Towards a Competitive Labour Market? Urban Workers, Rural Migrants, Redundancies and Hardships in China May 2001 Simon Appleton, John Knight, Lina Song and Qingjie Xia1 Corresponding to :
Dr. Lina Song Institute of Contemporary China Studies Economics and Geography Building University of Nottingham, NG7 2RD United Kingdom Tel: 44 115 846 6217 Fax: 44 115 846 6324 Email: [email protected]
Abstract This paper examines the traumatic effects on the urban labour market of the policy of large-scale redundancies intended to reform the state sector. It enables us to pose the question: is a competitive labour market emerging in urban China? This question addressed by means of a 1999 urban survey of 4500 households covering 13 cities in 6 provinces. Novel features of the data set are that it contains both urban residents and rural migrants of details of the labour market histories of workers who had left their employment in the previous eight years. It is therefore possible to examine the wage determination process for three groups: urban workers who have never been retrenched, retrenched urban workers and migrant workers. Which workers lost their jobs? Probit estimates are made to examine the determinants of redundancy. Which retrenched workers were re-employed? Again, probit functions are estimated to explain this process, as well as semi-parametric models of the duration of unemployment. This leads on to the estimation of earnings functions for the three groups of workers (both with and without correction for relevant selectivity). There are interesting differences in the earnings functions and in simulated mean earnings (the predicted earnings of each group if subject to the wage determination process of the others). The results can be interpreted as indicating that the traditional discrimination between urban residents and rural migrants has been much eroded, but that the retrenched urban workers have been flung onto an unwelcoming market. The majority remaining unemployed and those who are re-employed are in poorly-paid jobs which do not reward their productive characteristics. In the short run, at least, they appear to be the victims of imperfect information and constrained choice. JEL Key Words: China; Labour market; Redundancy; Re-employment; Wages.
Simon Appleton, Lina Song and Qingjie Xia are at the University of Nottingham, and John Knight is at the University of Oxford. The authors are grateful to the Department for International Development, United Kingdom, for their funding of the project (Escor grant R7526) and the Ford Foundation for their support of data collection.
The Chinese urban radical reform known as xia gang, first on trial in 1994 and finally launched in 1997, was intended to resolve the problem of inefficiency of the state sector by laying-off a quarter or more of its workers within four years (1997-2000). By the end of 1999, the official figure of the accumulated laid-off workers was 25.1million. At that date those who remained xia gang numbered 9.4 million. To this must be added the registered unemployed (5.4 million) to obtain the true measure of urban unemployment (then 9.7 per cent of the urban labour force). The sharp fall in urban employment by 27.7 million over the period of xia gang policy (from 148.5 million at the end of 1996 to 119.0 million in mid-2000) implies that the accumulated number of layoffs and the true unemployment rate will have continued to rise2. These developments have ended the ‘win-win’ phase of economic reform, and have brought it into a phase in which there are clear losers, at least in the short term. Knight and Song (1999a, 1999b), analysing a survey of workers in urban enterprises relating to 1995, found sharp segmentation between urban residents and rural migrants. This arose from the political and institutional arrangements which gave urban residents privileged access to secure employment at above market-clearing wages and which controlled the flow of peasants to the cities, allowing rural migrants to fill only the jobs that urban-dwellers did not want. Production function analysis showed that the wage of urban residents exceeded their marginal product, implying the existence of surplus labour in the enterprises, whereas the wage of migrants fell short of their marginal product, reflecting government restrictions on their recruitment. This snapshot of 1995 notwithstanding, our principal hypothesis is that the urban labour market has become much more competitive in the ensuing five years. The impetus has come from the policy of state-owned enterprise (SOE) reform and the associated redundancy policy. The nonstate sector, which is free to determine wages, has grown relative to the state sector, and even the SOEs have greater freedom to set their own wages than five years ago. Thus the rise in unemployment could have influenced the real wages even of workers who have retained their jobs. Redundant urban workers and urban labour market entrants faced much tougher labour market conditions than in the era of guaranteed employment, being thrown into competition not only among themselves but also with rural-urban migrants, some of whom were laying down roots and trying to settle in the urban areas. It is plausible, therefore, that the shock of SOE reform has unleashed competitive market forces. This competition should have narrowed the wage differences among equally productive workers. Our main purpose is to test whether the “law of one price” now holds in the Chinese urban labour market. Dong and Bowles (2000) found no significant differences among firms of different ownership in their returns to human capital. This is consistent with a degree of competition in the urban labour market across ownership categories. Our paper provides a similar test of competitiveness but across categories of labour market rather than across categories of firm. We wish to examine whether segmentation remains. Are there wage differences among workers who share the same characteristics? Are rural-urban migrants still at a wage disadvantage by comparison with urban residents, i.e. is there an unexplained wage difference between these two groups? Do the wages of those urban residents who retained their jobs exceed the wages of those who were made redundant and had to find new jobs? Have the laid-off workers been forced into 2
Derived from the National Bureau of Statistics, China Statistical Yearbook 2000, China Labour Statistical Yearbook 2000, and China Monthly Economic Indicators 2000.7. 2
competition with migrants, i.e. is there an unexplained wage difference between these two groups? The previously rigid and segmented nature of the urban labour system caused inefficiency in the allocation and utilisation of labour (Knight and Song 1991, 1994). The movement towards a competitive labour market should improve efficiency. We wish to examine some of the mechanisms of, and impediments to, efficiency gains. We ask, therefore: what kinds of worker are more likely to be laid-off as a result of the SOE reform policies? The SOE wage structure is generally tightly compressed: the productivity of individual workers is not fully rewarded – at least until recently. If wages do not correspond to productivity, then enterprises that lay-off less productive workers are likely to become more efficient. Despite the high female labour force participation rate, Chinese culture is very male-dominated. When firms are required to make workers redundant, are women more likely to be laid-off? Gender discrimination of this sort need serve the ends neither of efficiency nor of equity. The retrenchment of surplus labour in enterprises and its subsequent re-employment elsewhere should produce long run efficiency gains for the economy. There can, however, be short term costs, both to the retrenched workers and to the economy. How serious they are, and how long they continue, will depend on the flexibility of the labour market and on the growth of the economy in relation to the growth of the labour force. We wish to explore the costs of redundancy to retrenched workers. We estimate the costs as the difference between the income they could have received if they had retained their jobs and their current income. This cost can be decomposed into the loss of income during the period of unemployment and any fall in real wages after re-employment. This paper uses nationally representative survey data from the year 2000 to focus on three subgroups of the urban labour force. The first category, which we term ‘never retrenched’ is those with urban residential status (hukou) who have never been made redundant or have never lost their jobs. The second category is those with urban hukou that have been retrenched, whether currently unemployed or re-employed. The third category is the rural-urban migrants. They are usually excluded from the official counts of the Chinese urban labour force because they lack urban hukou status, although they live and work in the urban sector and they try to compete for urban jobs. Section 2 explains the data set used and the econometric methods applied to analyse it. Section 3 presents earnings functions for the four groups of workers, as well as estimates of the determinants of the probability of an urban worker being retrenched and of a retrenched worker being re-employed. Section 4 uses the earnings functions to perform two types of simulation. One, simulating the earnings of the retrenched workers had they kept their jobs, provides an estimate of the loss of earnings of retrenched workers. The other, simulating the earnings of urban residents and migrants under different earnings functions, considers whether there is evidence of discrimination against migrants.
To date, a major drawback in the analysis of the Chinese urban labour market has been the lack of information on the entire labour force. Most datasets do not include both urban residents and rural-urban migrants. This is because the “urban labour force" is officially defined to include only those who have urban residential status. Although working and living in the urban sector and competing for urban jobs, rural-urban migrants are not officially regarded as part of the urban labour force. We are fortunate in having survey data that include rural-urban migrants as well as urban residents in the labour force, whether working or unemployed. The survey contains three elements. One element (4000 households) is a nationally representative urban household survey that covers thirteen cities in six provinces. The second element (500 households) is an enlarged survey of households with laid-off workers in the same cities, designed to obtain more observations and more detailed information about this particular group3. The third element is a household survey of rural-migrants registered with residential neighbourhood committees in the thirteen cities. Not all rural-migrants are officially registered, so it is likely that our sample overrepresents the more settled migrants. The data (referred to as "1999 CASS Household Survey") are cross-sectional, with the questionnaires gathering information mainly about 19994. Taking these three elements together yields a sample of 4500 households with urban hukou and 800 rural-urban migrant households in 6 provinces. For the purpose of the econometric analysis, we classify the workforce into four categories: (1) Non-retrenched urban workers: residents with urban hukou who are both currently employed and have not been laid-off since 19925, (2) Retrenched urban workers: urban residents with hukou who have been laid-off since 1992, (3) Re-employed urban workers: those retrenched workers who have been laid-off since 1992 but were employed when this survey was conducted. Note that (3) is a subset of (4) (4) Rural-urban migrants: those working in the city without urban hukou - this group is usually excluded in surveys used by the government and by most researchers6. Of the 7994 workers in the labour force in our sample, 5770 are non-retrenched urban workers, 1159 are retrenched urban workers – 726 of this group still remaining unemployed and 433 being re-employed - and 1065 rural-urban migrants. 3
Because our sample over-represents households with laid-off workers, we assign individuals from such households a weight of 0.59. This weighting gives the same proportion of households with laid-off workers as in the representative survey. We use this weighting when reporting mean sample statistics but do not when estimating the econometric models. 4 The survey was designed by an international team that included the authors. Data collection took place at the start of 2000 by a joint team from the Department of Economics, University of Oxford, Chinese Academy of Social Sciences and the National Statistics Bureau of PR China. 5 This year was chosen because it marked the appointment of a new prime minister and the beginnings of significant moves to reform the SOE sector. Of the 1364 workers in the survey who had ever been laid-off, 96.1% had been laid off in 1992 or after. Twenty-five were laid-off in 1992, compared to nine in the year before and 55 in 1993. The tenyear SOE reform programme began officially in the 1994, when 127 workers in our sample were laid-off. 1998 saw the largest number of redundancies in our sample, with 305 workers losing their jobs. 6 Note that, although the migrants in our sample are working in urban areas, we reserve the term urban workers for non-migrants, ie those with urban hukou. 4
Table 1 provides a description of the different groups. Non-retrenched urban workers are the most educated group, with an average of 11.4 years of education compared with 9.9 years for the retrenched workers and 8.2 years for the migrants. The retrenched workers are oldest, averaging 23.5 years of potential work experience compared to 21.7 years for the non-retrenched workers and only 18.4 years for the migrants7. Women are more likely to have been retrenched - more women are found in the retrenched group (63%) than in the non-retrenched urban worker group (44%) and migrant group (41%). Retrenched workers also include fewer Communist Party members (proportionately, a third as many as are in the non-retrenched group). Even fewer migrants are Party members. Retrenched urban workers are more likely to have worked in urban collectives, and less likely to have been employed in centrally controlled SOEs than nonretrenched workers. They are also more concentrated in lower paid occupations, with a greater proportion of industrial, commercial and service workers. Migrants are less dependent on SOEs for employment – over half report being self-employed (the “urban individual” ownership classification). 2.2
Our main focus is on the determinants of earnings for the different categories of worker. We estimate semi-logarithmic functions for earnings, W, of each worker i in category k: (1)
LnWi = βkXi + εi
where X denotes a vector of explanatory variables, β a vector of associated coefficients, and ε is a stochastic term. The functions are estimated separately for the different categories of worker. The dependent variable is the log of earnings per eight-hour period of work. For brevity, we describe this as earnings per day, although in fact migrants worked more than eight hours each day. The migrants in our sample work much longer hours than is typical for urban workers, averaging about 60 hours per week. For non-retrenched urban workers, re-employed urban workers and migrants, we model actual earnings. However, we also include a “historic” earnings function where the dependent variable is the earnings in their previous jobs of retrenched urban workers. Historic earnings are revalued so as to be in 1999 prices. Comparing mean earnings, those for the non-retrenched urban workers are, unsurprisingly, the highest. However, it is notable that, on average, migrants earn more than re-employed urban workers and that re-employed workers earn more than retrenched workers did in their previous jobs. We follow a Mincerian specification with years of schooling and a quadratic for experience as explanatory variables of the log of earnings. Experience is defined as potential experience – years since completing full-time education. We include four dummy variables for personal characteristics: female sex, coming from a minority ethnic group, membership of the Communist Party, and reporting oneself to be in “bad health” (as opposed to the other possible response, being in “good health”). We also include dummy variables for the city of residence. This list of explanatory variables defines the most restrictive specification we estimate. We also include a fuller specification, with controls for sector (defined by ownership of employing institution) and for occupation. The more restrictive specification is arguably more useful for determining the 7
We define work experience as the number of years since a person finished full-time education. 5
full effect of human capital variables, since human capital will affect the occupations open to a worker and may also influence sector of employment.
Correcting for sample selectivity
Both the restrictive and full specification earnings functions are estimated by Ordinary Least Squares. However, we also concerned about the possible selectivity of our sub-samples. The issue is perhaps most acute in the case of migrants – these are clearing a non-random sample of those originating from rural areas. They are likely to more ambitious, hardworking and risktaking members of the rural labour force. Unfortunately, we cannot address this issue since we lack matching data on rural residents who did not migrate. Migrants may also differ from urban residents in many unobservable respects but these differences are not susceptible to the standard econometric correction since the process separating migrants and urban residents is not an endogenous economic process but rather an accident of birth. We find it more natural to regard migrants and urban residents as coming from different populations. We do note the caveat that earnings functions estimated for one population may not be meaningful for the other. By contrast, we can address the possible selectivity of the three different categories of urban resident. We model whether an individual has ever been retrenched (Ui =1) using a probit: (2)
U*i = αXi + áSSi + Vi Ui =1 if U*i>0 =0 otherwise
where Vi~N(0,1) Unlike most probits used to correct for the selectivity of earnings, this probit cannot be interpreted as reflecting a choice process by the individual worker. Instead, it captures whatever processes underlay the decision to retrench certain workers. One consideration is clearly the cost and productivity of the workers. Worker cost depends partly on their earnings, which may also mirror their productivity. Consequently, the potential determinants of retrenchment probabilities include the same variables, Xi, as those included in the full earnings functions (we only apply the selectivity correction to the full specification of the earnings function). However, what is important for such correction is to have an identifying instrument, Si: one that affects the retrenchment decision but does not affect the earnings of the worker. Official policy during retrenchment has been to take account of whether a worker has a fellow household member employed in the state sector and not to sack both of them. If this policy were implemented, one would expect that an SOE worker would have less chance of being retrenched from an SOE if their spouse was retrenched from an SOE. Consequently, we define Si as one if the individual was employed in an SOE and is married to a spouse who was employed in an SOE but was retrenched8. We expect this variable to have a negative coefficient (reduce the probability of retrenchment) in the probit but to have no direct effect on earnings. Unfortunately, we cannot identify a priori another variable that satisfies this exclusion restriction and so are unable to apply a Sargan test for the validity of the restriction in this case. When correcting for the possible selectivity of the historic earnings functions for retrenched workers, we apply the standard Heckman two-stage estimator. Since: 8
Specifically, the variable is 1 if a) the individual works in an SOE or worked in an SOE (and was retrenched since 1992); and b) has a spouse who worked in an SOE and was retrenched since 1992. 6
E(LnWi | Ui =1) = βkXi + E(εi | Ui =1) = βkXi + ñó å ö(αXi + áSSi) /Ô(αXi + áSSi)
where ñ is the correlation coefficient between εi and Vi and óå is the standard deviation of εI, we can obtain consistent estimators of βk by augmenting equation (1) with a correction for selectivity, ëi, defined by: (4)
ëi = ö(aXi + aSSi) /Ô(aXi + aSSi)
if Ui =1
where the parameters a are the estimates of á. The coefficient on this selectivity correction, βë , is an estimate of the covariance between εi and Vi. A similar correction can be used for the earnings function of the non-retrenched urban workers, viz: (4’)
ëi = -ö(aXi + aSSi) /[1-Ô(aXi + aSSi)] if Ui =0
For the re-employed workers, there is a double selectivity. To be re-employed, not only must the workers have been retrenched but they must also have found another job. We model the probability of re-employment among the sub-sample of the retrenched as another independent probit: (5)
R*i = ãXi + ãII i + çi Ri =1 if R*i>0 =0 otherwise
where çi~N(0,1) This probability will partly depend on the labour supply decisions of workers but also on the demand for particular kinds of workers. Wages will influence both the supply of and demand for labour. Since it is hard to identify an instrument that will affect wages and not participation directly, we adopt a reduced form approach. All the explanatory variables, X, included in the full earnings function are also included in the probit for re-employment. As identifying instruments, Ii, we include variables for the demographic composition of the household. These may affect labour supply decisions, particularly of women. In particular, where there are a number of young children in the household, one might expect female labour supply to be reduced. The presence of older household members – possibly retired – may offset this effect. To capture such interactions, we explored various specifications in preliminary regressions. From this probit for re-employment, we obtain a second correction for selectivity: (6)
ë2i = ö(gXi + gIIi) /Ô(gXi + gIIi)
where g are the estimates of the coefficients ã. We can obtain consistent estimates of the earnings function for the re-employed by using Ordinary Least Squares to estimate:
LnWi = βRXi + βë1 ëi + βë2 ë2i
Note that this correction assumes that the probabilities of retrenchment and re-employment are independent (see Maddala, 1983, p282). We tested this restriction by estimating a more general model, a bivariate probit, which allows for a correlation between the two error terms, çi and Ui. A bivariate probit was estimated in LIMDEP using the option for estimation with sample selectivity, the sample selectivity being that re-employment is not observed for those never retrenched. However, the estimate of the covariance between çi and Ui was not statistically significant at the 5% level, so we could not reject independence. Given that the corrections for the sample selectivity of earnings are much simpler under independence, we imposed this restriction.
The probability of retrenchment
We model the probability that an urban resident in the labour force has been retrenched since the start of 1992. We exclude those who have never been employed, such as new entrants to the labour market looking for work. After weighting, from our sample of 6929 urban workers, 1159 people had been retrenched, this implies a proportion of 15.9%. We do not attempt to model unemployment among migrants. Only 31 (2.7%) of the 1180 in our sample were unemployed at the time of the survey, too few to sustain econometric investigation. The probit results are given in Table 2.1. The model clearly identifies a number of determinants of the probability of retrenchment – most hypothesised explanatory variables are statistically significant. The goodness of fit can be measured by either the Likelihood Ratio, 0.18, or the percentage of correct predictions, 84%. Since most explanatory variables are dummy variables, one way of showing their quantitative importance is by reporting the predicted probabilities of the model evaluated at the (weighted) means of the other explanatory variables (Table 2.2). Due to the non-linearity of the probit model and the fact that it was not estimated with weights, the base probability of retrenchment – at the mean of the explanatory variables - is 23.5%, higher than the weighted mean proportion being retrenched. Consider first, the controls for ownership. The default category is employment in local government owned state enterprises. Such workers faced above-average risks of being retrenched, estimated at 27% at the mean of the explanatory variables. Workers in urban collectives faced an even greater risk, calculated at 35%. The lowest risk of being retrenched was faced by workers employed by the central government, including those employed in centrally run SOEs. The probability of such workers being retrenched, at the mean of the other explanatory variables, was 17%. Similarly low risks were faced by workers in micro-enterprises (less than eight employees); in enterprises listed on the stock market; and in enterprises benefiting from foreign investment. Within enterprises, retrenchment risks were substantially higher for certain categories of occupation. The default category of occupation was technical and professional. This category and the administrative and clerical categories all face low probabilities of retrenchment, estimated at around 20% when evaluated at the means of the other explanatory variables. By contrast, the retrenchment risks faced by other workers were in the range 26-30%.
Now consider the personal characteristics of the workers. Men were significantly less likely to be retrenched, with a 21% risk evaluated at the mean of the other explanatory variables compared to 26% for the female workers. Education reduced the risk of being retrenched. The estimated marginal effect of each year of education on the probability of being retrenched was 0.01, so that ten years of education would reduce the risk by ten percentage points. The risk of being retrenched had an inverse-U relationship with potential experience (defined as years since education was completed), peaking at 26 years of experience. The mean years of experience was 22, so the probit results imply that both younger and older workers faced lower risks of retrenchment. Being in bad health increased the probability by around five percentage points. Membership of the Communist Party reduced the risk by six points. A dummy for belonging to a minority ethnic group had an insignificant effect. One variable of particular interest to us is the dummy variable for whether the individual had a spouse who had been retrenched from the state sector. This variable was considered a priori to be an appropriate instrument for identifying the correction for the selectivity of retrenchment. Since the government has encouraged employers to ensure that married couples were not both laid-off, we expected an individual to be less vulnerable to dismissal from the state sector if their spouse had been retrenched from the state sector. As predicted, the variable has a negative and significant effect, consistent with implementation of the policy. At the mean of the other explanatory variables, the effect is non-negligible at around three percentage points.
The probability of re-employment
From the sub-sample of 1159 urban residents who have been retrenched, we modelled the probability of their having being re-employed. We estimate only 37% of retrenched workers had found re-employment by the time of the survey, consistent with most of the instances of retrenchment being relatively recent. Unlike the probit for the probability of retrenchment, most hypothesised explanatory variables in the probit for re-employment were insignificant. This includes all the controls for ownership and occupation which were excluded from the final model (in Table 3) on the basis of a likelihood ratio test. The model explains less than the model of retrenchment: the percentage of correct predictions is 69% and the likelihood ratio statistic 0.08. Variables that determined earnings often did not seem to affect the probability of re-employment. Education, experience and party membership all had no statistically significant effect on the probability of re-employment. The exceptions – aside from the city dummies – were the dummies for sex and for being in bad health. Table 3.2 reports the effects of various explanatory variables upon the probability of re-employment predicted at the means of the other explanatory variables. Men had a 44% chance of being re-employed compared to 39% for women. Individuals in bad health had only a 33% chance of re-employment compared to 42% for those in good health. Three variables for household demographic composition did influence the probability of re-employment. The number of young children was particularly important but its effect varied with the sex of the individual worker. For men, more young children increased the probability of re-employment; for women, more young children reduced the probability. This gender difference in the effects of young children may reflect a different balance of income and substitution effects on labour supply. Young children have an income effect, increasing child costs and encouraging households to increase their labour supply to meet those costs. However, they also have a substitution effect – reducing labour supply to the extent that child care is incompatible with employment. If women typically take responsibility for young children, this substitution effect may be confined to them. Interestingly, there was a positive interaction
between the number of children in the household, the number of people aged over 59 and the individual being female. We interpret this as elder, possibly retired, household members freeing women from the constraints of child-rearing. Analysing whether retrenched workers were re-employed at the time of the survey or not neglects the temporal aspect of unemployment. How long workers have been unemployed is important in affecting the probability of their re-employment and it is conceivable that neglecting this may bias the estimated effects of our hypothesised determinants of re-employment. To consider this, we also modelled the duration of unemployment for retrenched workers. From examination of the empirical hazard for re-employment, it was clear that there would be gains from not specifying a distribution for the baseline hazard. In particular, the empirical hazard had several peaks and would not be well fitted by a smooth distribution such as a Weibull or lognormal. Table 3.1 includes three semi-parametric models for the duration of unemployment: an ordered probit and the closely related Meyer model, with and without unobserved heterogeneity (Meyer, 1990)9. In our estimates, unobserved heterogeneity was not significant and the three duration models yielded broadly similar results to the simple binary probit for re-employment. Sex, health and household demographics were significant determinants of the duration of unemployment. Perhaps the most notable difference between the duration models and the binary probit concerns education. Education does not affect the binary probit but does reduce the duration of unemployment at around the 10% level of significance in the survival models. 3.3
We estimated earnings functions separately for non-retrenched urban workers, re-employed urban workers and migrants. We also estimated an earnings function for retrenched workers based on historic earnings prior to retrenchment. As noted previously, three types of earnings function were estimated: a full specification including controls for sector and occupation (Table 4); a restricted specification excluding these controls (Table 5); and a full specification augmented with a correction for possible sample selectivity (Table 6). Variable definitions are given in Appendix Table 1. We conducted Wald tests to explore whether any two earnings functions under the restricted specification could be pooled. In each case, the pooling of all the earnings functions was rejected at the 1% level. The four groups of worker do appear to be paid according to different standards. In that sense, we reject the hypothesis that the Chinese urban labour market is perfectly competitive. However, individual variables did often have similar coefficients in different earnings functions. We tested whether two coefficients were significantly different at the 5% level using the Wald test: (8)
(â1 – â2)/(Var(â1) + Var(â2)) ~ ÷2(1)
where â1 and â2 are coefficients on the same explanatory variable estimated in different earnings functions. Overall, the models identify a number of statistically significant determinants of earnings, but their explanatory power is limited. The models with the highest goodness of fit are those for the non-retrenched urban workers in Tables 4 and 6, where the adjusted R-squared is 0.33. The lowest goodness of fit is for the re-employed urban workers, where the adjusted R-squared in Table 5 is 0.135. The earnings functions exhibit heteroscedasticity according to Breusch-Pagan tests and so we use White’s heteroscedastic consistent standard errors. The OLS coefficients are 9
The ordered probit was estimated in LIMDEP; the Meyer models in STATA using the pgmhaz command developed by Stephen Jenkins. 10
inefficient in the presence of heteroscedasticity, but attempts to use estimate multiplicative heteroscedastic models were unsuccessful in most cases10. However, we were able to estimate such models for the non-retrenched urban workers and the migrants. These results are not reported here since the emphasis is on comparability across sub-samples but were not too far from the OLS estimates. We are slightly sceptical of the earnings functions corrected for selectivity reported in Table 6. The correction for the selectivity of re-employment is not significant at the 5% level. This is true whether the correction is derived from a binary probit for re-employment or from an ordered probit for the duration of unemployment. The corrections have positive signs, implying that the unobservables that raise the probability of re-employment are positively correlated with the unobservables that raise earnings. This is perhaps plausible on the grounds that the retrenched workers with productive characteristics observable to the recruiting firm but not to the researchers are more likely to be reemployed and are better paid. More troubling are the controls for the selectivity of retrenchment. The controls are statistically significant determinants of the earnings of non-retrenched and re-employed workers, but not significant for the historic earnings of retrenched workers in their previous jobs. However, the sign of the coefficients on the controls imply positive correlations between the unobservables that determine retrenchment and those that determine current earnings. This finding seems perverse, at least in so far as observables positively associated with the probability of retrenchment (low education, unskilled occupations, female sex, etc) are negatively associated with wages. Since we are unable to provide a convincing explanation for the perverse selectivity finding, we place more weight on the OLS results in what follows11. Turning to the observed explanatory variables, we consider first the returns to human capital as measured in the earnings functions. We have three variables that may proxy aspects of human capital: education, experience and health. In the restricted specification (Table 5), we find that the Mincerian returns to education were 6% for both non-retrenched urban workers and migrants. Once controls for occupation and sector are included (Table 4), the returns to education fall substantially, consistent with the hypothesis that some of the return to education comes via allocation to higher paying occupations and/or sectors. For the re-employed urban workers, education is never a statistically significant determinant of earnings. The historic earnings functions suggest that those retrenched did receive a return to education in their previous jobs. However, the return to their education was less than that currently received by non-retrenched workers – just over half as much (a 3.6% return) in the restricted specification. Moreover, unlike the returns to education estimated in other earnings functions, this effect is not robust to controls for selectivity. Education has no return in the historic earnings function in Table 6. Part of the explanation may be that the retrenched and the re-employed tend to be more concentrated in occupations where education may be less rewarded. Over a half are industrial workers compared to a third of non-retrenched workers. When we interacted years of schooling with occupational dummies for non-retrenched workers (the largest sample), all interactions except that with 10
The models used the HREG command of the LIMDEP econometric software, modelling the error variance as the exponent of the vector of explanatory variables and associated coefficients. For the re-employed and retrenched, estimation was aborted with the message that the variance was vanishing or exploding. 11 One explanation is that retrenchment allowed SOEs to dismiss workers who were in some sense being “overpaid”. Being “overpaid” might manifest itself in having higher earnings than are predicted from observables. However, this explanation is speculative. There seem no strong grounds for supposing that only observables and not unobservables in an earnings function are related to productivity. Moreover, the relationship between the unobservables determining retrenchment and earnings exists only for the contemporaneous earnings functions, not for the historic one. This implies that retrenched workers are currently “expensive”, given their observable characteristics, but were not expensive historically. 11
commercial occupations were negative, significantly so in the cases of industrial workers and administrative workers. This implies that education is more rewarded for professional and technical workers. An alternative or supplementary explanation is that the retrenched and the reemployed are workers who gained few useful skills from their education. However, this is speculative, since we have no data on educational qualifications or other indicators of human capital acquired from schooling. Current earnings vary with experience in the conventional inverse U-shaped pattern. In the restricted earnings functions (Table 5), the turning points come when the workers have around thirty years of potential experience. Experience has a monotonic positive effect on the earnings of the retrenched and is significant if the non-significant squared term is dropped from the model. To estimate the return to a year of experience, we evaluate it for a hypothetical worker with 20 years of potential experience, roughly the mean for the sample. For urban workers (never retrenched or re-employed), an extra year of potential experience raises earnings by around 1.2% to 1.3%. This is twice as large as the effect of age on migrant earnings but less than the result (1.6%) from the historic earnings functions for the retrenched. The final variable that may capture human capital is the dummy variable for workers who reported being in bad health. Self-reported data on health is probably subject to systematic biases compared to objective clinical assessment. However, some reviews have found it to be as good a predictor of mortality as more objective measures (Idler, 1991; Idler and Kasl, 1991). People who report themselves to be in bad health tend to earn less than others, ceteris paribus. However, this effect is only significant at the 5% level for non-retrenched urban workers. Among the other personal characteristics, that for sex is consistently significant and favourable to men. The quantitative effect of being male varies across the different sub-samples. It is stronger for migrants than for non-retrenched urban workers but interestingly it is weakest in the historic retrenched workers earnings function and strongest in the re-employed earnings function. This suggests that retrenchment may have significantly increased the “pure” gender gap in earnings. Party membership appears to be remunerative only among the non-retrenched urban workers and historically among the retrenched prior to retrenchment. Migrants and the re-employed may have their earnings determined by the market, where political status is unimportant. The size of the benefit falls if controls are made for occupation and sector. Coming from a minority (non-Han) ethnic group has no significant effect in any earnings function. The earnings of urban workers often vary significantly with the ownership of the enterprise in which they work. As shown in Table 1, most urban workers are employed either by central SOEs, by local SOEs or by urban collectives. Of these three categories, centrally controlled SOEs pay the highest wages and urban collectives the lowest. However, retrenched workers who are re-employed by centrally controlled SOEs and migrant workers appear not to enjoy the wage premium otherwise accruing to employees of such enterprises. These results are consistent with centrally controlled SOEs hiring re-employed urban workers and migrants on more competitive terms than those offered in the past or offered currently to non-retrenched. Indeed there are few significant earnings differentials by sector for migrant workers12. 12
Over half of the migrants in the sample are self-employed in urban enterprises. Some of the earnings of such workers may reflect a return on capital, rather than labour, so a comparison with the earnings of wage employees is problematic. However there is no significant difference, ceteris paribus, between the earnings of such self-employed migrants and migrants employed in most other kinds of enterprise.
In general, there are marked occupational differences in earnings, even after controlling for personal characteristics (such as education), sector and city. Ceteris paribus, the highest paid workers are those in administration and the default category of technical and professional workers. Industrial, commercial and service workers generally receive lower pay. For example, among urban workers who have never been retrenched, clerical workers earn around 7% less than professional and technical workers; industrial workers, commercial and service workers earn 20-22% less. Occupational differentials do vary across the four types of workers. In particular, re-employed clerical workers appear to suffer much lower wages than never retrenched clerical workers. Indeed, ceteris paribus, they are paid the same as industrial workers. Curiously, occupational wage differentials are not found in the historic earnings function for the retrenched. The generally negative and significant city dummy variables imply that workers in the default city, Beijing, receive an earnings premium. However, the premium does vary across sub-groups of workers. Pairwise Wald tests revealed that many of the city dummies have significantly different effects in the earnings functions for the non-retrenched urban workers compared to other earnings functions. To summarise, we find that the wage structure differs across the different categories of worker we identify: never retrenched urban workers, re-employed urban workers and migrants. Workers in centrally controlled SOEs, Communist party members, the educated and those in clerical occupations all receive wage premia if never retrenched. However, these premia vanish if the workers are retrenched and take other jobs. In some ways, migrants appear to be an intermediate category – they are rewarded for the education or occupation, as are never retrenched urban workers – but receive no premia for party membership or employment in a centrally run SOE.
Simulations from the earnings functions
Hardship generated from the redundancy
One measure of the hardship induced by retrenchment is to compare the actual incomes of those who have been retrenched with what they would otherwise have earned if never laid off. We focus on income in the form of earnings and unemployment benefits. We assume that income from other sources would not change if retrenched workers had not lost their jobs. Since our earnings functions were estimated in terms of earnings per eight hours of work, for convenience we consider figures in terms of yuan per day, where day refers to an eight-hour period of work. The mean earnings of the re-employed were 24.73 yuan per day. Retrenched workers who had not found re-employment did receive some payments for living expenses from their former work units and/or from local government. These unemployment benefits averaged 100.89 yuan per month or 4.59 yuan per working day. 62.6% of those who had been retrenched were still unemployed at the time of the survey. Consequently, the mean earnings plus unemployment benefits of all workers who had been retrenched were 12.11 yuan per (working) day [0.3736x24.73 + 0.6264x4.59=12.11]. How much would retrenched workers earn if they had never been retrenched? The most obvious counterfactual is that they would have been paid in a like manner to the non-retrenched urban workers who have never been retrenched. We can express this in numerical terms using the earnings function of the non-retrenched urban workers. For this and related exercises, we use the
parsimonious earnings functions (in Table 5) that control for personal characteristics and location only. Retrenchment may well lead to a change in sector and possibly occupation, so we do not try to control for those variables. We do not use the results with the selectivity correction since we find them unpersuasive. Since there is a positive covariance between the unobservables determining retrenchment and earnings, correcting for selectivity would lead us to predict higher counterfactual earnings for retrenched workers. Our estimates of the loss from retrenchment may therefore be underestimates. We predict the mean earnings of the re-employed if paid like the non-retrenched urban workers using the formula: E(W) = exp(βX+1/2 σ2) where W is earnings per day, β is the vector of coefficients from the earnings function for the non-retrenched urban workers, X is a vector of the means of explanatory variables for workers who have been retrenched, and σ is the standard error of the earnings function for the non-retrenched urban workers.
From this calculation, we estimate that re-employed workers would earn 27.99 yuan per day if paid like the non-retrenched urban workers (Table 7 refers). Unemployed workers are predicted to earn rather less, 25.87 yuan per day. Taken together, counterfactual earnings for the retrenched are 26.66 yuan (0.6264x25.87+0.3736x27.99). Consequently, one measure of the cost of retrenchment borne by the laid off workers is the 14.55 yuan per day shortfall between their actual income (earnings or unemployment benefits) and what they are predicted to earn if employed under the same pay schedule as non-retrenched workers. This implies that, in the year of the survey, retrenchment cost the dismissed workers 55% of their counterfactual earnings. This hardship caused by retrenchment has two components: firstly, the direct loss of earnings while unemployed; and secondly, a real wage cut when re-employed at a lower wage. However, the direct cost is much the larger of the two. Assuming that the unemployed would be paid like the never retrenched, the direct loss of earnings due to retrenchment accounts for over nine tenths of the overall cost of retrenchment (62.64% x [25.87-4.59]/14.55 = 92%). Although the reemployed have found new jobs, we estimate that these pay 12% less than they would be earning if paid according to the wage structure for the never retrenched (24.73 yuan compared to 27.99). This loss of earnings accounts for 8% of the overall cost of retrenchment (37.36%x[27.9924.73]/14.55 = 8%). Note that these costs are for the year of the survey only. We do not try to estimate costs prior to the survey – when some of the re-employed are likely to have lost earnings directly due to retrenchment. Nor do we try to predict the future cost. It is likely that more of the retrenched will find new work, reducing the cost of retrenchment. However, we do not know how long the indirect cost of retrenchment – the wage differential between the re-employed and the nonretrenched urban workers - will be maintained. It could be that the wages of the non-retrenched urban workers are bid down over time, although – as we shall see – there is little evidence of this so far. Alternatively, it could be that the wages of the re-employed rise more sharply than the real wages of the non-retrenched urban workers. This latter process is not automatic, however, as indicated by the absence of an effect of potential experience in the earnings function for the reemployed.
It may be questioned whether the retrenched could have been kept in employment and been paid like the non-retrenched urban workers. A more modest counterfactual is that the retrenched kept their jobs but merely had their pay schedules maintained in real terms. Wages under this assumption can be estimated by using the mean characteristics of those who have been retrenched and the historic earnings function of the retrenched. By using this estimate rather actual historic wages, we allow for a change in real wages due to increased experience. This calculation predicts mean daily earnings of 17.60 yuan for the retrenched workers. Note that this is much less than is predicted using the earnings functions of the non-retrenched urban workers. Urban workers have enjoyed a substantial rise in real wages during the period of retrenchment13. Under this scenario, the cost of retrenchment averages 31% of counterfactual earnings ([17.60 – 12.11]/17.60=0.31) or 5.5 yuan per day per worker. This costing is little more than a third as large, in absolute terms, as the previous estimate of 14.55 yuan per day per worker. It also implies a different decomposition of the cost of retrenchment into losses due to unemployment and those arising from lower wages. In particular, the historic earnings function for the retrenched predicts that re-employed workers would earn an average of 18.55 yuan per day. This is much less their actual mean earnings. All of the cost of retrenchment under this scenario is in the form of direct loss of earnings during unemployment. Re-employed workers may have lost out from retrenchment in the sense that they would probably have even higher real wages if they had remained in employment like the non-retrenched workers. However, their actual real wages are not lower than in their previous jobs. 4.2
Are migrants underpaid?
Our simulations above suggest that retrenched urban workers who have been re-employed are paid much less, given their characteristics, than non-retrenched urban workers. How do migrants fare? Recall that, in terms of mean earnings, migrants are paid less than non-retrenched urban workers but more than re-employed urban workers. These simple comparisons do not allow for the fact that migrants have different observable characteristics from urban workers. For example, migrants tend to have less education and experience. Such differences in observable characteristics may explain migrants’ lower rates of pay relative to non-retrenched urban workers. To explore this, we take migrant mean characteristics and the earnings function of the non-retrenched urban workers. From this, we predict migrants could earn an average of 24.06 yuan per day if paid like non-retrenched urban workers. This is 13% less than their actual earnings. Consequently, there is no evidence that on average migrants are underpaid or that they would receive higher earnings if they were non-retrenched urban workers. Interestingly, nonretrenched urban workers are predicted to receive less (27.36 yuan per day) if paid like migrants. Both groups of workers – migrants and non-retrenched urban workers - are paid more than they would be were their positions to be reversed. This result resembles Roy’s classic model of labour sorting according to comparative advantage (Roy, 1951). Although migrants cannot obtain the hukous required to become urban workers, wage differentials do not appear to be a reason for migrants wanting to acquire urban hukous if they could. Less surprisingly, while urban workers could conceivably compete with migrants in the same activities as the migrants currently undertake, there would be no monetary advantage to doing so. However, it should be noted that we are comparing earnings alone. Migrants still lack the numerous benefits – in housing, education, health care, pensions and social insurance – enjoyed by urban residents, and they may suffer from poorer working conditions. 13
According to official statistics, mean money wages for urban workers in 1999 were 83% higher than in 1994, whereas the urban CPI was only 29% higher, implying a 43% rise in real wages (authors’ computations from SSB, 1999, 2000). 15
Finally, note that although re-employed urban workers are paid less than migrants, this appears to be attributable to their observable characteristics. They are actually paid more, given their characteristics, than would be expected if paid like migrants of similar characteristics. Entering the mean characteristics of the re-employed into the earnings function for the migrants yields an expected daily wage of 23.93 yuan. This is somewhat lower than the actual earnings of the reemployed, 24.73 yuan per day.
The last decade has seen widespread redundancies in urban China: from a survey of 13 cities in 1999/2000, we estimate 16% of urban workers had been retrenched since 1992. The threat of redundancy has not affected all types of worker equally. Certain personal characteristics are associated with a greater risk of redundancy – these include lacking education, being female and being middle aged. Particular kinds of employment have also suffered more retrenchment – notably employment by local government or urban collectives, and manual and/or unskilled occupations. Some of these risk factors – such as lack of education – are consistent with retrenchment of less productive labour. However, others – notably female sex – are more suggestive of discriminatory retrenchment policies. We find few significant determinants of the probability of retrenched workers finding re-employment. Men in good health are more likely to be re-employed. Household demographics also play a role, with children inhibiting reemployment of women – an effect somewhat offset by the presence of older household members as alternative carers. Where retrenched workers are able to find re-employment, we find that they are paid according to very different criteria from those used either in their previous employment or those used for urban workers who have never been retrenched. Few observed characteristics are significantly related to the earnings of the re-employed. Men earn considerably more than women, ceteris paribus, and more so in other sub-samples. However, wage differentials associated with central government employment or party membership have disappeared. Such differentials do not exist for migrants and arguably their absence is to be expected in a competitive labour market. But the re-employed are unique among the four sub-groups we identify in being paid no return to their education. The results for these retrenched workers suggest that the labour market they face is a fierce one that is not yet working well for them. The lack of rewards for productive characteristics implies that their employers still lack information on their productivity, or that the workers are forced by limited opportunities to accept low-grade jobs in which their productive characteristics have little value. We used our earnings functions to calculate the loss of income during 1999 due to retrenchment. If the retrenched could have been paid according to the same standards as those urban workers who did not lose their jobs, then their loss of income in the year of the survey amounted to 55% of their counterfactual earnings. More than nine tenths of this cost is due to the direct loss of earnings during unemployment. The fact that the re-employed are paid less than non-retrenched urban workers with the same characteristics accounts for the remainder of the cost. On the more conservative assumption that retrenched workers’ real wages would have remained fixed, the loss of earnings due to retrenchment would have been 31% of counterfactual earnings. This loss is entirely due to the direct cost of retrenchment – the re-employed earn more than they did prior to retrenchment.
We also compared the earnings profiles for migrants and urban residents. Although nonretrenched urban workers are paid more than migrants, this is entirely attributable to the urban workers having more productive characteristics, such as more education. Our results suggest that migrants would not be better off if paid by the same standards as the non-retrenched urban workers. Neither would non-retrenched urban workers be better off if paid by the same standards as the migrants. This finding is consistent with a degree of competition between the two groups of workers. For example, the returns to education and experience are found to be the same for migrants and non-retrenched urban workers. This suggests that the historically segmented nature of urban employment in China is breaking down, with the jobs of the migrants and the urban residents becoming more interchangeable. However, we do not wish to overstate this conclusion – there remain significant differences between the earnings functions for non-retrenched urban workers and migrants. Among the non-retrenched urban workers, there remain wage differentials by sector, party affiliation and other characteristics that may reflect segmentation rather than differences in human capital. The re-employed urban workers appear to be facing the sharp end of competition between migrants and urban residents. Despite official attempts to protect retrenched workers, unemployment rates are high in our sample and the minority of retrenched workers who find re-employment is paid no better than migrants. Is a competitive labour market emerging in urban China? On the whole, the evidence suggests that it is. We could find little residual difference between the wages of the, traditionally privileged, non-retrenched urban workers and the, traditionally discriminated against, rural-urban migrants. Moreover, we found that human capital is now well, and equally, rewarded, for these groups. The outliers are the retrenched urban workers who have been flung into the market. In the short run, at least, they appear to be the victims of imperfect information and constrained choice.
References Dong and Bowles (2000) ""Segmentation and Discrimination in China's Emerging Industrial labour Market" (mimeo). Idler, Ellen L. (1991) "Self-assessed Health and Mortality: a Review of Studies" International Review of Health Psychology Idler, Ellen L. and Stanislav Kasl (1991) "Health Perception and Survival: Do Global Evaluations of Health Status Really Predict Mortality?" Journal of Gerontology Vol.46 No.2 S55-65. Knight, John and Lina Song (1991). ‘The Determinants of Urban Income Inequality in China’, Oxford Bulletin of Economics and Statistics, 53, 2, 123-54. Knight, John and Lina Song (1993). ‘Why Urban Wages Differ in China’, in Keith Griffin and Zhao Renwei (eds), The Distribution of Income in China, London: Macmillan, 216-84. Knight, John, Lina Song and Jia Huaibin (1999a). ‘Chinese Rural Migrants in Urban Enterprises: Three Perspectives’, Journal of Development Studies, 35, 73-104. Knight, John and Lina Song (1999b). ‘Employment Constraints and Sub-optimality in Chinese Enterprises’, Oxford Economic Papers, 51, 284-99. Maddala, G. S. (1983) Limited Dependent and Qualitative Variables in Economics, Cambridge University Press: New York Meyer, Bruce (1990) "Unemployment Insurance and Unemployment Spells" Econometrica Vol.58 p757-782
Roy, A. D. (1951), "Some thoughts on the distribution of earnings", Oxford Economic Papers 3:135-46 SSB (2001), China Statistical Yearbook 2000, China Labour Statistical Yearbook 2000, and China Monthly Economic Indicators 2000.7. Beijing, National Statistical Bureau of P.R. China.
Table 1: Descriptive Statistics - Means and Percentage by Type of Worker
No. of observations Earnings (yuan per 8 hour period of work) Male % Working experience (in years) Education (in years) Ethnic minority % Party membership % Not generally healthy % Ownership State-owned by (% ) central government State-owned by local government Urban collective Urban private Urban individual Joint venture and foreign investment Listed firms with the majority of state shares Other shareholding Rural private business or individual Other Occupation Professional or (%) technical Administrative Clerical Industrial worker Commercial staff Service worker Others City (%) Beijing Shenyang, Liaoning Jinzhou, Liaoning Nanjing, Jiangsu Xuzhou, Jiangsu Zhengzhou, Henan Kaifeng, Henan Pingdingshan, Henan Chengdu, Sichuan Zigong, Sichuan Nanchong, Sichuan Lanzhou, Gansu Pingliang, Gansu
(1) Nonretrenched urban worker 5770 33.45
(2) Retrenched urban worker 1159 17.16*
(5) Ruralurban migrant 1065 27.58
56.3 21.7 11.4 3.9 28.8 3.8 36.8
37.1 20.1 10.0 4.5 9.3 8.2 12.0
33.6 20.0 10.0 3.6 8.4 9.6 13.7
42.8 23.5 10.0 10.7 6.0 6.0 12.1
59.2 19.2 8.0 6.1 2.4 1.6 4.3
9.9 1.1 3.0 2.2
31.1 1.2 2.3 0.5
31.1 1.1 2.2 0.7
27.0 2.1 8.6 0.2
6.2 9.0 51.7 0.8
13.4 18.4 32.9 5.1 4.9 1.5 15.3 11.9
4.4 8.3 59.2 9.2 7.7 1.6 10.5 7.9
6.7 7.9 57.8 10.6 8.1 1.3 7.7 19.1
5.0 10.3 55.5 9.3 6.6 2.2 9.0 8.4
42.9 4.3 6.2 8.3 25.9 8.4 13.3 10.1
4.7 10.6 5.0 7.3 4.3 6.2
10.5 7.9 6.9 4.7 6.1 3.6
5.6 5.2 5.8 5.6 7.9 5.2
18.5 12.2 8.6 3.2 3.2 1.0
7.4 10.0 6.7 4.0 5.8 7.5
10.2 4.6 5.1 9.6 5.0
10.4 6.8 6.5 11.2 7.1
11.9 9.1 6.4 12.2 6.1
8.1 3.2 6.7 9.5 8.5
8.6 6.8 6.9 6.2 6.6
* =earnings in previous job, updated to allow for price inflation
Source: CASS 1999 Household Survey.
(3) Of (2), still unemployed
(4) Of (2) reemployed
Table 2.1: Probit Model for the Probability of Retrenchment Constant Male Education in years Experience Experience squared Minority ethnicity Party member Bad health Central SOE Urban collective Urban private Urban individual Foreign invested State listed Other listed Rural enterprises Other ownership Administrative Clerical Low skill industrial High skill industrial Unskilled industrial Commercial worker Service worker Other occupation Shenyang, Liaoning Jinzhou, Liaoning Nanjing, Jiangsu Xuzhou, Jiangsu Zhengzhou, Henan Kaifeng, Henan Pingdingshan, Henan Chengdu, Sichuan Zigong, Sichuan Nanchong, Sichuan Lanzhou, Gansu Pingliang, Gansu Dummy for SOE workers married to retrenched SOE workers Log-likelihood Restricted log-likelihood Likelihood ratio index Number of observations
Coefficient -1.036 -0.280 -0.044 0.039 -0.001 0.057 -0.353 0.267 -0.602 0.357 -0.073 -0.443 -0.551 -0.144 -0.549 0.252 -0.080 0.039 -0.008 0.571 0.379 0.524 0.511 0.387 0.362 -0.125 0.384 -0.099 0.270 -0.019 0.302 -0.120 0.233 0.286 0.287 0.260 0.541 -0.191
T-ratio -5.08 -6.74 -4.18 4.15 -3.67 0.55 -5.66 3.18 -10.99 6.64 -0.40 -3.53 -2.81 -1.15 -2.97 0.38 -0.46 0.42 -0.10 7.37 5.07 6.90 5.50 3.96 2.20 -1.46 4.15 -1.13 2.75 -0.19 2.97 -1.09 2.72 2.87 2.84 3.10 5.40 -2.10
*** *** *** *** *** *** *** *** *** *** *** ***
*** *** *** *** *** ** *** *** *** *** *** *** *** *** **
-2577.46 -3128.64 0.18 6929
Predicted 0 5650 1005 6655
Source: CASS 1999 Household Survey.
1 120 154 274
Table 2.2 Predicted Probabilities of Retrenchment Base probability (at mean of Xs) Personal characteristics Male Female Minority Han Party member Not party member Bad health Good health Ownership of establishment Local SOE Central SOE Urban collective Urban private Urban individual Foreign invested State listed Other listed Rural enterprises Other ownership Instrument (for SOE employees only) Spouse retrenched from SOE Spouse not retrenched from SOE Occupation Professional/technical Administrative Clerical Low skill industrial High kill industrial Unskilled industrial Commercial worker Service worker Other occupation City Beijing Shenyang, Liaoning Jinzhou, Liaoning Nanjing, Jiangsu Xuzhou, Jiangsu Zhengzhou, Henan Kaifeng, Henan Pingdingshan, Henan Chengdu, Sichuan Zigong, Sichuan Nanchong, Sichuan Lanzhou, Gansu Pingliang, Gansu
23.5 21.2 26.3 24.5 23.5 19.1 25.2 28.4 23.3 27.0 16.9 34.6 25.6 19.2 17.6 24.3 17.6 32.3 25.5 20.4 23.4 19.5 20.1 19.4 30.0 26.1 29.0 28.8 26.3 25.8 21.3 19.3 28.5 19.7 26.2 21.0 26.8 19.4 25.5 26.5 26.5 26.0 31.8
Source: Authors’ calculations from 1999 CASS Household Survey.
Table 3.1: Models for the Probability of Re-employment Models for the duration of unemployment
Binary probit for reemployment by time of survey Constant Male Education in years Experience Experience squared Minority ethnicity Party member Bad health Shenyang, Liaoning Jinzhou, Liaoning Nanjing, Jiangsu Xuzhou, Jiangsu Zhengzhou, Henan Kaifeng, Henan Pingdingshan, Henan Chengdu, Sichuan Zigong, Sichuan Nanchong, Sichuan Lanzhou, Gansu Pingliang, Gansu No. of children in household under 7 Number of children under7 in household x female worker No. children in household x no. old x female worker Log-likelihood Restricted loglikelihood Likelihood ratio index Number of observations
Coefficient -0.376 0.185 -0.019 0.012 0.000 0.184 0.090 -0.385 0.233 0.992 0.744 0.503 -0.269 -0.450 -0.829 -0.100 -0.516 0.144 -0.019 0.323 0.348 -0.775
T-ratio -0.97 2.02 -0.88 0.56 -0.88 0.94 0.63 -2.52 1.28 5.66 4.10 2.66 -1.21 -2.09 -2.74 -0.56 -2.46 0.74 -0.11 1.68 1.91
** *** *** *** ** *** **
Coefficient T-ratio 2.641 6.64 *** -0.177 -1.97 ** -0.034 -1.62 0.004 0.18 0.000 -0.89 0.065 0.37 0.065 0.44 0.323 2.31 ** -0.333 -1.87 * -0.492 -2.79 *** -0.786 -4.46 *** -0.341 -1.68 * 0.057 0.27 0.582 2.35 ** 0.552 1.51 -0.060 -0.33 0.333 1.50 -0.456 -2.27 ** -0.215 -1.20 -0.292 -1.42 -0.215 -1.22
Meyer model with no Meyer model with unobserved heterogengeity unobserved heterogengeity Coefficient T-ratio Coefficient T-ratio -0.206 -0.044 -0.01 0 -0.02 -0.044 0.563 -0.262 -0.694 -0.858 -0.557 0.426 0.772 1.013 0.04 0.651 -0.526 -0.227 -0.437 -0.337
-1.86 -1.67 -0.38 -0.24 -0.1 -0.27 2.7 -1.12 -3.42 -3.99 -2.37 1.36 2.38 1.92 0.17 2.01 -2.07 -0.99 -1.83 -1.66
-0.248 -0.052 -0.009 0 0.001 0.012 0.614 -0.346 -0.791 -1.062 -0.587 0.4 0.83 1.045 0.012 0.66 -0.608 -0.259 -0.486 -0.374
*** *** *** ** ** * * ** * *
Predicted 0 0 621 1 254
*** *** *** ** ** * * ** *
1 105 179
Source: Authors’ calculations from 1999 CASS Household Survey
-1.9 -1.71 -0.3 -0.38 0.01 0.06 2.63 -1.25 -3.19 -3.25 -2.25 1.18 2.35 1.89 0.05 1.93 -2.08 -1.03 -1.81 -1.59
Table 3.2 Predicted Probabilities of Re-employment Base probability (at mean of Xs) Personal characteristics Male Female Minority Han Party member Not party member Bad health Good health City Beijing Shenyang, Liaoning Jinzhou, Liaoning Nanjing, Jiangsu Xuzhou, Jiangsu Zhengzhou, Henan Kaifeng, Henan Pingdingshan, Henan Chengdu, Sichuan Zigong, Sichuan Nanchong, Sichuan Lanzhou, Gansu Pingliang, Gansu
41.1 43.9 39.4 45.4 40.9 43.1 40.9 32.9 41.9 38.0 43.6 62.3 56.3 50.3 32.0 28.1 21.1 35.7 26.7 41.4 37.5 45.8
Source: Authors’ calculations from 1999 CASS Household Survey.
Table 4: Earnings Functions of All Elements of Urban Labour Force (without correction for selectivity) Non-retrenched urban workers Coefficient T-ratio Constant 2.850 44.42 Male 0.135 9.75 Education in years 0.034 10.61 Experience 0.037 12.13 Experience squared -0.001 -9.40 Minority ethnicity 0.051 1.54 Party member 0.074 4.68 Bad health -0.085 -2.05 Central SOE 0.156 10.77 Urban collective -0.169 -6.86 Urban private 0.104 1.26 Urban individual -0.031 -0.59 Foreign invested 0.347 6.35 State listed 0.085 2.35 Other listed 0.064 1.21 Rural enterprises -0.257 -2.40 Other ownership -0.023 -0.36 Administrative 0.000 -0.02 Clerical -0.068 -3.43 Industrial worker -0.239 -12.68 Commercial worker -0.226 -6.57 Service worker -0.254 -7.46 Other occupation -0.346 -5.24 Shenyang, Liaoning -0.477 -17.64 Jinzhou, Liaoning -0.445 -11.80 Nanjing, Jiangsu -0.136 -5.23 Xuzhou, Jiangsu -0.341 -10.67 Zhengzhou, Henan -0.431 -14.09 Kaifeng, Henan -0.692 -19.88 Pingdingshan, Henan -0.477 -15.69 Chengdu, Sichuan -0.404 -14.86 Zigong, Sichuan -0.527 -15.53 Nanchong, Sichuan -0.585 -17.31 Lanzhou, Gansu -0.456 -16.08 Pingliang, Gansu -0.642 -21.16 Number of observations Mean of dependent variable Adjusted R-squared Standard error of equation
*** *** *** *** *** *** ** *** ***
*** ** **
*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
Coefficient T-ratio 2.778 8.72 *** 0.313 5.11 *** 0.007 0.44 0.052 3.99 *** -0.001 -3.67 *** -0.131 -1.24 -0.038 -0.23 -0.056 -0.55 0.057 0.60 -0.129 -1.83 * -0.117 -0.55 0.125 1.42 -0.380 -3.09 *** 0.000 0.00 0.187 0.83 -0.624 -6.50 *** -0.181 -1.04 0.016 0.12 -0.327 -2.34 ** -0.173 -1.72 * -0.123 -1.04 -0.346 -2.18 ** 0.124 0.56 -0.252 -1.61 -0.520 -4.12 *** -0.162 -1.21 -0.207 -1.46 -0.478 -2.47 ** -0.636 -3.32 *** -0.379 -1.55 -0.231 -1.70 * -0.754 -4.41 *** -0.511 -3.90 *** -0.336 -2.54 ** -0.691 -4.53 ***
Retrenched urban workers (historic earnings) Coefficient T-ratio 2.259 8.45 *** 0.064 2.05 ** 0.031 2.88 *** 0.012 1.32 0.000 0.41 -0.025 -0.35 0.094 2.14 ** -0.081 -1.58 0.121 2.45 ** -0.108 -3.12 *** -0.031 -0.29 -0.126 -0.60 0.073 0.54 -0.050 -0.43 0.254 1.74 * 0.538 6.08 *** 0.033 0.16 0.136 1.64 0.052 0.72 0.033 0.48 0.091 1.12 -0.037 -0.40 -0.297 -1.62 -0.114 -1.38 -0.168 -2.26 ** 0.066 0.87 -0.128 -1.36 -0.086 -0.99 -0.257 -3.21 *** -0.418 -3.89 *** -0.074 -0.89 -0.394 -4.38 *** -0.426 -4.82 *** -0.121 -1.53 -0.445 -4.60 ***
Migrants Coefficient T-ratio 2.498 14.49 *** 0.210 5.12 *** 0.045 4.64 *** 0.024 3.21 *** 0.000 -2.83 *** 0.112 1.22 0.178 0.88 -0.219 -1.49 0.029 0.23 0.124 1.11 0.074 0.81 0.094 1.14 0.490 2.47 ** -0.169 -0.92 0.342 2.21 ** 0.115 1.21 0.064 0.57 -0.052 -0.58 0.059 0.53 -0.199 -2.00 ** -0.340 -3.47 *** -0.422 -4.74 *** -0.233 -2.40 ** -0.121 -1.39 -0.238 -2.47 ** -0.324 -3.73 *** -0.149 -1.21 -0.537 -4.66 *** -0.444 -4.48 *** -0.706 -7.31 *** -0.197 -2.08 ** -0.566 -5.40 *** -0.400 -4.06 *** -0.304 -2.83 *** -0.824 -9.44 ***
Source: Authors’ calculations from 1999 CASS Household Survey. Note: T-ratios are calculated by using White's heteroscedasticity consistent standard errors.
Table 5: Earnings Functions with Human Capital Variables Only (no selectivity
correction) Non-retrenched urban workers Coefficient T-ratio Constant 2.445 43.79 *** Male 0.149 10.45 *** Education in years 0.060 20.61 *** Experience 0.036 11.67 *** Experience squared -0.001 -8.22 *** Minority ethnicity 0.056 1.59 Party member 0.147 9.61 *** Bad health -0.107 -2.53 ** Shenyang, Liaoning -0.490 -17.40 *** Jinzhou, Liaoning -0.516 -13.49 *** Nanjing, Jiangsu -0.192 -7.23 *** Xuzhou, Jiangsu -0.369 -11.08 *** Zhengzhou, Henan -0.428 -13.88 *** Kaifeng, Henan -0.719 -19.89 *** Pingdingshan, Henan -0.465 -14.90 *** Chengdu, Sichuan -0.403 -14.61 *** Zigong, Sichuan -0.593 -17.07 *** Nanchong, Sichuan -0.570 -15.95 *** Lanzhou, Gansu -0.434 -14.82 *** Pingliang, Gansu -0.585 -18.77 *** Adjusted R-squared Standard error of equation
Re-employed urban workers Coefficient T-ratio 2.452 9.07 *** 0.327 5.56 *** 0.017 1.20 0.048 3.68 *** -0.001 -3.21 *** -0.116 -1.11 -0.071 -0.44 -0.017 -0.15 -0.253 -1.61 -0.489 -4.26 *** -0.179 -1.42 -0.155 -1.17 -0.418 -2.34 ** -0.618 -3.44 *** -0.354 -1.71 * -0.162 -1.27 -0.702 -4.24 *** -0.389 -3.32 *** -0.249 -2.06 ** -0.584 -4.05 ***
Retrenched urban workers Coefficient T-ratio 2.223 11.60 *** 0.073 2.28 ** 0.036 4.08 *** 0.012 1.25 0.000 0.49 -0.026 -0.36 0.134 3.05 *** -0.087 -1.71 * -0.134 -1.72 * -0.201 -2.78 *** 0.020 0.28 -0.153 -1.60 -0.096 -1.13 -0.266 -3.51 *** -0.381 -3.57 *** -0.111 -1.41 -0.429 -5.10 *** -0.433 -5.08 *** -0.097 -1.21 -0.422 -4.56 ***
Source: Authors’ calculations from 1999 CASS Household Survey Note: T-ratios use White's heteroscedasticity consistent standard errors
Migrants Coefficient 2.244 0.232 0.056 0.027 -0.001 0.139 0.182 -0.214 -0.103 -0.250 -0.374 -0.122 -0.525 -0.415 -0.581 -0.156 -0.513 -0.365 -0.343 -0.783
T-ratio 15.34 5.52 5.73 3.48 -3.19 1.44 0.93 -1.37 -1.16 -2.76 -4.18 -1.00 -4.46 -3.87 -6.59 -1.68 -5.37 -3.78 -3.08 -9.79 0.153 0.691
*** *** *** *** ***
*** *** *** *** *** * *** *** *** ***
Table 6: Earnings Functions for Three Categories of the Urban Labour Force (with Selectivity Corrections) Non-retrenched urban workers Coefficient T-ratio Constant 2.917 43.21 Male 0.114 6.86 Education in years 0.031 8.88 Experience 0.040 13.95 Experience squared -0.001 -10.80 Minority ethnicity 0.055 1.60 Party member 0.054 2.82 Bad health -0.056 -1.52 Central SOE 0.120 5.45 Urban collective -0.122 -3.88 Urban private 0.098 1.52 Urban individual -0.064 -1.45 Foreign invested 0.314 6.41 State listed 0.074 1.67 Other listed 0.028 0.53 Rural enterprises -0.244 -0.98 Other ownership -0.029 -0.48 Administrative 0.004 0.15 Clerical -0.070 -3.38 Industrial worker -0.202 -7.88 Commercial worker -0.186 -4.87 Service worker -0.226 -6.16 Other occupation -0.321 -5.62 Shenyang, Liaoning -0.487 -18.74 Jinzhou, Liaoning -0.412 -11.08 Nanjing, Jiangsu -0.142 -5.33 Xuzhou, Jiangsu -0.322 -9.22 Zhengzhou, Henan -0.431 -14.37 Kaifeng, Henan -0.669 -17.99 Pingdingshan, Henan -0.483 -15.14 Chengdu, Sichuan -0.388 -13.98 Zigong, Sichuan -0.507 -13.83 Nanchong, Sichuan -0.564 -16.01 Lanzhou, Gansu -0.438 -15.34 Pingliang, Gansu -0.605 -15.96 Correction for 0.237 2.29 selectivity of retrenchment Correction for selectivity of reemployment
Number of observations Mean of dependent variable Adjusted R-squared Standard error of equation
*** *** *** *** *** *** *** ***
*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** **
Re-employed workers, selectivity via: Binary probit Ordered probit Coefficient T-ratio Coefficient T-ratio 1.950 4.53 *** 2.584 7.20 0.324 3.95 *** 0.283 3.82 -0.013 -0.81 -0.003 -0.16 0.066 4.82 *** 0.059 4.46 -0.001 -4.55 *** -0.001 -3.95 -0.206 -1.78 * -0.237 -2.04 0.042 0.25 -0.026 -0.16 -0.103 -0.79 -0.045 -0.37 -0.010 -0.10 -0.013 -0.13 -0.073 -0.98 -0.078 -1.05 -0.079 -0.39 -0.085 -0.42 0.120 1.34 0.120 1.34 -0.500 -3.77 *** -0.505 -3.73 0.032 0.17 0.015 0.08 0.194 0.96 0.191 0.91 -0.522 -4.99 *** -0.574 -5.74 -0.219 -1.28 -0.224 -1.31 0.083 0.64 0.077 0.60 -0.324 -2.35 ** -0.338 -2.45 -0.099 -0.98 -0.107 -1.05 -0.055 -0.45 -0.066 -0.54 -0.301 -1.92 * -0.306 -1.96 0.177 0.82 0.168 0.76 -0.221 -1.43 -0.235 -1.47 -0.179 -0.99 -0.354 -2.35 0.026 0.15 -0.038 -0.21 -0.029 -0.18 -0.110 -0.72 -0.584 -2.80 *** -0.512 -2.64 -0.758 -3.44 *** -0.675 -3.12 -0.671 -2.18 ** -0.487 -1.95 -0.241 -1.78 * -0.194 -1.43 -0.894 -4.45 *** -0.764 -4.30 -0.425 -3.22 *** -0.377 -2.47 -0.300 -2.31 ** -0.260 -1.93 -0.517 -3.20 *** -0.555 -3.37 0.282 2.23 ** 0.290 2.31
*** *** *** *** **
*** *** * *** ** * *** **
Retrenched urban workers (historic earnings) Coefficient T-ratio 2.215 5.53 *** 0.058 0.89 0.030 2.35 ** 0.013 1.35 0.000 0.27 -0.024 -0.31 0.085 0.94 -0.075 -1.00 0.106 0.81 -0.100 -1.29 -0.032 -0.22 -0.137 -1.01 0.058 0.26 -0.053 -0.49 0.240 1.16 0.543 1.05 0.031 0.22 0.137 1.56 0.052 0.73 0.045 0.41 0.103 0.83 -0.028 -0.26 -0.289 -1.97 ** -0.118 -1.54 -0.159 -1.64 0.064 0.85 -0.122 -1.39 -0.086 -1.05 -0.251 -2.65 *** -0.421 -4.44 *** -0.068 -0.84 -0.388 -4.25 *** -0.419 -4.59 *** -0.115 -1.45 -0.433 -3.46 *** 0.031 0.12
Source: Authors’ calculations from 1999 CASS Household Survey. Note: T-ratios are calculated by using White's heteroscedasticity consistent standard errors.
Table 7: Simulated wage earnings
Mean earnings & benefits (yuan per 8 hours of work) Actual If paid like nonretrenched workers If paid historical real wages of retrenched If paid like migrants
Nonretrenched urban workers 33.45
Retrenched urban workers Re-employed
Source: Authors’ calculations from 1999 CASS Household Survey. Note: counterfactual earnings simulated using the actual mean observable characteristics and counterfactual earnings function; see equation (9) in the text for the relevant formula.
[Appendix] Table 1: Variable Definitions Used for the Analysis (1999 CASS Household Survey) Variable name
Male Education in years Experience Experience squared Minority ethnicity Party member Bad health Sectoral 0-1 variables (default is employment by local government, including locally controlled SOE) Central SOE Urban collective Urban private Urban individual Foreign invested State listed Other listed Rural enterprises Other ownership Occupation 0-1 variables (default is professional or technical) Administrative Clerical Industrial worker Commercial worker Service worker Other occupation Low skill industrial High kill industrial Unskilled industrial City 0-1 dummy variables Shenyang, Liaoning Jinzhou, Liaoning Nanjing, Jiangsu Xuzhou, Jiangsu Zhengzhou, Henan Kaifeng, Henan Pingdingshan, Henan Chengdu, Sichuan Zigong, Sichuan Nanchong, Sichuan Lanzhou, Gansu Pingliang, Gansu Controls for selectivity Correction for selectivity of retrenchment
1 if male, 0 otherwise Years of education Years of potential experience (age - education - six) Experienced squared 1 if not Han Chinese, 0 otherwise 1 if member of Communist Party, 0 otherwise 1 if describe self as in "bad health", 0 otherwise
Correction for selectivity of re-employment Instruments for selectivity corrections Dummy for SOE workers married to retrenched SOE workers No. of children in household under 7 Number of children under7 in household x female worker No. children in household x no. old x female worker
1 if employed by central government, including centrally controlled SOE 1 if employed by urban collective 1 if employed by urban private company 1 if self-employed 1 if employed by a company with foreign investment 1 if employed by a state enterprise listed in the stock market 1 if employed by non-state enterprise listed in the stock market 1 if employed by rural enterprise 1 if employed in type of enterprise not listed above
1 if employed in an administrative or managerial position 1 if employed in a clerical position 1 if employed as an industrial worker 1 if employed as a commercial worker 1 if employed as a service worker 1 if employed in another occupation 1 if employed as a low skilled industrial worker 1 if employed as a high skilled industrial worker 1 if employed as a unskilled industrial worker 1 if in Shenyang 1 if in Jinzhou 1 if in Nanjing 1 if in Xuzhou 1 if in Zhengzhou 1 if in Kaifeng 1 if in Pingdingshan 1 if in Chengdu 1 if in Zigong 1 if in Nanchong 1 if in Lanzhou 1 if in Pingliang Inverse Mills ratio from the probit in Table 2; see equation (4) in text for definition. Inverse Mills ratio from the probit in Table 3 (or correction calculated from ordered probit); see equation (6) in text for definition. 1 if working for state sector (or used to work for state sector) and spouse was retrenched from state sector; 0 otherwise. Number of children aged under seven in the household Number of children aged under seven in the household interacted with dummy for individual being female Number of children in the household multiplied by number of people aged over 59 years old in the household multiplied by a dummy for individual being female