The Determinants of Rural Migrants’ Employment Choice in China: Results from Joint Estimation

Daehoon Nahm*, Yuling Cui*, and Massimiliano Tani**

Abstract This paper investigates the determinants of employment choice of rural migrant workers across state-owned and privately-owned enterprises by applying a nested logit model to take into account the unobservable random parts of utility that link the choice to migrate with the choice of employer. We find strong evidence that the decisions for migration and the type of employer are related, suggesting that estimates should be obtained from data on migrants and stayers when these are available, and that caution should be used when forming policy recommendations from disjoint analyses of migration and employment choices.

JEL classification: C35, J21, J61 Keywords: rural migrant workers, employment choice, SOEs, non-SOEs, nested logit model

* Department of Economics, Macquarie University, Sydney, Australia Email: [email protected]

** School of Business, UNSW Canberra, Australia and IZA, Bonn, Germany Email: [email protected]

Acknowledgement: this project was partly funded by the Ministry of Education in China (MOE) Project of Humanities and Social Sciences (Project No. 14XJC790003)

1

1. Introduction The emergence of China’s urban labour markets since 1978 has progressed along two clear streams of reforms. The first has focused on enabling a growing over-supply of labour from the countryside to be redeployed in urban centres. The reported volume of rural migrants in China’s cities has increased from 3,000 in 1978 to about 120 million in 2004, and 244 million in 2011 (Wei and Han 2006, Yang and Yang 2009). The second has centred on diversifying labour demand away from state-owned enterprises in favour of a productivity-driven labour demand where private domestic and foreign employers can also operate. Although related, migration and employment choices across different ownerships are hardly jointly studied. More commonly, the choice to migrate is analysed separately from the choice of employment (e.g. Roberts 2001, Wang 2005, Démurger et al. 2009, Gagnon et al. 2009, Zhao 2002). This however can be problematic as the estimates obtained not only may suffer from omitted variable bias, but might also point towards the introduction of migration and employment policies in direct contrast to each other. The purpose of this paper is to analyse the effects of various personal characteristics and alternative attributes on the decisions for migration and the choice of employer by explicitly taking into account unobserved factors that relate the migration decision with employment choices. To do so, we apply a nested logit approach to model the determinants of employment choice between state-owned enterprises (SOEs) and various subtypes of non-state owned enterprises (non-SOEs) using data from the 2002 China Household Income Survey (CHIP). This database collects information from

2

rural movers and stayers, hence enabling one to jointly analyse the migrationemployment decisions faced by rural residents. The results confirm the existence of a strong link between the error terms in the migration and employment equations, suggesting that one should use both data on rural migrants and stayers when these are available, and that caution should be used when forming policy recommendations from disjoint analyses. The rest of the paper is organised as follows: section 2 provides a brief background to labour market reforms since 1978; section 3 discusses the methodology. Section 4 presents the data. Section 5 discusses the results, and section 6 concludes. 2. Background to labour market reforms Since the late 1950s, the mobility of the Chinese population between rural and urban areas, as well as across regions, has been shaped by the strict household registration (hukou) system. This system registers each person at a specific location and defines an individual’s household type (Goodkind and West 2002). Individuals have rural-hukou if they live off agricultural production and their household registrations are in rural areas, while they have urban-hukou if they live in cities. The main institutional barrier to rural-urban migration is the exclusion of rural migrants from the urban welfare system, which covers job allocation, housing, education, pension, medical care and other services (Cai 2001). The introduction of the household responsibility system (HRS)1 provided incentives to increase agricultural output per person through improved technology and productivity but at the same time created a surplus of rural labour. This triggered progressive relaxations of the hukou system since 1979, which included providing 1

The household responsibility system denotes that rural labor can contract land and other resources with local authorities. As a result, they can obtain output quotas. This system broke the collectivization of agriculture and improved agricultural productivity (Lin 1988).

3

rural migrant workers with food and housing, information on employment opportunities, relevant vocational training as well as some freedom to move permanently to medium-sized cities 2 and some provincial capitals. Despite the reforms and the substantial flow of rural migrants towards the cities, enormous social and economic gaps between urban residents and rural migrants underpinned by the hukou system still exist (Zhao 2004, Chan and Zhang 1999). With reference to labour demand, the only type of ownership existing in China prior to 1978 was the “socialist public ownership” (SOEs). The government operated a centralized mechanism in each SOE, covering labor allocation, wage determination and benefits. SOEs did not have the right to freely hire workers, as these were allocated according to an annual quota given by government departments. Once assigned to a job, employees could not be dismissed unless they committed criminal offences (Meng 2000). Wage determination depended heavily on job titles and ranks3 rather than performance. For benefits, SOEs operated as mini-benefit states providing employees with housing, medical care, and pension, amongst others (Gu 2001). These benefits substantially compensated for a prevailing low-nominal wage (Lin, Cai, and Li 1998). Such a centralized operation mechanism fostered underemployment and low productivity, let alone worsening the economic burden associated with the costs of providing benefits (Fan 2001, Brooks and Tao 2003). In 1978, specific measures were established to encourage the development of nonSOEs, such as enabling them to widen the scope from retail and wholesale industries to consulting services; to invest, import and export; and to face lower taxes. As a

2

Medium-sized cities are defined as those whose population is between 200 thousand and 500 thousand. The grade wage system was introduced as the standard of wage distribution in SOEs in the early 1950s. This system comprised of eight distinct levels for workers and technicians (working-class wage ranking) in ascending order, as well as 24 levels for administrative and managerial workers (cadres’ wage ranking) in descending order (Wang and Li 1995). For example, a university graduate who obtained a cadres’ level-23 wage for the first working year would automatically be promoted to level 22 one year later. 3

4

result, various subtypes of non-SOEs gradually developed, including collectiveowned enterprises (COEs), self-employed enterprises, private, foreign, and stateprivate joint-venture enterprises. At the same time, since the 1980s, SOEs were granted more autonomy in hiring workers and a new labor contract system (laodong hetongzhi) was introduced on a pilot basis to replace the socialist-style lifelong employment system (Seeborg, Jin, and Zhu 2000). As a result of these reforms, the percentage of urban employment in SOEs decreased from about 77% in 1978 to 32% in 2002. Wages nowadays include both fixed (guding) and variable (huo) portions, the latter depending on the individual’s productivity and the enterprise’s profitability. 3. Methodology As a simple model of joint migration and employment choice, we consider the employment choices of rural stayers and migrants using a nested logit (NL) model. This approach provides us with two advantages. First, we avoid selectivity issues related to the analysis of migrants only, as we use information on both those who selfselect into migration as well as those who do not. Second, we can partially relax the assumption of independence from irrelevant alternatives (IIA), which is inherent in multinominal logit models, as correlations between the unobservable utility components within a group of choices and the variances across groups are allowed to differ. Furthermore, the structure of the nested logit model satisfies the conditions for the underlying random utility maximisation. We assume that the indirect utility function for a rural worker i’s choosing employment alternative j that belongs to the set (branch) J is given by: Uij  Xi '  j  Zij '  j  Yi ' J  uiJ  u i, j|J

5

= Vij + ViJ+ εij

for i = 1, .., N, and j  BJ

(1)

where Xi and Yi are vectors of personal characteristics of worker i, which are invariant across choices; Zij is vector of the attributes of choices that vary across choices and workers; BJ is the set of the choices in branch J; and βj, γj, and αJ are the coefficient vectors. The random error term εij represents random component of utility, which consists of two parts; namely, the part that captures the utility element common to all choices within each branch, uiJ, and the part that represents the random utility element that is uncorrelated between the choices within the branch, ui,j|J. The random element that is common to all choices within a branch, uiJ, makes random errors of the utility functions in a branch correlated with each other. When the random errors are independently (over i) distributed as a special type of generalised extreme-value (GEV) distribution, maximising random utility leads to a nested logit model. Under the random utility maximisation framework, the inclusive value (IV) of branch J, which is defined as ln  e jBJ

Vij / J

where  J is a parameter,

represents the maximum utility that the decision-maker can expect to enjoy from the choice among the alternatives in branch J. It has been shown that the coefficient for the IV index term at the upper level,  J , is in fact the ratio of the standard deviation of uj|J, which is the same for all j  BJ, to the standard deviation of εj (i.e., uJ + uj|J). As the standard deviation of an extreme-value distribution is inversely proportional to the scale parameter, denoted λ, the IV coefficient is equal to the ratio of the upper-level scale parameter to the lower-level scale parameter, λJ/λj (noting that λj is constant for all j  BJ). Since the variance of εj cannot be smaller than the variance of uj|J, which is a component of εj, the IV coefficients (λJ/λj), which is denoted as  J in the present

6

model, for all branches must lie within the interval (0,1] for the model to be consistent with random utility maximisation. When λJ are normalised to 1 for identification,4 it is implied that λj must be greater than or equal to unity. It is also shown that the correlation between the random components of any pair of utility functions within J is 1 minus τJ2 (see Louviere et al. (2000, pp146−7). If all the IV coefficients are equal to unity, the error variances of all elemental utility functions are the same and hence the NL model collapses into the MNL model. If this were the case in the present study, it would imply that separate analyses of migration choice and employment choice would not lead to biased estimates. 4. Data The data used is drawn from the 2002 China Household Income Project (CHIP), which features distinct samples for urban, migrant, and rural China. We only use information from the individual (rather than the household) survey from the rural and migrant sample to generate the set of choices faced by an individual living in a rural area and considering whether to stay or emigrate. We include demographic information, migration history, family situation before leaving the home village, and socioeconomic characteristics of the labor market. The CHIP distinguishes various forms of SOEs and non-SOEs enterprises in the rural, migrant and urban files. We classify all employment types into six groups by the ownership type of the work unit and whether or not the worker has migrated to another city or region to work. The six employment choices available to a worker are: working in a privately-owned work unit without migrating (Private-Non-migrant),

Normalising λJ is preferred to normalising λj as the former guarantees that the model under any specification will be consistent with global utility maximisation; see Hensher and Greene (2002). 4

7

working in a state-owned enterprise without migrating (SOE-Non-migrant), working in a work unit other than privately-owned or SOE without migrating (Other-Nonmigrant); and working as a migrant worker in a privately-owned work unit (PrivateMigrant), in an SOE (SOE-Migrant), and in other types (Other-Migrant). Table 1 provides summary statistics of the explanatory variables in the final data set. Table 1: Summary Statistics of the Explanatory Variables* Characteristics of worker Gender (Male) Married Minority CPC Healthy School Year%

Mean (s.d.) 0.871 (0.335) 0.934 (0.249) 0.085 (0.278) 0.096 (0.295) 0.906 (0.292) 7.986 (2.581) 32.833 (6.370)

Attribute of choices: Income (obs.) Income+ (2,861) Private Non-migrant (156) Private Migrant (1,408) SOE Non-migrant (836) SOE Migrant (118) Other Non-migrant (161) Other Migrant (182)

Mean (s.d.) 8.146 (11.475) 2.471 ( 3.015) 11.108 (14.865) 5.082 ( 5.735) 7.449 ( 4.720) 4.811 ( 4.360) 7.571 ( 4.719)

*: total number of observations = 2,861 +: in thousands of yuans per year, before the normalisation and averaging. %: the year of starting the present employment minus 1962.

All the variables except Year, Income and Schooling are dummy variables. The proportions of the observations with value 1 for the dummy variables show that most of the workers in the sample are male, married, not belonging to an ethnic minority, not a member of the Communist Party, and regard themselves as healthy. On average, a worker had studied for 8 years, and started the current work in 1994/1995 (i.e. had been working in the current job for 7 to 8 years at the time of the survey in 2002). Average incomes for different employment choices show that migrant workers earn much more money than non-migrant workers regardless of which type of ownership they work for. In particular, a typical migrant worker in a privately-owned enterprise earns about four and half times what a non-migrant worker earns in the same type of company. For migrant workers, a privately-owned company is the best choice in terms of 8

average income they can earn, but it is the worst choice for non-migrant workers. When migrant and non-migrant workers are combined, the average income for privately-owned enterprises (10.247) is the highest, followed by other types (6.275) and then SOE (5.375). Excluding the observations with missing values and those in the regions where the classification of alternatives is incomplete or inconsistent leaves the final data set with 2,861 cases. We employ a discrete-choice model for the analysis of the effects of various characteristics and attributes on the choice between these six alternatives. With the present data set, the structure that meets all the requirements and yields the most significant estimates is a two-level tree of which the partition is based on the type of ownership, namely, partitioning as Private (Non-migrants, Migrants), SOE (Nonmigrants, Migrants), and Other (Non-migrants, Migrants). The model does not necessarily imply that the decisions are made sequentially, as noted by Hensher et al. (2005), but the partition is plausible because the choices within each group are likely to share common unobservable utility elements. For the present analysis, the vector of individual characteristics, Xi, includes gender (=1 if male), marital status (Married), ethnic minority, membership of the Communist Party (PCC), and year (the year of starting the present employment minus 1962); while the vector of alternative attributes, Zij, includes income (average for each region-and-alternative combination, expressed as a difference from the average income of private-non-migrants for each region in thousands of Yuans per year). All workers within a region face the same set of average incomes for six employment choices. For the upper-level utility functions, Yi includes choice-specific constants, schooling (number of years), and the dummy variable for health. The partition of the 9

variables into the two levels was dictated by the requirements for the IV coefficients, and the significance and plausibility of the estimates. A caveat for the present study worthy of mention, however, is that in the case when rural and migrant workers have significantly different utility patterns, and as a result, the distributional characteristics of the random utility parts, other than the variance, are different from each other, the results obtained from the above model could be biased. This is the case because random utility maximisation results in a nested logit model only if all individual workers’ random utilities are distributed as the same type of distribution, that is, an extreme-value distribution with different variances across branches.5 With this important limitation in mind we proceed to discuss the results. 5. Results The nested model has been estimated using the full-information maximum likelihood (FIML) method with the private-non-migrant group as the base choice. The estimation results are reported in Table 2.

5

We thank an anonymous referee for raising this point. 10

Table 2: Coefficient Estimates$ Level

Variable Gender Married

Lower Level

Minority PCC Year Income

Private Migrants −0.264 (0.181) 0.055 (0.102) −0.137 (0.098) −0.357 (0.220) −0.004 (0.004) 0.103* (0.059)

SOE NonMigrants migrants 1.430*** 1.417*** (0.304) (0.326) 0.864*** 0.859*** (0.255) (0.292) −0.460*** −0.454* (0.172) (0.260) 1.031*** 1.029*** (0.236) (0.210) 0.014** 0.014* (0.007) (0.008) −0.002 −0.001 (0.052) (0.040)

Other NonMigrants migrants 1.346** −0.029 (0.604) (0.333) 1.273* −0.178 (0.755) (0.328) −0.166 0.525** (0.353) (0.267) 1.487*** −0.195 (0.348) (0.697) 0.002 0.030** (0.014) (0.015) −0.090 0.367*** (0.100) (0.121)

−2.077*** (0.612) −0.024 (0.018) −0.349** (0.146)

−3.833*** (0.819) 0.016 (0.022) −0.565*** (0.192)

0.008 (0.243)

0.983** (0.460)

Constant School Upper Level

Healthy IV ( τ)

0.237* (0.139)

No. of cases

2,861

Log L Overall Significance

−3606.729 χ2(39) = 3038.99 (p-value = 0.00000)

McFadden R2

0.296

$: Standard errors are in parentheses ***: Significant at 1%, **: significant at 5%, *: significant at 10%

About half of the coefficients are significant at least at the 10% significance level. McFadden’s pseudo R2 shows that the model performs better than the model with intercepts only by 29.6% in terms of the log likelihood values, and this difference is highly statistically significant according to the likelihood-ratio (LR) statistic for the overall significance test. Some of the coefficients for income are negative, which is against the a priori expectation that a higher income from an alternative would lead to a higher probability of choosing it. However, all three coefficients with the wrong sign are 11

statistically insignificant while those with the correct sign are significant. The LR test on the null hypothesis that all three IV coefficients equal to 1.0 strongly rejects the null hypothesis, implying that the full IIA assumption underlying the MNL model is invalid. 6 This result in turn implies that random errors of the utility functions for migration and employer choice are correlated and hence the two decisions are related with each other. The upper panel of Table 3 shows that the predicted probabilities are similar to the actual proportions.7 The marginal effects, or the discrete changes in the probabilities in the case of dummy variables, are averaged over observations and reported in the lower panel of the same table.

6

The value of the χ2(3) statistic is 30.915 with the p-value lower than 0.00000. This is not surprising because the predicted probabilities are the averages over observations. If the MNL model were used the average predicted probabilities would be exactly the same as the actual proportions. 7

12

Table 3: Probabilities and Marginal Effects$# Choice Variable Actual Numbers (%) Predicted Probabilities: Conditional Marginal Joint Marginal Effects: Gender Marriage Minority PCC School Year Healthy Income 2 Income 3 Income 4 Income 5 Income 6

Private

SOE

Other

Nonmigrants

Migrants

Nonmigrants

Migrants

Nonmigrants

Migrants

156 (5.45%)

1408 (49.21%)

836 (29.22%)

118 (4.12%)

161 (5.63%)

182 (6.36%)

87.72%

84.55%

15.45%

46.75%

53.25%

12.28%

54.67% 5.80%

0.0243 −0.0344 0.0393 0.0318 0.0004 0.0003 0.0112 0.0000 0.0003 −0.0191 0.0000 −0.0013

48.87%

−0.2991 −0.1034 −0.0166 −0.3118 0.0026 −0.0045 0.0839 0.0002 0.0021 0.0398 0.0000 −0.0120

33.35% 29.31%

0.2531 0.1360 −0.0996 0.2069 −0.0050 0.0020 −0.0396 −0.0074 0.0018 −0.0130 0.0055 −0.0060

11.99% 4.04%

−0.0034 0.0019 0.0187 0.0182 −0.0007 0.0009 −0.0062 0.0071 0.0002 −0.0020 −0.0055 −0.0009

5.70%

6.29%

0.0398 0.0353 0.0015 0.0840 0.0013 −0.0002 −0.0221 0.0000 −0.0048 −0.0024 0.0000 −0.0013

−0.0147 −0.0353 0.0568 −0.0290 0.0013 0.0016 −0.0273 0.0000 0.0003 −0.0034 0.0000 0.0215

$: Averages over observations. #: Difference in the predicted probabilities for a dummy variable.

A male rural worker is less likely to migrate and more likely to work for an SOE than a female worker with similar characteristics. On the other hand, a female rural worker is much more likely to migrate and work in a privately-owned work unit than a male worker. One possible interpretation for this outcome is associated with the timing of migration, whereby young single males move first in search of better employment opportunities as economic reform takes place (mostly throughout the 1980s and 1990s), followed by females, partners and singles, at a later stage. As our results arise

13

from a cross-section measured in 2002, well after economic reforms had started to change China’s economic landscape, this explanation seems plausible. A married worker is about 14% more likely to work for an SOE and that much less likely to work in a privately-owned unit than a worker who is not married. This result supports the idea that attachment to SOE also reflects non-pay incentives, such as the provision of medical and unemployment insurance and pensions, which may appeal especially to some groups of the workforce. Marital status, however, does not have much impact on the decision to migrate. Having an ethnic-minority background increases the probability of migrating or of being employed in another type of ownership, but only slightly, in line with the prior that minorities may face limited employment opportunities in their places of residence. A communist party (PCC) member is much more likely to work for an SOE and much less likely to migrate than a non-member, as SOE employment is strictly associated with PCC membership. As the number of years of schooling increases, the probability of working for an SOE decreases, reflecting a wider choice of enterprises’ ownerships in China and a stronger emphasis on rewarding human capital rather than seniority in privatelyowned and foreign-capital firms. The trend since 1962 is that the probability of working for an SOE increases by 0.3% per year and the probability of migrating decreases by 0.2% per year on average. As one would expect, healthy workers are more likely to migrate and to work in a privately-owned unit than are unhealthy workers. As noted above, the signs of the coefficients for income are incorrect in the utility functions for SOE-Non-migrant, SOE-Migrant, and Other-Non-migrant. Hence, interpreting the marginal effects of an increase in the average income from those choices would be pointless. Interpreting 14

those with the correct sign, an increase in the average income from Private-Migrant by 1,000 Yuans increases the probability of choosing that category by 4% and decreases all the other probabilities. Similarly, an increase in the average income from Other-Migrant by 1,000 Yuans increases the probability of the alternative being chosen by 2.2% while decreasing all the other probabilities. 6. Conclusions The results support the hypothesis that the employment preferences of rural migrant workers across different ownership enterprises after the economic reform in China are related to unobservables entering the decision to migrate. As a result, the decision to migrate and which type of firm to work for are components of a broader aim to improve one’s economic circumstances. This outcome has implications for both researchers and policy makers. For researchers, it highlights that separate analyses of migration and employment choices could result in biased estimates and hence misleading information. For policy makers, it suggests that reforms in the labour market, for example by enabling private and foreign entrepreneurs to set up businesses, ought not to be carried out independently from reforms affecting workers mobility. The danger of independent reforms is the emergence of subgroups of the working population that respond to economic incentives without necessarily having well-defined legal status and rights, with opportunities for exploitation and illegal activities, some possibly related to survival. In the case of China, the expansion of labour demand following the economic reforms of 1978 have been met by large flows of rural-urban immigrants that have often moved to cities without the ability to fully participate to the urban labour market (urban hukou), contributing to a segmented labour market and persistent inequality

15

References Brooks, R. and Tao, R. (2003). China's labor market performance and challenges. IMF Working Paper WP/03/210. Washington, DC: International Monetary Fund. Cai, F. (2001). Institutional barriers in two processes of rural labor migration in China. Working Paper Series No. 9. Beijing: Institute of Population Studies, Chinese Academy of Social Sciences. Chan, K. W. and Zhang, L. (1999). “The hukou system and rural-urban migration in China: Processes and changes.” The China Quarterly, 160: 818-855. Chen, G., and S. Hamori, (2009). “Solution to the dilemma of the migrant labor shortage and the rural labor surplus in China”. China and World Economy, 17(4): 53−71 Démurger, S., Gurgand, M., Li, S. and Yue, X. (2009). “Migrants as second-class workers in urban China? A decomposition analysis.” Journal of Comparative Economics, 37: 610-628. Fan, C. C. (2001). “Migration and labor-market returns in urban China: Results from a recent survey in Guangzhou.” Environment and Planning A, 33: 479-508. Gagnon, J., Xenogiani, T. and Xing, C. (2009). Are all migrants really worse off in urban labour markets? New Empirical Evidence from China. IARIW-SAIM Conference Working paper 278. Goodkind, D. and West, L. A. (2002). “China's floating population: Definitions, data and recent findings.” Urban Studies, 39: 2237-2250. Gu, E. (2001). “Beyond the property rights approach: Welfare policy and the reform of state-owned enterprises in China.” Development and Change, 32: 129-150. Hensher, D.A., and W.H. Greene, (2002), “Specification and estimation of the nested logit model: alternative normalisations”. Transportation Research Part B, 36: 1−17 Hensher, D.A., J.M. Rose, and W.H. Greene, (2005). Applied Choice Analysis: A Primer, New York, Cambridge University Press Lianos, T., and A. Pseiridis, (2009). “On the occupational choices of return migrants”. Entrepreneurship and Regional Development, 21(2): 155−181 Lin, J. Y. (1988). “The household responsibility system in China's agricultural reform: A theoretical and empirical study.” Economic Development and Cultural Change, 36: S199-S224. Lin, J. Y., Cai, F. and Li, Z. (1998). “Competition, policy burdens, and state-owned enterprise reform.” American Economic Review, 88: 422-427.

16

Meng, X. (2000). Labour market reform in China, Cambridge: Cambridge University Press. Roberts, K. D. (2001). “The determinants of job choice by rural labor migrants in Shanghai.” China Economic Review, 12: 15-39. Seeborg, M., Jin, Z. and Zhu, Y. (2000). “The new rural-urban labor mobility in China: Causes and implications.” Journal of Socio-Economics, 29: 39-56. Terza, J.V. (2009). “Parametric nonlinear regression with endogenous switching.” Econometric Reviews, 28(6): 555–580 Wang, M. (2005). “Employment opportunities and wage gaps in the urban labor market: A study of the employment and wages of migrant laborers.” Journal of China Social Science, 5: 36-46 (in Chinese). Wang, H. and Li, S. (1995). Industrialization and Economic Reform in China, Beijing: New World Press. Wei, L. and Han, C. (2006). The research report of migrant workers in China, Project Team of Research Office of the State Council, Beijing, Chinese Yanshi Press (in Chinese). Yang, C. and Yang, L. (2009). Investigation on the movement of contemporary Chinese migrant workers movement. Chinese Sociology Network (in Chinese). www.sociology.cass.cn. Zhao, Z. (2004). “Rural-urban migration in China-What do we know and what do we need to know?” China Economic Quarterly, 3: 517-536.

17