This article presents an empirical analysis of the flow of individuals along sequences of jobs, what can be called career mobility. The analysis follows the career mobility of a single cohort through a large corporation over a 13-year period, using the official personnel records. Because of the extensive checking procedures used to verify these records, these data provide an unusually accurate source of time-series information on career mobility.
The conceptual focus of this article is the issue of whether early career positions and changes in status affect later careers, apart from the intervening positions held. This issue has been discussed in the status attainment, Markov, and organization-career literatures. One version of a historical model-the tournament model-is described, and a number of hypotheses derived from it are tested. The empirical analysis supports the historical-effects position, finding that very early job moves are related to subsequent mobility even a decade later after employees have moved on to second and third jobs. Indeed, mobility in the earliest stage of one's career bears an unequivocal relationship with one's later career, predicting many of the most important parameters of later moves: career ceiling, career floor, as well as probabilities of promotion and demotion in each successive period.
Theoretical Background Historical Antecedents of Mobility
The central issue of the present inquiry is whether individuals' career histories are associated with their future careers independent of their current positions. This issue is being argued on several fronts. In some status attainment research, a historical model is proposed that shows income and occupation at a given period as being directly influenced by income and occupation at all previous periods; Featherman's (1971) analyses provide some support for this model. Kelley (1973a, p. 492) offered empirical analyses challenging the historical model and concluded that "as a man's career progresses, past failures are forgiven and past successes forgotten." Subsequent exchanges between Feather-man (1973) and Kelley (1973b) elaborate on the issue, but the question of historical effects remains unresolved. Both, however, agreed that investigation of sequences of jobs and job characteristics would be useful in resolving the question.
Markov models provide another way to analyze historical effects by using simple transition matrices that show the probabilities of moving from each time 1 position to each time 2 position. In Markov analyses, inferences are made about the consequences of transition matrices based on simple assumptions. As Mayer (1972, pp. 312-313) states: "The basic principle which distinguishes Markov models asserts that the status category a person will occupy in the future depends only on the status category he occupies at present and not at all on the categories he has previously occupied. This is sometimes referred to as the principle of path independence. ... [If two people] have different status histories, but their status levels at the time recorded are identical... a Markov model would make identical predictions about their future [mobility]."
Tests of path independence are rare because of the paucity of mobility data at three points in time. Consequently, some major work in this area admits that this may be an unrealistic assumption, but then stresses models based on this assumption because the mathematics is more tractable (Bartholomew 1968, p. 9). Hodge investigated the path-independence assumption using several sets of data on intergenerational and intragenerational occupational change. Each analysis showed slight departures from path independence, particularly for intragenerational occupational changes; but for the most part his findings, like Kelley's (1973a), supported the path-independence assumption. March and March (1977), analyzing careers of school superintendents, also found support for Markov models.
One troubling finding has been a tendency for Markov models to underpredict the number of nonmobile individuals (Blumen, Kogan, and McCarthy 1955; Hodge 1966). Although this finding has not been used to question the path-independence assumption (which is some-times called "the Markov property"), it has led to adaptations of the Markov model that modify its other two assumptions. The stationarity assumption asserts that the probability of mobility over a fixed time internal is independent of when it occurs in that interval, and it depends only on the length of the time interval (McGinnis 1968; Ginsberg 1971; Mayer 1972; Serensen 1975). The homogeneity assumption asserts that all persons have identical transition probabilities (Blumen, Kogan, and McCarthy 1955; Goodman 1961; Bartholomew 1967; McFarland 1970; Spiler-man 1972).
It is not clear why semi-Markov models have ignored career histories. Although history is incorporated in a limited sense in the age-decline and duration-decline models, none of the semi-Markov models questions the path-independence assumption. A path-dependence semi-Markov model would posit that an individual's early mobility history influences later mobility and so might be called, for simplicity, a historical model (although non-stationarity models are also historical in a more limited sense). Of course, historical effects are difficult to study, but detailed analyses of historical effects may contribute to more realistic assumptions and to more appropriate applications of models.
Much research on organizational careers has dealt with the question of historical effects, and its conclusions contradict the path-independence assumption in the status attainment and Markov literatures. Since the 1950s, many researchers have studied job mobility, and they have emphasized the crucial importance of career paths. Organizations quickly "size up" new employees and allocate them to different training and socialization experiences (Becker and Strauss 1956; Berlew and Hall 1966; Peres 1966; Campbell 1968; Van Maanen 1977; Schein 1978). Furthermore, this selection and allocation is repeated, with later assessments being based not only on age and duration (nonstationarity) (Roth 1963; Jennings 1971; Bray, Campbell, and Grant 1974; Kanter 1977), but also on one's particular career history (Dalton 1951; Glaser 1964; Jennings 1971; Faulkner 1974) and one's advancement on the "correct" path (Kanter 1977).
Unfortunately, although these studies have described selection and socialization processes in extensive detail, they have been far less thorough in examining actual long-term career patterns. Patterns of career moves are rarely described in detail, and many studies (e.g., Martin and Strauss 1959) have relied on respondent reports so that they seem to be descriptions of ideal careers rather than actual job moves (which might be less simple to report). The difficulty of obtaining appropriate longitudinal data is understandable and may explain why these micro level studies of organizational careers have never been integrated with macro level studies in the status attainment and Markov literatures. Systematic longitudinal studies of actual career paths would permit a more direct test of the path-dependence issue in organizations.
