Journal of International Business Studies (2006) 37, 148–161
& 2006 Academy of International Business All rights reserved 0047-2506 $30.00 www.jibs.net
Japanese career progress: an empirical examination George Graen1, Ravi Dharwadkar2, Rajdeep Grewal3 and Mitsuru Wakabayashi4 1 University of Louisiana, Louisiana, USA; 2Martin J. Whitman School of Management, Syracuse University, Syracuse, NY, USA; 3Smeal College of Business Administration, Pennsylvania State University, Pennsylvania, USA; 4Faculty of Management, Aichi Gakuin University, Nagoya, Japan
Correspondence: Dr R Dharwadkar, Martin J Whitman School of Management, Syracuse University, Syracuse, NY 13244, USA. Tel: þ 1 315 443 3386; Fax: þ 1 315 443 5457; E-mail:
[email protected]
Received: 16 October 2002 Revised: 5 May 2005 Accepted: 30 May 2005 Online publication date: 24 November 2005
Abstract This article investigates managerial career progress in a major Japanese multinational corporation over a 23-year period. We contrast our one-stage model of career progress, in which early career experiences can predict longterm career progress, with the well-accepted two-stage model of career progress, in which performance at the end of the first stage determines longterm career progress. In addition, we compare the predictions of our early screening model with those of the tournament model. Our early screening model is as good as or better than the other models in isolation or combination. Overall, we provide preliminary evidence that career progress in Japanese organizations during the high growth period of the 1970s and 1980s was perhaps more complex than shown by the two-stage career progress model that has enjoyed widespread acceptance. Journal of International Business Studies (2006) 37, 148–161. doi:10.1057/palgrave.jibs.8400175 Keywords: career progress; Japan; duration models
Introduction In major Japanese corporations with lifetime employment systems, career progress and promotions are very important to both the organization and its white-collar employees (Yoshino, 1968; Chen, 1995). Many researchers have argued that the long-term career progress of white-collar employees in Japanese corporations relies on late screening and involves essentially a two-stage process, in which most employees are treated equally during the first stage and then differentiated during the second (Clark, 1979; Pucik, 1989; Koike, 1995). Koike (1995) specifically emphasizes the late-screening model to argue that careers in Japanese firms have two stages, and that in the first stage promotion is based largely on seniority. According to Koike (1995), the increasing gap in career progression during the second stage results in extremely competitive behavior in the seniority/age-based promotion system of Japanese internal labor markets (the nenko joretsu system). Similarly, Evans et al. (2002) describe the two-stage process as an elite cohort approach to career progress based on late screening. They also concur that the two-stage process includes an initial trial period ‘characterized by equal opportunity in the sense that there are no immediate sanctions for poor performancey the company has a clear idea from the trial as to who are the sheep and who are the goats. The rules change and career progression becomes a tournament with winners and losers’ (Evans et al., 2002, 372).
Japanese career progress: an empirical examination
G Graen et al
149
However, other studies find that career progression in Japanese firms, at least in the shorter term (i.e., 7–12 years), is based on early screening, in that competition during the first stage is more indicative of future career progress. For example, Wakabayashi and co-workers (Wakabayashi and Graen, 1984; Wakabayashi et al., 1988) examine earlyscreening variables, such as university reputation, quality of superior–subordinate relationships, and the influence of these variables on employees’ mobility into middle management, and conclude that experiences during the first 3 years of employment predict mobility into middle management. The main goal of this article is to re-examine whether career progress in Japanese organizations in the long term is unrelated to early career experiences, as proposed by Koike (1995) and others, or based on early screening, as proposed by Wakabayashi and co-workers (Wakabayashi and Graen, 1984; Wakabayashi et al., 1988). If we find support for the early-screening model, we would need to re-examine late-screening models after accounting for the early-screening variables and attempt to understand the complex dynamics of a single-stage model that may appear to have two stages. A secondary goal of this article is to compare two early-screening models that characterize a singlestage process: sponsored mobility and tournament mobility. Limited research has addressed these concepts in a Japanese context. The sponsored mobility model focuses on the employee’s background and other personal characteristics that may facilitate his or her career progress: thus, it mirrors the single-stage process (Turner, 1960; Bowman, 1981; Ishida, 1993). According to this model, early career characteristics (e.g., university background, cognitive ability, socioeconomic status) are associated with, and therefore vital for understanding, long-term career progress. Proponents of tournament mobility, in contrast, focus on the outcomes of career progress during the employee’s first few promotions and use this information to predict long-term career progress (Rosenbaum, 1979, 1984; Forbes, 1987). They suggest that career progress is a ‘historically dependent’ process by which an early first promotion critically influences the employee’s long-term career progress. Examining these two early-screening models simultaneously is important for two reasons. First, though predicting subsequent career progress using early career progress results in the inherent problem of using predictors that are part of the process being
investigated, it still enables us to verify whether the first promotion provides any additional information about career progress. Second, the early career progress school cannot provide managerial recommendations to either the organization or its employees. In contrast, an in-depth examination of the early career experiences school may provide valuable guidelines for ensuring the commitment of employees to their organizations as they progress through their careers. With that in view, this article is organized as follows. First, we develop a hypothesis on the basis of the pertinent literature for the early-screening career progress model and contrast it with an alternative hypothesis from the late-screening career progress model. Second, we consider two alternatives of the early-screening model (sponsored mobility and tournament mobility), as well as their implications for subsequent career progress, and then assess a combined model with both factors. Third, we review duration analysis and use it to test the various models. Fourth, we present an empirical analysis of career progress in a Japanese context using a cohort of 65 Japanese graduates who joined the same corporation upon graduation in 1972 and remained there for 23 years. Fifth, and finally, we present our conclusions.
Hypotheses To understand career progression in Japanese firms, it is important to study the evolution of Japanese internal labor markets from both historical and economic perspectives. Historically, according to Clark (1979), management training and career progression in large Japanese firms can be traced to the training techniques of the old merchant houses in Japan. Most internal labor market participants tend to focus on learning the business from within the organization, and spend time in many different parts of the firm so that they may become familiar with its holistic functioning. The end products of such training programs are very different from those of Western countries, in which there is a tendency to think of firms merely as contexts in which individual skills can be applied. As employees begin their careers with this worldview, managing internal labor markets (ILMs) becomes a key function for Japanese organizations. From an economic perspective, ILMs became feasible during the tremendous growth in the Japanese economy in the 1950s and 1960s, which occurred because of the emergence of large Japanese corporations that focused on market share and
Journal of International Business Studies
Japanese career progress: an empirical examination
G Graen et al
150
long-term growth (Ariga et al., 2000). Eventually, large-scale manufacturing necessitated the use of both simple and complex technologies. By developing ILMs, Japanese firms were able not only to move employees from simple to complex tasks through on-the-job training but also to generate additional savings in skill formation by creating steep ILMs for highly skilled workers and garnering compliance and commitment from their employees (Cole, 1979; Lincoln and Kalleberg, 1990). Over time, the ILM system has become institutionalized with comparable staff designations. The English equivalents of these titles, according to Clark (1979), are as follows: ordinary company member, team head, subsection head, section head, deputy department head, department head, director, managing director, senior managing director, vice president, and president. Whereas those ranks up to department head are considered employees, directors and above are classified according to a different legal status. In addition, rank has implications for retirement age and therefore fosters competition (Clark, 1979). Many Japanese organizations prefer to employ their higher-cost, older employee labor force at smaller subsidiaries and consequently force early retirement between the ages of 55 and 60 years. However, if an employee is promoted to a department headship or directorship, the retirement age rises to 65 years. Owing to this well-developed ILM context, there has been a widespread acceptance of the late-screening model of career progress, according to which the competition among middle managers – to ensure that they eventually get promoted – intensifies as they continue through the first stages of the process (Koike, 1995). In other words, the first stage of the process is considered rather benign, and the movements through the initial ladders of the ILM are considered more as rites of passage. Thus, performance near the end of the first stage is of critical importance, because automatic promotions come to an end while the competitive dynamics of the process come to the fore. In support of this latescreening hypothesis, Koike (1995: 186–187) notes that The screening that determines assignment to management positions occurs late, around fifteen years after a worker enter[s] a company. Careers thus have two stages. The first covers the years before screening[,] during which most white collar graduates are promoted y the assertion that promotion in Japanese companies is based on seniority corresponds fairly closely to the situation during this stagey. In the second stage, the pace of selection speeds
Journal of International Business Studies
up and the gap grows y the assumption that Japanese companies do not differentiate between workers on basis of ability is only true if the second stage is ignored.
In contrast to this late-screening, two-stage model, Wakabayashi and co-workers (Wakabayashi and Graen, 1984; Wakabayashi et al., 1988) argue that initial screening is more critical for determining short-term career success. Their model combines variables from the sponsored mobility model with early career experiences in an organization. We appropriate three factors from that model to proxy individual ability – namely, the university that the employee attended, the employee’s early job performance, and the employee’s ability to form high-quality social exchange relationships with multiple supervisors – and link them collectively to long-term career progress. Japanese corporations prefer to hire college graduates but do not place the same emphasis on applicants with advanced degrees. In addition, the road to the Japanese university is extremely competitive; only the very best make it to the top schools after strenuous entrance exams (Ono, 2001). Cole (1979) also notes the strong emphasis on passing university entrance exams; once a student is accepted into a good university, the work required to graduate may be less than that expected of him or her in high school. Since an education from a leading educational institution signals the external labor market and increases opportunities for employment in large Japanese firms, there is tremendous competition to attend such institutions (Ono, 2001). In addition, attending an elite university can compound a person’s social capital by providing access to ‘old boy’ networks. Both Clark (1979) and Rohlen (1974) highlight the influence of university cliques, in that alumni from the same university help one another’s careers in large Japanese organizations. We therefore suggest that employees who attend more highly ranked universities will be of higher ability and, at the same time, have better access to internal networks on the basis of school ties, prestige, and competence, which will continue to assist them as they advance in their careers. Researchers also have argued for the important role of early job performance for career progress but have not found support for these variables with respect to short-term career progress (Wakabayashi and Graen, 1984; Wakabayashi et al., 1988). Owing to the lifetime employment systems in many large Japanese organizations, employees are hired at the onset of their career on the basis of their potential.
Japanese career progress: an empirical examination
G Graen et al
151
With a view to developing this potential in the first few years of their careers, new employees are trained to work in various jobs (Baba et al., 1995). Consistently high performance across the various jobs is an important condition for advancement in the tight ILMs that provide for few hires at nonentry levels. For example, Evans (1989) and Evans et al. (2002) describe how the entering cohort of recruits at large Japanese corporations is rotated through various responsibilities, intensively trained in the first few years, and extensively monitored and evaluated every few months by the human resource managers. Consequently, the organization has multiple assessments of an employee’s early performance in multiple jobs; these assessments are used as important indicators of both future performance and potential. Evans (1989) states that a failure in this early stage is tantamount to a ceiling for the employee in the race to make it to the position of a division head. As the employee spends his or her first few years training and rotating jobs, consistent performance across multiple assignments leads to more avenues for future progress and the development of networks of peers and superiors that assist that employee in his or her long-term career progress. Another aspect of the Japanese ILM system that sets it apart from Western ILMs is its supervisory paternalism. Scholars have noted how vertical ties enable the development of a collective consciousness in Japanese organizations (Cole, 1979; Lincoln and Kalleberg, 1990). Such paternalism is considered vertical bonding in both patron–client (oyabun kobun) and senior–junior (sempai kohai) relationships. According to Nakane (1970), in Japanese organizations, social integration results from strong bonds between status unequals rather than among peer groups. Cole (1979) and Lincoln and Kalleberg (1990) also provide evidence of the importance of informal relationships between superiors and subordinates in Japanese work settings. Forming effective social bonds then becomes crucial in the case of Japanese organizations, because new employees tend to have multiple supervisors during their first few years of training and therefore can develop multiple supervisory relationships. Those who are able to form highquality relationships with their supervisors consistently have access to more information and networks in the long run. These recruits also develop more professional skills, thanks to their multiple role-making opportunities and their ability to go above and beyond the call of duty. In the
Japanese context, Wakabayashi and colleagues (Wakabayashi and Graen, 1984; Wakabayashi et al., 1988) use the concept of leader–member exchange (LMX) to measure the quality of these superior–subordinate relationships. Research on LMX suggests that recruits who consistently develop high-quality relationships with their various trainers/supervisors are more likely to continue to do so with all subsequent supervisors. If this finding is true, new employees who are effective in developing such relationships will progress faster through their careers. We also propose that the factors contained in the early-screening model (i.e., job performance, university ranking, and LMX) will contribute to longterm career progress. In addition, we propose that each of our main effects will interact. For example, university ranking and job performance jointly will reduce the duration of time between promotions, because those who come from elite schools and perform well are likely to progress faster than, say, those who come from elite schools but fall short in their performance. Similarly, those who come from elite schools, perform well, and have better relationships with their supervisors are likely to encounter more opportunities for progress than are those who come from elite schools and perform well but lack effective working relationships with their supervisors. However, if the late-screening versions of the two-stage career progress models are true, the factors from the early-screening model should not be able to predict long-term career success, especially because screening occurs only in the latter part of an employee’s career. Consequently, we propose the following: H1: According to the early-screening model, early career experiences (university reputation, job performance, LMX) will be negatively associated with the duration of time between promotions during an employee’s career. H2: According to the late-screening model, early career experiences (university reputation, job performance, LMX) will not be associated with the duration of time between promotions during an employee’s career. We now turn our attention to the two types of early-screening models. The first type resembles the early-screening model that we described previously and is often referred to as the sponsored mobility model (Turner, 1960). In sponsored mobi-
Journal of International Business Studies
Japanese career progress: an empirical examination
G Graen et al
152
lity models, factors such as university education and social background are givens, and aid in long-term career progress. We add early experiences in the form of job performance and the quality of relationship formation to arrive at our own earlyscreening model. In addition, another stream of research on early-screening models, referred to as tournament models, emphasizes the importance of early success in an employee’s overall career development (Rosenbaum, 1979, 1984). Specifically, Rosenbaum (1979) suggests that managerial careers develop in a way similar to a tournament, in which organizational members who have early career success (faster first promotion) experience an advantage in their subsequent careers. The results of his analysis of the career movement of a cohort of employees in a large organization over a 13-year period indicate that early promotions influence subsequent career progress through career ceilings, career floors, and probabilities of promotion and demotion in successive periods. Other researchers have found similar results and, in some cases, demonstrated that employees could make up for lost opportunities in the early stages of their careers (Forbes, 1987; Sheridan et al., 1990). On the basis of this literature, we might argue that early career progress plays a critical role in an employee’s subsequent career (Rosenbaum, 1979, 1984; Forbes, 1987; Dreher and Bretz, 1991; Sheridan et al., 1990). However, only limited research in the Japanese context considers these models simultaneously. Ono (2001) studies students’ progress from middle school to college and finds that both institutional tracking and social origins influence their access to and attendance at higher universities in Japan. His institutional tracking arguments mirror those of the tournament models; he finds that institutional tracking matters, and that the end winners are selected in the early rounds. In addition, Ono (2001: 190) finds that social origin also matters, and that students from higher social backgrounds have more access to prestigious universities, in that ‘Students have a better chance to advance to high school and college if they are smart; but it certainly helps if they’re from a good family.’ We extend these arguments to the post-university stage of a person’s life. For example, Ishida (1993) states that though few Japanese universities are deemed prestigious (as compared with US universities), the influence of college prestige on a person’s career continues for a longer period in
Journal of International Business Studies
Japan because, in the United States, it is supplanted earlier by labor market experience. The very nature of Japanese culture and the organizational personnel system of continuous tracking may intersect and provide longer historical memories in Japanese organizations. If these memories exist, the sponsorship model will predict long-term career progress. We similarly can extend tournament model arguments and posit that early career progress will result in the quickened pace of longterm career progress (Ono, 2001). In general, Ono (2001) finds that both tournament and sponsored mobility variables matter independently for predicting access to higher education. Therefore, we examine whether there are independent effects of the two models in our setting of interest. We first hypothesize, on the basis of the tournament model, that early career progress is related to long-term career progress. We then hypothesize that the tournament model should explain long-term career progress, beyond that explained by the sponsored mobility models. H3: Early career progress will be negatively associated with the duration of time between promotions during an employee’s career. H4: Early career progress will be negatively associated with the duration of time between promotions during an employee’s career, over and above the contribution of early career experiences (university reputation, job performance, LMX).
Method Sample and setting The sample consists of 65 employees who had continued to work for a large Japanese organization since joining it as part of a cohort group in 1972. All the cohort members had passed the organization’s assessment center evaluation and started their jobs at the same level in the organization. In this sample, 85% of the cohort had degrees in social sciences and 15% had degrees in engineering. In the initial 3 years of the study (1972–1974), we obtained multiple assessments of their relationships with supervisors from the employees, and of their job performance data and college information from the company files. In 1995, we obtained data about the promotions of this cohort group from the personnel files of the organization.
Japanese career progress: an empirical examination
G Graen et al
153
Measures Some variables of the early-screening model were measured, using scales that we detail subsequently, in the initial 3 years of the study (1972–74). Job performance (JP) was averaged over this 3-year period using data collected on six different occasions. Nine items constituted the JP scale: accountability, alertness, interpersonal skills, planning, technical skills, know-how, level of contribution, interpersonal attraction, and willingness to contribute to the company. Supervisors rated the job behaviors of subordinates on these nine dimensions using a five-point scale, and the employee’s overall rating was a composite of the nine items. This measure had high reliability (Cronbach’s a¼0.92). In addition, the six overall performance ratings were aggregated to measure the employee’s average performance during his or her first 3 years with the company. During this period, the cohort rotated extensively from one department to another, and several supervisors evaluated each employee’s JP. Thus, our measure of initial JP represents a performance appraisal over a 3-year period, which results in an extremely representative sampling of JP. The quality of the employee’s relationship with his or her supervisor was assessed using the LMX approach with the following items on a five-point scale: approachability and flexibility of the supervisor toward the newcomer, the supervisor’s willingness to use authority to help the newcomer, clarity of the supervisor’s expectations and feedback to newcomers, the newcomer’s latitude to influence the supervisor to change his or her role situation, and after-hours interaction between the supervisor and subordinates with respect to social and leisure activities. Both newcomers and supervisors were asked to respond to this scale, and an aggregate scale was created after each administration. This scale thereby assesses the effectiveness of the working relationship between the new recruit and the supervisor. The new recruits and their respective supervisors were asked to respond to these questions on six separate occasions during a 3-year period. The results indicate that the scale has high internal consistency (Cronbach’s a¼0.90). During the 3-year period, each recruit was rotated and had three to four different supervisors. According to the averaged early LMX, the new recruits tended to maintain similar types of relationship with their different supervisors (correlation of 0.6 over the multiple relationships).
University ranking (UR) was based on the quality of the education provided, the student body, and the experiences by which students became socialized into business careers. A total of 22 different universities located in Tokyo and Osaka areas were represented in our sample. Two students and one faculty member rated the prestige of each university on a four-point scale. The agreement correlation coefficients among the raters were 0.92, 0.90, and 0.88, respectively. Finally, the speed of progress was assessed in terms of the duration of time (in months) that an employee spent in a particular position before being promoted. Graduates initially were promoted to level 6. The first set of promotions took place in March 1978. Since then, 25 employees in the cohort group have made it to level 10 (i.e., have been promoted on five occasions). Two employees remain at level 7, 13 employees at level 8, and 25 employees are at level 9. Thus, we have information on all employees for all five time periods after they entered the organization (i.e., 325 duration periods or periods between promotions).
Analytic procedures Owing to the nature of our dependent variable – namely, the time until an employee is promoted – we use a duration analysis approach (also called survival analysis) (cf. Kalbfleisch and Prentice, 1980; Kiefer, 1988; Greene, 1993; Morita et al., 1993; Allison, 1997). This method is extremely useful for studying events, including job terminations, strikes, promotions, retirements, marriages, divorces, and arrests, and we use it for three main reasons (Allison, 1997). First, we believe that our dependent variable (i.e., time until promotion) can be used appropriately in a duration analysis. According to Morita et al. (1993), researchers have become more interested in simultaneously predicting both whether an event occurs (in our case, a promotion) and when the event occurs (in our case, time in months before the promotion). We appreciate the utility of this technique over conventional ordinary least-squares (OLS) regression techniques, as described by Morita et al. (1993, 1433): For instance, a researcher might be interested in the duration between organizational entry (the starting point) and departure (the event of interest) for a sample of cohorts. From this set of durations, the probabilities of individuals experiencing the event can be estimated; they either survive in the organization or terminate. An advantage of survival analysis over many other techniques (such as OLS
Journal of International Business Studies
Japanese career progress: an empirical examination
G Graen et al
154
regression) is that it allows the probability of an event at a point in time to differ from the probability of that event at a different point in time. Thus, these probabilities are not necessarily constant but instead can be described as a function of time. Thus, the nature of our dependent variable offers a unique modeling opportunity that overcomes the shortcomings of traditional OLS regression techniques. Second, if we were to use a conventional regression approach with our dependent variable, we would not be able to account for employees who were not promoted (Allison, 1997). For example, only 25 employees were promoted five times: therefore, at least 40 employees were still at level 9 or below, which means we have 40 censored cases. We might discard the censored cases, but that method would mean discarding a large proportion of the data. Instead, we use a procedure that combines the information in the censored and uncensored cases to produce consistent estimates of the parameters of interest (Allison, 1997). Thus, duration models can account for censored observations, unlike an OLS regression approach (Allison, 1997). Third, parametric duration models are useful when theory has recommendations regarding the shape of the hazard function (Morita et al., 1993). To comprehend the utility of the shape of the hazard function fully, we first must define a hazard function and then examine a few duration densities. The hazard function defines the conditional probability of observing an event (e.g., promotion) at a particular point in time, say t, given that the event has not occurred until time t1. The shape of the hazard function defines how this conditional probability varies over time. Monotonically increasing hazard functions for employee promotion, such as Weibull, would imply that the likelihood of an employee getting promoted in period t, given that the employee has not been promoted until period t1, is less than the probability of the employee getting promoted in period t þ 1, given that the employee has not been promoted until period t. Non-monotonic hazard functions, such as lognormal, capture the complex events for which we expect hazard probabilities to increase (decrease) up to a point in time and then decrease (increase). Again, an OLS regression approach does not incorporate the shape of the hazard function with respect to promotion opportunities and therefore would force us to use categorical data in the
Journal of International Business Studies
analysis. An OLS approach also would use the number of promotions as the dependent variable and, in the process, information would be lost. The main questions on which we focus in this article are as follows: (1) How much time (i.e., duration) does an employee spend at a particular level in an organization? (2) How are early career experiences (university reputation, job performance, LMX) associated with these durations? The dependent measure in duration models is the time until an event occurs (in our case, time until promotion) and is referred to as the failure time (Kalbfleisch and Prentice, 1980; Kiefer, 1988). In some cases, however, instead of observing the failure time, the data are censored, so we observe only the survival time (here, an employee would not be promoted for the duration of the study). Although these observations are censored, they contain limited information: that is, the employee did not get promoted (survived) for the observed duration of time. This limited information can be incorporated in the likelihood function, by which we obtain the parameter estimates. Next, we select the functional form for the probability density function on the basis of both theoretical and empirical considerations. Empirically, the base model (with the constant as the only explanatory variable) with the highest log-likelihood (LL) is considered suitable (Kiefer, 1988). Thus, the first step is to consider various functional forms of the probability density functions and, on the basis of the LL criterion, identify the best functional form. In step 2, using the recommendations of step 1, we choose the relevant probability density function and introduce the covariates (early career experience variables) into the equation (H1 and H2). In step 3, we use the first time period as the only covariate (to represent early career progress) in the equation to predict the duration in periods 2, 3, 4, and 5 (H3). In this way, we verify the utility of the tournament models. In step 4, we use all the covariates from step 2 and predict the duration in periods 2–5 (H4). Finally, in step 5, we use the tournament model covariate (i.e., first time period) and all the covariates from step 2 simultaneously to predict the duration in all the subsequent periods (H4; see also Table 1).
Results The LLs for the base model of exponential, Weibull, log-logistic, and log-normal functions were 309.35, 112.67, 99.40, and 94.17, respec-
Japanese career progress: an empirical examination
G Graen et al
155
Table 1
Summary of the analytic procedure
Analysis step
Model
Functional form
Evaluation criteria
Step 1
Baseline hazard (five time periods) Hazard with all predictive variables and their interactions Hazard with first duration as predictor variable Hazard with all predictors and their interactions except first duration Hazard with all predictors
Exponential, Weibull, log-normal, loglogistic Best from step 1
Log-likelihood
Step 2 Step 3 Step 4
Step 5
Table 2
Best from step 1 Best from step 1
Best from step 1
Coefficients and signs of independent variables Coefficients and signs of independent variables Coefficients and signs of predictor variables Compare step 3 and 4 models with step 5 using the likelihood-ratio test. Select the best model.
Estimated survival distribution in months: percentage of employees not promoted
Promotions Trainee to Assistant Section Manager (ASM) ASM to Senior ASM Senior ASM to Section Manager (SM) SM to Senior SM Senior SM to Department Head Baseline (average across all durations)
95%
75%
50%
25%
58.70 34.17 28.94 45.63 27.92 30.75
63.80 40.18 36.98 56.87 37.26 42.10
67.59 44.97 43.83 66.27 45.52 52.35
71.62 50.32 51.90 77.22 55.62 65.10
tively. On the basis of the LL, we retain the lognormal form for further analysis because it has the highest value. Our further exploration of the data, based on the log-normal form, confirms some trends in career progression in the organization. We study the five duration periods independently and consider the survival times of employees in each level.
Nature of progression In terms of the first promotion (from trainee to assistant section manager), 5% of the cohort had been promoted after approximately 5 years (see Table 2). By end of the 6th year, 75% of the cohort had been promoted to the next level. Thus, the fairly lengthy duration to the first promotion for most of the cohort lasted 5–6 years. This finding is consistent with the rigorous initialization and socialization processes commonly found in large Japanese organizations. The survival distributions also indicate that the next two promotions (from assistant section manager to senior assistant section manager to section manager) followed similar patterns. In contrast with the 5-year wait for the first promotions, some
employees received their second and third promotions as early as 2–3 years later, and most employees were promoted to the next level by the 4th year. The process picked up rigor again as some employees from the cohort started approaching senior positions. The initial promotions at this level (i.e., the fourth promotion from section manager to senior section manager) came within 4 years, but half of the employees took 5.6 years to clear this obstacle. Thus there was systematic pressure to push the very best into the next level. The final promotion (from senior section manager to department head) was based on the employee’s performance both on the job and in an examination-type assessment. Employees who passed the exam were readily promoted to the next level, as is clearly reflected in the decreased duration of the first few promotions at this stage. Some employees were promoted within 2 years, but on average the trends after that were similar to previous duration periods.
Career progression In Table 3 we show the correlates of career progression, including the moderate correlation of the early experience model variables. The estimated
Journal of International Business Studies
Japanese career progress: an empirical examination
G Graen et al
156
Table 3
Descriptive statistics: means, standard deviations, and correlation coefficients
Variable
Mean
Standard deviation
Min
Max
Duration
LMX
JP
Duration LMX JP UR
51.59 32.81 33.58 2.92
19.23 5.63 3.90 0.93
1 16 25 1
108 46 43 4
(1) 0.089 0.125* 0.067
(0.90) 0.398** 0.243**
(0.92) 0.032
UR
*Po0.10; **Po0.05. Notes: The minimum and maximum are the reported scores. The possible minimum scores were 9, 1, and 12 for JP, UR, and LMX, respectively; the possible maximum scores were 45, 4, and 60 for JP, UR, and LMX, respectively. Reliabilities appear across the diagonal, where applicable.
Table 4
Estimation results for the duration modelsa
Covariates Constant D2 D3 D4 D5 LMX JP UR LMX JP LMX UR JP UR LMX JP UR l P¼1/s Log-likelihood
Baseline log-normal
Log-normal
3.9579 (0.01921)
12.354 (3.749) 0.40757 (0.06912) 0.43232 (0.06697) 0.019659 (0.06860) 0.54084 (0.07050) 12.286 (5.737)* 9.6244 (4.754)** 10.576 (4.688)** 14.672 (7.232)** 16.459 (6.934)** 12.742 (5.971)** 20.063 (8.798)** 0.01944 (0.00035) 4.6558 (0.17381) 13.95238
0.01910 (0.00037) 3.09292 (0.14619) 94.17252
a Standard errors are in parenthesis. D2–D5 are dummy variables to incorporate promotion-level heterogeneity. The three explanatory variables were scaled between 0 to 1 by dividing each by the maximum to obtain convergence in the non-linear optimization algorithms used to obtain the maximum likelihood estimates. *Po0.10; **Po0.05.
survival distributions for the different promotion levels provide an interesting descriptive picture of the process. However, the differences in the survival distributions also create heterogeneity for data analysis. For example, according to the baseline distribution, 5% of the cohort was promoted after approximately 2.6 years, whereas 25% continued to be in the same position after almost 5.6 years. To incorporate this heterogeneity into the level of promotions (i.e., varying rates across different levels), we add four dummy variables (for five promotions, i.e., n1; D2–D5 in Table 4), coded for the respective levels, in which the first promotion represents the base level. We use the three variables that represent early career experiences as independent covariates, and all are significant at the 0.05 level. Next, we add the interactions to the covari-
Journal of International Business Studies
ates, and all terms are significant at the 0.05 level. We then use the likelihood ratio test to determine whether the two models – main effects and main effects plus interactions – differ. The difference in w2 values between the two models is 7.82 with 4 df, which indicates that the model with interactions is significantly different from the main effects model, as we show in Table 4. We retain the interaction terms for further analysis. As our three-way interaction is significant, we interpret all lower-order interactions and main effects in conditional terms in the presence of the higher-order interactions (Aiken and West, 1991). According to Aiken and West (1991), the two-way interactions must be evaluated at the mean of the third term (e.g., the conditional LMX JP interaction at the mean of UR). For example, when UR equals 0, the joint effect of LMX JP is 14.672. However, at the mean value of UR, the total conditional effect is positive (14.67220.063 0.5¼4.64), which suggests a slower promotion. In other words, for the twoway effects, the early experience variables may substitute for each other, which does not support our hypothesis. In contrast, with regard to the main effects (e.g., main effect of LMX at the mean of JP and UR), each is associated with faster promotions. The main effect of LMX at the mean of JP and UR would be 12.286 þ 14.642(0.5) þ 16.458(0.5) 20.063(0.5)(0.5)¼1.75. Hence greater LMX is associated with faster promotions. Overall, our interpretation of the main effects and the twoway interaction effects, along with the significant three-way interaction term, provides support for H1 and the early-screening, one-stage model. To test the predictions of the two types of earlyscreening model separately, we use the first time period as a covariate for predicting promotion duration in the subsequent time periods (see step 3 in Table 1), in line with the predictions of the tournament models. Our results (Table 5) show that the first time period is positively and significantly
Japanese career progress: an empirical examination
G Graen et al
157
Table 5
Estimation results for the duration modelsa
Covariates
Constant D3 D4 D5 LMX JP UR LMX JP LMX UR JP UR LMX JP UR Duration in first position l P¼1/s Log-likelihood
Log-normal with first promotion duration as independent variable 1.2026 0.025 0.3943 0.1342
(0.5768) (0.058) (0.0582) (0.0611)
0.6178 (0.1310)** 0.0209 (0.0004) 4.0229 (0.2002) 23.48512
Log-normal with all independent variables except first promotion 14.121 0.0246 0.3916 0.1208 15.610 12.389 13.475 18.906 20.838 16.472 25.708
(5.063) (0.0542) (0.0558) (0.058) (7.75)** (6.419)* (6.289)** (9.776)* (9.312)** (8.000)** (11.81)**
0.0208 (0.00039) 4.1482 (0.2000) 16.98478
Log-normal with all independent variables
12.354 (5.057) 0.0247 (0.0553) 0.3969 (0.0570) 0.1145 (0.0605) 15.105 (7.717)* 12.073 (6.391)* 13.107 (6.209)** 18.498 (9.753)* 20.158 (9.236)** 16.194 (7.912)** 25.097 (11.72)** 0.3286 (0.1648)** 0.0208 (0.0004) 4.1675 (0.2015) 15.73338
a Standard errors are in parenthesis. D3–D5 are dummy variables to incorporate promotion-level heterogeneity. The three explanatory variables were scaled between 0 to 1 by dividing each by the maximum to obtain convergence in the non-linear optimization algorithms used to obtain the maximum likelihood estimates. *Po0.10; **Po0.05.
related to subsequent time periods, which suggests that the more time an employee spends in the first duration, the more likely it is that he or she will be delayed in getting subsequent promotions. In step 4, we run a duration analysis using the same dependent variables as in step 3, but we include only our hypothesized explanatory variables (from the early-screening models of sponsored mobility) as covariates. The results suggest that all three main effects (LMX, JP, UR), three two-way interaction effects (LMX JP, LMX UR, UR JP), and the triple interaction effect (LMX JP UR) are significant in the appropriate directions and consistent with our findings in step 2. The findings of step 5, in which we combine the covariates of steps 3 and 4, also are consistent with prior results. Next, we use the likelihood ratio test to determine whether the models in steps 3 and 4 differ significantly from the model in step 5. The two w2 values are 15.50 and 2.50, with 7 and 1 df, respectively. The first is significant at the 2.5% level, whereas the second is not significant at the 10% level. This finding implies that the full model (step 5) has more explanatory power than the tournament mobility model (step 3), but not more than our proposed model (step 4). Thus, our proposed model (sponsored mobility based on early career experiences) is more effective than the
tournament mobility model and the combined model, according to parsimony.
Discussion and conclusion The major contributions of this study are threefold. First, we find support for our early-screening, onestage model. Whereas conventional wisdom favors a two-stage model of career progress, we believe that the promotion die may be cast early on, and the apparent use of the two-stage process may be a management technique to encourage competitive behavior in the long run (akin to the contest mobility models of career progress). Second, we do not find an independent effect of initial career progress on long-term career progress. Although the first promotion matters and influences subsequent promotions, early experiences and the first promotion form an endogenous system, in that early experiences affect the first promotion. Thus, the first promotion provides no new information about long-term career progress. Third, we demonstrate that parametric duration modeling can be used to study career mobility and should be the method of choice for examining long-term career progress. We begin by discussing our first major contribution. Why, after 23 years, do early career experiences continue to predict long-term career success?
Journal of International Business Studies
Japanese career progress: an empirical examination
G Graen et al
158
We believe that our three measures capture the broad range of a person’s ability that stands him or her in good stead as he or she progresses through his or her career. In other words, the organizational decision-makers are fairly cognizant of who is going to succeed in the longer term, but they manage to give the impression that the game is not over yet. We have captured different aspects of a person’s ability and how they influence career progress. The advantage of university reputation can be explained fairly easily in this context. Competition to get into the top schools is so intense that a Darwinian selection based on ability occurs at the start. Thus universities constitute efficient screening systems that select students on the basis of their intellectual abilities. In addition, university networks provide benefits within the organization. Access to such influential networks can facilitate desirable appointments or enhance managerial effectiveness by helping alumni solve job problems, succeed at difficult assignments, and so forth. Early job performance represents the performance assessments of multiple supervisors during the first 3 years. These job performance ratings are directed toward the newcomer’s behavior in entry positions that involve low skill and knowledge tasks. Consistent low performance results in career ceilings, as mentioned by Evans (1989). Consistent high performance across multiple tasks and multiple supervisors, in contrast, may open avenues for faster career progress. In this particular case, few members of the cohort were terminated for poor performance, which leaves a ceiling on career progress as the only organizational option. Finally, we expected that high-quality relationships with supervisors in the first 3 years on the job would aid an employee’s long-term career progress. Given the Japanese system of supervisory paternalism, this expectation may seem obvious, but to our knowledge it has not been tested sufficiently. Since social integration with supervisors is important in the ILM system, we believe that employees become comfortable with the role-making process that eventually evolves into high-quality relationships with their supervisors. Extensive research on LMX suggests that the ability to build consistent, effective relationships is an important professional skill and prepares recruits to manage multiple relationships with various supervisors during their careers (Graen, 2003). Furthermore, our measure of the newcomer’s evaluation of the quality of his or her working relationship with an immediate supervisor may be unique to Japanese career progress
Journal of International Business Studies
studies. During the first 3 years, each newcomer had three or four different supervisors, and the ratings therefore are the average relationship across these supervisors. In some cases, the supervisors had not graduated from college, were approaching the end of their careers, and saw their mission as teaching the newcomers the value of hard work and sacrifice for the organization. Many reported that the most important signal they gave their young charges about their potential was simply the amount of time they spent with them – the more the better. It did not matter if the time was spent criticizing or praising, but it was usually the former. However, not all newcomers appreciated this ‘tough love’ approach. Their supervisors were assistant section managers who had reached the top of their promotion ladder. Thus early LMX was based on building effective working relationships with a series of tough bosses. In summary, the three factors in our earlyscreening model have predictive power over 23 years, which provides some evidence in contrast to the two-stage career progress model, which posits that the first stage may be based on seniority and eventually devolves into a competitive stage. If that were the case, the first 3 years should have limited predictive power in terms of long-term career progress. However in effect, they do not. Early screening that appears like late screening may be the norm in Japanese organizations, and the contrary illusion may be necessary to encourage competitive behavior in the long run. Our second major contribution sheds light on the sponsored vs tournament mobility models that have characterized early-screening research. Rosenbaum’s (1979) tournament model is both supported and unsupported in this investigation. Our finding that the speed of the first promotion predicts later promotions supports it but not as a prediction increment of the sponsored mobility model. Regardless, the results of the first promotion are not the first means to predict career progress (Sheridan et al., 1990). Career progress predictions based on the quality of the supervisor–subordinate relationships and early job performance during the first 3 years, along with university reputation, are not enhanced by including the time until first promotion. Moreover, first promotion is part of the criterion system of promotions. Similar to our findings, Sheridan et al. (1990) show that the power of the recruit’s department and his or her job training program are important correlates of career progress. On the basis of our results, we conclude
Japanese career progress: an empirical examination
G Graen et al
159
that early career progress (tournament model) as measured by the speed of promotions is a stable system and can predict subsequent promotions. Building on our research and that of Sheridan et al. (1990), we believe that further studies should examine interpersonal (i.e., LMX, job performance) and contextual (i.e., departmental power, company training) correlates of career progression simultaneously. Finally, on a methodological note, we use the parametric duration modeling technique to study career mobility. As we mentioned previously, the time that an employee stays in a particular position has a non-negative indicator function. The very nature of the distribution of this dependent variable lends itself to this modeling technique. Herein, we have outlined a summary of the analytic procedure that can be used to perform such analyses in the future. In addition, parametric duration modeling can be used in a predictive manner that is not possible with non-parametric duration models. We strongly recommend that this approach be the method of choice in future career progress studies. It also behooves us to mention some strengths and limitations of this study. We had access to survey data for a group of employees who joined the firm at the same time, and the organization was kind enough to let us update the promotion data on a regular basis. Owing to the nature of this study design, we have a fairly controlled setting, in that the firm, time, and gender context (all male employees) remain constant. However, variance in these dimensions may result in different career trajectories. For example, employees who join the firm during high growth times may have faster career trajectories than those who join during a recession. In this way, the very advantages of our context also turn out to be its limitations, as we elucidate next. The first major limitation of this research is the small sample size and the single organizational context. This narrow context may limit generalizability, because our findings may be specific to the organization under consideration. We have no reason to believe that our organizational context differed greatly from those of other Japanese organizations within the vicinity of this particular organization, but we also have no empirical data to show this similarity. Therefore, we advise caution when extrapolating the results to other contexts. The second limitation of this study is that it focused on a particular cohort of employees, all of whom entered the organization at the same time.
The underlying characteristics of this particular cohort could be unique and thus affect the nature of career progress within the organization. Similarly, alternative explanations might suggest that the single- and two-stage models evolve within the overall economic context. For example, during the explosive growth of the 1970s and 1980s, the twostage model may have been replaced by the singlestage model owing to the scarcity of managerial talent in the booming economic context. Slower economic growth in earlier time periods may have necessitated the use of the two-stage model, especially if promotion opportunities were limited, in which case human resource managers may have established the first stage with equalized opportunities for promotions. As evidence of the potential role of the economic context, we note that after the ‘bubble economy’ of high land and stock market prices collapsed in the late 1980s, Japanese firms were forced to change their internal career systems to provide alternate career progress routes. Despite these limitations, our study is among the first to question the two-stage model of Japanese career progress. Although a single study does not provide conclusive evidence, we believe that we have provided preliminary reasons to re-examine prior empirical findings. Therefore, we recommend the replication of our study in a larger setting, with more employees, and across multiple firms. At a broader level, it would behoove researchers to examine how managers in charge of personnel systems in Japanese organizations encourage competition and whether they portray career progression as a two-stage process. Also, it would be interesting to examine the processes that lead to high-quality early career experiences. For example, what characterizes those employees who develop effective superior–subordinate relationships with multiple supervisors? Do they continue to have similar relationships with supervisors during their entire careers? What other early experience factors may be important for predicting long-term career progress? Moreover, Japan has experienced major changes on the economic front during the past 15 years, which has influenced the ILMs of many Japanese firms. What are the implications of such economic issues for internal career progression, and have the implications of variables in our model changed? These are a few unanswered questions that should be investigated. Nonetheless, our study suggests that career progress in Japanese organizations during the high growth period of the 1970s and 1980s was perhaps
Journal of International Business Studies
Japanese career progress: an empirical examination
G Graen et al
160
more complex than the two-stage career progress model that has enjoyed widespread acceptance. Our study also identifies some important predictors related to the varying rates of career progress. Additional research should study the processes that underlie these determinants, and we encourage further longitudinal research in this direction.
Acknowledgements We thank Professor Arie Lewin and the two anonymous reviewers for their insightful feedback. We also thank Hal Angle, Pamela Brandes, Diya Das, and Maria Goranova for reading previous versions of this manuscript, and Keio University, Nagoya University, and Syracuse University for their research support.
References Aiken, L.S. and West, S.G. (1991) Multiple Regression: Testing and Interpreting Interactions, Sage Publications: Newbury Park. Allison, P.D. (1997) Survival Analysis Using the SAS System, SAS Institute: Cary, NC. Ariga, K., Brunello, G. and Ohkusa, Y. (2000) Internal Labor Markets in Japan, Cambridge University Press: Cambridge. Baba, M., Granrose, C.S. and Bird, A.C. (1995) ‘Career planning and career development of managers in Japanese firms and US subsidiaries in Japan’, Journal of Asian Business 11: 71–96. Bowman, M. (1981) Educational Choice and Labor Markets in Japan, University of Chicago Press: Chicago. Chen, M. (1995) Asian Management Systems: Chinese, Japanese, and Korean Styles of Business, Routledge: New York. Clark, R. (1979) The Japanese Company, Yale University Press: New Haven, CT. Cole, R.E. (1979) Work, Mobility, and Participation, University of California Press: Berkeley. Dreher, G.F. and Bretz Jr, R.D. (1991) ‘Cognitive ability and career attainment: moderating effects of early career success’, Journal of Applied Psychology 76: 392–397. Evans, P. (1989) ‘Managing human resources in the international firm’, in C.A. Bartlett and S. Ghoshal (eds.) Transnational Management, Irwin: Chicago, pp: 649–666. Evans, P., Pucik, V. and Barsoux, J. (2002) The Global Challenge: Frameworks for International Human Resource Management, McGraw-Hill: Boston, MA. Forbes, J.B. (1987) ‘Early intraorganizational mobility: patterns and influences’, Academy of Management Journal 30(1): 110–125. Graen, G.B. (2003) ‘Interpersonal workplace theory at the CROSSROADS: LMX and transformational theory as special cases of role making in work organizations’, in G.B. Graen (ed.) Dealing with Diversity, LMX Leadership: The Series, Vol. 1, Information Age Publishing: Greenwich, CT, pp: 145–182. Greene, W.H. (1993) Econometric Analysis, (2nd edn), Prentice Hall: Englewood Cliffs, NJ. Ishida, H. (1993) Social Mobility in Contemporary Japan, Stanford University Press: Stanford. Kalbfleisch, J.D. and Prentice, R.L. (1980) The Statistical Analysis of Failure Time Data, John Wiley & Sons: New York. Kiefer, N.M. (1988) ‘Economic duration data and hazard functions’, Journal of Economic Literature 26(2): 646–679. Koike, K. (1995) The Economics of Work in Japan, LTCB International Library Selection 3, LTCB Library Foundations. Tokyo, Japan. Lincoln, J.R. and Kalleberg, A.L. (1990) Culture, Control, and Commitment: A Study of Work Organization and Work Attitudes in the United States and Japan, Cambridge University Press: Cambridge. Morita, J.G., Lee, T.W. and Mowday, R.T. (1993) ‘The regression analog to survival analysis: a selected application to turnover research’, Academy of Management Journal 36(6): 1430–1464. Nakane, C. (1970) Japanese Society, University of California Press: Berkeley.
Journal of International Business Studies
Ono, H. (2001) ‘Who goes to college? Features of institutional tracking in Japanese higher education’, American Journal of Education 109(2): 161–195. Pucik, V. (1989) ‘Managerial career progression in large Japanese manufacturing firms’, in A. Need, G.R. Ferris and K.M. Rolan (eds.) Research in Personnel and Human Resource Management, JAI Press: Greenwich, CT, pp: 257–276. Rohlen, T.P. (1974) For Harmony and Strength: Japanese WhiteCollar Organization in Anthropological Perspective, University of California Press: Berkeley. Rosenbaum, J.E. (1979) ‘Tournament mobility: career patterns in a corporation’, Administrative Science Quarterly 24(2): 220–241. Rosenbaum, J.E. (1984) Career Mobility in a Corporate Hierarchy, Academic Press: Orlando, FL. Sheridan, J.E., Slocum Jr, J.W., Buda, R. and Thompson, R.C. (1990) ‘Effects of corporate sponsorship and departmental power on career tournaments’, Academy of Management Journal 33(3): 578–602. Turner, R. (1960) ‘Sponsored and contest mobility and the school system’, American Sociological Review 25: 855–867. Wakabayashi, M. and Graen, G.B. (1984) ‘The Japanese career progress study: a 7-year follow-up’, Journal of Applied Psychology 69: 603–614. Wakabayashi, M., Graen, G.B., Graen, M.R. and Graen, M.G. (1988) ‘Japanese management progress: mobility into middle management’, Journal of Applied Psychology 73(2): 217–227. Yoshino, M. (1968) Japan’s Managerial System, MIT Press: Cambridge, MA.
About the authors George B. Graen, a 1967 University of Minnesota PhD, taught at Illinois, Champaign-Urbana, Cincinnati, Louisiana, Keio, Nagoya, HK University Science and Technology, and has published over 150 research articles in journals such as Journal of Applied Psychology, Academy of Management Journal, Administrative Science Quarterly, and books including LMX Leadership: The Series (2003, 2004, 2005). Ravi Dharwadkar is an Associate Professor of Management at the Whitman School of Management at Syracuse University. His research in the areas of organizational behavior and organization theory appears in the Academy of Management Review, Academy of Management Journal, and Academy of Management Executive.
Japanese career progress: an empirical examination
G Graen et al
161
Rajdeep Grewal is an Associate Professor of Marketing at the Smeal College of Business Administration at the Pennsylvania State University. His research in the areas of consumer behavior, managerial marketing, and research methods appears in the Journal of Marketing, Journal of Marketing Research, and Marketing Science.
Mitsuru Wakabayashi is a Professor of Organizational Behavior and International Management at the Faculty of Management, Aichi Gakuin University, Nagoya, Japan. He has published widely in the area of leadership in journals such as Journal of Applied Psychology, Group and Organization Management, and International Human Resource Management.
Accepted by Arie Y Lewin, Editor-in-Chief, 30 May 2005. This paper has been with the author for two revisions.
Journal of International Business Studies
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
