of Vietnam. After data cleaning procedures – including the removal of firms with missing key variables, negative or zero value added and extreme outliers in capital and labour inputs – the final sample consists of 15,703 firms and 44,780 firm-year observations. Firms are classified by ownership (state-owned, private and foreign-invested), by sub-sector (textile and garment) and by firm size following the official classification of the General Statistics Office.

The first data set is sourced from the Annual Enterprise Survey conducted by the General Statistics Office of Vietnam. This is one of the most comprehensive and reliable data sets available for empirical research on Vietnamese enterprises. It contains detailed firm-level information on human resources, fixed assets, revenue, cost structures, profits, technological adoption, market access and export–import activities. Importantly, it also includes indicators related to value-added composition, innovation inputs, technological investment and participation in global supply chains. The breadth and depth of the survey enable nuanced assessments of enterprise performance across ownership types (state, private and Foreign Direct Investment), sectors, and regions. The richness of this data makes it a critical foundation for evaluating firm-level productivity, informing evidence-based policymaking and tracking structural transformation within the Vietnamese economy (Table 1).

The second data set used is the Provincial Competitiveness Index (PCI), obtained from the Vietnam Chamber of Commerce and Industry and accessible at Link to the cited article Developed in collaboration with the United States Agency for International Development, the PCI has been published annually since 2005. The index serves as a diagnostic tool to assess the quality of sub-national economic governance across Vietnam’s provinces and municipalities. PCI dimensions include access to land, transparency in governance, informal cost burdens, administrative procedure efficiency, dynamism and proactivity of local leadership, public service support for private sector development and infrastructure quality. These metrics reflect not only administrative effectiveness but also institutional commitment to private sector development and regulatory reform. The PCI data are increasingly used in empirical analyses to evaluate how regional institutional environments influence firm performance and competitiveness.

Together, these two data sets provide a robust empirical foundation for analysing how internal firm characteristics and external institutional contexts jointly influence labour productivity in Vietnam’s textile and garment industry.

The selection of explanatory variables is grounded in both productivity theory and empirical evidence. Capital intensity (capital–labour ratio) captures the role of capital deepening and technological adoption, consistent with the Cobb–Douglas production framework (Islam and Syed Shazali, 2011; Fu, 2015). Labour size reflects scale effects and potential co-ordination inefficiencies in labour-intensive production. Labour quality, proxied by average wage per worker, represents human capital and incentive effects (Islam and Syed Shazali, 2011; Ephrem, 2019). Financial leverage is included to capture firms’ access to external finance and investment capacity (Njagi et al., 2017). Market share reflects competitive positioning and economies of scale, while provincial institutional quality (PCI) captures the external business environment affecting firm productivity (Le, 2020).

Descriptive statistics of the variables in the study are presented in Table 2, providing an overview of the distribution and variability of the variables.

This study uses a set of continuous variables derived from enterprise- and province-level data to analyse the determinants of labour productivity in Vietnam’s textile and garment industry. All key variables are log-transformed to reduce skewness and enable elasticity-based interpretation in the regression analysis.

Ln(LP) (log of labour productivity): The mean value is 5.565 with a standard deviation of 1.296, indicating substantial variation in productivity across enterprises. Labour productivity, calculated as value added per worker, reflects differences in operational efficiency, labour quality, technological intensity and firm size.

Ln(K/L) (log of capital–labour ratio): With a mean of 5.836 and standard deviation of 1.316, this variable shows heterogeneity in capital intensity across firms. A higher capital–labour ratio is typically associated with greater technological adoption and automation, while lower values suggest labour-intensive production.

Ln(L) (log of labour size): The mean value is 3.056 with a relatively high standard deviation of 1.769. This reflects wide variation in enterprise scale, from micro-firms with one worker (minimum of 0 in log scale) to large firms with substantial employment (maximum log value of 7.516).

Ln(LQ) (log of labour quality, proxied by average wage): The mean is 4.198 with a standard deviation of 0.615, implying relatively moderate dispersion in wage levels across firms. Although variation exists, it is narrower than other indicators, suggesting more uniformity in compensation structures, possibly tied to labour market regulations or prevailing industry wage levels.

Ln(DR) (log of debt ratio): The mean debt ratio is 3.584 with a standard deviation of 1.270, signifying considerable variation in the use of financial leverage. This highlights differing capital structures across firms, from debt-dependent to equity-financed operations.

Ln(TP) (log of market share at five-digit industry level): This variable has a mean of 2.170 and a standard deviation of 1.977, indicating significant variation in market dominance. Larger values denote leading firms in their respective niche markets, whereas smaller firms face more competitive pressures.

Ln(PCI) (log of PCI): The mean PCI score is 4.159 with a very low standard deviation (0.049), reflecting a relatively uniform institutional environment across provinces in the data set. This suggests that most localities exhibit similar levels of business-friendliness and regulatory quality.

Overall, the observed dispersion in key variables such as Ln(LP), Ln(K/L), Ln(DR) and Ln(TP) suggests high heterogeneity across firms in terms of productivity, capital intensity, financial structure and market positioning. This variability underscores the necessity of controlling for firm- and province-level characteristics in the econometric analysis to obtain consistent and unbiased estimators.

The pairwise correlation matrix for the variables used in this study is presented in Table 3, illustrating the linear relationships between labour productivity [Ln(LP)] and key explanatory variables, including capital intensity [Ln(K/L)], labour size [Ln(L)], labour quality [Ln(LQ)], financial leverage [Ln(DR)], market share [Ln(TP)] and the provincial institutional environment [Ln(PCI)].

All correlation coefficients are statistically significant at the 1% level (p-value < 0.01), confirming strong linear associations between labour productivity and the selected covariates. The positive correlations suggest that higher capital intensity, greater firm size, better labour quality, increased market share and improved provincial competitiveness are generally associated with higher labour productivity. In addition, the significance of the correlation between financial leverage and productivity highlights the role of capital structure in shaping firm performance.

However, while the coefficients indicate meaningful relationships, they do not account for potential multicollinearity or causality. Therefore, these findings serve as a preliminary step for the regression analysis, which will further isolate the independent effects of each factor on labour productivity while controlling for firm- and province-level heterogeneity.

The positive correlation between Ln(LP) and Ln(K/L) (0.585) suggests that as the capital–labour ratio increases, labour productivity also tends to increase. This is consistent with economic theory that additional investment in capital (machinery and technology) can increase labour productivity.

However, the negative correlation between Ln(LP) and Ln(L) (−0.168) suggests that as the size of the labour force increases, labour productivity tends to decrease. This may reflect the diminishing returns to scale or the problem of decreasing returns to scale of inputs implied by the Cobb–Douglas production constraint under the assumption of competitive markets.

The positive correlation between Ln(LP) and Ln(LQ) (0.254) suggests that labour quality (as represented by average wages) is closely related to labour productivity. This suggests that firms that pay higher wages, possibly due to having a higher quality workforce, tend to have better labour productivity.

The relationship between Ln(LP) and Ln(DR) (0.251) suggests that firms with higher financial leverage (high debt ratio) tend to have better labour productivity. This may imply that efficient use of debt helps firms invest in factors of production, thereby increasing productivity.

The strong positive correlation between Ln(LP) and Ln(TP) (0.458) suggests that firms with large market shares tend to have higher labour productivity. This may reflect that larger firms with high market shares can take advantage of economies of scale to improve productivity.

Finally, the relationship between Ln(LP) and Ln(PCI) (0.198) is also positive, suggesting that provincial competitiveness is related to labour productivity. Provinces with high PCI scores may create a more favourable business environment, helping enterprises operate more efficiently and improve productivity.

The results from the correlation matrix provide some important implications for the economic structure of the sample enterprises. Firstly, increasing capital investment per worker (K/L) and improving labour quality (LQ) are both closely related to improving labour productivity, which emphasises the importance of investing in capital and developing human resources. Secondly, the relationship between labour size and productivity tends to be negative, suggesting that scale efficiency may decrease as enterprise size increases. Finally, enterprises with large market shares and operating in a favourable business environment (high PCI) have an advantage in labour productivity, which suggests a positive impact of the competitive environment and local support policies.

To test for multicollinearity among the independent variables, the variance inflation factor (VIF) was calculated and reported in Table 4. VIF measures the extent to which an independent variable is affected by correlations with other independent variables in the model. A VIF value higher than 10 indicates a serious multicollinearity problem (Alisson, 1999), while a value below 4 is generally considered acceptable.

Among all variables, Ln(L) exhibits the highest VIF at 4.320, indicating a relatively elevated degree of multicollinearity. This finding is consistent with the correlation matrix, where Ln(L) shows a strong positive correlation with Ln(TP) (0.759) and a negative correlation with Ln(K/L) (−0.490). These correlations imply that firms with a larger labour force are more likely to capture greater market share while exhibiting lower capital intensity, thereby increasing the collinearity of Ln(L) with these two variables.

Ln(TP) also has a relatively high VIF value of 3.538, further supporting the presence of multicollinearity between market share and labour size. The strong correlation between these two variables may cause estimation inefficiencies, although the VIF values remain below the critical threshold of 10.

Ln(K/L), with a VIF of 1.774, indicates a moderate degree of correlation with other variables, notably its inverse relationship with Ln(L). As the number of employees increases, capital per labour unit tends to decline. Nevertheless, the VIF remains within the acceptable range and does not suggest serious multicollinearity concerns.

The remaining explanatory variables – including Ln(DR), Ln(PCI) and Ln(LQ) – exhibit VIF values well below 2 (1.202, 1.102 and 1.094, respectively), indicating negligible multicollinearity. These variables can be safely retained in the model without adjustment.

The mean VIF across all variables is 2.172, which is well below the conventional threshold of concern. Although Ln(L) and Ln(TP) display higher VIF values compared with others, both remain under the commonly accepted threshold of 5 and far below the critical level of 10 (Allison, 1999). Therefore, the model does not exhibit problematic multicollinearity, and no variable exclusion is deemed necessary at this stage.

The results of estimating the labour productivity equation from three pooled OLS models, FE, RE and selection tests are presented in Table 5.

For the Ln(K/L) variable, the coefficients in the OLS (0.180), FE (0.032) and RE (0.070) models are all positive and statistically significant (p  < 0.01). However, the coefficients in FE are significantly lower than those in OLS and RE, suggesting that the FE method better controls for unobserved fixed factors. The Ln(L) variable (labour size) has negative and statistically significant coefficients in all models, with values of −0.744 (OLS), −0.927 (FE) and −0.874 (RE), respectively. This suggests that larger labour size may be associated with lower labour productivity, especially in the FE model. The coefficient of the variable Ln(LQ) (labour quality) is positive and significant in all three methods, but the estimated value in the FE (0.027) and RE (0.035) models is lower than that of OLS (0.097), reflecting the difference in controlling for fixed or random factors. The variable Ln(TP) (market share) shows the largest and most positive coefficient among all variables, especially in FE (0.896) and RE (0.887), suggesting that market share plays an important role in increasing labour productivity. The variables Ln(DR) (debt ratio) and Ln(PCI) (provincial competitiveness index) also have positive and significant coefficients in all three models, but their coefficients in FE and RE are lower than those in OLS, suggesting the difference in how the methods handle fixed factors.

The significant differences between the three models suggest the choice of the most appropriate model for analysis. The Wald test is used to compare the FE and OLS models. The p -value of the Wald test is 0.000, indicating that the FE model is more suitable than the OLS. This suggests that unobserved fixed factors between enterprises strongly affect labour productivity, and the use of the FE model is necessary. The Breusch–Pagan test compares the RE and OLS models. The test result with a p-value of 0.000 shows that the RE model is more suitable than the OLS. This suggests the existence of random factors affecting labour productivity. The Hausman test compares the FE and RE models, with a p-value of the test being 0.000, indicating that the FE model is more suitable than the RE. This result suggests that fixed factors between firms are important and should not be ignored using the RE model.

A comparison across model specifications indicates that coefficient magnitudes are generally smaller in the FE model than in the pooled OLS and RE estimations. This suggests that cross-sectional estimators tend to overstate productivity effects by failing to control for unobserved firm-specific characteristics. The FE estimates therefore provide more reliable evidence on within-firm productivity dynamics over time, which is particularly relevant in the context of Vietnam’s textile and garment industry, where managerial practices and organisational structures vary substantially across firms.

Thus, the FE model is the most suitable model to estimate the relationship between independent variables and labour productivity. The FE model better controls for fixed factors that do not change over time, which helps to minimise the bias caused by these factors, thereby providing more accurate results than OLS and RE. Beyond the statistical evidence from the Hausman test, the FE model is particularly appropriate in the context of Vietnam’s textile and garment industry, where firms differ substantially in unobservable characteristics that are likely to be correlated with input choices. Factors such as managerial practices, production organisation and long-term buyer relationships are largely time-invariant but play a crucial role in productivity performance. By controlling for these firm-specific effects, the FE estimator provides more credible estimates of within-firm productivity changes over time. However, as pointed out in the descriptive analysis, these results can be strongly influenced by underlying heterogeneity. Therefore, heterogeneity analysis is performed in the next section.

Although the FE model is a better model than OLS and RE, the effects of heteroscedasticity and autocorrelation in random errors can still strongly affect the standard errors of the estimated coefficients. To control for the potential effects of these phenomena, the study uses clustered standard errors (Arellano, 1987) at the firm level. With 15,703 firms, the number of clusters is large enough for the clustered standard error estimates to give results that are close to the true values.

The estimation results with clustered standard errors by firm that account for the effects of heteroscedasticity and autocorrelation in random errors are presented in Table 6. In which, Column (1) is estimated on the entire sample; Columns (2) and (3) are estimated for the textile and garment sub-sectors; Columns (4), (5) and (6) are estimated for the economic sector groups; Columns (7) and (8) are estimated for firm size; and the last two columns are estimated on the lifetime-specific sample of firms. All are based on FE estimates with clustered standard errors.

In the estimated model based on the entire sample (Column 1), all independent variables are statistically significant with suitable directions and show a clear relationship with labour productivity.

Ln(K/L): The capital-to-labour ratio has a positive impact (0.032) and is statistically significant at the 1% level (***) on labour productivity. Specifically, when the capital-to-labour ratio increases by 1%, the labour productivity of enterprises increases by an average of 0.032%. This shows that investment in machinery, technology and facilities can improve labour productivity in enterprises. Increasing capital can help improve the efficiency of workers, especially in industries that require the support of technology and automation. Therefore, enterprises should consider increasing investment in fixed assets and technology to improve production efficiency. However, the relatively modest impact size shows that the role of capital in the production process of the textile industry is quite small, the majority of the impact belongs to the labour factor. It implies that the production technology used in the industry is still mainly labour-intensive technology, somewhat outdated.

Ln(L): Labour size has a negative impact (−0.927) and is statistically significant at the 1% level (***) on labour productivity. When the number of employees increases by 1%, the labour productivity of the enterprise decreases by an average of 0.927%. The negative coefficient on firm size contrasts with the conventional expectation of economies of scale but is consistent with the labour-intensive nature of the textile and garment industry. Rapid expansion in the number of workers, without commensurate improvements in production technology and managerial capacity, may lead to co-ordination inefficiencies and declining average labour productivity. Similar findings are reported by Ha et al. (2023), who show that excessive labour growth can reduce productivity in Vietnamese manufacturing firms.

Ln(LQ): Labour quality, represented by the average wage per worker, has a positive (0.027) and statistically significant (***) impact on labour productivity. When labour quality increases by 1%, the labour productivity of enterprises increases by 0.027% on average. This result confirms that improving labour quality through higher wages and improving work skills can bring direct benefits to labour productivity. This may include skills training, improved working environment and better welfare policies. Therefore, enterprises can improve overall productivity through investing in human resources. This result is in line with Ha et al. (2023), who emphasise the importance of labour quality and wage incentives in enhancing productivity in labour-intensive manufacturing sectors in Vietnam.

Ln(DR): Financial leverage (debt/total assets ratio) has a positive impact (0.012) and is statistically significant at the 1% level (***) on labour productivity. When the debt ratio increases by 1%, labour productivity increases by an average of 0.012%. This shows that a reasonable level of financial leverage can support enterprises in increasing investment in production resources and improving productivity. However, the not-so-significant impact shows that debt only plays a supporting role, and in the context of labour-intensive production technology, textile and garment enterprises need to be cautious in using leverage to avoid high financial risks.

Ln(TP): The market share of a firm has a very strong positive impact (0.896) and is statistically significant at the 1% level (***) on labour productivity. When market share increases by 1%, labour productivity increases by an average of 0.896%. Firms with large market shares tend to have higher productivity, possibly due to economies of scale, industry experience and better control of production costs. Expanding market share not only increases revenue but also has the potential to optimise operating efficiency. This suggests that firms should seek to increase market share through competitive strategies, product improvement or market expansion to achieve higher efficiency in production operations.

Ln(PCI): Provincial competitiveness index has a positive (0.084) and statistically significant impact at the 5% level (**) on labour productivity. For every 1% increase in PCI, labour productivity increases by 0.084% on average. The competitiveness of the province where enterprises operate plays a certain role in improving labour productivity. Factors such as business support policies, infrastructure and favourable business environment can all promote production efficiency. This suggests that enterprises will choose localities with good business environments, and local governments should create more favourable conditions to attract investment and improve business performance. The positive impact of institutional quality is consistent with Le (2020), who find that improvements in provincial governance and business-support institutions significantly enhance firm performance in Vietnam.

Columns (2)–(10) represent the same estimation models as Column (1) but on sub-samples to control for the effects of potential heterogeneity. Most of the results show consistent effects on labour productivity, consistent with the results in Column (1), such as capital-to-labour ratio, market share and labour quality. These factors are all positive and statistically significant, confirming that investing in capital, improving labour quality and increasing market share are sustainable strategies to improve labour productivity. In addition, the debt ratio has different effects across industries and business groups. This indicates that expansion or the use of financial leverage needs to be carefully managed and considered to avoid negative impacts on business performance.

From a policy perspective, the results suggest that productivity improvement in the textile and garment industry should not rely primarily on labour expansion. Instead, policies should prioritise skill upgrading, managerial capacity building and improvements in the local institutional environment. Strengthening vocational training systems and enhancing provincial governance can generate more sustainable productivity gains than firm scale expansion alone.

This study examines the determinants of labour productivity in Vietnam’s textile and garment industry using firm-level panel data and econometric analysis. Beyond reporting empirical relationships, the findings contribute to the productivity literature by providing evidence from a labour-intensive export-oriented sector in a developing economy, where productivity dynamics are shaped not only by firm-level inputs but also by the institutional environment. By integrating capital intensity, labour characteristics, market structure and provincial institutional quality within a unified empirical framework, the study extends previous research that has largely focused on isolated productivity drivers.

From a theoretical perspective, the results highlight the limited role of firm size expansion in improving labour productivity in labour-intensive industries. The negative association between labour scale and productivity underscores the importance of organisational efficiency and managerial capacity, supporting theoretical arguments that scale effects are conditional on complementary inputs such as technology and management quality. At the same time, the positive effects of labour quality and institutional quality reinforce human capital theory and institutional economics, suggesting that productivity gains depend on both internal firm capabilities and external governance conditions.

The findings also carry important policy implications. Productivity growth in Vietnam’s textile and garment industry should not rely primarily on expanding the labour force. Instead, policies should prioritise skill upgrading, managerial capability development and improvements in the local business environment. Strengthening vocational training systems, promoting firm-level management training and enhancing provincial governance and regulatory transparency can create conditions conducive to sustainable productivity growth. These results provide empirical support for industrial upgrading strategies that emphasise quality, efficiency and institutional reform rather than scale expansion alone.

In terms of practical implications, the study offers actionable insights for key stakeholders. For enterprise managers, the results suggest that investments in workforce skills, production organisation and management practices are more effective in improving productivity than rapid increases in employment. For policymakers, improving provincial-level institutions, reducing administrative burdens and strengthening business support services can enhance firm performance and competitiveness. For labour training institutions, closer alignment between training programmes and industry skill requirements is essential to meet firms’ evolving productivity needs.

Despite its contributions, this study has several limitations. Some variables, such as labour quality and institutional quality, are proxied by aggregate indicators, which may involve measurement error. Although the panel data approach controls for time-invariant firm heterogeneity, potential endogeneity issues cannot be fully eliminated. In addition, the focus on the textile and garment industry may limit the generalisability of the findings to other manufacturing sectors.

Future research could address these limitations by incorporating more direct measures of technology adoption and managerial quality, or by applying dynamic panel methods to better address endogeneity concerns. Comparative analyses across manufacturing sectors or cross-country studies within Southeast Asia would further enhance understanding of productivity dynamics in labour-intensive industries.