Upload
lyhuong
View
217
Download
0
Embed Size (px)
Citation preview
Determinants of salary growth in Shenzhen, China: An analysis of formal education, on-the-job training, and adult education with a three-level model
Jin Xiao
Faculty of Education
the Chinese University of Hong Kong, Hong Kong Sha Tin, N.T. Hong Kong Special Administration Region, China
Email: [email protected]
Received 19 March 1999; Accepted 6 June 2001
Abstract Using 1996 surveyed data of 1,023 employees in Shenzhen, China, this study estimated the effects of three forms of human capital on employee salary, namely formal education, on-the-job training provided by employers, and adult education pursued by employees. Using a hierarchical linear model, the analysis estimated employee monthly salary growth over a maximum of six years due to (a) such temporal factors as work experience and improved performance, (b) individual-level characteristics, and (c) firm-level characteristics. This study found that (a) pre-work formal education was positively associated with salary only at hiring, (b) employees’ experience in changing production technology as well as on-the-job training were positively associated with salary increases through improved technical proficiency, formal education was not; (c) manufacturing firms introduced more new production technology than the service sector and provided more on-the-job training, thus improving workers’ performance and increasing their salary. JEL classification: [C00, I20, J31] Keywords: Salary growth, Educational economics, Human capital, Economic development, Productivity
1
Determinants of salary growth in Shenzhen, China: An analysis of formal education, on-the-job training, and adult education with a three-level model
1. Introduction
Human capital theory suggests that education or training raises the productivity of workers by
imparting useful knowledge and skills, hence raising workers’ future income by increasing their lifetime
earnings (Becker, 1964). Becker (1964) and Mincer (1974) provide an explanation that links investment in
training with workers’ wages. In particular, their theory draws a crucial distinction between general
education and firm-specific training. Over the past thirty years or so, hundreds of studies have been
conducted to estimate rates of return to education (RORE); most such studies show that formal schooling is
a crucial factor in explaining variations of salary and wages in well developed countries (Cohn & Addison,
1998). Comparative studies have been conducted in some less developed countries, focusing on investment
in formal education (Psacharopoulos, 1985, 1994).
While formal education has expanded rapidly in many countries, a large portion of human capital
accumulation in the forms of on-the-job training and other modes for working adults actually take place
both inside and outside the workplace. Adult education development in developed countries in recent years
has focused on a strengthening of vocational training to meet the needs of skill development across all
occupational strata in the global economy (Belanger & Tuijnman, 1997). Studies in some developing
countries find that a mix of education and training is available for skill acquisition and there are multiple
paths to skill development for a given occupation (Middleton, Ziderman, & Adams, 1993; Ziderman and
Horn, 1995. A study of education provision in Shenzhen, China shows that both firm-provided on-the-job
training and adult education financed by employees offer substantial means to develop vocational/technical
skills (Xiao & Tsang, 1994), and provided about 2.07 million head counts of education and training to the
workforce of 2.5 million during the period of 1980-1996 (Xiao, 1998a:13). Given that education and
training programs for working adults have experienced significant expansion, it is important that they be
included in estimations of returns to education and training. This paper attempts to estimate the effects of
formal education, on-the-job training, and adult education both on employees’ performance and salary
growth with data from a survey conducted in Shenzhen in 1996 (Xiao & Tsang, 1999).
This study was conducted in the municipality of Shenzhen, China, which is situated on the border
with Hong Kong. Prior to 1980, Shenzhen was a small, poor county. The county’s population of 0.3
million was mainly engaged in farming and fishing. In 1980, the State Council, China, inaugurated the
Shenzhen Special Economic Zone, which was designed as a prototype for economic development. The
policies of reform and opening to the outside world were implemented to experiment market economy and
help boost the Chinese economy. Over the last two decades, Shenzhen has developed into a large
industrialized area with an emerging tertiary sector. Its economic growth is rapid. For example, in 1980,
Shenzhen’s per-capita real GDP (in 1978 prices) was RMB 719.8, less than one-third that of Shanghai
(RMB 2,360.3). In 1994, Shenzhen’s per-capita real GDP reached RMB 4,032, which exceeded that of
2
Shanghai by RMB 1,410, ranking first in China ever since (Xiao, 1998a).
Economic reforms brought substantial change to the economic system. In addition to the two
major forms of ownership in a planned economy, namely state-owned and collective firms, there have
emerged other forms of ownership. They include cooperatives and firms with sole-investment from and
joint ventures with firms in Hong Kong, Macau, and Taiwan, other countries, a large number of local
private firms and corporate firms. These firms have created a competitive market and they make businesses
in both domestic and international markets. In the period of 1980 and 1996, these firms had provided about
2.3 million new jobs. These new occupations required that employees have new knowledge, skills, and
attitudes/values. Employees had to learn to treat their clients as “kings” in order to win market share; time
became as valuable as gold and efficiency was vital for a firm’s survival. Advanced production technology
was imported to replace outdated hand-operated equipment.
The local workforce was unable to meet the human-capital demands of the Shenzhen economy in
the early 1980s. In addition to expanding its formal education system, Shenzhen has cultivated and
nurtured the development of workplace training and adult education over time. Two types of institutions
provide education and training to working adults (Xiao & Tsang, 1994). First, employee-oriented adult
education/training centers, set up and financed by large firms, provide training to their own employees.
Training is job-related, corresponding to the productivity requirements of the firm. The second type of
training is run by community organizations that provide education and training to adults in the local
community. This community-oriented education/training provide job-related practical skills programs,
vocational/technical certificate programs and credential programs. Participants usually voluntarily attend
the adult course and pay for the courses themselves, though some may receive subsidies from their firms.
Xiao and Tsang (1999) found that among the 4,002 sampled employees, 59 percent received on-the-job
training, and 31.1 percent attended adult education/training programs. This study distinguished three types
of education and training programs: pre-job formal education, firm-based on-the-job training (OJT)
provided by employers, and self-financed adult education/training (AET) outside the firm.
By using a three-level hierarchical linear growth model, this study incorporates both individual and
firm-level factors to estimate their impact on salary growth over time. The remainder of this paper is
divided into five sections. The next section provides a framework to interpret the rising demand for job-
related training and adult education in the fast growing economy of Shenzhen. Followed is a discussion on
conceptualized issues with a three-level analytic method. The fourth presents the survey and data sets. The
fifth section examines empirical results. Finally, the concluding section presents a discussion on findings
and their implication in development policy.
2. A framework for human capital development in the workplace
While Becker (1964) suggests that education or training raise the productivity of workers by imparting
useful knowledge and skills, others provide different explanations for how education is related to worker
3
productivity. One is based on the argument that the higher earnings of educated workers simply reflect their
superior ability acquired during the process of education, rather than through skills and knowledge. Spence
(1973) argues that education is used as a market signal to indicate the potential productivity of workers.
Thurow (1975) maintains that productivity is largely characteristic of jobs rather than of workers;
employers use education credentials to select workers because better-educated workers can be trained for
specific jobs more quickly and at a lower cost than their less-educated peers. Schultz (1975) suggests that
education enhance an individual’s ability to successfully deal with disequilibria in changing economic
conditions. Such ability includes that of perceiving a given disequilibrium, analyzing information, and
reallocating resources to act. Another argument is based on the conditions of production. For instance,
Levin (1987) argues that the organization of production, such as the extent of discretion, participation in
decision-making, responsibility sharing, and information available to employees, all affect employees’
utilization of their ability to act. Levin and Kelley (1994) suggest that education can improve productivity
only if complementary inputs exit, which include training, contract terms, and management practices; they
point out that economists and other social scientists have overestimated the payoffs resulted from increased
formal education while they have ignored the complementary inputs and conditions. Recently, Hall and
Jones (1999) maintain that differences in capital accumulation, productivity, and therefore output per
worker are fundamentally related to differences in social infrastructure across countries. Such social
infrastructure includes the institutions and government policies that determine the economic environment,
within which individuals accumulate skills and firms accumulate capital and produce output. Lack of these
conditions would cause loss in production. Tsang (1987) found that mismatch of workers characteristics
such as over-education caused dissatisfaction among workers and this was associated with a loss of over 8%
in an output for the $57 billion dominant U.S. telecommunication industry AT & T in 1981.
In this article, I offer an interpretation for the large demand of learning in a workplace in a modern,
fast changing economic state of Shenzhen, China. I argue that job-related training provided to working
adults is a strategy to re-equilibrate in the changing economy. The economic reform, and new technologies
have resulted in qualitative economic changes, which aim to boost economic development in Shenzhen.
Employees, nested in a given production organization, face such changes together with their firms; both
firms and employees are involved in an equilibrating process to increase productivity.
As technological innovations and economic reforms have been continuously adopted in the
workplace in order to speed up growth, the organization of production, and management practices of the
firm that formerly worked well in a planned economy are now obsolete. Such qualitative changes create
disequilibrium, causing a technical discrepancy between the firms’ new investments and their employees’
competence; employees are unable to perform to meet the new demand. They may have accumulated
significant knowledge, skills, and attitudes/values (KSA) through both formal schooling and work
experience, but as change accelerated, KSA gained in previous learning gradually became also largely
obsolete.
4
In a market-oriented economy, the competitiveness of a firm depends on its stock of both physical
and human capitals as well as the uniqueness of management practices that can bring the firm’s capacity
into full play. In response, OJT provided by employers has thus become a firm strategy to develop human
capital in order to adapt to changes in the workplace. OJT, while upgrading job skills of employees, also
develops shared values and ways of working together to strengthen a firm’s unique competitiveness in a
transforming economy. Therefore, in-firm training, as opposed to formal education, as a strategy to develop
human capital can over time compensate for competence gaps of the internal market and the increasingly
complex technological demands of the workplace. By receiving OJT, the competence of employees is to be
re-established.
In Shenzhen, individual employees are free to change jobs whenever they find better opportunities
to realized their personal potentials. If their job performance is not up to an expected level over time,
employees may also be dismissed. Employees are cognizant of both changes in the firm and the increased
discrepancy in their KSA and of opportunities outside the firm. They seek learning to close the gap, either
to remain competent in their current jobs, or to change to new jobs. When employer-provided training does
not fulfill the employees’ personal expectations, external open-market AET becomes an option. By taking
AET, employees can close up the gap between competence and expectations.
Employees look to OJT or AET to enable them to disengage from obsolete sets of KSA and to
regain job competence. Given that firms constitute an economic setting where human capital is utilized,
training and education related to the job setting for employees develop unique sets of KSA that can engage
them in organized production. Therefore it is argued that in a fast-changing economic context, education
and training programs related to jobs are a means to readjust to the changing workplace and improve
productivity.
3. Measuring salary growth with a three-level model
An estimation of human capital effects is complicated by the existence of different types of human
capital (Chapman, 1993:69-72). Training can be taken in the pre-work form of formal education.
Individuals can also train themselves through learning-by-doing in their jobs. The Mincerian method has
been commonly used to estimate RORE to formal education in the earning function equation. It might
measure the second type of human capital by using job experience, as proxied by years of working.
However, various vocational adult education and training have been widely used in both OECD countries
and other less developed countries to develop human capital. Recently Cohn and Addison (1998)
conducted a comprehensive review of the literature on returns to both formal schooling and various
vocational training programs for youth and adults. Regarding the former, they concluded that RORE is
substantial across levels of schooling. Regarding the latter, they found that returns to training investment
are mixed. Of course, education and training take place in more heterogeneous forms, namely, provided by
different agencies to different age and occupations groups or delivered in various modes and settings. In
5
addition, the sample selection problems present greater ambiguity for an empirical investigation of training
for adults than for formal education. It is unclear to what extent ability factors endogenous, determined by
training and experience, and to what extent they are exogenous to other influences such as the market
(Chapman, 1993:71-72). To date, most studies have used the three conventional methods to estimate
returns of investment in education and training, namely short-cut methods, internal RORE or the Mincerian
approach (see Cohn & Addison, 1998; Psacharopoulos and Woodhall, 1985). By focusing on the end result
of earnings, few methods make allowances for the endogeneity of training decisions in a workplace setting.1
In order to understand the occurrence and benefits of training, it is interesting to combine an
analysis of the decisions of employers and employees regarding various education and training programs
with the presence of influences of workplace characteristics within a single study. Such a study draws on
recent advances in multilevel statistical theory by Bryk & Raudenbush (1992) and attempts to measure
changes in earnings due to the effects of different types of education and training.
Since employees are not randomly assigned to firms, the task of measuring changes in earnings
becomes challenging. For instance, hirings and earnings increase over time are firm decisions affected by
the firm’s characteristics, as well as by individual characteristics pre-embodied and carried to the firm (e.g.,
formal schooling, sex, age, previous work experience, etc.) or developed while working (e.g., improved job
skills and then performance through OJT). An aggregation bias can occur when a variable takes on
different meanings, and therefore it may have a different effect at a different level of analysis. For example,
receiving OJT may have an impact on one’s job skills at the individual level, which can be measured by
performance assessment and increased earning increase. At the firm level, providing OJT is a proxy
measure of a firm’s capacity and normative management practices. The recent statistics advancement in
hierarchical linear models (HLM) resolves this problem by facilitating a decomposition of any observed
relationship between variables into different level components (Bryk & Raudenbush, 1992). Again, within
the firm, employees may develop dependence among themselves while working in the same normative
environment. HLM incorporates a unique random effect for each organization unit and the variability in
these random effects is taken into account to estimate standard errors. HLM also resolves the problem of
heterogeneity of regression by estimating a separate set of regression coefficients for each organizational
unit and by then modeling variations among firms in their sets of coefficients. The other advantage of HLM
is that it accommodates multiple-time-point observations of an individual over time.
The analysis employs a three-level growth model (Bryk & Raudenbush, 1992:130-154, 185-196),
which offers an integrated approach for studying determinants of salary growth in an organized structure
and presents both individual and firm predictors of salary growth from a multiple-time-point design. The
analysis consists models at three levels: (a) salary observations at three time points, (b) employee factors,
and (c) firm factors. At Level-1 (L1), the within-individual level, such variables as employee salary growth
within a span of a maximum of six years at three observations and gains in one’s technical proficiency level
at corresponding observation time points are examined. At Level-2 (L2), the individual level, such
variables as sex, age, formal schooling, OJT and AET, and experience of changes in the firm are
6
considered. At Level-3 (L3), such firm characteristics as firm size, location, ownership, industrial sector
and firms’ capacity to provide OJT are included in the analysis.
It is then assumed that Ytij, the observed log salary at time t for employee i, is a function of a
systematic growth curve plus random error. The systematic growth over time is represented as a polynomial
of degree P. Then, the L1 Model is:
Ytij = π0ij + π1ij (OBSERVATION) tij + π2ij (OBSERVATION2) tij + π3ij (OBSERVATION3) tij + π4ij(PERF
GAIN)tij + etij (1)
For i = 1,…., 1,023 employees in firm j (j =1,… 71), where
Ytij is the outcome variable of log monthly salary at time t for employee i in
firm j; the total t consists of three observations, the first in the initial working year, the second in the
third year, and the third in 1996 (see Table 1);
π0ij is the mean log salary of employeeij at the second observation, which is coded as 0 (see discussion on
data coding in the next section);
π1ij is the parameter of mean annual salary growth rate for employeeij;
π2ij is the parameter of the accelerated rate of salary growth based on a squared term;
π3ij is the parameter of the accelerated rate of salary growth based on a cubic term;
π4ij is the parameter of a gain in employee’s technical proficiency level;
etii is the error which is independent, with a mean of zero and normally distributed with
a common variance σ2. It is assumed that each etij is independently and normally distributed with a mean of
zero and constant variance, σ2.
Equation 1 is the assumption that the growth parameters vary across employees. At L2, the
parameters of L1 (Equation 1) become outcomes variables and the L2 model represents the growth
variation due to individual factors as:
Q p
πpij = β p 0 j + Σ βpqj Xqij + rpij (2) q=1
βp0j is the intercept term for firm j in modeling the employee effect πpij;
Xqij is a measured characteristic of the individual background for employee i in
firm j (e.g., sex, formal education, OJT, or AET, etc.).
βpqj represents the effect of Xqij on the pth growth parameter; and
γpij is a random effect with a mean of zero. The set of P + 1 random effects for employee
i are assumed multivariate normally distributed with full covariance matrix, T, dimensioned (P +
1) x (P + 1).
7
At L3, a similar modeling process is repeated for firm factors. Each L2 outcome (i.e., each βpqj
coefficient) may be predicted by some firm-level characteristics as,
Spq
βpqj = γpq0 + Σ γpqs Wsj + upqj (3) s = 1 where
γpq0 is the intercept term in the firm-level model for βpqj;
Wsj is a firm characteristic used as a predictor for the firm effect on βpqj (note that each βpq may have a
unique set of the L3 predictor, Wsj s = 1, …, Spq);
γpqs is the corresponding L3 coefficient that represents the direction and strength of association between
firm characteristic Wsj and βpqj ; and
upqj is a L3 random effect that represents the deviation of firm j’s coefficient, βpej, from its predicted value
based on a firm-level model.
Note that for each form there are ΣPp = 0 (Q + 1) equations in the L3 model. The residuals from
these equations are assumed multivariate normally distributed. Each is assumed to have a mean of zero,
some variance, and covariance among all pairs of elements.
4. The survey and data set
In order to investigate the extent of OJT and AET provided to working adults and their effects,
Xiao and Tsang (1999) conducted a reverse tracer study survey in 1996. The reverse tracer technique
focuses on the analysis of employees who are currently employed in certain occupations and traces back the
education and training histories pursued by employees (Ziderman & Horn, 1995). The reverse tracer study
in Shenzhen began with the current job destinations and sought to identify each major alternative education
and training route pursued by employees to reach the current destination in the previous five years.
The 1996 survey questionnaire consisted of five groups of questions on: (a) an employee’s pre-job
formal education, (b) technological changes experienced in the workplace, (c) OJT provided by firms to
employees; (d) AET courses that an employee attended outside the firm; and (e) an employee’s position
technical proficiency level and salary at three time points over six years. This questionnaire was
administrated to a stratified random sample of 6,200 employees, slightly less than one percent of the
registered workforce in Shenzhen.2 The sample included (a) firms in both manufacturing and service
sector; (b) firms of eight types of ownership;3 (c) firms of different sizes, and (d) one or two major
production lines in the firms, which included all the personnel from managers, clerk, technicians to front
workers.
In both manufacturing and service sectors, three large-size firms (see Table 1 for definition), two
medium-size firms and one to two small-size firms in each of the 8 types of ownership were to be sampled
(96 firms). The Yearbook of Registered Firms (Shenzhen AE, 1996) in these classifications was obtained
8
from the Association for Shenzhen Enterprises. Representatives of industrial and service firms were
randomly selected. Managing directors and personnel offices were contacted about ownership, size, and
acceptance of the study in their firms. Replacement was made with another randomly selected firm when
either of the condition was not agreed. After deleting some types, which did not exist (large private service
and manufacturing, for instance), finally 76 firms were selected and agreed to participate in the study. The
survey was conducted during later 1996. Information was collected from 4,002 workers in these 76 firms,
corresponding to a 65% response rate. Among the returned questionnaires, it is found that the state-owned
firms were under-represented while corporation firms were over-sampled due to that their production lines
were big. Weights were then applied to analysis in order to arrived at estimation for a representative sample
of the overall workforce.
As the first work year in the current firm for each employees varied from 1980 to 1996, this study
selected a sub-sample of 1,023 employees who were hired by their current firm in 1990, 1991, 1992, or
1993.4 One advantage of HLM is that it allows observation time points and interval lengths to vary,
especially those associated with growth and change. With employees hired within a short span of four years
in the beginning of the third economic development stage in Shenzhen5 and observations over a time span
of a minimum of four years and maximum of six years, biases in growth rates should be reduced to a
minimum. A sub-sample with all newly employed workers allows us to examine the effects of formal
education, OJT, and AET on performance improvement and salary growth in the first few years of work.
Table 1 presents descriptions of temporal variables that reflect individual change over time for the
L1 model. The sub-sample of 66 firms, with an average of 15.5 employees in each, had three observations
of salary and gain in technical proficiency level: for the first work year, the end of the third year, and the
end of 1996 for each employee.6
OBSERVATIONS. OBSERVATIONS are the number of years that elapsed from the initial
recruiting year to the time when salary and performance data were reported (see Table 1). To avoid
collinearity in the analysis that occurred with other temporal variables in the estimation, the secondary
observation is coded as 0, and the first occasion is coded in negative number of years that elapsed from the
initial to the second occasion. The third occasion is coded in the number of years that elapsed from the
second to the third occasion.
SALARY. SALARY refers to actual monthly salary in 1996 price. Shenzhen was the first city in
China to restructure salaries to suit the market-oriented economy in 1985 (Shenzhen ESRC, 1989). Salary
consists of three components: basic salary, seniority, and position-based salary. Basic salary is mostly fixed
at RMB 75 (1985 price) for all managerial /professionals, supporting staff, as well as workers (Shenzhen
Government, 1989: 341-366). Seniority is rewarded at about RMB 10.5 for every additional year of work.
Position-based salary can correspond to at least 60 percent of the total monthly payment. It consists of two
categories, performance-based and profitability-based. Performance-based salary relates to the annual
assessment of performance of task accomplishment, including technical skills, quality, quota, attendance,
and security. Salary can be reduced if one’s job tasks are not accomplished. The profitability-based salary
9
can be zero if the firm fails to generate profits, there is no ceiling because it is based on firm profitability.
Fast economic growth in Shenzhen led to rapid profitability-based salary increases. The Shenzhen
Statistical Bureau has created salary index to indicate the rate of salary increase not due to personnel factors
(e.g., performance improvement, promotion, etc.), but due to increases in bonuses and overall firm profits.
Therefore, SALARY in this study is transformed into actual salaries in 1996 price, with both a price index
and salary index control (Shenzhen Statistical and Information Yearbook Committee, 1997:335) and log
SALARY is used in the analysis.
(Insert Table 1 about here)
PERF GAIN. PERF GAIN refers to a gain in job position’s technical proficiency level from the
previous observation time point, which is a reference for performance-based salary. In China, there are
three payroll categories: cadre/managerial/professional staff, supporting staff, and workers. Among the
three categories, managerial/professional ranks were the most prestigious followed by supporting staff and
the workers. Firms in Shenzhen, use three levels to define one’s technical proficiency: entry, intermediate,
and senior for each category. Shenzhen has adopted performance assessments to determine one’s
performance, which are carried out annually. The assessment includes the technical skills that the employee
could perform, quality of work accomplished (e.g., scrap rates), quota accomplished, ability to deal with
diverse tasks, attendance, security records, technical and innovative suggestions made during a year, attitude
and co-operativeness.7 The level of one’s technical proficiency is an official record of the assessment
results and is reflected in performance-based salary.
In this study, PERF GAIN for the initial observation at the hiring trial point is coded as zero. A
gain during the first two observation points is coded as 1, two gains codes as 2, etc., for the second
observation point. For the third observation point, any gains during the second interval of observations are
added to values at the second observation point. An outstanding performance may give one a promotion
from worker category to staff, or from staff to managerial/professional category. In such case, 4 is coded as
the former across-category promotion occurs, and 5 is coded as the latter across-category promotion occurs.
Table 2 contains individual-level variables. Shenzhen is a newly industrialized. Generally, female
employees outnumber male employees, and the workforce is very young.8 EXPERIENCE refers to whether
an employee had any work experience prior to work in the current firm. POSITION refers to the current
job position. Front-line workers are those employees who work on the floor. They can be either unskilled
or skilled workers. Supporting staff includes clerical personnel and salespersons. Managerial/professionals
staff refers to managers, senior supervisors, engineers, and senior technicians.
(Insert Table 2 about here)
CHANGE. CHAGNE refers to the amount of change employees experienced on the job. In the
10
survey, employees were asked if they had experienced changes of three kinds: the introduction of new
production technology, production of a new product, and new requirements for increased job skills. If they
experienced a single kind, they were coded 1; two kinds were coded 2; and three kinds of changes on the
job were coded 3.
EDUCATION. EDUCATION refers to formal schooling before the first job. PRE-JOB
TRAINING refers to short-term vocational/technical training before applying for the first job. ADULT
EDUCATIN refers to AET attended by employees at a local community-oriented center. TRAINING is a
dummy variable with 1 referring to having received OJT and 0 to no. Many employees had OJT more than
once. TRAINING AMOUNT ranks the amount of OJT received in the firm, with 0 for having received no
OJT, 1 for one session, and 6 for six sessions.
(Insert Table 3 about here)
Table 3 presents firm-level variables. SECTOR, and LOCATION are dummy variables. Firm
SIZE is coded in an ordered manner, with small firms coded as 0, medium as 1, and large as 2.
OWNERSHIP refers to types of investment owners. Local private and collective firms, coded as 0, state-
owned and newly created corporate firms, mostly coming from the state-owned sector, are coded as 1.
Those firms with sole-investment from Hong Kong, Macau, and Taiwan, and other countries, or joint-
ventures with firms from outside the PRC, are coded as 2. The first two types are the least formal in terms
of personnel and production management. The second two types have formal management, and the last two
types are most formal in management, in terms of recruitment, selection for training, performance
assessment, cost-analysis, and production control. This ranking accords with per employee productivity
(see Xiao, 1996b).9 TRAINING EXTENT is an aggregated proxy variable made from the overall OJT that
employees received in the firm. It is coded as 0 if less than one-third of the employees in a firm received
OJT, 1 if between one- and two-thirds had received OJT, and 2 if over two-thirds received OJT in the job.
5. Empirical results
The results are presented in three groups. An analysis of salary growth over time as a baseline
model is conducted first, in which variance is left randomly in the L2 model and L3 model. Thereafter,
follows an explanatory model that allows for an estimation of the separate effects of employee characteristic
variables on employee initial salary, the annual salary increase rate, and PERF GAIN. Finally, the L3
model estimates the explanatory effects of the firm’s characteristics variables to define employee
performance in the firm context.
5.1 Unconditional model
11
Employees mature in performance and their salaries tend to increase over time. The L1 model (see
Equation 1) estimates salary growth of an employee over time. There are no predictions in the L2 and L3
models on salary growth. This provides useful empirical evidence to determine a proper specification of the
individual growth equation and baseline statistics for evaluating the effects of subsequent the L2 and L3
models later (Bryk & Raudenbush, 1992:135). The unconditional models can partition variability in the
individual growth parameters into L2 and L3 components:
Level-2 Model,
π0ij = β00 j + r0ij (2.0a)
π1ij = β10 j + r1ij (2.1a)
π2ij = β20 j (2.2a)
π3ij = β30 j (2.3a)
π4ij = β40 j (2.4a)
and the Level-3 Model,
β00j = γ000 + u00j (3.00a)
β10j = γ100 + u10j (3.10a)
β20j = γ200 (3.20a)
β30j = γ300 (3.30a)
β40j = γ400 + u40j (3.40a)
Where πpij become the outcome variables in the L2 Model, and βpqj become the outcome variables in L3
Model. β00j is the mean salary within firm j at the initial occasion while γ000 is the grand mean salary of
1,023 employees in 66 firms at the initial observation occasion. β10j is the mean annual growth rate of
salary within firm j, while γ100 is the grand mean growth rate for all employees; and β40j is the mean rate of
performance increase within firm j, while γ400 is the grand mean rate of performance gain among 1,023
employees. β20 and β40 are the mean accelerated rates of salary increase within firm j in squared and cubic
terms respectively; and γ200 and γ400 are the grand mean accelerating rates in squared and cubic terms
respectively.
Because there are only three occasions in the observations, variance at the individual level can
only allow t-1 random variance for freedom in calculation, in this case, two Level-2 random effects, r0ij and
r1ij with variances τ00j and τ11j, respectively, and with a covariance of τ01j. Level-2 random effects represent
the deviation of employee ij’s coefficients (πpij) from its predicted value based on the individual level
model. There are three random effects, τ00j , τ10j, and τ20j in the L3 model, which represent the deviation of
firm, j’s coefficients (βpqj) from its predicted value based on the firm-level model.
(Insert table 4 about here)
12
Table 4 presents the results of salary growth with unconditional modes at Level-2 and Level-3.
The fixed effects, as presented in the top panel, indicate a strong positive salary growth trajectory. The
estimated grand mean SALARY for π0 parameter in the linear growth model represents the true average
salary of the employee at the second time point, (e.g., the third year in the current firm, which is coded as 0)
^
γ000 is ln 6.89 (about RMB 979.54; t = 246.998 and p =0.000), significantly different from that at the initial
and third observation time points and therefore, the true initial salaries are γ000 minus three times of γ100 .10
Over time, SALARY appears as an positive growing trend. The average annual SALARY increase was
estimated a 14.3 percent (i.e., γ100 ), with another increase of 7.7 percent if an employee had a gain in one’s
technical proficiency level (i.e., γ400). The estimate of the growth rate in a cubic acceleration
(γ300 < 0) indicates that salary growth is in a nonlinear function over times. The growth of SALARY over
time is a parabola in a concave downward direction.
Estimates for the random effects testing appear in the lower panel of the table, which separate the
variance in employee initial salary and mean growth rate into within-firm and between-firm components.
Residual variance at the growth level (σ2) is .0700. The residual variance among employees, τπ, is .12072
(R 0ij) for average initial salary and .00533 (R1ij) for annual growth slope. The residual variance between
firms, τβ is .03095 (U00j) for average initial salary, 0.00163 (U10j) for annual growth, and .00321 (U40j) for
FERF GAIN. The χ2 statistics accompanying these variances indicate that there is significant variation
among employees within firms in initial salary and annual growth (π0ij, and π1ij). There is also significant
variation between firms in terms of mean initial salary, annual salary increase, and gain in proficiency level
(i.e., β00j, β10j and β40j). Analysis at L2 should explain these variances.
Table 4 shows that the reliability for the initial salary intercept is .760, based on an average of 15.5
employees in each firm, and .451 for the annual salary increase slope, based on the three temporal
observations. Due to the small number of observations, the annual salary increase appears to be less
reliable, but it is still fairly acceptable. The reliability of the initial salary intercept (i.e., .580) and the
annual increase rate intercept (i.e.,0.466) at the firm level are acceptable; but the reliability of the
performance improvement intercept (i.e., .351) seems a bit low with only three observations, but still can be
considered to be fair.11
5.2 Conditional L2 model with individual characteristics as predictors
Now, the L1 model being the same as in Equation 1, the parameters in L1 model become outcome
variables in the L2 Model, and their variability will be predicted by the employee characteristic variables.
The specific L2 model is:
π0ij = β00j + β01j (SEX)ij + β02j (EDUCATION)ij + β03j (POSITION)ij
+ β04j (EXPERIENCE)ij + β05j (PRE-JOB TRAINING)ij + r0ij (2.0b)
π1ij = β10j + β11j (AGE)ij + β12j (EDUCATION)ij + β13j (POSTION)ij
13
+ β14j (ADUTL EDUCATION)ij + β15j (TRAINING)ij + r1ij (2.1b)
π2ij = β20j (2.2b)
π3ij = β30j (2.3b)
π4ij = β40j + β41j (EDUCATION)ij + β42j (CHANGE)ij + β43j (TRAINING6)ij
+ β44j (ADULT EDUCAITON)ij (2.4b).
It is hypothesized (1) that sex, years of formal education, job position assigned, work experience, and pre-
job vocational/technical training are related to initial salary; (2) that age, formal education, job position,
adult education, and training are all related to annual salary increases over time; and (3) that sex, formal
education, amount of change experienced in the job, amount of on-the-job training received within the firm,
and adult education pursued outside the firm are all related to the performance gain. We leave the L3
model still unpredicted at this stage.
Table 5 shows the results with the individual explanatory variables in the L2 model. The
SALARY parameter remains positive and significant. SEX, FORMAL EDUCATION, and POSITION
show a significantly positive effect on the grand mean SALARY, the initial hiring salary, but
EXPERIENCE and PRE-JOB TRAINING do not.
^ On average, an employee earned ln RMB 6.514 (γ000, t = 77.319 and p =.000) at the third year of
working in the firm (about RMB 674.56). However, male employees on average start with 5.9 percent
(γ010) more than their female counterparts, indicating a difference in the initial salary between male and
female employees, all other things being equal. Formal education has a significant positive effect on being
hired. For every additional year of education, one can receive about 2.37 percent (γ020) more initial salary
when hired by a firm. This means that formal education is an important factor that affects employers’ hiring
decisions. The assigned job position (γ030) has a significantly positive effect on initial salary. Support staff
are likely to received 10 percent more salary at time of hire than front-line workers, and
professional/managerial staff receive another 10 percent more.
(Insert Table 5 about here)
Work experience (γ040) before coming to the current firm does not have an effect on initial salary.
As discussed earlier, most jobs in Shenzhen resulted from the economic changes, thus they were new jobs.
Previous work experience may have been less relevant for the current jobs. Prospective employees and
employers often did not find the right match for the specific jobs in the market. We will see below that on-
the-job training has an effect on current job skill improvements. “Vocational training before hiring” became
a national labor policy from 1993 (China CC & SC, 1993). Some employees received pre-job vocational
training before coming to the firm. However, the coefficient indicates that, PRE-JOB TRAINING (γ050) did
not have a job preparation effect on the firms’ hiring decision.
14
The salary increase parameter (γ100) had an average of 18.7 percent slope, which is substantially
high. Age group (γ110) had no effect on the annual salary increase. Shenzhen has a very young workforce
and the salary reform favors those with job competence instead of seniority or age. Though formal
education (γ120) had a strong positive effect on one’s initial salary, it showed a negative effect of 0.45
percent less salary increase for employees with every additional year of formal education. In Shenzhen,
salary scheme is set up in such a way (see discussion on data set) that formal education only counts in
computing the fixed basic salary and employers can decide on the salary increase based on performance.
This statistic reflects the effect of the implementation of such salary scheme. It indicates that employees
with less formal education tend to have a higher annual salary increase slope than those with more years of
education do though they have a higher salary at the time of hiring. The position assigned (γ130 = .0129, t
=2.624, p = .009) had a significantly positive effect on the annual salary increase. Support staff received a
1.3 percent more annual salary increase and managerial/professional staff received another 1.3 percent more
than front-line workers. Therefore, the positions, as determined by the firm, have an effect on salary. Adult
education (γ140) selected by employees outside the firm shows no relation to the annual salary increase, nor
does on-the-job training (γ150) have an effect on the annual salary increase. Although the actual salary is
used to control for salary inflation, the annual salary increase is still substantial. Nevertheless, the annual
increase is not due to formal education, on-the-job training, or adult education per se.
EDUCATION (γ410), CHANGE (γ420), TRAINING AMOUNT (γ423), and ADULT EDUCATION
(γ440) were used to predict PERF GAIN (π1ij). First, formal education prior to the job (γ410) offered marginal
or no effect on gains in firm recognized proficiency level (γ410 = .00655, p = .056). The amount of change
experienced in the job (γ420 = .0122, p = .007) showed a positive and significant association with a gain in
proficiency level, about a 1.2 percent increase in salary for every one more change experience in the job.
This indicates that change promotes performance improvement through learning by doing. The amount of
OJT (γ430 = .013, p = .005) also showed a positive and significant association with a gain in proficiency
level, thus associated with increased salary. With each OJT received in the workplace, employees are likely
to obtain some gain in job’s technical proficiency level, in association with 1.3 percent salary increase.
AET received outside the firm showed no association with job performance. AET chosen by employees
outside the firm may largely suit employees’ individual needs, or needs perceived by employees themselves.
Such needs may not be relevant to specific job requirements, though AET can be a means to change jobs.
Thus, those who received AET showed no gain in firm recognized technical proficiency level, thus no
salary increase.
With CHANGE and TRAINING AMOUNT as predicting variables, the PERF GAIN slope now
becomes non-significant (γ400 = -.0471, t = -1.129, p = .260), which means that statistically, CHANGE and
TRAINING AMOUNT have explained all the variations of PERF GAIN among individuals within firms.
CHANGE and TRAINING AMOUNT are powerful predictors and have contributed to explain a gain in
firm recognized technical proficiency in association with salary increase in the positive direction. OJT and
15
experience of change in the workplace are associated with increased salary through improved performance,
as recognized by firms with a gain in proficiency level.
In short, formal education and job position largely determine the initial salary. Sex discrimination
exists; females receive 5.9 percent less initial salary than their male counterparts. Annual salary growth is
substantial and also due to the effect of position; formal education had no effect on annual salary increases.
A gain in proficiency is strongly associated with increased salary, but such gains are associated with an
employee’s experience of change in their firms, a type of learning-by-doing in technical changes in the job,
and with receiving OJT.
It is interesting to note that formal pre-job education has a positive effect on initial salary, but no
positive effect on annual salary increase, nor on improved job performance. This supports the Spence’s
(1973) argument that in hiring, formal education provides a signal in the labor market that employees with
more education might have higher productivity. With respect to productivity, more educated employees can
be trained at a lower cost (Thurow, 1975). Therefore, only the initial salary is in accord with formal
education.
(Insert Table 6 about here)
For random effects testing, Table 6 shows that the residual variance among employees (R0ij) is
now reduced from 0.12072 in the baseline model to 0.10754 for the average initial salary, and from .00533
to .00508 for annual growth (R1ij). About 11 percent of the variance in initial salary and 4.7 percent of the
variance in annual salary growth among employees within firms were explained by predictors in the L2
model. The residual variance is .03095 (U00j) for the average initial salary between firms, 0.00163 (U10j) for
annual growth, and .00321 (U40j) for PERF GAIN in the baseline model. With the individual predictors
included in the L2 model, about 31 percent, -11.7 percent, and 44.2 percent of variance at the firm was
explained in association with the personal characteristics of employees. For the variance in mean growth
(U10j) across firms, that the variance among individuals within firms (R1ij) was reduced made the variance
between firms appear to be larger. The variance due to firm impact, which was confused, is decomposed by
the individual variables.
It is interesting to note that the individual characteristics of employees explain more for the firm-
level difference than for the individual-level difference. This indicates that individual variables affect salary
due to the employees being nested in a firm and due to firm decisions about individuals. Thus, productivity
is largely a workplace characteristic; and individual characteristics promote productivity in a collective
manner according to how production is organized in the firm context rather than in an individual manner.
Nevertheless, the corresponding χ2 statistics accompanying these variance components remain
significant, except variance for PERF GAIN (U40j). This indicates that there is still significant variation
(R0ij and R1ij) among employees within firms in terms of mean initial salary and mean annual growth as well
16
as significant variance (U00j, U10j ) between firms for mean initial salary, and mean annual salary increase.
Individual predictors in the L2 model explain all the variance of PERF GAIN (U40j).
5.3 Conditional L3 model with firm characteristics as predictors
The L1 model and L2 model being the same, the L3 model will use firm characteristic variables to
predict six Level-2 parameters and to present their variability between firms. In the following equations,
the firms’ characteristic variables serve as explanatory variables to define the association of employees’
performance in the firm context:
β00j = γ000 + γ001 (SIZE)j + u00 j (3.00c)
β01j = γ010 + γ011 (OWNERSHIP)j (3.01c)
β02j = γ020 + γ021 (SECTOR)j (3.02c)
β03j = γ030 (3.03c)
β04 j = γ040 (3.04c)
β05 j = γ050 (3.05c)
β10j = γ100 + γ101(TRAINING EXTENT)j + γ102(OWNERSHIP)j + u10 j (3.10c)
β11j = γ110 (3.11c)
β12 j = γ120 (3.12c)
β13j = γ130 (3.13c)
β14j = γ140 (3.14c)
β15j = γ150 (3.15c)
β20j = γ200 (3.20c)
β30j = γ300 (3.30c)
β40j = γ400 + u40 j (3.40c)
β41j = γ410 (3.41c)
β42j = γ420 + γ421(SIZE)j + γ422 (LOCATION)j + γ423 (SECTOR) j
+ γ424 (OWNERSHIP)j (3.42c)
β43j = γ430 + γ431 (SIZE)j + γ432 (LOCATION)j + γ433 (SECTION)j
+ γ434 (OWNERSHIP)j (3.43c)
β44j = γ440 (3.44c).
It is hypothesized these firm level predictors (Wsj) have effect on their corresponding level 2
parameters (βpqj).
(Insert Table 7 about here)
17
Table 7 presents estimations of the L3 model. Look at the fixed effects first. When hiring
employees, firms of a smaller size (γ001) pay about 9 percent more than larger size firms.12 Small firms are
usually at the periphery in the market and their working conditions are far less attractive. Consequently,
they offer higher salaries to attract prospective employees but their annual salary increase rate does not
differ from other firms. Changes occurred with respect to the effect of sex on the initial salary. With firm
ownership as a predictor, it was found that male employees in the more formally managed state-
owned/corporate firms received on average about 10 percent more than female employees. Male employees
in firms with outside investment on average received another 10 percent more.13 Generally speaking, males
received higher salaries than females. Across firms of different ownership, firms with investment from
outside China ranked females at bottom and males at the top in terms of salary. This indicates that
discrimination in salary exists and is most prevalent in those firms with investment from outside China.
Regarding the firms’ ability to pay, the local private/collective firms ranked the lowest and this is believed
to be associated with productivity (see Endnote 9). Though hypothesized, sectors had no difference in
hiring preference with respect to formal education.
Regarding the salary growth slope (γ100), the capacity to provide OJT (γ101) to employees did not
have a direct effect on the annual salary increase. This is consistent with estimation for OJT received by
employees at an individual level (β15j). However, when OWNERSHIP (γ102) is considered, employees in
state-owned/corporate firms tend to have a 2.5 percent higher annual salary increase than those in local
private/collective firms. Employees in firms with investment from outside China tend to have another 2.5
percent higher annual increase. These estimations reflect findings in previous studies (Xiao, 1996a; see
Endnote 13). Well-managed firms tend to generate much higher productivity and thus are able to provide
higher salary increase.
Now consider PERF GAIN and its effect on salary with firm-level variables.14 SIZE (γ421),
LOCATION (γ422), and SECTOR (γ423) did not show any effect on the amount of technical change (β42) that
employees experienced in the job. However, the firm’s ownership (γ424) did have an effect, indicating that
state-owned/corporate firms presented individual employees with fewer changes in the job. In the same
manner, firms with investment from outside presented fewer changes compared to state-owned/corporate
firms.
Regarding the amount of OJT that an employee received (β43), SIZE (γ431) and LOCATION (γ432)
had no effect. SECTOR (γ433 = -0.0189, t = -2.140, and p = 0.032) showed a negative but significant effect
on the amount of OJT that employers provided to employees. It suggested that service industries provided
less training while manufacturing provided more training, something associated with a gain in proficiency,
thus salary increases. Firms in manufacturing were at the forefront of economic development in Shenzhen
through the introduction of new production technology [see (Li, 1995); (Liu, 1985, 1992)]. Provision of
OJT enables employees to continue to learn new skills. Though firms with investment from outside and
state-owned/corporate firms presented fewer technological changes to individual employees, they provided
18
more OJT (γ434 = 0.015, t = 1.981, and p = 0.047), suggesting a positive association with skill improvement,
thus about a 1.5 percent increase in salary.
All these findings about improved performance due to technological changes and OJT suggest an
interesting relationship. For instance, OJT on its own did not contribute to a salary increase, either at the
individual (β15 = 0.01 and p =.173) or at the firm level (γ101 = 0.032 and p = .251). OJT contributed to a
salary increase through firm-recognized gain in proficiency level (β43), indicating that firms did associate
performance with salary decisions. Ownership has an effect on changes in experienced by employees and
on the provision of OJT. This indicates that firms are rational in organizing their production, considering
both individual and firm factors, which in turn have effects on skill improvements. Firms in the
manufacturing sector, the spearhead of Shenzhen development strategy--to accumulate capital and
technology through manufacturing--provided more OJT to their employees. It is suggested that employers
have used workplace training as a complementary strategy to upgrade human capital in abreast of replacing
of physical capital, thus improving productivity. The findings of this analysis indicate that receiving OJT
and learning-by-doing have a strong and positive association with firm recognized skill improvement, and
thus subsequent salary increases.
The lower panel of Table 7 shows the tests of the random effects χ2. With the firm-level variables
added in the L3 model, the corresponding χ2 statistics accompanying the variance components among
employees (R0ij and R1ij) and between firms (U00j and U10j) still remain significant (Ps < 0.000). This
indicates that there is still significant variation among employees within firms in terms of mean initial salary
and mean annual salary growth as well as between firms in terms of mean initial salary and mean annual
salary increase.
Table 6 presents a comparison of the explanatory power of the three-level models. For variance
between firms, L3 Model predicts another 20 percent of the variance in the mean initial salary difference
(U00j), another 25 percent in the salary increase slope (U10j), and 27 percent in gain in proficiency (U20j). In
total, the three-level models explain 46 percent of the variance between firms for initial salary, 16 percent of
the variance for annual salary increase, and 60 percent of the variance in gain in technical proficiency. By
contrast, for the within-firm variance, 12 percent and 5.4 percent were explained for the initial salary and
the annual salary increase slope, respectively. Rowan, Raudenbush, and Kang (1991:261) point out that
between-unit variation in HLM models is accounted for by unit-level as well as by within-unit individual
characteristics. Given that individuals are nested in their firms, firm production is an interactive process:
employees contribute to productivity with their characteristics and the firms organize production in a way
that makes use of employees’ potentials. Therefore, studying individuals in their social context helps to
explain how employees behaves due to both their own individual factors as well as the firm factors.
HLM also allows us to detect proportions of variation existing among employees within firms and
among firms. The low panel of Table 6 shows that in the baseline model, 80 percent of the variance in
initial salary lies among employees within firms while 20 percent of the variance lies among firms. For the
19
growth slope, 77 percent of the variance in salary lies among employees within firms and 23 percent across
firms. As the individual variables and firms variables were put into L2 and L3 models, variation among
firms tends to decrease while that among employees tends to increase.
6. CONCLUDING DISCUSSION
This study analyses how three different human capital development strategies contribute to salary
growth over time. In the workplace, employees are nested in their firms; individual characteristics are
promoted or constrained as the firm makes production decisions. If this process is neglected in estimates of
rates of return, for instance including only individual variables and pre-job characteristics, the human
capital concept will be flawed theoretically.15 This is because other alternatives that impart skills and
knowledge to individuals are omitted and other complements in the workplace are neglected. The most
important of these neglected conditions involves the process of how knowledge and learnt skills is
transferred into productivity, through many complementary and necessary conditions, in the context of firm.
The analysis in this study has attempted to discover what is going on in this “black box” by revealing
several important findings in the context of the emerging economy of Shenzhen. I argue that qualitative
changes in a transitional economy create disequilibria in the workplace. As changes accelerate and become
a constant, OJT and AET serve as complementary strategies to regain equilibrium. OJT provided in the
workplace strengthens a firm’s capacities in a market economy, while AET suits individual expectations of
the external market. Therefore, both OJT and AET engage employees in the changing economic process
and upgrade their human capital. The major findings of this study support this argument and can be
summarized by the following points.
First, formal education has a significant impact on employers’ hiring decisions, and the initial
salary. However, additional years of formal education are not associated with annual salary increases or
with technical proficiency improvements recognized by the firm. Second, OJT provided by employers in
the workplace does not automatically contribute to annual salary growth at either the individual or the firm
level. OJT contributes to salary increases only through firm-recognized improved job performance, which
is positively associated with an increase in salary. Firms in Shenzhen associate job performance salary.
This finding shows that OJT has a positive impact on productivity. Third, firms with more formalized
management, such as firms with investment from outside China and state-owned/corporate firms, provide
more OJT to their employees. These firms are able to pay higher salaries, both at the initial hiring point and
during annual increases; but they are associated with gender discrimination to female employees at hiring.
Fourth, manufacturing firms, which are at the forefront of economic development, provide significantly
more OJT to match their human capital with physical capital. This indicates that firms are rational in
utilizing OJT to keep abreast of changes in the workplace. Fifth, technological changes in the workplace
provide as learning-by-doing and contribute to job performance improvements, thus they are associated
with salary increases. Nevertheless, compared with more formally managed firms such as those with
20
investment from outside China and state-owned/corporate firms, local private/collective firms tend to let
their employees experience more job changes and they are less likely to provide their employees formal
OJT. Their ability to pay is also the lowest both at the hiring point and during the annual salary increases.
Finally, voluntary AET outside of the firms by individual employees does not have an impact on job
performance or on salary increases. While AET may suit individual expectations of the external market,
firms do not yet associate it with salary decisions.
The findings on formal education and OJT are in accord with Spence’s (1973) assumption that
formal education is a signal of potential productivity, which Thurow (1975) refers to as trainability at lower
cost. The findings also indicate that employers are rational in matching individual characteristics to the
changing characteristics of the firms: they provided OJT to close the skill gap when changes resulted in
disequilibria. They also associate OJT with salary through firm-recognized job proficiency. Learning
through OJT in the workplace is located in the social context of the firm, a collective and intentional
process. Learning in such a mode makes content relevant, thus giving performance application and
meaning (Xiao, 1999). The association of OJT with firm-recognized job proficiency further creates
incentives for the transfer of training. The findings on ownership, which is a proxy for the formality of
management, show that a firm with formal management can provide higher pay and more OJT. All these
findings confirm that complementary inputs in the workplace promote the transfer of training. Organized
learning, OJT in this study, in the changing context of appropriate technology is found to improve skills and
increase salary.
With respect to formal vocational/technical education, Xiao and Tsang (1999, Table 3), found that
secondary vocational/technical graduates and tertiary graduates did not receive less on-the-job training than
general education graduates. Formal VTE is expensive and costs more than general education (Tsang,
1997). Furthermore, research findings from China have thus far not favored VTE graduates (Lo and Lee,
1996; Yang, 1997; Min & Tsang, 1990). These studies suggest that the Chinese government would be well
advised to re-examine its national education policy, which favors a western pattern of formal vocational
education, whereby VTE enrolment constitutes 50 per cent or more of the upper-secondary level (China
SEC, 1996).
Because of the contextual nature of the workplace, it is doubtful that formal education will be able
to accommodate the specific, ever-changing demands for workplace competence. It is thus not surprising
that this study found no effect of formal education on specific job skills improvement. Moreover, the
human resource strategy of continuing learning that has been observed in the early industrialization process
in the developed world (Gospel, 1991) has also been observed at different stages of development during the
twentieth century in China: from the early revolution movement in the 1920s through the socialist
modernization program in the 1950s and early 1960s (Zang, 1985), as well as in other nations at the present
time (see discussion in the introduction section). Continuing learning is a strategy incorporated in the
process in development, even though it has been marginalized in educational investment policies. The
literature findings suggest that learning in the workplace, as in Shenzhen, does not supplement gaps in
21
education credentials due to the absence of formal education, as was also the case in the early history of
adult education in China (Xiao, 1998b). The continual upgrading of human capital in order to reduce
discrepancies due to constant changes in the workplace has been an effective strategy in Shenzhen to regain
an “equilibrium”. The experience of Shenzhen is relevant to other fast changing regions in China.
There are two important policy implications in the Chinese context. As part of the socialist
tradition, adult education and workplace learning have played a dual role in the education system in China.
It has attempted to close gaps in education among the population since 1949. Due to the recent focus on
economic development adult learning now plays an important, complementary role to upgrade job skills in
the workplace (Xiao, 1999; this study). Considering relevance and efficiency, education and training for
working adults should be integrated into social policies, along with an expansion of formal education. In
the last two decades, economic reforms in China have focused on macro policies. Improvements in firm
management have received little attention. Because firm polices are complementary to the utilization of
human capital and thus productivity, efforts should be made to improve firm-level management, thus
promoting the transfer of training.
Given that the purpose of this study is to survey the educational and training histories of employees
in firms, it does not examine the personal characteristics developed through formal education and/or
associated with family background as well as the impact of AET on other aspects of one’s life (e.g.,
changing one’s job). Much variance at the individual level is still unexplained. Since an ability to deal with
disequilibria is critical to increasing productivity, as argued by Shultz (1975), future studies should explore
how employees with different personal characteristics deal with disequilibria in the workplace. Levin
(1987) suggests that individual discretion in the workplace is important to allow educated employees to
allocate resources and improve productivity. Therefore, it is important to examine if the involvement of
employees in decision making will improve productivity in China’s new market economy. These factors
might reflect how individuals behave in the workplace, thus explaining more variance at the individual
level.
Acknowledgements This study is part of a research project entitled “Evaluation of the External Efficiency of Vocational/technical Education in Shenzhen” funded by the Chinese University of Hong Kong’s Direct Grant (SSEP AC No. 2020287). The author is indebted to Professor Brian Rowan, School of Education, University of Michigan, for providing the analytical method as well as suggestions for revision. The author also acknowledges the collaborative efforts of the Shenzhen Institute of Educational Research, especially the assistance of Ms. Zhao Peifeng in data collection; Dr. Chan Wing Shing, of the Education & Manpower Bureau, Hong Kong Special Administrative Region, for technical advice; Dr. Michael Agelasto for commenting on the manuscript; and Ms. Nancy Hearst, Fairbank Center for East Asian Research, Harvard University for copyediting. Finally, thanks go to the reviewers and editor of this journal for suggestions and comments.
22
Notes 1 Some studies have allowed for endogeneity of training decisions (Heckman, 1979), occupational changes (Greenhalgh and Stewart, 1987), and experiments in participation and non-participation in training (Lalonde, 1986). 2 The total workforce in Shenzhen is 2.45 million. Among it, there are 0.46 million town individuals, 1.1 million village labor force, 0.89 million registered employees, and .05 million others. Firms and organizations that registered their employees with the government labor department or personnel department hire the third group. They are considered as registered workforce. 3 These eight types of ownership refer to state-owned, collectively owned, joint-venture with firms from Hong Kong and Macau, and Taiwan, joint-ventures with firms from other countries, sole-investment firms from Hong Kong, Macau, and Taiwan, sole-investment firms from other countries, local private firms, and corporate firms. There first two types are considered typical in the planned economy and the six latter ones are new forms of ownership in the transferred economy after 1980. 4 The minimum individual cases within a unit for analysis with HLM are 12. On the average, for the total sample, each production line/group contained 52 employees. Each year, a firm would hire a few new employees, about five to six in this sample in the early 1990s. If only one year is used as a sub-sample, there are not enough individual cases for a reliable estimation of a firm unit. Therefore, a sub-sample with employees recruited in four years meets the statistical requirement of at least 12 individual observations in a unit. Eleven firms with less than 12 observations were automatically excluded from the analysis. 5 Development in Shenzhen Special Economic Zone took place in three stages. The first stage (1980 to mid-1980s) was the take-off stage, characterized by a priority on infrastructure construction to attract investment. The second stage (mid-1980s to late-1980s) was devoted to expanding labor-intensive manufacturing for the international market to build up an industrial base and accumulate capital. Since 1990, the development strategy has shifted from manufacturing toward a diversified economy, with an increased emphasis on services (Li, 1995; Liu, 1985, 1992). 6 For data sets arranging, please see Appendix A for “Determinants of salary growth in Shenzhen, China” at www.fed.cuhk.edu.hk/eap/people/xiaoj.html. 7 Performance assessments were conducted systematically in those firms with investment from outside China, the state-owned and newly created corporate firms. And decisions on skill proficiency were largely made on technical grounds. In some firms, particularly the local collectives and private firms, performance assessment were not well conducted. In such case, non-technical factors as subjectivity, personal relation or even political consideration (mostly in the state-owned firms) might make an influence. Nevertheless, PERF GAIN is the best available indicator an employee’s job performance in China. 8 The average age was 28.7 years for the 1 million permanent residents and 26 years for the 2.5 million temporary residents in 1996 (Shenzhen SIYC, 1997: 111-112). 9 Xiao (1996b) finds that in 1992 firms with investment from the outside ranked highest in per employee industrial product (RMB 74,357.00 in 1978 price); while state-owned firms ranked second (RMB 29,581.00); corporations ranked third (RMB 11,949.00), and collective and local private firms tanked last (RMB 3,262). Productivity is due to management practices, and training, all else being the same (Xiao, 1996a). 10 The intercept parameter, π0, is the true salary of each employee at some fixed time point. The specific time point depends on the scaling of the observations. In this model, the second observation time point is coded as 0 to avoid collinearity with other temporal variables.
23
11 It is important to examine the reliability of the ordinary least squares (OLS) estimate at individual and firm levels and the correlation among the growth parameters. The reliability of L2 and L3 outcome variables will help ensure that the data can detect systematic relations between growth parameters and personal-level variables, and between personal and firm variables (Bryk & Raudenbush, 1992: 69, 137, 177-178). For each employee ij at L2, ^
reliability (π 0ij ) = τπ / [ τπ + σ2/tjk] (5) is the reliability of individual mean for use in discrimination among employees within the firm. For any firm j at L3,
^ τβ reliability (β 00j ) = (6)
τβ + {Σ [ τπ + σ2/tjk] –1 }–1
is the reliability of the firm’s sample mean as an estimate of its true mean. The averages of these reliabilities across employees and firms (Equation 6) provide summary measures of the reliability of the employee and firm means, respectively. With multi-wave data, the correlation of the true initial status and true change can be obtained with more consistence than a simple pre-test and post-test design or two points comparison (Bryk & Raudenbush, 1992:137-138). Under a linear individual growth model, this correlation is just the correlation between π0ij, and π1ij,
^ ^ ^ ^ ρ (π0ij, π1ij ) = τπ01 / [ τπ00 + τπ11 ] 1/2 (4)
For the salary data, the estimated correlation between initial salary status and true annual growth rate was –0.122. This means that employees whose initial salary was low when entering the firm tended to have a salary increase at a mildly steeper slope over time. This can be inferred as a true negative relationship and not a spurious result of the measurement process. 12 Size is not included in the model because it is not significant for predicting salary growth slope, γ100 because it reduced the power to explain the variability. Appendix B portrays salary trends predicted by firm size, which is available at www.fed.cuhk.edu.hk.eap/people/xiaoj.html. 13 Appendix C delineates the estimation trends and is available at www.fed.cuhk.edu.hk.eap/people/xiaoj.html. 14 To test if association of PERF GAIN with salary increase is true, a three-level model was run, excluding PERF GAIN, thus all its L2 and L3 predictors. Estimations of variables remaining the essentially the same, σ2 increased by 2 per cent, R0ij increased by 8 per cent, R1ij increased by .8 per cent, U00j by 62 per cent and U10j by 9 per cent. Excluding PERF GAIN only increased variance across firms for grand mean initial salary (γ000). This shows the PERF GAIN does not have collinearity with OBSERVATIONS, and its L2 and L3 predictors shared little “the same effects” with other L2 and L3 predictors for other L1 parameters. 15 A comparison of the Mincerian method is provided with the third salary data in 1996 (see Appendix D at www.fed.cuhk.edu.hk.eap/people/xiaoj.html). The ANOVA only provide variance in a single sum. It is also observed that with only one year of salary data, sex as a predictor appeared not to be significant, contrary to the estimations of the three-level salary growth model. The three-level model finds significant effect of gender on salary decisions (see Tables 6 and Table 8) and it tends to increase the disparity as portrayed in Appendix C for the salary growth trajectory. With only one year of salary data, the estimation of the Mincerian method failed to detect such a gender effect on salary.
24
REFERENCES
Becker, G. S. (1964). Human capital. New York: Columbia University Press. Belanger, P. & Tuijnman, A. (1997). The “silent explosion” of adult learning. In: P. Belanger, & A.
Tuijinman (Ed.), New patterns of adult learning: A six-country comparative study (pp. 1-16). Oxford: Pergamon.
Bryk, A. S. & Raudenbush, S. W. (1992). Hierarchical linear model: Applications and
data analysis methods. Newburg Park, CA: Sage Publications.
Chapman, P. G. (1993). The economics of training. New York: Harvester Wheatsheaf. China Central Committee and State Council [CC & SC] (1993). Guideline on China’s education reform
and development decisions on strengthening employee Training. Beijing. China State Education Commission [SEC] (1996). The ninth five-year plan of the national education and development plan for 2010. In: Policy report of State Education Commission No 6 (pp. 219-227). Beijing: China State Education Commission Cohn, E. & Addison, J. (1998). The economic returns to lifelong learning in OECD countries. Education
Economics 6 (3), 253-307. Gospel, H. F. (1991). Industrial training and technological innovation: A comparative and historical Study. London and New York: Routledge. Greenhalgh, C. & Stewart, M. (1987). The effects and determinants of training. Oxford
Bulletin of Economics and Statistics 49 (2), 171-190. Hall. R. E. & Jones, C. I. (1999). Why do some countries produce much more output per worker than
others? The Quarterly Journal of Economics 114 (1), 83-116. Heckman, J. J. (1979). Sample selection bias as a specification error. Econometirca 47, 153-162. Lalonde, R. J. (1986). Evaluating the econometric evaluations of training programs. The Amercian Economic Review 76 (4), 604-620. Levin, H. M. & Kelley, C. (1994). Can education do it along? Economics of Education Review
13 (2), 97-108.
Levin, H. M. (1987). Improving production through education and technology. In G. Brude, and R. W. Rumberger (Ed.), The future impact of technology on work and education (pp. 194-214). London: The Falmer Press. Li, Q. S., (1995). Foreign investment on a large scale. In Shenzhen special economic zone yearbook 1995
(pp. 88-91). Shenzhen: Shenzhen Special Economic Zone Publishing House, Shenzhen. Liu, G. G. (1985). Research on development strategies of Shenzhen special economic zone. Hong Kong: Hong Kong Economic Guide Press and Shenzhen Research Centre for Economics. Liu, G. G. (1992). Strategies of Economic Development for Shenzhen Special Economic Zone in 1990s: Building up a prototype of socialist market economy and an internationally integrated city. Beijing: Economic Management Press.
25
Lo, L. N. K., and Lee, C. H. (1996). In rural China: which road to relevant education? Educational Leadership 35 (8), 60-63.
Middleton, J., Ziderman, A., & Adams, A. V. (1993). Skills for productivity: Vocational education and
training in developing countries. New York: Oxford University Press. Min. W. F. & Tsang, M. C. (1990). Vocational education and productivity: A case study of the Beijing
General Auto Industry Company. Economics of Education Review 9 (4): 351-364. Mincer, J. (1974). Schooling, experience, and earnings. New York: Columbia University Press.
Psacharopoulos, G. (1985). Returns to Education: A further international update and implications. Journal
of Human Resources 20, 583-604.
Psacharopoulos, G. (1994). Return to investment in education: A global update. World Development 22 (9), 1325-1343.
Psacharopoulos, G. & Woodhall, M. (1985). Education for development: Analysis of investment
choices. New York: Oxford University Press. Rowan, B, Raudenbush, S. and Kang . J. (1991). Organizational design in high schools: A multilevel
analysis. American Journal of Education 99 (fall), 238-266. Schultz, T. (1975). The value of the ability to deal with disequilibira. Journal of Economic
Literature 13 (3), 827-846. Shenzhen Association for Enterprises [AE] (1996). Yearbook of Shenzhen registered firms. Shenzhen.
Shenzhen Economic System Reform Commission [ESRC](1989). Section Five: Labor, salary,
and security system. Shenzhen special economic zone selected documents on reform, (pp.333-363).
Shenzhen Government (1989). Salary reform scheme for Shenzhen state-owned and non-profit Units (1984); Salary reform scheme for manufacturing firms and factories (1984); Decisions on firm salary reform (1985); Temporary ordinance for salary tax in Shenzhen Special Economic Zone (1985). In: Shenzhen Economic Reform Commission, Selected documents of Shenzhen special economic zone (pp. 341-366).
Shenzhen Statistical and Information Yearbook Committee[SIYC] (1997). Statistical and
Information Yearbook of Shenzhen. Beijing: China Statistics Press. Spence, M. (1973). Job market signaling. Quarterly Journal of Economics 87 (3),
355-375.
Thurow. L. (1975). Generating inequality. New York: Basic Books, Inc. Tsang, M. C. (1987). The Impact of underutilization of education on productivity: A case study of the U.S. Bell Companies. Economics of Education Review 6 (3): 239-254. Tsang, M. C. (1997). The costs of vocational training. International Journal of Manpower 18 (1/2): 63-89. Xiao, J. (1996a). A study of the relationship between organizational factors and the transfer of
training in the electronics industry in Shenzhen, China. Human Resource Development Quarterly 7 (1), 55-73.
Xiao, J. (1996b). Multi-ownership, management and productivity: A case of Shenzhen. Paper
26
presented at the International Conference on Co-ordinated Development Among Regions, in China’s Economic Reform and Social Development, Dec. 11-12, 1996, Hong Kong. Available in the conference proceedings (pp. 421-431).
Xiao, J. (1998a). Education expansion in Shenzhen, China: Its interface with economic development. International Journal of Educational Development 18 (1), 3-19. Xiao, J. (1998b). Higher adult education in China: Redefining its roles. In: M. Agelasto & B. Adamson (Ed.), Higher education in the post-Mao era (pp. 189-210). Hong Kong: Hong Kong
University Press. Xiao, J. (1999). Alternative learning approaches in an Emerging Economy: An Experience of Shenzhen, China. Educational Practice and Theory 21 (1), 27-49. Xiao, J. & Tsang, M. C. (1994). Costs and financing of adult education in Shenzhen, China
International Journal of Educational Development 14 (1), 55-73. Xiao, J. & Tsang, M. C. (1999). Human capital development in an emerging economy:
The experience of Shenzhen, China. China Quarterly 157 (March), 72-114. Yang, J. (1997). The interaction between the socialist market economy and technical and vocational
education and training in the People’s Republic of China. Ph.D Dissertation, University of Manchester and Bolton Institute, U. K.
Zang, Y. C. (1985). China’s history of employee education [Zhongguo zhigong jioayu shigao].
Lianning People’s Press. Ziderman A. & Horn, R. (1995). Many paths to skilled employment: A reverse tracer study of seven occupations in Colombia. Education Economics 3 (1), 61-79.
Table 1 Descriptions of level-1 variables Variable Definition Mean Std Dev Min. Max. N OBSERVATIONS: Years elapsed between observation time points
entry year 0.00 0.00 0.00 0.00 1023 the second point 2.99 0.08 2.00 3.00 1023 the third point 5.25 1.01 4.00 6.00 1023
SALARY: Actual salary observations in 1996 price (RMB) entry salary 793.76 562.63 150.63 4752.60 1023 salary at the second point 1191.64 875.45 187.44 7764.15 1023 salary at the third point 1625.75 1182.90 199.19 8872.78 1023
PERF GAIN: Gain in job performance between observation time points (level) entry year 0.00 0.00 0.00 0.00 1023 the second point 0.94 0.96 0.00 4.00 1023 the third point 2.26 1.02 1.00 5.00 1023
Table 2 Descriptions of level-2 variables (1,023 employees)
Variables Definition Coding % SEX Sex
female 0 51.1 male 1 46.6
missing missing 2.1 AGE Age in group
16-25 0 52.8 26-35 1 34.6 36 and above 2 12.7
EXPERIENCE Had work experience before the current job no 0 58.0 yes 1 42.0
POSITION Types of current positions front-line workers 0 56.7 support staff 1 19.1 managerial/professional staff 2 24.2
CHANGE Extent of technical changes experienced in job none 0 29.6 once 1 14.6 two times 2 22.6 three times 3 33.2
EDUCATION Education attainment in years missing missing 1.1 9 years 9 25.1 12 years 12 58.7 13 years 13 3.6 14 years 14 3.7 16 years 16 6.4 18 years or more 18 1.4
PRE-JOB TRAINING Vocational/technical training before current job no 0 91.3 yes 1 8.7
TRAINING Received OJT or not
yes 1 66.7 no 0 33.3
TRAINING AMOUNT Amount of OJT received none 0 33.3 once 1 29.5 twice 2 18.9 three times 3 10.9 four times 4 5.2 five times 5 1.2 six times 6 1.1
ADULT EDUCATION Attended AET outside firm no 0 69.0 yes 1 31.0
Table 3 Descriptions of level-3 variables (66 firms) Variable Definition Code Percentage SECTOR Economic sector Manufacturing 0 46.6 Service 1 53.5 SIZE Size of the firm Small, fewer than 300 employees 0 29.6 Medium, 301 to 800 employees 1 28.2 Large, over 800 employees 2 42.3 LOCATION Location of the firm Rural 0 26.8 Urban 1 73.2 OWNERSHIP Ownership of firms Local private and collective 0 11.3 State-owned and corporate 1 32.4 Joint-venture and sole-foreign investment 2 56.3 TRAINING EXTENT Training as human resource strategy Provided to less than 1/3 of employees 0 15.5 Provided to over 1/2 to 2/3 of employees 1 42.3 Provided to more than 2/3 of employees 2 42.3
Table 4 Unconditional mModel of gGrowth in Shenzhen eEmployee sSalaries
The outcome variable: Log actual salary in 1996 price Fixed eEffect Coefficient SE T-ratio P-value
For SALARY INTRCPT1 π0ij
For INTRCPT2 β00j INTRCEPT3 γ000 6.887084 0.027883 246.998 0.000 For OBSERVATION slope π1ij For INTRCPT2 β10j
INTRCPT3 γ100 0.143218 0.012354 11.593 0.000 For OBSERVATION2 slope π2ij For INTRCPT2 β20j INTRCPT3 γ200 0.01275 0.001318 0.967 0.334 For OBSERVATION3 slope π3ij For INTRCPT2 β30j INTRCPT3 γ300 -0.003563 0.001089 -3.271 0.001 For PERF GAIN slope π4ij For INTRCPT2 β40j INTRCPT3 γ400 0.076592 0.011075 6.916 0.000 Standard Variance Random effect dDeviation cComponent df Chi-square P-value
Level-1 σ2
Temporal vVariation Etij 0.26458 0.07000
Level-2(Employees within firms)τπ INTERCEPT R0ij 0.34745 0.12072 957 2681.64954 0.000 OBSERV slope SLOPE R1ij 0.07302 0.00533 957 1645.82241 0.000
Level-3 (Between firms)τβ INTERCEPT U00j 0.17593 0.03095 63 214.66090 0.000 OBSERV interceptNTERCEPT U10j 0.04035 0.00163 63 165.10610 0.000 PERF GAIN intercept U40j 0.05669 0.00321 63 98.66105 0.003 Reliability of OLS rRegression cCoefficient eEstimate Random level-1 coefficient Reliability estimate INTRCPT1, P0 0.760 OBSV, P1 0.451 Random level-2 coefficient Reliability estimate INTRCPT1/INTRCPT2, B00 0.580 OBSV/INTRCPT2, B10 0.466 PERFGAIN/INTRCPT2, B40 0.351
Table 5 Effects of employee characteristics on salary growth
Outcome variable: Log actual salary in 1996 price Fixed effect Coefficient SE T-ratio P-value
For SALARY INTRCPT1 π0ij For INTRCPT2 β00j INTRCPT3 γ000 6.514054 0.084249 77.319 0.000 For SEX β01j INTRCPT3 γ010 0.059128 0.025311 2.336 0.020 For EDUCATION β02j INTRCPT3 γ020 0.023716 0.007347 3.228 0.002 For POSITION β03j INTRCPT3 γ030 0.101371 0.017259 5.873 0.000 For EXPERIENCE β04j INTRCPT3 γ040 0.018402 0.027176 0.677 0.498 For PRE-JOB TRAINING β05j
INTRCPT3 γ050 -0.033248 0.043904 -0.757 0.449 For OBSERVATION slope π1ij For INTRCPT2 β1ij INTRCPT3 γ100 0.187027 0.029199 6.405 0.000 For AGE β11j
INTRCPT3 γ110 0.002298 0.005526 0.416 0.677 For EDUCATION β12j INTRCPT3 γ120 -0.004514 0.002262 -1.995 0.046 For POSITION β13j INTRCPT3 γ130 0.012948 0.004934 2.624 0.009 For ADULT EDUCATION β14j INTRCPT3 γ140 -0.003553 0.009433 -0.377 0.706 For TRAINING β15j INTRCPT3 γ150 0.010234 0.007429 1.378 0.168 For OBSERVATION2 slope π2ij For INTRCPT2 β20j INTRCPT3 γ200 0.001380 0.001318 1.047 0.295 For OBSERVATION3 slope π3ij For INTRCPT2 β30j INTRCPT3 γ300 -0.003949 0.001087 -3.633 0.001 For PERF GAIN slope π4ij For INTRCPT2 β40j INTRCPT3 γ400 -0.047102 0.041737 -1.129 0.260 For EDUCATION β41j INTRCPT3 γ410 0.006550 0.003435 1.907 0.056 For CHANGE β42j INTRCPT3 γ420 0.012249 0.004540 2.698 0.007 For TRAINING AMOUNT β43j
INTRCPT3 γ430 0.013096 0.004571 2.865 0.005 For ADULT EDUCATION β44j INTRCPT3 γ440 -0.015939 0.014629 -1.090 0.276 Standard Variance Random effect deviation component df Chi-square P-value
Level-1 σ2
Temporal variation Εtij 0.26472 0.07008 Level-2 (Employees within firms) τπ INTERCEPT R0ij 0.32793 0.10754 952 2831.12405 0.000 OBSERV slope R1ij 0.07130 0.00508 952 1635.74473 0.000 Level-3 (Between firms)τβ INERCEPT U00j 0.14524 0.02109 63 170.13323 0.000 OBSERV intercept U10j 0.04268 0.00182 63 179.92424 0.000 PERF GAIN intercept U40j 0.04234 0.00179 63 81.09605 0.062
Table 6 Comparison of variance
Level-1 Level-2 Level-3
Additional Total Conditional Variance Conditional variance variance Unconditional on employee explained on firm explained explained
Variance component model characteristics (%) characteristics (%) (%)
Level-1 (growth level) σ2 0.07000 0.07008 -0.1 0.06984 0.3 0.2 Level-2 (within firm) τπ Initial salary R0ij 0.12072 0.10754 10.9 0.10612 1.3 12.1
Growth slope R1ij 0.00533 0.00508 4.7 0.00504 0.8 5.4
Level-3 (between firms) τβ
Firm mean salary U00j 0.03095 0.02109 31.9 0.01687 20.0 45.5
Firm mean growth U10j 0.00163 0.00182 -11.7 0.00137 24.7 16.0
Firm skill slope U40j 0.00321 0.00179 44.2 0.00130 27.4 59.5
Variance partitioning
Variance among employees within firm (τπpp/(τβpp + τπpp)) Initial salary 0.80 0.84 0.86 Growth slope 0.77 0.74 0.79 Variance among firm (τβpp/(τβpp + τπpp )) Initial salary 0.20 0.16 0.14 Growth slope 0.23 0.26 0.21
Table 7 Effects of employee and firm characteristics on salary growth
Outcome variable: Log actual salary in 1996 price Fixed effect Coefficient SE T-ratio P-value
For SALARY INTRCPT1 π0ij For INTRCPT2 β00j INTRCPT3 γ000 6.617126 0.087686 75.464 0.000 SIZE γ001 -0.090304 0.028554 -3.163 0.002 For SEX β01j INTRCPT3 γ010 -0.087679 0.056416 -1.554 0.120 OWNERSHIP γ011 0.100965 0.033578 3.007 0.003 For EDUCATION β02j INTRCPT3 γ020 0.026550 0.007503 3.538 0.001 SECTOR γ021 -0.003502 0.003821 -0.917 0.360 For POSITION β03j INTRCPT3 γ030 0.098908 0.017173 5.759 0.000 For EXPERIENCE β04j INTRCPT3 γ040 0.021861 0.027107 0.806 0.420 For PRE-JOB TRAINING β05j INTRCPT3 γ050 -0.038646 0.043663 -0.885 0.376 For OVSERVATION slope π1ij For INTRCPT2 β10j INTRCPT3 γ100 0.131476 0.034271 3.836 0.000 TRAINING γ101 0.031566 0.027452 1.150 0.251 OWNERSHIP γ102 0.024758 0.008715 2.841 0.005 For AGE β11j INTRCPT3 γ110 0.004042 0.005464 0.740 0.459 For EDUCATION β12j INTRCPT3 γ120 -0.004815 0.002250 -2.140 0.032 For POSITION β13j INTRCPT3 γ130 0.013133 0.004896 2.683 0.008 For ADULT EDUCATION β14j INTRCPT3 γ140 -0.002941 0.009393 -0.313 0.754 For TRAINING β15j INTRCPT3 γ150 0.010065 0.007389 1.362 0.173 For OBSERVATION2slope π2ij For INTRCPT2 β20j INTRCPT3 γ200 0.001399 0.001316 1.063 0.288 For OBSERVATION3slope π3ij For INTRCPT2 β30j INTRCPT3 γ300 -0.003866 0.001082 -3.572 0.001 For PERF GAIN slope π4ij For INTRCPT2 β40j INTRCPT3 γ400 -0.043315 0.041491 -1.044 0.297 For EDUCATION β41j INTRCPT3 γ410 0.006258 0.003420 1.830 0.067 For CHANGE β42j INTRCPT3 γ420 0.025995 0.013282 1.957 0.050 SIZE γ421 0.006222 0.005334 1.166 0.244 LOCATION γ422 0.005507 0.009974 0.552 0.580 SECTOR γ423 -0.014039 0.008005 -1.754 0.079 OWNERSHIP γ424 -0.013820 0.006471 -2.136 0.032 For TRAINING AMOUNT β43j INTRCPT3 γ430 0.026946 0.014005 1.924 0.054 SIZE γ431 -0.008842 0.005769 -1.533 0.125 LOCATION γ432 -0.014623 0.009578 -1.527 0.127 SECTOR γ433 -0.018870 0.008817 -2.140 0.032 OWHERSHIP γ434 0.015214 0.007679 1.981 0.047 For ADULT EDUCATION β44j INTERCEPT3 γ440 -0.017135 0.014653 -1.169 0.243
Table 7 (Continued) Standard Variance
Random effect deviation component df Chi-square P-value
Level-1 σ2
Temporal variation, Εtij 0.26428 0.06984 Level-2 (Employees within firms) τπ INTERCEPT R0ij 0.32576 0.10612 952 2961.94930 0.000 OBSERV slope R1ij 0.07103 0.00504 952 1707.90088 0.000
Level-3 (Between firms)τβ INTERCEPT U00j 0.12987 0.01687 62 144.29059 0.000 OBSERV intercept U10j 0.03701 0.00137 61 149.77398 0.000 PERF GAIN intercept U40j 0.03602 0.00130 63 78.16548 0.094