Working hours in Chinese enterprises: evidence from matched employer–employee data

Abstract

The purpose of this article is to explore the factors that are correlated with hours worked in China. A distinguishing feature of the study is that we used representative-matched employer and employee data. Hence, in addition to the usual worker characteristics examined in conventional economic models of labour supply, we also took into account the influence of firm characteristics and policies on the number of hours worked. The results suggested that in addition to the hourly wage rate, labour supply characteristics and human capital characteristics of the individual, firm-level differences are important in explaining variation in weekly hours worked in Chinese firms. In particular, our results suggested that there is a norm of longer working hours in firms that employ a high proportion of female and migrant workers, that hours worked are less in firms which pay overtime and that hours worked are less in firms in which labour disputes have disrupted production. The policy implications of Chinese firms reducing hours worked were discussed.

1 Introduction

The International Labour Organisation (ILO) points out that despite rapid economic growth in Asia, workers in Asian countries are still generally working longer hours than most of their counterparts in developed countries (Lee et al., 2007). This is particularly true for China. China has had one of the highest rates of economic growth in the world over the last three decades, but the average working week in China remains one of the highest in the world. The OECD (2011) found that the Chinese ranked fourth in the world in terms of time spent at work, after Mexico, Japan and South Korea. The main beneficiaries of a decline in the length of the working week in OECD countries over the course of the 20th century were unskilled blue-collar workers (Golden, 2008). However, in China, it has been long working hours by unskilled blue-collar workers and, in particular, rural–urban migrants, who has fuelled the engine room that has been responsible for China's high growth rate.

The purpose of this article is to explore the factors that are correlated with hours worked in China. A distinguishing feature of the study is that we use representative-matched employer and employee data. Hence, in addition to the usual worker characteristics examined in conventional economic models of labour supply, we also take account of the influence of firm characteristics and policies on the number of hours worked. By contrast, the limited existing literature on hours worked in China has primarily used a conventional labour supply function in which the employer dimension is ignored (see e.g. Li and Zax, 2003; Maurer-Fazio et al., 2011). The demise of allocated, lifelong jobs in the push towards a market economy has resulted in the materialisation of a competitive labour market in urban China (Warner, 1996). Increased competition among private sector employers and the freedom to diverge from a state administered labour system have led to increased job turnover and mobility as firms vie to attract and retain skilled staff, and China's skilled workforce aims to maximise employment opportunities. In an environment such as this in which there is a high level of competition between employers to attract and retain the best staff, workplace characteristics and policies can be expected to be important in influencing labour supply.

2 Existing Literature

An extensive literature exists that uses a conventional model of labour supply to estimate the labour supply of men and women. Most of these studies are primarily interested in estimating how hours worked respond to variations in wages (see, e.g. Killingsworth, 1983; Pencavel, 1986, for surveys). While the labour supply model often forms the basis for such studies, disciplines other than economics have also examined various aspects of hours worked including determinants of non-standard working hours (evenings, overtime, shift work) (Presser, 1995) and mismatch between actual and preferred hours of work (Reynolds, 2003). Few studies examined the impact of both firm and worker characteristics on hours worked. One such study is Bryan's (2007), which uses matched employer and employee data to examine the effect of firm and worker characteristics on number of hours worked in the UK. However, a limitation of Bryan's (2007) study is that he did not have data on wages.

There is little evidence on the determinants of hours worked in China. There is a small literature on labour supply in Chinese urban labour markets (Chau et al., 2007; Li and Zax, 2003) including studies that focus on the labour supply responses of married females (Maurer-Fazio et al., 2011). However, these studies focus exclusively on the supply side and do not consider the relevance of firm characteristics or policies. One existing study for China that does consider the employer dimension is Smyth et al.'s (2012). That study uses matched employer and employee data from the Fair Labor Association (FLA) to examine the correlates of excessive hours worked in Chinese and Thai supply chain factories.

Our study differs from Smyth et al. (2012) in four respects. First, that study focused on the determinants of excessive hours worked, defined as in excess of 60 hours per week, while we look at the determinants of hours worked more generally. Second, that study was restricted to 15 factories in one industry. However, supply chain factories in the main have long working hours, and hours in such enterprises are not representative of working hours more generally. In this current study, we look at determinants of working hours across a range of industries. Third, conventional economic models of labour supply suggest that the wage rate per hour is a key determinant of hours worked. While Smyth et al.'s study does not have data on wages, the wages is included in this study. Fourth, Smyth et al.'s study does not contain any measure of family or non-labour income, while the current study does incorporate this measure.

3 Data

The sample that we used was from a matched worker–firm data set from Minhang district in Shanghai, which was originally collected in 2007 by the Institute of Population and Labour Economics in the Chinese Academy of Social Sciences (CASS). The CASS researchers worked in close cooperation with the Shanghai municipal government in collecting the data. The data set was selected using probability proportion to size sampling according to a list of all manufacturing firms in Minhang district with annual sales of at least 5 million RMB. The representativeness of the sample in terms of number of employees, sales revenue, profits and average wages are considered in Table 1. The firms in the sample are representative of firms in Minhang District and Shanghai as a whole.

Table 1. Representativeness of sample

	Sample	Minhang district	Shanghai
Source: The data for Minhang District and Shanghai are from SBS (2008).
Number of employees (person)	182.82	202.83	190.38
Sales revenue (10,000 RMB)	8,896.69	11,974.22	12,445.22
Profits (10,000 RMB)	675.27	800.10	866.94
Average wage of employees (RMB/month)	2,145.55	2,383.42	2,423.25

The data set contains information on 784 workers across 78 firms (on average, 10.05 workers per firm). Once missing observations were removed, we had data on the variables of interest in this study for 587 workers across 72 firms. Table 2 provides descriptive statistics for the variables used in the study based on the 587 workers from these 72 firms. The average number of hours worked is 46.14 hours per week and the average hourly wage rate is 10.29 RMB. Among labour supply and human capital characteristics, 54 per cent of respondents were male, 74.62 per cent were married, 10.9 per cent were members of the Chinese Communist Party, 56.45 per cent held a non-agricultural hukou (household registration), the average age was 34.51 years and average years of schooling was 11.48. The firm characteristics are the proportion of female workers (average is 39 per cent), proportion of migrant workers (average is 37 per cent), whether there is a trade union presence in the firm (52.78 per cent of firms have a trade union), whether labour disputes affect production (4.17 per cent of the time) and whether the firm pays overtime (97.22 per cent). The data set also contains information on the industry in which the firm is located (one of 22 manufacturing industries based on the classification system employed by the China State Statistical Bureau, 2008) as well as the ownership of the firm (state-owned, public, foreign or privately owned).

Table 2. Descriptive statistics

Variable	Mean (%)	Standard deviation (%)
Number of hours worked per week	46.14	7.87
Hourly wage rate (RMB)	10.29	7.78
Labour supply characteristics
Gender (male = 1)	Male = 317 (54)
Marital status (married = 1)	Married = 438 (74.62)
Family income (RMB per month)	2,816.75	7,426.88
Communist Party membership (Yes = 1)	Yes = 64 (10.90)
Hukou status (non-agriculture = 1)	Non-agriculture = 349 (59.45)
Health status	Ordinary = 123 (20.95)
	Good = 173 (29.47)
	Very good = 291 (49.57)
Trade union membership (member = 1)	Member = 213 (36.29)
Human capital characteristics
Age (years)	34.51	10.78
Age squared	1,306.67	815.28
Years of education	11.48	2.97
Language proficiency (standard = 1)	Standard = 399 (67.97)
Length of time with firm (years)	5.85	3.33
Certification	No title = 459 (78.19)
	Elementary = 87 (14.82)
	Junior/senior = 41 (6.98)
Occupation	Professional and technical personnel = 136 (23.17)
	Production transport and related worker = 135 (23.00)
	Business service personnel = 94 (16.01)
	Equipment operators and related workers = 222 (3.39)
Received on job training (yes = 1)	Yes = 258 (43.95)
Feels under pressure to meet deadlines (yes = 1)	Yes = 190 (32.37)
Firm characteristics and policies
Number of employees	182.82	319.35
Proportion of female employees	0.39	0.23
Proportion of migrant workers	0.37	0.34
Firm has trade union (yes = 1)	Yes = 38 (52.78)
Labour disputes affected production (yes = 1)	Yes = 3 (4.17)
Will be paid overtime (yes = 1)	Yes = 70 (97.22)
Other
Firm ownership	State-owned firms = 6 (8.33)
	Public firms = 25 (34.72)
	Foreign firms = 27 (37.50)
	Private firms = 14 (19.44)
Industries	Processing of food = 1 (1.39)
	Foods = 3 (4.17)
	Textile = 3 (4.17)
	Textile wearing apparel, footwear and caps = 5 (6.94)
	Furniture = 3 (4.17)
	Paper and paper products = 5 (6.94)
	Printing, reproduction of recording media = 1 (1.39)
	Raw chemical materials and chemical products = 7 (9.72)
	Rubber = 4 (5.56)
	Plastics = 6 (8.33)
	Non-metallic mineral products = 3 (4.17)
	Smelting and pressing of ferrous metals = 1 (1.39)
	Smelting and pressing of non-ferrous metals = 1 (1.39)
	Mental products = 1 (1.39)
	General purpose machinery = 2 (2.78)
	Special purpose machinery = 11 (15.28)
	Transport equipment = 1 (1.39)
	Electrical machinery and equipment = 10 (13.89)
	Communication equipment, computers and other electrical equipment = 3 (4.17)
	Recycling and disposal of waste = 1 (1.39)
Number of respondents	587
Number of firms	72

Table 3 shows the overall distribution in number of hours worked per week, plus the distribution in number of hours worked according to gender and the hukou status of the respondents. While the vast majority of the respondents (66.27 per cent) worked between 41 and 45 hours per week, 26 per cent of the respondents worked in excess of 50 hours. There is little difference in the distribution of hours worked between genders; however, those with an agricultural hukou worked longer hours than those with non-agricultural hukou. Table 4 shows the share of variation in individual characteristics due to workplace affiliation. The top row of Table 4 shows that 33 per cent of the variation in hours worked can be attributed to workplace affiliation. Overall, this figure implies that weekly hours worked by an individual are fairly closely correlated with the workplace to which they belong. This is consistent with the premise earlier that firm characteristics matter for hours worked. One would expect, though, that the share due to workplace effects would decline when worker characteristics are added because some of these other variables may not be randomly distributed across workplaces (Bryan, 2007). The second row shows that 35 per cent of the variation in the hourly wage rate can be attributed to workplace affiliation. Among the labour supply and human capital characteristics of respondents, the shares attributable to workplace affiliation vary between 4 per cent (certification) and 58 per cent (trade union membership) and are predominantly in the range of 20–40 per cent. These figures do not, however, control for other determinants of working hours, for which a multivariate decomposition is needed.

Table 3. Distribution of number of hours worked per week

Hours worked per week	Male (%)	Female (%)	Agriculture hukou (%)	Non-agricultural hukou (%)	Overall (%)
<40	14 (4.42)	8 (2.96)	9 (3.78)	13 (3.72)	22 (3.75)
41–45	209 (65.93)	180 (66.67)	126 (52.94)	263 (75.36)	389 (66.27)
46–50	4 (1.26)	21 (7.78)	6 (2.52)	19 (5.44)	25 (4.26)
51–55	59 (18.61)	43 (15.93)	57 (23.95)	45 (12.89)	102 (17.38)
56–60	7 (2.21)	7 (2.59)	11 (4.62)	3 (0.86)	14 (2.39)
>60	24 (7.57)	11 (4.07)	29 (12.18)	6 (1.72)	35 (5.96)

Table 4. Share of variation in individual characteristics due to workplace affiliation

Dependent variable (worker level outcome)	Proportion of variation due to workplace affiliation
Number of hours worked per week	0.33
Hourly wage rate (RMB)	0.35
Labour supply characteristics
Gender (male = 1)	0.15
Marital status (married = 1)	0.21
Communist Party membership (yes = 1)	0.14
Hukou status (non-agriculture = 1)	0.27
Health status	0.26
Trade union membership (member = 1)	0.58
Human capital characteristics
Age (years)	0.33
Years of education	0.33
Language proficiency (standard = 1)	0.29
Length of time with firm (years)	0.19
Certification	0.04
Received on job training (yes = 1)	0.28
Feels under pressure to meet deadlines (yes = 1)	0.21

4 Empirical framework

4.1 Main empirical framework

The empirical framework used to estimate the main results is the conventional labour supply function; this is extended to accommodate firm characteristics and policies. This framework posits that the natural log of number of hours worked per week is a function of the natural log of the hourly wage rate and socio-demographic variables correlated with hours worked. We distinguish between labour supply characteristics (family income, gender, marital status, communist party membership, hukou status, health and trade union membership) and human capital characteristics (education, age, language proficiency, tenure with the firm, certification, received on-the-job training and feels under pressure to meet deadlines).

One important characteristic that has not received much attention in traditional analysis of working time at the individual worker level is workplace policies or norms (Bryan, 2007). A firm's policies on paid overtime and whether the firm has a trade union will formulate a set of norms around what firms expect. Norms are also related to the types of workers that firms employ. In high-income countries, the ideal worker norm exists among highly educated managers and professionals, with such individuals expecting themselves and others in similar positions to work long hours for years or even decades (Drago et al., 2009). In manufacturing and service sector enterprises in developing countries, though, the ideal worker norm is likely to apply to less educated female and migrant workers who firms will expect to work long hours (Smyth et al., 2012). Hence, the proportion of female staff and proportion of migrant workers that the firm employs is likely to have a significant impact on the organisational culture of the firm and underlying norms governing employer expectations of hours worked.

Bringing this together, we express the natural log of hours worked ln(HW) as a function of the following: the natural log of the hourly wage rate ln(W); labour supply characteristics of workers (LS); human capital characteristics of workers (HC); and firm characteristics (FC). Taking the natural log of hours worked and wages follows MaCurdy (1981). This can be expressed as follows where ε is the error term, reflecting unobserved random factors:

• (1)

The effect of wages on hours worked is ambiguous and depends on the magnitude of the income and substitution effect. The substitution effect is predicted to exert a positive effect on hours worked in response to a wage increase. The income effect is predicted to exert a negative effect on hours worked in response to a wage increase, assuming leisure is a normal good. If the income effect outweighs the substitution effect, the individual will work less in response to a wage increase; otherwise, the individual will work more.

Human capital characteristics such as education and training relate to differences in the productivity of leisure and work across employees. Age and the square of age may capture variations in preferences for work as well as changes in family responsibilities impacting on hours worked over the course of the life cycle (Li and Zax, 2003).

Of the worker characteristics, we expect that individuals with higher non-wage family income will work fewer hours. Moreover, we expect that females will work fewer hours because of traditional familial responsibilities (see references cited in Maurer-Fazio et al., 2011). Similarly, we expect that workers who are married will be more likely to want to synchronise home time with their partner and, hence, have less flexibility to work longer hours. We expect that those with a non-agricultural hukou will work fewer hours than those with an agricultural hukou. We expect that health status will be positively correlated with hours worked because people in better health will have greater capacity to work longer hours.

The sign on the coefficient for individuals who are members of the Chinese Communist Party is ambiguous. If party membership is a proxy for talent (Bishop and Liu, 2008), individuals who are members of the Chinese Communist Party can be expected to work less because they are more productive and complete tasks faster. Alternatively, party members might work more than non-members if ideology exhorts such individuals to exceptionally high work effort (Li and Zax, 2003). The expected sign on trade union membership is not clear-cut. Studies in Western contexts have found that being a member of a trade union will be negatively correlated with hours worked (see e.g. Bryan, 2007). In China, trade unions have traditionally played a subordinate role in resolving labour disputes and have typically acted as a mediator between employer and employee rather than as a representative of labour.

Among the firm characteristics, we expect that firms that employ a higher proportion of females and/or migrant workers will have norms of longer working hours over and above individual characteristics (Smyth et al., 2012). However, the effect of union presence in the firm on hours worked is unclear. Perloff and Sickles (1987), Earle and Pencavel (1990) and DiNardo (1991) all found that unionisation reduces the number of hours worked in studies with US data. As mentioned earlier, unions in China have played a different role than in the West. Finally, the expected sign on paid overtime is also unclear. If firms pay overtime, this might mean that employers will work longer hours because they are responding to the monetary incentive to do so. Alternatively, if firms pay overtime, this may indicate better labour management practices more generally (Seo, 2011). If so, such firms may be better able to schedule their workload and reduce excessive overtime, hence reducing hours at work.

Because the survey did not contain data on the hourly wage rate, the only way to obtain a measure of this variable was to divide reported monthly earnings by the monthly number of hours worked. Deriving the wage rate in this manner means that any errors in the measurement of monthly hours worked would be repeated in the derivation of the respondent's wage rate. Hence, estimation of Equation (1) using ordinary least squares (OLS) results in a spurious, inverse correlation between measurement errors in the wage rate and the error term (Hall, 1973; Schultz, 1980). The spurious correlation biases the estimate of the wage correlation downward (Killingsworth, 1983). To overcome the problem of biased and inconsistent estimates using OLS, a standard approach is to use instrumental variable estimation (IV) made popular by Hall (1973). The practical difficulty with IV estimation is finding an instrument or set of instruments that are significantly correlated with wages but also orthogonal to the residuals of the main equation (hours worked).

The existing literature relies mainly on worker characteristics for IVs that are normally excluded from the hours worked equation, such as higher order terms of age or education and age interacted with education (Chau et al., 2007; Fortin and LaCroix, 1997; Li and Zax, 2003; Mroz, 1987; Sahn and Alderman, 1996). We used the square of years of schooling and age interacted with education, which are common IVs in hours worked equations.

It should be noted that components of non-wage family income are also potentially endogenous. For instance, transfer payments from the government or from individuals outside the family may depend on hours worked (Li and Zax, 2003). To address this issue, Li and Zax (2003) use lagged values of non-wage family income to instrument for current non-wage family income. However, we do not have this information for our data set or other appropriate IVs for non-wage family income. Thus, while recognising the problem, we follow Sahn and Alderman (1996) and treat non-wage family income as being exogenous.

4.2 Multi-level modelling

A potential problem with the OLS and IV estimates is that workers from the same firm may be more alike in their attitudes towards hours worked than workers chosen at random from a variety of firms. This problem occurs because many data sets, including our data set, have a hierarchical or clustered structure. Hence, as a robust check on the OLS and IV estimates, we also employed a multi-level model, which recognises the existence of such data hierarchies by allowing for residual components at each level in the hierarchy. In the two-level hierarchy, as in our case, the residual variance is partitioned into a between-firm component (the variance of the firm-level residuals) and a within-firm component (the variance of the worker-level residuals). One must choose between the random intercept and random slope alternatives. To do so, we relied on a Likelihood Ratio (LR) test. If the LR test is significant, this suggests that the random slope alternative is to be preferred. While the multi-level mixed effects model has the advantage that it allows for residual components at the employer and employee level, it has the important limitation that it can not account for the endogeneity of wages.

4.3 Determinants of long hours

While our main purpose is to consider the determinants of hours worked more generally, we also considered the factors correlated with long working hours. For this purpose, we employed a logit model in which the dependent variable is a binary variable set equal to one if the respondent works above a certain threshold. In separate models, we employed two thresholds for number of hours worked. The first is 44 hours, which is the normal working week in China. Thus, the binary variable was set equal to one if the individual works greater than or equal to 45 hours per week. The second is 60 hours, which is the maximum working week, including overtime, according to ILO regulations. Thus, the binary variable was set equal to one if the individual works greater than 60 hours per week. The explanatory variables, which are the same as in the conventional labour supply model, are discussed earlier.

5 Results

5.1 Main results

Table 5 contains the results for the OLS estimates for Equation (1). Hours worked are significantly and negatively correlated with wages, suggesting the existence of a backward bending labour supply curve. Those with higher human capital are more likely to work longer hours. Among the labour supply characteristics of workers, males work longer hours than females, while males and females with a non-agricultural hukou status work fewer hours. Among the variables denoting firm characteristics and policies, the coefficients on proportion of female employees and proportion of migrant workers are always significant.

Table 5. OLS estimates

ln(Number of hours worked per week)	(1)	(2)	(3)	(4)	(5)
*Denotes statistical significance at 1%; **denotes statistical significance at 5%; ***denotes statistical significance at 10%.
ln[Hourly wage rate (RMB)]	−0.168* (−9.794)	−0.176* (−10.29)	−0.181* (−10.49)	−0.182* (−10.47)	−0.179* (−10.25)
Labour supply characteristics
Gender (male = 1)	0.0364** (2.427)	0.0510* (3.314)	0.0542* (3.493)	0.0543* (3.462)	0.0553* (3.619)
Marital status (married = 1)	−0.00118 (−0.0498)	0.00612 (0.263)	0.00687 (0.295)	0.00728 (0.311)	0.00998 (0.432)
ln(family income)	−0.00205 (−1.011)	−0.00208 (−1.038)	−0.00205 (−1.018)	−0.00193 (−0.953)	−0.00294 (−1.441)
Communist Party membership (yes = 1)	−0.0141 (−0.575)	−0.00964 (−0.399)	−0.00865 (−0.358)	−0.00720 (−0.295)	−0.00886 (−0.365)
Hukou status (non-agricultural = 1)	−0.0655* (−3.765)	−0.0520* (−2.958)	−0.0533* (−3.021)	−0.053* (−2.989)	−0.052* (−2.980)
Health status (ordinary = 1)
Good	0.0201 (0.994)	0.0254 (1.276)	0.0256 (1.283)	0.0244 (1.215)	0.0276 (1.389)
Very good	0.0311 (1.575)	0.0303 (1.561)	0.0322*** (1.658)	0.0316 (1.612)	0.0329*** (1.699)
Trade union membership (member = 1)	−0.00253 (−0.163)	0.00543 (0.285)	0.00376 (0.198)	0.00454 (0.236)	0.0156 (0.815)
Human capital characteristics
Age	1.674* (2.957)	1.332** (2.374)	1.321** (2.353)	1.303** (2.311)	1.162** (2.053)
Age^2	−0.239* (−2.980)	−0.190** (−2.398)	−0.189** (−2.379)	−0.186** (−2.333)	−0.169** (−2.109)
Education	0.0162* (4.895)	0.0167* (5.068)	0.0152* (4.500)	0.0151* (4.425)	0.0145* (4.256)
Language proficiency (standard = 1)	−0.0257 (−1.509)	−0.0221 (−1.321)	−0.0235 (−1.401)	−0.0243 (−1.429)	−0.0184 (−1.073)
Length of time with firm	−0.00405*** (−1.953)	−0.00352*** (−1.734)	−0.00323 (−1.587)	−0.00322 (−1.552)	−0.00241 (−1.149)
Certification (no title = 1)
Elementary	−0.00414 (−0.209)	−0.0207 (−1.050)	−0.0273 (−1.368)	−0.0270 (−1.346)	−0.0326 (−1.626)
Junior/senior	0.0671** (2.344)	0.0609** (2.168)	0.0576** (2.049)	0.0563** (1.985)	0.0503*** (1.801)
Received on job training (yes = 1)	−0.00506 (−0.353)	0.00662 (0.462)	0.00712 (0.493)	0.00819 (0.562)	−0.00509 (−0.343)
Feels under pressure to meet deadlines (yes = 1)	0.0211 (1.417)	0.0175 (1.173)	0.0183 (1.227)	0.0173 (1.142)	0.0178 (1.163)
Firm characteristics and policies
Firm size (number of employees)		−3.07e-05 (−1.199)	−3.27e-05 (−1.274)	−3.36e-05 (−1.290)	−2.38e-05 (−0.867)
Proportion of female employees		0.120* (3.591)	0.121* (3.598)	0.113* (3.161)	0.123* (2.754)
Proportion of migrant workers		0.0675* (2.892)	0.0616* (2.613)	0.0630* (2.629)	0.0510*** (1.859)
Firm has trade union (yes = 1)		0.0150 (0.848)	0.0154 (0.869)	0.0173 (0.964)	0.00574 (0.295)
Labour disputes affected production (yes = 1)		−0.102* (−2.969)	−0.0972* (−2.832)	−0.095* (−2.720)	−0.0450 (−0.939)
Will be paid overtime (yes = 1)		−0.0988** (−2.326)	−0.110** (−2.555)	−0.107** (−2.456)	−0.0213 (−0.422)
Other dummies
Occupation dummies	No	No	Yes	Yes	Yes
Ownership dummies	No	No	No	Yes	Yes
Industry dummies	No	No	No	No	Yes
Constant	1.127 (1.148)	1.733*** (1.780)	1.812*** (1.858)	1.829*** (1.868)	2.024** (2.051)
Observations	587	587	587	587	587
R-squared	0.220	0.268	0.275	0.275	0.350

Table 6 contains the results for the IV estimates for Equation (1). Diagnostics at the bottom of Table 6 indicate that our instruments are valid. The coefficient on the hourly wage rate is negative in all specifications, suggesting that the income effect outweighs the substitution effect and that the labour supply curve is backward bending. The IV estimates suggest that for a 1 per cent increase in wages, the number of hours worked per week will decrease between 0.22 per cent and 0.26 per cent. Among the labour supply characteristics, gender and hukou status are statistically significant in each specification. The males work between 4.4 per cent and 6.8 per cent longer hours than the females. The males and females with a non-agricultural hukou work between 4.9 per cent and 6.3 per cent fewer hours than do those with an agricultural hukou, depending on the exact specification. The other labour supply characteristics, including non-wage family income, are statistically insignificant. In terms of human capital characteristics, additional years of schooling is associated with longer hours of work. For each additional year of schooling, the respondents worked 1.7 to 2.1 per cent longer hours. As in the OLS results, the effects of age on hours worked is non-linear. The other human capital characteristic that is statistically significant is certification. Workers with junior or senior certification worked between 6.1 per cent and 7.5 per cent longer hours than those with no certification. Other human capital characteristics are insignificant, with the exception of tenure with the firm shown in the second column.

Table 6. IV estimates

ln(Number of hours worked per week)	(1)	(2)	(3)	(4)	(5)
*Denotes statistical significance at 1%; **denotes statistical significance at 5%; ***denotes statistical significance at 10%.
ln[Hourly wage rate (RMB)]	−0.215* (−2.612)	−0.238** (−2.542)	−0.255** (−2.504)	−0.258** (−2.512)	−0.245* (−2.620)
Labour supply characteristics
Gender (male = 1)	0.0437** (2.257)	0.0617* (2.796)	0.0677* (2.812)	0.0680* (2.833)	0.0666* (3.089)
Marital Status (married = 1)	−0.000832 (−0.0356)	0.00761 (0.329)	0.00898 (0.386)	0.00949 (0.407)	0.0124 (0.545)
ln(family income)	−0.00193 (−0.954)	−0.00201 (−1.014)	−0.00197 (−0.985)	−0.00184 (−0.915)	−0.00303 (−1.527)
Communist Party membership (yes = 1)	−0.00760 (−0.286)	0.000365 (0.0129)	0.00338 (0.116)	0.00597 (0.200)	0.00189 (0.0679)
Hukou status (non-agricultural = 1)	−0.0627* (−3.507)	−0.0504* (−2.870)	−0.0517* (−2.936)	−0.051* (−2.910)	−0.049* (−2.834)
Health status (ordinary = 1)
Good	0.0173 (0.840)	0.0222 (1.096)	0.0216 (1.054)	0.0201 (0.972)	0.0267 (1.381)
Very good	0.0294 (1.488)	0.0284 (1.467)	0.0305 (1.572)	0.0296 (1.512)	0.0333*** (1.774)
Trade union membership (member = 1)	−0.00247 (−0.160)	0.0103 (0.510)	0.00890 (0.442)	0.0102 (0.498)	0.0236 (1.090)
Human capital characteristics
Age	2.013** (2.508)	1.757** (2.084)	1.814** (2.074)	1.812** (2.064)	1.590** (1.961)
Age^2	−0.286** (−2.534)	−0.250** (−2.105)	−0.258** (−2.094)	−0.258** (−2.081)	−0.229** (−2.001)
Education	0.0194* (3.054)	0.0205* (3.135)	0.0192* (3.013)	0.0193* (2.977)	0.0177* (3.201)
Language proficiency (standard = 1)	−0.0266 (−1.571)	−0.0234 (−1.402)	−0.0256 (−1.517)	−0.0268 (−1.564)	−0.0205 (−1.217)
Length of time with firm	−0.00399*** (−1.939)	−0.00337*** (−1.662)	−0.00294 (−1.429)	−0.00285 (−1.348)	−0.00233 (−1.143)
Certification (no title = 1)
Elementary	−0.00228 (−0.115)	−0.0192 (−0.976)	−0.0276 (−1.392)	−0.0273 (−1.374)	−0.0346*** (−1.765)
Junior/senior	0.0748** (2.398)	0.0721** (2.222)	0.0698** (2.149)	0.0683** (2.115)	0.0601** (1.983)
Received on job training (yes = 1)	−0.000478 (−0.0295)	0.0129 (0.759)	0.0149 (0.836)	0.0167 (0.910)	0.00215 (0.123)
Feels under pressure to meet deadlines (yes = 1)	0.0238 (1.542)	0.0187 (1.258)	0.0200 (1.336)	0.0191 (1.259)	0.0201 (1.326)
Firm characteristics and policies
Firm size (number of employees)		−3.72e-05 (−1.370)	−4.10e-05 (−1.471)	−4.23e-05 (−1.497)	−3.16e-05 (−1.101)
Proportion of female employees		0.132* (3.519)	0.134* (3.530)	0.125* (3.218)	0.127* (2.913)
Proportion of migrant workers		0.0618** (2.506)	0.0534** (2.064)	0.0546** (2.084)	0.0373 (1.137)
Firm has trade union (yes = 1)		0.0105 (0.565)	0.0107 (0.573)	0.0130 (0.699)	−0.000898 (−0.0428)
Labour disputes affected production (yes = 1)		−0.105* (−3.061)	−0.0998* (−2.919)	−0.098* (−2.818)	−0.0571 (−1.156)
Will be paid overtime (yes = 1)		−0.124** (−2.201)	−0.141** (−2.342)	−0.139** (−2.295)	−0.0423 (−0.742)
Other dummies
Occupation dummies	No	No	Yes	Yes	Yes
Ownership dummies	No	No	No	Yes	Yes
Industry dummies	No	No	No	No	Yes
Constant	0.584 (0.436)	1.085 (0.795)	1.080 (0.775)	1.066 (0.760)	1.385 (1.060)
Observations	587	587	587	587	587
R-squared	0.209	0.252	0.251	0.250	0.333
Regression diagnostics
F test of excluded instruments (p value)	0.000	0.000	0.000	0.000	0.000
Under-identification test (Anderson canon. corr. LM statistic)	24.84*	19.25*	16.68*	16.46*	19.24*
Instrument exogeneity (Sargan test) (p value)	0.171	0.367	0.261	0.295	0.270

The relationship between hours worked and firm characteristics and policies are explored in the second to fifth columns of Table 6. The proportion of female workers and the proportion of migrant workers are both generally significant and positively correlated with hours worked. For each additional 0.1 increase in the proportion of female workers in a firm, hours worked increases between 12.5 per cent and 13.4 per cent. For each 0.1 increase in the proportion of migrant workers in the firm, the corresponding figure is 5.3–6.2 per cent. The results suggest that if labour disputes have affected production, hours worked are 9.8 per cent to 10.5 per cent less. In firms that paid overtime, employees worked 12.4 to 14.1 per cent less.

5.2 Multi-level modelling results

The results of the multi-level linear model are reported in Table 7. All the specifications are based on a random intercept model. We conducted an LR test between the random intercept model and random slope model. The results are not reported, but the LR test statistic was insignificant in all cases. One can also use an LR test to ascertain if the multi-level mixed effects linear model is preferable to a linear model. The LR statistic reported at the bottom of Table 7 is significant for each specification, suggesting that the multi-level mixed effects linear model is preferable to the OLS model (although because endogeneity of wages is not addressed, not necessarily the IV model). The results for the multi-level linear model are very similar to the OLS and IV results reported in Tables 5 and 6 in terms of signs and significance of the coefficients. This suggests that the earlier reported results are generally robust. One difference is that non-wage family income is significant with a negative sign in the multi-level linear model. For each additional 1 per cent increase in non-wage family income, the respondents worked 0.4 per cent fewer hours. The other difference is that being paid overtime is significant with a negative coefficient in only one specification (column 3) in the multi-level linear model compared with three specifications in the IV estimates.

Table 7. Multi-level linear (random intercept) model estimates

ln(Number of hours worked per week)	(1)	(2)	(3)	(4)	(5)
*Denotes statistical significance at 1%; **denotes statistical significance at 5%; ***denotes statistical significance at 10%.
ln[Hourly wage rate (RMB)]	−0.191* (−11.23)	−0.192* (−11.37)	−0.198* (−11.50)	−0.198* (−11.51)	−0.194* (−11.39)
Labour supply characteristics
Gender (male = 1)	0.0517* (3.753)	0.0562* (4.041)	0.0588* (4.199)	0.0591* (4.201)	0.0588* (4.190)
Marital status (married = 1)	0.0219 (1.032)	0.0226 (1.067)	0.0224 (1.061)	0.0225 (1.065)	0.0203 (0.959)
ln(family income)	−0.00419** (−2.217)	−0.00400** (−2.126)	−0.00389** (−2.057)	−0.0038** (−2.024)	−0.0040** (−2.099)
Communist Party membership (yes = 1)	0.00161 (0.0728)	0.00236 (0.107)	0.00181 (0.0826)	0.00206 (0.0938)	−0.00164 (−0.0739)
Hukou status (non-agricultural = 1)	−0.0341** (−2.132)	−0.0315** (−1.966)	−0.0321** (−2.000)	−0.0320** (−1.995)	−0.0364** (−2.265)
Health status (ordinary = 1)
Good	0.0133 (0.716)	0.0154 (0.829)	0.0163 (0.879)	0.0158 (0.850)	0.0203 (1.093)
Very good	0.0271 (1.458)	0.0259 (1.403)	0.0283 (1.536)	0.0280 (1.514)	0.0308*** (1.674)
Trade union membership (member = 1)	−0.00698 (−0.400)	−0.00260 (−0.138)	−0.00303 (−0.161)	−0.00279 (−0.148)	0.00519 (0.278)
Human capital characteristics
Age	1.114** (2.142)	1.034** (1.991)	1.027** (1.972)	1.022** (1.963)	1.013*** (1.934)
Age^2	−0.161** (−2.197)	−0.150** (−2.041)	−0.149** (−2.022)	−0.148** (−2.009)	−0.147** (−1.994)
Education	0.0157* (5.116)	0.0161* (5.229)	0.0149* (4.755)	0.0149* (4.723)	0.0147* (4.664)
Language proficiency (standard = 1)	−0.00272 (−0.165)	−0.00277 (−0.169)	−0.00328 (−0.200)	−0.00362 (−0.220)	−0.00582 (−0.355)
Length of time with firm	−0.00313 (−1.582)	−0.00304 (−1.553)	−0.00285 (−1.454)	−0.00288 (−1.464)	−0.00260 (−1.317)
Certification (no title = 1)
Elementary	−0.0260 (−1.434)	−0.0310*** (−1.707)	−0.0367** (−1.997)	−0.0365** (−1.984)	−0.0371** (−2.010)
Junior/senior	0.0504** (2.004)	0.0483*** (1.924)	0.0462*** (1.848)	0.0460*** (1.835)	0.0456*** (1.806)
Received on job training (yes = 1)	0.0135 (0.962)	0.0171 (1.222)	0.0159 (1.132)	0.0163 (1.155)	0.00731 (0.516)
Feels under pressure to meet deadlines (yes = 1)	0.0230 (1.623)	0.0213 (1.506)	0.0220 (1.559)	0.0214 (1.506)	0.0210 (1.468)
Firm characteristics and policies
Firm size (number of employees)		−3.31e-05 (−0.904)	−3.32e-05 (−0.908)	−3.41e-05 (−0.920)	−2.59e-05 (−0.755)
Proportion of female employees		0.131* (2.608)	0.131* (2.622)	0.123** (2.296)	0.129** (2.200)
Proportion of migrant workers		0.0753** (2.177)	0.0693** (2.001)	0.0702** (2.010)	0.0542 (1.539)
Firm has trade union (yes = 1)		0.0191 (0.773)	0.0185 (0.750)	0.0200 (0.796)	0.01000 (0.417)
Labour disputes affected production (yes = 1)		−0.109** (−2.029)	−0.105*** (−1.951)	−0.103*** (−1.881)	−0.0582 (−0.924)
Will be paid overtime (yes = 1)		−0.102 (−1.499)	−0.114*** (−1.682)	−0.109 (−1.575)	−0.0325 (−0.481)
Other dummies
Occupation dummies	No	No	Yes	Yes	Yes
Ownership dummies	No	No	No	Yes	Yes
Industry dummies	No	No	No	No	Yes
Constant	2.137** (2.365)	2.282** (2.525)	2.347* (2.591)	2.346* (2.587)	2.295** (2.509)
Observations	587	587	587	587	587
Number of groups	72	72	72	72	72
LR test: ChiBar2	82.60*	61.51*	61.07*	60.77*	22.74*

5.3 Determinants of long hours

Table 8 reports results from a logit model in which the dependent variable is that the respondent works 45 hours per week or more (longer than a standard working week). Many of the results from the logit model reported in Table 8 are similar to the IV estimates for hours worked more generally in Table 6. Specifically, an increase in the wage rate reduces the probability of working 45 hours or more, both males and females with junior/senior certification are more likely to work 45 hours or more, both males and females with a non-agricultural hukou are less likely to work 45 hours or more and there is a non-linear relationship between age and working 45 hours or more. Employees in firms with a higher proportion of female and migrant workers are more likely to work 45 hours or more, while employees in firms in which labour disputes have affected production and in firms that pay overtime are less likely to work longer than the standard working week.

Table 8. Logit model estimates (DV = hours of work per week ≥ 45)

DV = 1 if hours of work per week > 45, 0 if otherwise	(1)	(2)	(3)	(4)	(5)
Marginal effects reported. *Denotes statistical significance at 1%; **denotes statistical significance at 5%; ***denotes statistical significance at 10%.
ln[hourly wage rate (RMB)]	−0.460* (−7.712)	−0.467* (−7.836)	−0.483* (−7.982)	−0.481* (−8.008)	−0.455* (−7.398)
Labour supply characteristics
Gender (male = 1)	0.0400 (0.994)	0.0751*** (1.798)	0.0822*** (1.959)	0.0923** (2.199)	0.0929** (2.228)
Marital status (married = 1)	−0.153** (−2.464)	−0.137** (−2.210)	−0.130** (−2.099)	−0.120*** (−1.953)	−0.0889 (−1.454)
ln(family income)	0.00189 (0.330)	0.00335 (0.586)	0.00304 (0.530)	0.00365 (0.648)	0.00334 (0.595)
Communist Party membership (yes = 1)	0.0166 (0.214)	0.0107 (0.137)	0.0114 (0.146)	−0.00448 (−0.0574)	−0.0225 (−0.292)
Hukou status (non-agricultural = 1)	−0.141* (−3.058)	−0.0914** (−1.974)	−0.0991** (−2.132)	−0.0910** (−1.975)	−0.0909** (−2.001)
Health status (ordinary = 1)
Good	0.0855 (1.480)	0.0973*** (1.712)	0.0933 (1.635)	0.0876 (1.548)	0.0741 (1.286)
Very good	0.0278 (0.496)	0.0253 (0.473)	0.0367 (0.684)	0.0341 (0.639)	0.0290 (0.543)
Trade union membership (member = 1)	−0.0904** (−2.080)	−0.0754 (−1.469)	−0.0839 (−1.633)	−0.0787 (−1.552)	−0.111** (−2.107)
Human capital characteristics
Age	6.567* (4.215)	5.420* (3.480)	5.462* (3.496)	5.185* (3.348)	4.404* (2.895)
Age^2	−0.935* (−4.216)	−0.770* (−3.480)	−0.780* (−3.511)	−0.739* (−3.356)	−0.632* (−2.923)
Education	0.0166*** (1.815)	0.0187** (2.056)	0.0146 (1.593)	0.0126 (1.375)	0.0127 (1.368)
Language proficiency (standard = 1)	−0.0459 (−0.997)	−0.0327 (−0.711)	−0.0402 (−0.873)	−0.0389 (−0.850)	−0.0431 (−0.938)
Length of time with firm	−0.00781 (−1.350)	−0.00789 (−1.387)	−0.00669 (−1.170)	−0.00810 (−1.412)	−0.00841 (−1.472)
Certification (no title = 1)
Elementary	0.117** (2.176)	0.0521 (0.955)	0.0345 (0.628)	0.0286 (0.515)	0.0309 (0.538)
Junior/senior	0.196** (2.329)	0.188** (2.207)	0.187** (2.192)	0.176** (2.107)	0.193** (2.372)
Received on job training (Yes = 1)	−0.0892** (−2.260)	−0.0518 (−1.304)	−0.0397 (−0.991)	−0.0404 (−1.015)	−0.0693*** (−1.700)
Feels under pressure to meet deadlines (yes = 1)	0.0215 (0.532)	0.0147 (0.358)	0.0159 (0.386)	0.00764 (0.187)	0.0141 (0.344)
Firm characteristics and policies
Firm size (number of employees)		−0.00014 (−1.466)	−0.00016*** (−1.650)	−0.00021** (−1.990)	−0.000175 (−1.639)
Proportion of female employees		0.205** (2.257)	0.200** (2.201)	0.170*** (1.791)	0.0635 (0.657)
Proportion of migrant workers		0.283* (4.573)	0.272* (4.363)	0.294* (4.630)	0.312* (4.676)
Firm has trade union (yes = 1)		0.0657 (1.396)	0.0781 (1.644)	0.0976** (2.014)	0.139* (2.765)
Labour disputes affected production (yes = 1)		−0.468* (−2.690)	−0.464* (−2.653)	−0.458* (−2.604)	−0.384** (−2.275)
Will be paid overtime (yes = 1)		−0.481* (−3.709)	−0.478* (−3.568)	−0.453* (−3.438)	−0.522* (−3.540)
Other dummies
Occupation dummies	No	No	Yes	Yes	Yes
Ownership dummies	No	No	No	Yes	Yes
Industry dummies	No	No	No	No	Yes
Observations	587	587	587	587	587

Table 9 reports the results of a logit model in which the dependent variable is that the respondent works in excess of 60 hours. Again, the results are generally consistent with the IV results reported in Table 6 for the wage rate, gender, hukou status, education and proportion of migrant workers in the firm. Age, certification and proportion of female workers are insignificant in Table 9. The results in Table 9 are affected by the fact that just 5.96 per cent of respondents worked in excess of 60 hours per week. This reduces the level of variation in some of the explanatory variables and forced us to drop variables for ‘labour disputes affected production’ and ‘will be paid overtime’. The results of a ReLogit model (King and Zeng, 1999), which are not reported, are similar in terms of signs and significance.

Table 9. Logit model estimates (DV = hours of work per week > 60)

DV = 1 if hours of work per week > 60, 0 if otherwise	(1)	(2)	(3)	(4)	(5)
Marginal effects reported. *Denotes statistical significance at 1%; **denotes statistical significance at 5%; ***denotes statistical significance at 10%.
ln[hourly wage rate (RMB)]	−0.0580* (−3.327)	−0.0603* (−3.082)	−0.0587* (−2.985)	−0.067* (−3.026)	−0.060* (−2.821)
Labour supply characteristics
Gender (male = 1)	0.0157*** (1.785)	0.0180*** (1.821)	0.0171*** (1.819)	0.0191*** (1.783)	0.0179*** (1.842)
Marital status (married = 1)	−0.00545 (−0.492)	−0.00581 (−0.467)	−0.00855 (−0.759)	−0.0101 (−0.772)	−0.00918 (−0.813)
ln(family income)	−0.00124 (−1.070)	−0.00117 (−0.952)	−0.000989 (−0.894)	−0.00125 (−0.950)	−0.00103 (−0.898)
Communist Party membership (yes = 1)	–	–	–	–	–
	–	–	–	–	–
Hukou status (non-agricultural = 1)	−0.0350* (−2.623)	−0.0301** (−2.253)	−0.0277** (−2.146)	−0.0313** (−2.113)	−0.0278** (−2.099)
Health status (ordinary = 1)
Good	−0.0202 (−1.329)	−0.0184 (−1.148)	−0.0160 (−1.080)	−0.0201 (−1.175)	−0.0210 (−1.337)
Very good	0.0177 (1.416)	0.0226*** (1.691)	0.0221*** (1.721)	0.0229 (1.593)	0.0188 (1.463)
Trade union membership (member = 1)	0.00651 (0.750)	0.00247 (0.218)	0.00230 (0.222)	0.00157 (0.129)	−0.00170 (−0.157)
Human capital characteristics
Age	0.578*** (1.702)	0.458 (1.293)	0.420 (1.287)	0.491 (1.325)	0.517 (1.509)
Age^2	−0.0809*** (−1.667)	−0.0636 (−1.260)	−0.0577 (−1.243)	−0.0674 (−1.278)	−0.0709 (−1.455)
Education	0.00442** (1.960)	0.00543** (2.153)	0.00489** (2.113)	0.00577** (2.107)	0.00551** (2.137)
Language proficiency (standard = 1)	−0.0126 (−1.439)	−0.0135 (−1.397)	−0.0125 (−1.361)	−0.0162 (−1.505)	−0.0148 (−1.492)
Length of time with firm	−0.000836 (−0.679)	−0.000871 (−0.670)	−0.000508 (−0.423)	−0.000514 (−0.366)	−0.000499 (−0.401)
Certification (No title = 1)
Elementary	−0.00498 (−0.402)	−0.0121 (−0.889)	−0.0123 (−0.955)	−0.0119 (−0.790)	−0.00896 (−0.663)
Junior/senior	0.0211 (1.284)	0.0124 (0.706)	0.0118 (0.765)	0.0130 (0.727)	0.00919 (0.614)
Received on job training (Yes = 1)	−0.00208 (−0.251)	0.00352 (0.393)	0.00303 (0.365)	0.00405 (0.419)	0.00253 (0.287)
Feels under pressure to meet deadlines (yes = 1)	0.00955 (1.170)	0.0138 (1.550)	0.0151*** (1.760)	0.0168*** (1.704)	0.00987 (1.142)
Firm characteristics and policies
Firm size (number of employees)		−1.99e-05 (−0.874)	−1.88e-05 (−0.841)	−1.59e-05 (−0.608)	−1.78e-05 (−0.754)
Proportion of female employees		0.0122 (0.642)	0.00922 (0.529)	0.0131 (0.595)	0.00567 (0.271)
Proportion of migrant workers		0.0277*** (1.837)	0.0243*** (1.741)	0.0231 (1.473)	0.0278*** (1.751)
Firm has trade union (yes = 1)		0.00671 (0.675)	0.00751 (0.799)	0.00609 (0.539)	0.00662 (0.639)
Labour disputes affected production (yes = 1)	–	–	–	–	–
	–	–	–	–	–
Will be paid overtime (yes = 1)	–	–	–	–	–
	–	–	–	–	–
Other dummies
Occupation dummies	No	No	Yes	Yes	Yes
Ownership dummies	No	No	No	Yes	Yes
Industry dummies	No	No	No	No	Yes
observations	587	587	587	587	587

6 Discussion

In this section, we discuss in more detail the main findings and compare the results of this current study with those of existing studies in the literature. We begin with wages. A consistent result across all the tables is that wage elasticity is negative. For a long time, the accepted wisdom was that wage elasticity for females is large and positive, while wage elasticity for males is small and negative (Killingsworth and Heckman, 1986; Pencavel, 1986). The former proposition, however, has been challenged in a series of studies of female labour supply, which have found wage elasticity to be negative (see e.g. Nakamura and Nakamura, 1981; 1983; Nakamura et al., 1979; Robinson and Tomes, 1985). At low wage levels (typical in urban China), one might expect wage elasticity to be positive because uncompensated wage increases are likely to represent large changes in the relative price of leisure, while the income effect might be expected to be comparatively small (Li and Zax, 2003).

However, the existing evidence for urban China is mixed. Using data from the 1995 China Household Income Project on 21,700 individuals living in urban areas across 11 provinces, Li and Zax (2003) found the wage elasticity to be small and positive. On the other hand, using data from 1615 urban households across five cities collected in 2002, Chau et al. (2007) found wage elasticity for husbands and wives to be negative. In Shanghai, income is high relative to the rest of China with Shanghai urban residents receiving relatively large wage increases compared with the rest of the country in the market reform period. In terms of previous studies for urban China, our finding is consistent with Chau et al. (2007) but differs from Li and Zax (2003). It suggests that with relatively high wages and long working hours in Shanghai, the relative value of extra hours in terms of leisure is valued more than extra income. This result is consistent with recent increases in living standards in China, in particular in large cities such as Shanghai where workers prioritise leisure over extra income.

A second finding across the tables is that higher human capital in the form of education and certification/training is positively correlated with hours worked. At first, it might seem as if the effect of education on hours worked is ambiguous. This is because increases in human capital increase the wage rate, and an increase in the wage rate produces opposing income and substitution effects. However, this is only part of the story. Higher human capital will increase the productivity of work and leisure (see Michael, 1972, for evidence that human capital increases the productivity of leisure). Beginning with Heckman (1976), in many economic models, human capital is assumed to affect the productivity of leisure in the same proportion that it affects the productivity of work. However, as Trostel and Walker (2000: 8) noted, ‘… it seems unlikely that human capital affects the productivity of work and leisure in the same proportion. Because there is much less scope for specialisation, limits on leisure productivity seem much more likely than limits on work productivity’. Trostel and Walker (2000) showed that in a two-period model, provided human capital affects productivity of work more than leisure, there will be a positive correlation between education and hours worked; this is consistent with the results reported here.

More generally, one reason we find that the workers with some form of certification work more than workers with no certification is that workers with low skills generally are engaged with tasks that have a fixed number of working hours. Meanwhile, higher skilled workers have supervisory/managerial roles and, hence, end up working more hours (due to added responsibility and/or accountability associated with their job). Moreover, higher skilled workers might face higher performance pressures than lower skilled workers. For example, a worker on the factory floor may have almost the same performance as another worker. However, different supervisors may not have same level of performance and, hence, may try to outperform each other in realising benefits such as incentive payments or promotion.

A third consistent finding is that those with a non-agricultural hukou work shorter hours than those with an agricultural hukou. This result is consistent with the widespread evidence that rural–urban migrants in China face urban labour market discrimination and work longer hours relative to their urban counterparts. This is also consistent with rural–urban migrants working much longer hours than their urban counterparts in the survey used in this study (see Table 3). That rural migrants who receive lower hourly earnings also tend to work longer hours may imply a strong income effect in labour supply behaviour, consistent with the finding of a negative wage elasticity, but may also result from working constraints imposed by employers on workers with limited negotiating power (Demurger et al., 2009).

Among firm characteristics, a consistent finding is that employees work longer hours in firms employing a higher proportion of female and migrant workers. This finding is consistent with the results in Smyth et al. (2012) for supply chain factories. Firms that employ a high proportion of female employees are typically in textile, clothing and footwear industries. That females are disproportionately represented in such factories is because of their comparative advantage in the sort of work that is required and generally because they are compliant (Pun, 2007). On the basis of interviews in China in the 1990s and 2000s, Chan and Siu (2010: 172) stated: ‘managers explained that young female workers have nimble hands, are more obedient and easier to manage and are faster and more meticulous’. Women working in textile, clothing and footwear industries are typically young and single and have few family responsibilities––characteristics that reinforce the norm of long working hours.

Firms employ migrant workers in a variety of blue-collar occupations. The norm of long hours in firms that employ a high proportion of migrant workers is reinforced by the fact that most migrants live on site in factory dormitories. Hence, the distinction between work and leisure is clouded. Chan and Siu (2010) also found that migrant workers in the garment sector in China worked very long hours relative to other industries. The main reason for long working hours in this sector is that most workers are unskilled and are paid by piece rates. Edgren (1995) reported that it is common for norms of long working hours to be reinforced by graphs displaying daily production achievements attached to their bench and competitions to set new records of productivity. This sets up an internal tournament in which workers compete. Workers accept long hours to compensate for low wages. The problem is often compounded by poor internal production systems with tight delivery schedules (Seo, 2011).

Among other workplace characteristics, a consistent finding is that if labour disputes have affected production, hours worked are less. This reflects that labour disputes represent a disruption to production in that employees are not working. It may also be that those employees who work longer hours are more likely to protest against working longer hours. This conjecture is consistent with studies that suggest that long working hours is positively correlated with absenteeism (e.g. Barmby et al., 2002). According to one report, one of the main causes of labour disputes in Shanghai is long working hours (China Daily, 2001).

Another finding is that in firms that paid overtime, employees work fewer hours. This result is consistent with payment of overtime being a proxy for enhanced labour management practices more generally in which firms were able to better schedule their workload and demands on workers (Seo, 2011). There is a substantial literature suggesting that a reduction in excessive working hours improves productivity (e.g. Bosch and Lehndorff, 2001). Workers in manufacturing enterprises with better internal quality and management systems can work shorter hours without reducing wages because they typically make more pieces in standard hours (Seo, 2011). This finding is generally consistent with those reported in Smyth et al. (2012) and, in particular, the conclusion in that study that hours worked in Chinese and Thai supply chain factories are positively correlated with poor work organisation and scheduling.

7 Conclusion

Conventional economic analysis of hours worked at the individual worker level has tended to ignore the employer dimension. This omission results from the fact that conventional labour supply estimates model hours worked as a function of the hourly wage rate and employee preferences for work. There are relatively few studies that have considered the role of employer policies and characteristics in modelling hours worked, and those studies that have considered these factors omit the hourly wage rate. This potentially biases the estimates. In this study, we have used a representative-matched employer–employee data set from Shanghai to analyse the correlates of hours worked at the individual worker level, while taking account of firm characteristics, in China's rapidly evolving urban labour market.

Our main findings are as follows. First, the wage elasticity is negative. Second, human capital in the form of education and training is positively correlated with hours worked. Third, those with a non-agricultural hukou work shorter hours than those with an agricultural hukou. Fourth, employees work longer hours in firms employing a higher proportion of female and migrant workers. Fifth, employees work shorter hours in firms that pay overtime. Sixth, while trade unions are not influential in terms of hours worked, labour disputes are relevant. This latter result reflects the fact that the traditional role of the trade unions in China is to mediate disputes between employer and employee and not to act as a representative of employees. Consequently, employers typically agitate for reform outside the trade union movement, including through the factory dormitories. It is generally accepted that the number of labour disputes has increased despite trade unions. Hence, one sees a disconnect between the results for trade unions and labour disputes with respect to hours worked.

Organisations such as the FLA and ILO have long argued for the benefits of a reduction in long working hours in countries such as China to both employers and employees as part of their ‘better work’ agendas. There are some broad conclusions from our findings that have practical implications for reducing work hours focused on firm policies, which can potentially inform these policy agendas. Based on our results, three practical avenues for reducing work hours in Chinese manufacturing firms are increase wage rates, address norms of long working hours in firms that employ a high proportion of migrant and female workers and employ better labour and management practices. These findings are consistent with the ILO's Conditions on Work and Employment Programme, which provides for a framework on working time that balances workers’ needs with business requirements (Lee et al., 2007). We conclude by briefly elaborating on how these findings could be implemented in the context of the wider policy debates on the best means to reduce hours at work.

The ILO has emphasised that from an employee perspective, one of the main reasons for working overtime is low hourly wages (ILO, 2009; Lee et al., 2007). For example, in one survey of 142 Chinese garment factories conducted by Verite (2004), 53 per cent of respondents stated that the main reason for working long hours was to earn extra income. Similar findings have been reported in surveys in Cambodia and the Philippines (Kang and Dannet, 2010; Mehran, 2005). Our finding that increasing the wage rate would reduce work hours is consistent with ILO recommendations for reducing work hours, which are based on the existence of a large income effect (see e.g. ILO, 2009; Lee et al., 2007; Seo, 2011). There are several studies suggesting that wages are tied to productivity (see e.g. Hellerstein et al., 1999). Hence, there are also likely to be secondary benefits with higher wages resulting in higher productivity while the employee is at work, reducing the need for longer hours.

As discussed earlier, the finding that in firms that paid overtime employees worked fewer hours is consistent with paid overtime acting as a proxy for better management and labour practices. The ILO has emphasised that one of the root causes of long working hours in Chinese manufacturing is poor internal production systems with tight delivery schedules that squeeze workers (Seo, 2011). While there is no direct evidence on this point from our survey, firms that pay overtime often also have better internal quality and productivity management systems as part of a suite of high performance work practices (Seo, 2011).1 There is a lot of evidence that high-performance work practices and better internal management practices have a positive effect on productivity and lead to a reduction in working hours without reducing wages (see e.g. Frenkel and Scott, 2002; Huselid, 1995).

Finally, our results suggest that changing norms of long working hours in firms that employ a high proportion of female and migrant workers can reduce work hours. This generally entails changes in work organisation and methods of production (including the optimum number of hours, optimal beginning and finishing times, optimal rest breaks and budgeting methods) (Seo, 2011). These changes should be combined with incentives and proper targets for line managers to reorganise work methods and replace what Seo (2011) called ‘the long work culture’ with ‘smart working’. As long hours of work are positively correlated with absenteeism and staff turnover, moving to ‘smart working’ can also benefit firms in terms of reduced absenteeism and lower turnover (Seo, 2011).

Footnotes

1. 1

   Frenkel and Scott (2002) and Smyth et al. (2012) provide evidence on this in Chinese manufacturing firms.