Institutional characteristics of Chinese unions

The institutional characteristics of Chinese unions are shaped by the historical evolution of China’s labour relations and national governance framework, with a core positioning rooted in serving workers and advancing national development – reflecting both historical logic and realistic demands. In the pre-1978 planned economy era, unions embedded in SOEs acted as a key link between management and workers, focusing on mobilising production and distributing welfare (e.g. housing, medical care) to ensure basic living security. With working hours uniformly regulated by state plans, overtime management was not a core task. The 1978 reform and opening-up triggered profound changes in labour relations. With SOE restructuring, the emergence of private/foreign-invested enterprises, and labour market decentralisation, many enterprises began using overtime to control costs, leading to widespread uncompensated overtime and excessive working hours. In response, Chinese unions – under the leadership of the CPC and guidance of the ACFTU – gradually shifted from ‘production mobilisation’ to ‘rights protection’. This transformation was legally solidified, especially the 2008 Labor Contract Law and 2021 revised Trade Union Law provided specific guarantees for overtime regulation. Notably, workers’ union establishment applications are supported by local ACFTU branches with professional guidance, differing from Western bottom-up models (Fan et al Reference Fan, Dong, Hu and Luo2018).

Three core pillars shape the operational characteristics of contemporary Chinese unions. First, CPC leadership ensures alignment with national goals, making unions a bridge between the Party and workers, – their overtime regulation balances individual rights, enterprise development, and social stability. Second, legalisation provides clear standards: Labor Law of the People’s Republic of China (Revised in 2018) stipulates tiered overtime pay (150% of base wage for weekdays, 200% for rest days without compensatory leave, and 300% for statutory holidays); Trade Union Law of the People’s Republic of China (Amended in 2021) mandates independent unions for enterprises with ≥25 employees, while smaller firms may form joint committees or appoint organisers (Ge Reference Ge2014). Third, flexible models (e.g. specialised unions for platform workers) expand coverage, with our sample’s 57.42% union presence tied to policy implementation.

In practice, unions offer targeted services: collective negotiation for labour-intensive sectors such as manufacturing and construction workers to safeguard overtime rights (Budd et al Reference Budd, Chi, Wang and Xie2014); hidden overtime prevention and mental health counselling for professionals in knowledge-intensive industries, such as finance and IT (Tsai et al Reference Tsai, Nitta, Kim and Wang2016); and focused support for migrant workers – laying the groundwork for the discussion below on heterogeneous analysis. Unlike Western unions centred on collective bargaining autonomy, Chinese unions prioritise ‘interest coordination and harmony’, embodying their unique institutional value in labour governance.

The ‘party-state face’ of unions in China and its impacts on overtime

Mechanisms of unions’ overtime-reducing effects in China

Within the ACFTU framework and guided by the orientation of harmonious labour relations (as mandated by relevant laws), Chinese unions reduce overtime through three context-specific, non-confrontational mechanisms rooted in legal provisions and grassroots practices:

The protection effects of unions for unprivileged workers

Unprivileged workers are those who hold low job status, have poor job stability, and possess lower bargaining power within an organisation (Tsai et al Reference Tsai, Nitta, Kim and Wang2016). They are often short-term temporary employees or individuals in low-skilled positions, making them more easily replaceable. In China, these workers are primarily concentrated in the private sector due to labour market segregation. Unfortunately, their lower bargaining power leaves them vulnerable to exploitation, such as being pressured to accept longer overtime working hours without proper compensation, as documented by Yu (Reference Yu2008).

This paper argues that unions in China are inclined to provide more protection for unprivileged workers, particularly in addressing illegal overtime practices. This is because Chinese unions prioritise social stability. Yet, unprivileged workers are more prone to conflicts and dissatisfaction due to their low work status, low income, heavy workload, job instability, and limited advancement opportunities (Qian et al Reference Qian, Li, Song and Wang2020). From this perspective, unions have a strong incentive to safeguard the rights and interests of unprivileged workers, including protecting them from overtime beyond legal limits. This not only addresses their grievances but also contributes to social stability. In contrast, for workers with stable job positions and less replaceable skills, companies may adopt other measures, such as higher pay and future job promotions, to maintain their job satisfaction. As a result, unions focused on stability may be less motivated to prioritise protecting these workers from overtime issues.

Therefore, this paper proposes that Chinese unions provide stronger protection against overtime, especially overtime beyond legal limits for unprivileged workers, compared to workers with more secure long-term contracts and lower job substitutability. This proposition aligns with previous research indicating that Chinese unions primarily concentrate on improving the ‘bottom-line welfare’ of workers, especially migrant workers. In so doing, unions seek to ensure that they receive minimum wages, obtain social benefits, and are protected from forced labour (Zhu Reference Zhu2004). These efforts reflect the role of Chinese unions in safeguarding the fundamental interests of vulnerable workers in order to maintain social stability.

Data

This paper undertakes an empirical analysis using data from the CLDS. The CLDS dataset is derived from a nationwide multi-year survey that employs a stratified random sampling approach. It comprehensively captures information on respondents’ educational attainment, employment status, migration patterns, health conditions, social engagement, and economic activities, among other aspects.

The labour-related information provided by the CLDS, including data on labour unions and overtime work, allows us to conduct a viable empirical study. Specifically, this paper uses the pooled data from three waves of the CLDS (2012, 2014, and 2016) for the analysis. The initial sample size was 61,293.

To ensure the comparability of the workers in the sample, we imposed several restrictions. First, we limited the sample to individuals aged between 16 and 65 years. As defined by the Labor Contract Law of the People’s Republic of China, this age group has the legal qualification to establish labour relations.Footnote ² This restriction led to the exclusion of 12,131 respondents. Second, we excluded those engaged in self-employed agricultural work (in China, most farming is carried out on a family or individual basis), self-employed business activities (such as a sandwich vendor), freelancing (such as a freelance writer), those working in non-profit organisations, and those identifying themselves as employers. Finally, we removed 11,825 observations with missing values of control variables, most notably workplace type, education, organisational ownership type, and industry.Footnote ³ As a result, the final sample for this analysis comprises 13,929 observations.

Variables

Union-related variables: Consistent with previous studies, this paper measures the union using three variables: ‘union’, ‘union efficacy’, and ‘union member’. These three variables collectively provide a comprehensive understanding of the union in the context of our study. They allow us to assess union establishment, perceived efficacy, and union membership, enabling a comprehensive examination of the relationship between unions and overtime in the analysis. The three union-related variables are constructed as follows:

1. (1) Union. This variable is created from the survey question, ‘Does your institute/company have a union?’ If the respondent answers affirmatively, then it assigns a value of 1 to the union variable, 0 otherwise. In the sample of this analysis, 57.42% of the employers’ units have a union established

2. (2) Union efficacy. This variable is a dummy variable, created from the survey question ‘Do you think the union is helpful to you?’. If the respondent answers affirmatively, then it assigns a value of 1 to the union efficacy variable, 0 otherwise. 70.08% of workers in this sample believe the union is useful.

3. (3) Union member. This variable is created from the survey question ‘Do you join the union?’ If the respondent answers affirmatively, then it assigns a value of 1 to the union member variable, 0 otherwise. Only the 2016 survey included a question on individual union membership, which limits our ability to generalise the findings across all three waves.

Overtime-related variables: Overtime is measured by three dummy variables: overtime, overtime exceeding legal limits, and uncompensated overtime. These variables collectively provide a comprehensive understanding of the various aspects of illegal overtime work in the context of this study. The overtime-related variables are constructed as follows:

1. (1) Overtime. This dummy variable indicates whether a worker has worked overtime or not. It takes a value of 1 if the worker has worked overtime in the last month and 0 otherwise. In the sample of this analysis, 33.87% of workers reported that they worked overtime.

2. (2) Overtime exceeding legal limits. This dummy variable determines whether the worker has worked overtime beyond the legal limits, i.e. work overtime maximum overtime hours of law limits. In this analysis, the legal limits of overtime are set as 3 hours per day or 36 hours per month, as specified by China’s Labor Contract Law. This variable takes a value of 1 if the worker has exceeded the legal limits in the last month, and 0 otherwise. In the sample of this analysis, 8.31% of the workers’ overtime exceeds the legal limits.

3. (3) Uncompensated overtime. This dummy variable identifies whether the worker has received compensation for their overtime work. The CLDS questionnaire asks respondents: ‘How many hours of overtime did you work last month? How many of those hours were compensated? (If no compensation, please enter 0 hours.)’ We identify respondents who report 0 compensated hours in the last month as having performed uncompensated overtime and assign a value of 1, while 0 indicates they have been compensated. In the sample, 14.16% of workers reported that their overtime work is uncompensated.

Instrumental variables: To control for the endogeneity issue in this analysis, two instrumental variables (IVs) ‘union committee’ and ‘union committee members more than 25’ are created. The instrumental variables are constructed as follows:

1. (1) Union committee. This variable is based on the union policy requirements of each province in 1998 in China. A value of 1 is assigned if the province mandates the establishment of a union committee within the organisation, and 0 otherwise.

2. (2) Union committee members more than 25. This variable takes a value of 1 if the province requires the number of union committee members to be greater than 25 and 0 if such a requirement is not specified.

These two IVs are derived from provincial-level regulatory documents formulated in 1998 to implement China’s 1992 Trade Union Law, with clear practical requirements and enforcement mechanisms that underpin their validity as instruments. Specifically, these 1998 provincial regulations include two core provisions: (1) Mandatory scope: Enterprises with 25 or more employees are legally required to establish a grassroots union committee; enterprises with fewer than 25 employees may form a joint union committee or set up union groups – consistent with the national standard for trade union organisation building at the time. (2) Enforcement mechanism: Local labour departments and provincial branches of the ACFTU jointly conduct compliance inspections. Enterprises failing to meet the requirements face administrative reminders, rectification orders, or restricted access to government procurement and policy subsidies.

In practice, these regulations create heterogeneous compliance pressure across regions (e.g. variations in enforcement intensity between provinces) and directly drive union establishment at the firm level. Eligible enterprises (≥25 employees) are significantly more likely to form union committees to avoid regulatory risks and access policy incentives. This ensures the IVs satisfy two key validity conditions: (1) Relevance: Variation in policy implementation (e.g. whether a province mandates union committees, or requires ≥25 committee members) is positively correlated with firm-level union presence – the endogenous explanatory variable. (2) Exogeneity: As macro-level institutional arrangements, these 1998 provincial policies were formulated based on regional labour market governance needs, not influenced by individual firms’ illegal overtime practices (the dependent variable). This eliminates reverse causality concerns, as single enterprises cannot alter provincial regulatory frameworks.

Unprivileged Workers: Two variables ‘short-term labour contract’ and ‘task replaceability’ are used to identify workers who are considered unprivileged due to the nature of their short labour contracts and the ease of their replaceability within the organisation. The two unprivileged workers variables are constructed as follows:

1. (1) Short-term labour contract. This variable is defined as 1 if the worker has signed a short-term or temporary contract with the company. It is defined as 0 if the worker has a long-term labour contract.

2. (2) Task replaceability. This variable reflects whether the tasks performed by a worker can be easily carried out by others. It is assigned a value of 1 if the worker takes less than a year to master the required skills or does not have a subordinate within the organisation, and 0 otherwise (Ge Reference Ge2014; Tsai et al Reference Tsai, Nitta, Kim and Wang2016). This 1-year cut-off aligns with academic consensus on Chinese labour markets and practical labour regulations, effectively capturing the skill cycle feature of high-replaceability roles.

In addition, this analysis controls for individual demographic characteristics that may influence the effects of unions on overtime, including age, gender, hukou (China’s household registration system, which formally distinguishes rural from urban residents and affects access to local social services – such as healthcare, housing – and labour protection), and education level. The analysis also controls for the impacts of different skill levels on the worker’s overtime by including whether the individual receives occupational training and whether the individual obtains an occupational certificate.

To capture the effects of job tasks on the worker’s overtime behaviour, this analysis controls for whether the worker is monthly paid, the organisation type, and income level. Moreover, this analysis uses the number of times the worker has signed a labour contract with the same employer to reflect job stability and controls for its effect on illegal overtime.

Furthermore, the analysis integrates organisation-type dummies, industry dummies, and year dummies as control variables. The organisation-type dummies serve to indicate the ownership and administrative forms of the employer, reflecting whether the employer is state-owned, privately-owned, foreign-invested, or a public institute. The industry dummies are employed to account for the sector of activity in which the employers operate, such as in manufacturing or services.

The variable definitions and descriptive statistics for the CLDS sample in 2012, 2014, and 2016 used in the analysis are shown in Table 1.

Table 1. Variable definitions and descriptive statistics (Year 2012, 2014, 2016)

Note: The union membership information is only reported by CLDS in year 2016, therefore, the sample size for this variable is 4488.

Empirical models

The Probit model used in this paper to estimate the effects of Chinese unions on overtime is constructed as follows:

(1) ${\rm{\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;prob}}\left( {{\rm{o}}{{\rm{t}}_i} = 1} \right) = {\rm{\Phi }}({{\rm{\alpha }}_{\rm{i}}} + {{\rm{\beta }}_1}unio{n_i} + {\beta _2}yea{r_i} + {\rm{\gamma }}{{\rm{X}}_{\rm{i}}} + {{\rm{\varepsilon }}_{\rm{i}}})$

where ${\rm{o}}{{\rm{t}}_i}$ represents whether the respondent $i\;$ had overtime/overtime exceeding legal limits/uncompensated overtime last month. $unio{n_i}$ represents union membership or union efficacy. $yea{r_i}$ are the year dummies. ${X_i}$ , a vector of control variables, consists of individual and their employer’s characteristics, including age, hukou, education level, workplace dummies, industry dummies, and ownership dummies. A detailed description of the control variables is shown in Table 1.

Moreover, recognising the potential endogeneity issue, this paper employs an instrumental variables (IV) estimation strategy and applies the two-stage least squares (2SLS) method using Heckman’s (Reference Heckman1979) two-stage model to estimate the effects of unions on overtime. The estimation model is as follows:

(2) $union_i^{\rm{*}} = \partial polic{y_i} + {\rm{\alpha }}yea{r_i} + {\rm{\beta }}{{\rm{X}}_{\rm{i}}} + {{\rm{\varepsilon }}_{\rm{i}}}$

(3) ${\rm{prob}}(o{t_i} = 1) = {{\rm{\beta }}_{\rm{i}}}\widetilde {{\rm{unio}}{{\rm{n}}_{\rm{i}}}} + {\rm{\alpha }}yea{r_i} + {\rm{\gamma }}{{\rm{X}}_{\rm{i}}} + {{\rm{\varepsilon }}_{\rm{i}}}$

where $polic{y_i}$ in Equation (2) represents two instrumental variables: union committee and union committee members more than 25. The endogenous variable – union – is represented as $union_i^*$ . The dependent variable $o{t_i}$ in Equation (3) is the same as in Equation (1).

To further investigate the heterogeneous effects of Chinese unions on workers’ overtime, the analysis includes the interaction term between unions and the moderating variables, i.e. labour contract and task replaceability. Then, the analysis utilises the Probit model to estimate as follows:

(4) ${\rm{prob}}(o{t_i} = 1) = {{\rm{\alpha }}_{\rm{i}}} + {\rm{\theta }}\left( {unio{n_i} \times {M_i}} \right) + {{\rm{\beta }}_1}unio{n_i} + {{\rm{\beta }}_2}{M_i} + {\rm{\alpha }}yea{r_i} + {\rm{\gamma }}{{\rm{X}}_{\rm{i}}} + {{\rm{\varepsilon }}_{\rm{i}}}$

where $o{t_i}$ and $unio{n_i}$ are the same as in Equation (1). ${M_i}$ represents two variables related to heterogeneity effects: short labour contract and task replaceability. $unio{n_i} \times {M_i}$ denotes the interaction term of independent variable and these two variables, which captures the heterogeneous effects of the union on preventing overtime across various groups of workers, including workers in the standard or nonstandard workers, workers with or without task replaceability. $yea{r_i}$ and ${X_i}$ are the same as in Equation (1).

Baseline results

Table 2 presents the Probit and Ordinary Least Squares (OLS) estimations of the relationship between Chinese unions and overtime. The results of Table 2 show that in models (1) to (6), the estimated coefficients of union and union efficacy on workers’ overtime experiences are all significantly negative. This indicates that the establishment of a union in an organisation and the degree of union efficacy can significantly reduce workers’ overtime experiences, including overtime exceeding maximum overtime hours of law limits and uncompensated overtime. Specifically, the estimated marginal effects of unions on overtime exceeding legal limits and uncompensated overtime are −0.025 and −0.173, respectively. This means that the establishment of a union can reduce the likelihood of workers experiencing overlong overtime work and overtime work without compensation by 2.5% and 17.3% (relative to the sample mean), respectively, compared to those without a union.

Table 2. The effects of unions on workers’ overtime (Year 2012, 2014, 2016)

Note: The marginal effects of Probit regression are estimated at the sample means and reported. Standard errors in parentheses. ***Significant at 1%; **significant at 5%; *significant at 10%. We also tested the effects on voluntary overtime and found results consistent with those for total overtime. The dependent variable – overtime exceeding legal limits – is defined solely based on exceeding the monthly limit of 36 hours.

Comparing union efficacy and union coefficients, union efficacy (worker-evaluated) shows stronger negative effects on overtime, especially beyond legal limits and uncompensated overtime. If workers perceive the union as effective, the probability of overtime exceeding legal limits drops by 35.5% and uncompensated overtime by 26.7%.

In addition, the estimated coefficients of control variables in Table 2 indicate that the presence of overtime exceeding legal limits is more common among groups with low education levels and rural hukou. However, these groups are less likely to engage in uncompensated overtime work, and the uncompensated overtime working hours are also shorter.

Robustness checks

The analysis conducts a robustness check by using union membership as a measure to further investigate the effects of Chinese unions on overtime. Since the CLDS only reported union membership information in the 2016 wave, this paper utilises the 2016 data for the analysis. The results are reported in Table 3. As shown in Models 1 to 3, the estimated coefficients of union membership are significantly negative for overtime, overtime exceeding legal limits, and uncompensated overtime. This indicates that unions can protect their members from engaging in overtime, which is consistent with the findings from Table 2. However, comparing the magnitudes of the coefficients of union membership in Table 3 with those of union and union efficacy in Table 2, the results show that the protection effects of union membership are much stronger.

Table 3. The effects of union membership on workers’ overtime (Year 2016)

Note: Due to a space limitation, the estimations of control variables same as Table 2 are not reported. Standard errors in parentheses. ***Significant at 1%; **Significant at 5%; *Significant at 10%.

The analysis also estimates the effects of the interaction term between union membership and union efficacy on overtime to examine the additional effects of union membership beyond union efficacy. Results of Models 4 to 6 in Table 3 demonstrate that for workers who believe their unions are efficient, being union members can provide extra protection against overtime possibilities. However, there is no such additional effects on overtime exceeding legal limits and uncompensated overtime, indicating that an efficient union offers the same level of protection for both union members and non-union members.

We further verify the robustness of core findings through subgroup regressions across gender, hukou status, and occupation type (see Appendix Table 1). Results show unions exert significant protective effects on overtime, illegal overtime, and uncompensated overtime across all subgroups. Notably, union effects are consistent between men and women (union coefficient: −0.089 for males and −0.082 for females), with a slightly stronger union efficacy effect for women (−0.135 vs. −0.103), potentially due to gender differences in industry distribution. Additionally, unions provide stronger protection for rural-hukou and blue-collar workers – aligning with our core finding that unions prioritise unprivileged groups.

IV Estimations: Potential endogeneity of unions

The effects of unions on overtime may suffer from an endogeneity issue. For example, workers who establish unions may be more inclined to obey the law or value employee welfare, leading to less overtime working. To address this issue, this paper utilises the instrumental variable (IV) method to estimate the effects of unions on overtime.

Two dummy variables, ‘union committee’ and ‘union committee members more than 25’, are constructed as the instrumental variables for the endogenous variable ‘union’. Provincial union regulations concerning union committee formation differ across provinces: some mandate that all organisations establish a local union committee, while others impose this requirement only when the organisation has more than 25 union members. These regulatory variations serve as potential instrumental variables for union establishment, as they are likely correlated with union formation but exogenous to workers’ overtime.

These two policy variables satisfy the key validity requirements of instrumental variables. First, the regression estimates demonstrate a significant correlation between the union committee and union committee members more than 25 with the union. Second, they show a minimal correlation with the random disturbance term in the model. The IVs have passed the under-identification test and the weak identification test, as indicated by the statistics of Anderson LM and Cragg-Donald Wald F in Model 1 of Table 4.

Table 4. The effects of unions on overtime based on IV-Probit Model (Year 2012, 2014, 2016)

Note: The estimations of control variables same as Table 2 are not reported. Standard errors in parentheses. ***Significant at 1%; **Significant at 5%; *Significant at 10%. Marginal effects are not reported due to literature suggesting that IV-Probit models are sensitive to covariate distributions (Evdokimov et al Reference Evdokimov, Kalnina and Zeleneev2025; Rivers and Vuong Reference Rivers and Vuong1988). Compared to Probit model, IV estimation in IV-probit inflates variance, increasing uncertainty (Andrews et al Reference Andrews, Stock and Sun2019).

In the second-stage regression, the coefficient estimates for unions are significantly higher than the corresponding baseline estimates in columns (2) to (4) of Table 2. These results indicate that the endogeneity issue leads the baseline regression to underestimate the negative impacts of unions on workers’ overtime. This confirms that Chinese unions are effective in reducing overtime, including overtime exceeding legal limits and uncompensated overtime. In other words, the results remain robust when using the IV estimation.

The heterogeneous effects of unions in preventing excessive overtime

To examine the heterogeneous effects of Chinese unions, this paper introduced two variables: short-term labour contract and task replaceability, to measure the unprivileged workers in the analysis. By including the interaction terms of union/union efficacy with these two variables, the analysis estimates the different protection effects of unions on these two different worker groups – privileged and unprivileged workers. The estimation results of the interactive effects of unions and short-term labour contract and task replaceability are reported in Table 5.

Table 5. Heterogeneous effects of unions in preventing overtime (Year 2012, 2014, 2016)

Note: Due to space limitations, the estimations of control variables same as Table 2 are not reported. Standard errors in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.

First, the results of Model 1 to Model 6 in Table 5 show that the interaction terms between unprivileged workers and unions/union efficacy are generally negatively significant in predicting overtime, overtime exceeding legal limits, and uncompensated overtime. This indicates that unions are more protective of unprivileged workers. Unprivileged workers whose employers have established a union reported a much lower probability of overtime (b = −0.191, p < 0.01 in panel A, b = −0.348, p < 0.01 in panel B), overtime exceeding legal limits (b = −0.120, p < 0.1 in panel A, b = −0.271, p < 0.01 in panel B), and uncompensated overtime (b = −0.231, p < 0.01 in panel B). Moreover, those who reported high union efficacy also experienced a lower probability of overtime, overtime exceeding legal limits, and uncompensated overtime.

Second, comparing the two measures of unprivileged workers, the magnitude of estimated coefficients of interaction terms (task replaceability with unions/union efficacy) in Panel B (ranging from −0.231 to −0.348) is higher than those (short-term labour contract with unions/union efficacy) in Panel A (ranging from −0.060 to −0.219). These results suggest that the union protection effects for unprivileged workers are strongest for those who are in easily replaceable jobs. In comparison, for those who have short-term labour contracts, although unions can provide additional protection against overtime, the protection effects are relatively smaller.

Third, when comparing the estimates between union interactions and union efficacy interactions, we find notable differences in the heterogeneous protection effects. Specifically, the heterogeneous protection effects of union efficacy (Models 4 to 6 in Table 5) are far more significant than those of union establishment (Models 1 to 3 in Table 5). To be more precise, both the magnitude and significance of the coefficients for union interactions are substantially smaller than those for union efficacy interactions. In Panel A, the estimated coefficients of the union interaction term in Models 2 and 3 turn out to be insignificant. This lack of significance is revealing that while an efficiently-operated union can offer stronger protection to unprivileged workers, merely establishing a union may not necessarily provide additional protection for workers with unstable employment relationships.

Propensity score matching estimations

Furthermore, to address self-selection biases, this paper adopts propensity score matching (PSM) method to explore the unions’ prevention effects against overtime among privileged and unprivileged workers. The first step is to calculate the propensity score for each worker using a logit model. The second step is to match the employers where workers work with/without the union according to the obtained propensity scores and to compare the differentials of overtime exceeding legal limits and uncompensated overtime between the two groups of workers with similar propensity scores. This two-step procedure is undertaken for workers having long-term or short-term labour contracts, for workers having easily replaceable jobs or not, as well as for workers employed in the private sector or public sector. The analysis verifies the overlap condition to ensure that the propensity scores for employers with and without a union lie in a common support and conduct balancing tests to determine whether there are significant differences between the two worker groups after matching.

We employed a nearest-neighbour PSM approach. Each treated unit was matched to one control unit with the closest propensity score, subject to a calliper restriction that limited matches to pairs whose scores differed by no more than 0.05. This ensured that treated and control observations were only compared when they were sufficiently similar in terms of their predicted probability of treatment. To avoid extrapolation, the analysis was restricted to the common support region, meaning that treated units without comparable controls (and vice versa) were excluded. When multiple controls were equally close, ties were resolved randomly. Uniform weights were applied to matched controls, and the resulting estimates reflect the average treatment effect across the matched sample. Common support region and balance tests are all conducted to confirm effective balancing.

Following the approach of Smith and Todd (Reference Smith and Todd2005), this paper uses a set of t-tests of equal means between the employers with/without a union. The previous conclusions hold for all samples and different subgroup samples. For all samples, the results of Table 6 show that there are significant differences between employers with a union and without a union regarding overtime exceeding legal limits (b = −0.032, p < 0.001). This indicates that unions do have a significant impact on reducing overtime.

Table 6. PSM with different subsamples: Union as the treatment

Note: Due to space limitations, the results regarding union efficacy as the treatment variable are not reported. *** Significant at 1%; ** significant at 5%; * significant at 10%. Common support region, balance tests, Rosenbaum bounds sensitivity analysis are also conducted.

Moreover, for subgroups divided according to privileged and unprivileged working status, the differences in overtime outcomes between workers employed by employers with and without a union vary. For instance, in Panel A of Table 6, for workers with short-term labour contracts, the difference between employers with/without a union is significant (−0.047, p < 0.05), while for workers with long-term labour contracts, the difference is not significant (−0.008, p > 0.1). This suggests that the unions’ prevention effects against overtime exceeding legal limits only apply to short-term contract workers. Similarly, in Panel B of Table 6, for workers with easily replaceable tasks, the difference between employers with/without a union is significant (−0.043, p < 0.01), while for workers with tasks that are difficult to replace, the difference is not significant (−0.006, p > 0.1). This suggests that the unions’ prevention effects against uncompensated overtime only applies to workers with easily replaceable job tasks. Rosenbaum bounds sensitivity analysis is also conducted.

The analysis also examines the effects of union efficacy using PSM. The results are consistent with the findings reported in Tables 2 and 5.