The Hukou system and institutional background

Hukou is a system of household registration in China. Similar to an internal passport, it is used by the government to regulate population mobility between rural and urban areas (Reference Zhao and Howden -ChapmanZhao and Howden -Chapman, 2010). Hukou is closely connected to access to most of the public services offered by the local city. Without a local Hukou, rural migrants are less likely to gain urban welfare benefits and need to pay more for social services (Reference WuWu, 2002).

Accordingly, to reduce restrictions on Hukou, many cities have already relaxed residency requirements to attract domestic migrants – especially young graduates (Reference Wu and ZhangWu and Zhang, 2018; Reference Zhang and TaoZhang and Tao, 2012). These incentives, however, tend to favour young and educated workers rather than the rural migrants without higher education qualifications. Many of these rural migrants are unable to reach those criteria and have less likelihood to obtain access to urban welfare benefits. ^{Footnote 2} The Hukou system still acts as the main institutional barrier that restricts access the basic urban welfare and social service programmes to the disadvantage of rural migrants in modern China (Reference Fang and ZhangFang and Zhang, 2016).

The segregated health insurance in the Hukou system may induce different commuting influences on sickness absence among rural migrants and urban citizens. The health insurance system in urban is mainly consisted of Urban Employee Basic Medical Insurance (UEBMI), which was established in 1988 and fully implemented nationwide in 1998. UEBMI is a compulsory medical insurance which is jointly financed by employers (6%) and employees (2%). However, rural migrants have lower UEBMI coverage than their urban counterparts. According to Reference MengMeng (2017), only 29% of rural migrant workers participated in UEBMI, whereas, nearly 92% of urban citizens enrolled in it.

The Hukou system exposes rural migrants to a disadvantaged position, giving them an unfairly restricted share of public health insurance and compelling them to pay high out-of-pocket expenses for health services (Reference Peng, Chang and ZhouPeng et al., 2010). When these workers get sick, they may be involved in a series of ineffective health-seeking behaviours, such as unsupervised self-medication, obtaining medical advice from unlicensed private clinics or ‘just holding on’ without enjoying health services for minor illnesses (Reference Hong, Li and StantonHong et al., 2006; Reference Liang, Zheng and HeLiang et al., 2010; Reference Peng, Chang and ZhouPeng et al., 2010).

Conceptual framework

In line with previous studies of Reference Van Ommeren and Gutiérrez-i-PuigarnauVan Ommeren and Gutiérrez-i-Puigarnau (2011) and Reference Giménez-Nadal, Molina and VelillaGiménez-Nadal et al. (2018a, Reference Giménez-Nadal, Molina and Velilla2018b), the present study addresses the positive effect between commuting time and sickness absence, and the nature of the transmission mechanism. The effect of commuting time on sick leave may be transmitted by health status. A number of studies have shown that long commuting was accompanied by lower subjective/psychological well-being and increased stress (Reference Novaco, Gonzales and Amichai-HamburgerNovaco and Gonzales, 2009; Reference Stutzer and FreyStutzer and Frey, 2008; Reference Van Ommeren and Gutiérrez-i-PuigarnauVan Ommeren and Gutiérrez-i-Puigarnau, 2011), and that such stress impairs the employee’s health and thereby induces absenteeism.

However, unique institutional arrangements, especially the Hukou system, may be another transmission mechanism that has a different influence on worker’s commuting and health-seeking activities. With insufficient health insurance for the reasons outlined above, rural migrants may have to give up the right to sick leave or to seek professional hospital services once they experience health problems (Reference Peng, Chang and ZhouPeng et al., 2010). The impact of commuting on sickness absence may therefore vary by Hukou status, that is, commuting time induces absenteeism for urban citizens, while it has no significant effect on rural migrants. The potential mechanism channel we claim is still the health status. Commuting time has both significant effect on the health status of urban citizens and rural migrants. But with insufficient health insurance, rural migrants who suffer health-related illness have less access to ask for formal absence permission, whereas urban citizens with illness find it relatively easier to get an official certificate to ask for legal sickness absence. Thus, unequal and discriminatory health insurance may block the transmission channel linking commuting time and sickness absence, which may lead to a commuting–absenteeism effect that varies according to Hukou status (Figure 1).

Figure 1. Research framework of this study.

Sample introduction

This study uses data from CMEES, which is conducted by the School of Labour and Human Resources, Renmin University of China. The samples were selected from an enterprise listing according to 2008 national economic census data, which were drawn using the two-stage stratified random sampling method. The enterprises in the private and public sectors must have 20 employees or above, and if one enterprise refused to be interviewed, it will be replaced by another with the same size in the same industry and regions.

Unfortunately, the CMEES conducted a commute times survey only in 2013, and so our study could not use more recent data. CMEES 2013 includes 4532 employees in 444 companies and covers 12 cities across the country. CMEES 2013 collects rich information on the demography, employment traits and company characteristics, while it also provides detailed information about both sickness absence days and commuting time. Thus, it provides useful data for analysing the effect of commuting on sickness absence.

Descriptive statistics

The distributions of the main variables are shown in Table 1. Rural migrants occupied 21.74% of the full sample, while urban citizens amounted to 78.26%. The dependent variable ‘sickness absence’ was defined from the question ‘In the past year, how many days have you asked for leave due to illness?’, which was defined to include sickness absence with and without official sickness certificate. The average sickness absence days in the full sample were nearly 2.46, with a standard deviation of 7.42. In addition, 60.26% of employees had not been absent for illness during the past year. Moreover, the duration of sickness absence of rural migrants was slightly lower than that for urban citizens. Figure 2 portrays the full distribution of sickness absence days.

Table 1. Descriptive statistics.

SD: standard deviation.

Figure 2. Frequency distribution of absent days.

Commuting time is the focal variable in this study, which is defined as minutes spent in one-way daily. The average one-way commuting time was approximately 26 minutes in the full sample, whereas rural migrants’ commute time was less than urban citizens, with a gap of 6.2 minutes. The probable explanation is that most rural migrants have less chance to be homeowners. Many have to live in employer-provided dormitories or ‘Urban Villages’ with a short commute distance from the workplace (Reference Li and DudaLi and Duda, 2010). Figure 3 portrays the full distribution of commuting times.

Figure 3. Frequency distribution of commuting time.

The control variables included individual-level variables (personal characteristics and job-related traits), company-level variables (company types and sectors) and city-level variables. Rural migrants were younger (a mean age of 29 years) than urban citizens (a mean age of 34 years), and only 47% of rural migrants were married, whereas 72% of their urban counterparts were married. Although rural migrants had lower educational levels than urban citizens, most of them had completed the 9-year compulsory education.

Most of the rural migrants seemed to be in a weaker position than urban citizens. As shown in Table 1, 80% of employees worked in domestic private enterprise and about 40% of them were employed in the manufacturing sector. A majority of the samples came from second-tier cities, with only 19.70% and 18.56% chosen at the level of first- and third-tier cities. As indicated in endnote 2, over 849 of the samples were thus from most developed areas in China; 2661 of them were collected in well-developed cities, and 800 of the samples were from other prefecture-level cities excluding the first- and second-tier cities in China.

Econometric method

Sickness absence days are a count variable (0, 1, 2, 3, etc.). When modelling a count variable, there are several count model options. As shown in Figure 3, the distribution of sickness absence days in our sample was heavily skewed to the right and contained a large proportion of zeroes. The number of zeroes might seem excessive given the processes that could lead to a response variable value of zero. In this case, an employee might be absent zero days during the last year if he never got sick and requested sick leave. Another employee might be absent 0 days because their supervisor disagreed with their request for sick leave as well as because they insisted on going to work every day regardless of illness. These two employees will look identical in the response variable, but they have arrived at the same outcome for different reasons. The first employee potentially could have been absent during the last year, but in actuality was not. The second employee was certain to be absent 0 days, and will be referred to from this point forward as a ‘certain zero’. Thus, the number of zeroes may be inflated and the number of sickness but absent for 0 days cannot be explained in the same manner as the number of employees that were absent for more than zero days. The standard negative binomial (NB) and Poisson regression (PR) models will not distinguish and deal with this problem, but a zero-inflated model allows for and accommodates this complication.

The count-fit function in Stata software was used to explore the goodness of fit for all four count models. Both the results of the residuals and a set of fit statistics from the tested models, including AIC, BIC and Vuong tests, indicate that ZINB model is better than others. The results are depicted in Table 2 and Figure 4.

Table 2. Tests and fit statistics for count-data model.

BIC: Bayesian information criterion; AIC: Akaike’s information criterion; ZINB: zero-inflated negative binomial. PRM: Poisson Regression Model(PRM ) ; NBRM:The Negative Binomial Regression Model (NBRM) ZIP: Zero-Inflated Poisson(ZIP); LRX2:likelihood-ratio X² statistic(LRX2)

Figure 4. Observed–predicted deviations for count-data models.PRM: Poisson Regression Model; NBRM:Negative Binomial Regression Model;ZIP: Zero-Inflated Poisson(ZIP);ZINB:Zero-Inflated Negative Binomial Rural migrants accessing

The ZINB regression model generates two separate models and then combines them. First, a logit model is generated for the ‘certain zero’ cases (described above) to predict whether or not a worker would be in this group. Then, an NB model is generated that predicts the counts for those workers who are not certain zeroes. Finally, the two models are combined. For more information about the specific introduction of ZINB regression model, see Reference HilbeHilbe (2007).

Empirical results: Baseline model

Three ZINB regressions were applied to explore the relationships among commuting, illness and absence. The first estimation was conducted using the full sample to confirm the positive nexus between commuting time and the number of sickness absence days, while the subgroups of rural migrants and urban citizens were estimated separately to identify the heterogeneity effects. The results are shown in Table 3. The Vuong test in the bottom of Table 3 also proved that a ZINB model was a better fit for the data than standard NB.

Table 3. Estimation results of commuting time and sickness absence.

The selection of variables is almost consistent with the NB (negative binomial) regression model.

Robust standard errors in parentheses:

* p < 0.1,

** p < 0.05,

*** p < 0.01.

The estimation in the full sample reveals that commuting time is significantly positively related to the number of sickness absence days. It is consistent with the conclusion of Reference Van Ommeren and Gutiérrez-i-PuigarnauVan Ommeren and Gutiérrez-i-Puigarnau (2011) and Reference Giménez-Nadal, Molina and VelillaGiménez et al. (2018a), who found evidence of a positive relationship between commuting time and sickness absence in Germany and America. Furthermore, the estimation coefficients demonstrate that the longer an employee’s commuting time is, the more predicted days absent for employee. When the commute time increases by 1 minute, the expected number of days absent will increase by a factor of 1.003807 (exp(0.0038)) while holding all other variables in the model constant.

When we divided the samples into two subgroups by Hukou status (rural migrants and urban citizens), the separate estimations in Table 3 indicate that the commuting–absenteeism effect varies across different employee groups. Commuting time exerts a positive effect on absenteeism for urban citizens, but a very different picture emerges for rural migrants, for whom sickness absence do not seem to be significantly associated with commuting time. The result is also analogous with Reference Giménez-Nadal, Molina and VelillaGiménez-Nadal et al. (2018a), who found the evidence of another heterogeneity effect. Reference Giménez-Nadal, Molina and VelillaGiménez-Nadal et al. (2018a) concluded that the link between commuting and sick-day absences varies by gender. In other words, the commute is associated with male worker’s sick-day absences, but in the case of women, its relationship is not significant.

The estimations of control variables in the full-sample model are consistent as expected. Migrants have less likelihood to be absent due to sickness: the expected number of days absent in a year for migrants is exp(–0.2767) = 0.7583 times the expected number of days for urban residents while holding all other variables in the model constant. This finding reveals that rural migrants may suffer more discrimination in labour right protections, exposing them in a vulnerable position. When they suffer sickness, they may have less access to absence permission.

Higher wages will weaken the likelihood of being absent when sick. With a 1% increase in the wage, the expected number of days absent in a year will decrease 0.7368% (exp(–0.3055)) while holding all other variables in the model constant. This result is consistent with the proposition that an increase in sick leave costs will reduce the demand for sick leave. Moreover, the results also indicate that socio-economic status, such as age, marital status, occupational class and health status are important determinants of sickness absence.

Robustness checks

We now develop additional analyses to check the robustness of our results. The robustness checks involve omitting variables, possible measurement error and other additional variables, which are estimated through three employees groups: full samples, rural migrants and urban citizens.

Robustness model I explores the effect of extreme values. An extreme value of commute time may result from potential self-reporting error, which will induce a bias estimation. Thus, robustness model I drops the samples whose absenteeism was more than 180 days during the past year from the estimation.

Furthermore, to correct the measurement error of sickness absence, we also define sickness absence as a dummy variable (it equals 1, if the individual took sick leave during the past year) to make regression (robustness model II). To identify the effect difference between long and short commuters, we also divide the commuters into three groups, short-time commuters, middle-time commuters and long-time commuters. The employees whose time spent on one-way daily travel is less than 10 minutes (i.e. 0 ⩽ CT ⩽ 10 minutes) are defined as short-time commuters, those who travel to work over 10 minutes and less than 26 minutes are middle-time commuters (i.e. 10 < CT ⩽ 26 minutes), while the rest of the employees who travel over 26 minutes are long-time commuters (i.e. CT > 26 minutes).

Robustness model IV is re-estimated for those individuals who have not been injured during work because injury can affect both the probability of becoming a commuter and of being absent from work. One can imagine that an injured employee may receive a large number of days absent for recuperation and unwillingness to experience a longer commute. In this case, it may bring a few outliers. Hence, robustness model IV excludes observations for those who have been injured at work during the past year.

In a further robustness check, robustness model V excludes observation of workers who reported that their medical expenditure in the past year was more than 10,000 Yuan. The mean of medical expenditure for the full sample is 1083 Yuan. An employee whose medical expenditure is nearly 10 times greater than the mean value could suffer a serious illness and poor health, which may cause estimation bias. In a further robustness check, model VI, the modes of transportation such as active and passive modes were added to the estimation. The former refers to those who walk or cycle to work, while the latter includes those who drive cars or use public transportation.

The estimations from robustness checks are shown in Table 4. It is clear that commuting time has a significant positive effect on sickness absence in all six robustness estimations, which indicate that a higher commute burden may induce more sickness absenteeism. More importantly, the conclusion that the effect of commute time on sickness absence is varied by Hukou status is still robust. In all six estimations, the significant positive effect only emerges in the group of urban citizens, whereas there is no significant evidence for rural migrants.

Table 4. Robustness checks.

Model: Zero inflated negative binomial regressions (ZINB) regressions are used in all models except for robust (2) (logit). Only the coefficients for the commuting time (CT) variables are reported in NB model.

Robust standard errors in parentheses:

* p < 0.1,

** p < 0.05,

*** p < 0.01.

Transmission mechanism analysis

Given the previous findings for the baseline model and robustness checks, the conclusion can be stated that longer commuting leads to higher absence for urban citizens but not for rural migrants. It is a surprising outcome, calling for an explanation of why rural migrants and urban citizens suffer this different commuting–absenteeism effect. The potential mechanism we claim is health status, which implies that commuting time affects absence by impairing the worker’s subjective and objective health status, leading to additional health-related absence, which is recognised as involuntary absenteeism.

To explore the transmission channel that links the commute to sickness absence, we estimate the effects of commuting on a set of health-related outcomes, such as self-rated health status, degree of depression, whether obese or not. Table 5 presents the results from three groups. The results reported in Columns 1 and 2 show that commuting time has a significant positive influence on health outcomes for both the migrants and urban citizens. Longer commute times for both the migrants and urban citizens will induce lower self-related health status and higher depression. Meanwhile, migrants and urban citizens with longer commute time are both more likely to be obese.

Table 5. Mechanism analysis: health-related outcomes.

Only the coefficients for commuting time (CT), interaction of commuting time and migrants and migrants variables are reported. Individual-level variables are included: age, male, married, education year, manager, skilled worker, job strain, Overtime, training time, job tenure, job security, injury, wage (log), manufacture.

Robust standard errors in parentheses:

* p < 0.1,

** p < 0.05,

*** p < 0.01

We also explore the differentiated commuting–health effect through adding an interaction term between rural migrants and commuting time into the estimation. In Table 5, the results are shown in Column 3. The interaction term in the full-sampled model presents positive effect but these are not significant. Those findings suggest that commuting has a negative effect on the health of both rural migrants and urban citizens, while there is also no obvious commuting–health effect difference between these two groups.

There is a consensus that health status is related to involuntary sickness absence. Poorer health status will induce higher sickness absence (Reference Kyrolainen, Hakkinen and KautiainenKyrolainen et al., 2008; Reference ParingerParinger, 1983). In this case, commuting time may be significantly related to the sickness absence of both rural migrants and urban citizens, as we have confirmed that commuting has a significant effect on the health-related outcomes for both groups. But the previous findings from the baseline model and robustness checks have demonstrated that longer commuting induces higher absence only for urban citizens but not for rural migrants.

Why was the transmission channel blocked? It may be the unequal social health insurance induced by the Hukou system. The Hukou system remains a fundamental institutional constraint in the process of migrant integration in China. Without a local urban Hukou, rural migrants can move to a new place but cannot obtain the same public services and welfare as local urban citizens. They are less likely to participate in the UEBMI and have less access to share the formal public health care as local urban citizens. Therefore, the informal and insufficient medical service such as unsupervised self-medication, obtaining medical advice from unlicensed private clinics, or ‘just holding on’ without any health services for minor illnesses has become a common choice for rural migrants. Rural migrants accessing informal medical services will obtain no official certificate from the doctor. Consequently, it is difficult for rural migrants to ask employers for legitimate sickness absence without an official sickness certificate. If absent without permission, they will be punished for large economic losses. That is, why the commute induces the same health problems for migrants and urban citizens, but has no significant effect on the sickness absence for migrants.