Reformulating Turner's ideal types, there have been proposed a tournament mobility model which is a historical model (Rosenbaum 1976). In the tournament mobility model, careers are conceptualized as a sequence of competitions, each of which has implications for an individual's mobility chances in all subsequent selections. Although tournaments can be constructed with numerous variants in the rules, the central principle involves an important distinction between winners and losers at each selection point. Winners have the opportunity to compete for high levels, but they have no assurance of attaining them; losers are permitted to compete only for low levels or are denied the opportunity to compete any further at all. As in a contest model, winners must continue competing in order to attain high levels, for there is no assurance of continual advancement; but, as in a sponsored model, early selections have irreversible consequences for losers. The tournament results in a progressively greater winnowing down of the winner's cohort at each successive stage. In contrast, the losers are relegated to "minor tournaments" for lesser positions. Unlike sports tournaments which eliminate losers, the career tournament retains losers but lowers and narrows their options.
The tournament model grew out of research on mobility patterns in a school track system and in students' postgraduate careers. The research 4Turner proposed his models as ideal types to describe normative systems, not actual mobility patterns. However, one kind of research suggested by his ideas "is an exploration of different channels of mobility ... to discover the extent to which mobility corresponds to each of these types" (Turner 1960, p. 865). The present article is an attempt to pursue this suggestion.
Theoretical Background
In some ways, the tournament model is an abstract formulation of some of the observations in the organizational careers literature. It posits that assessments in an employee's first few years have profound and enduring effects on later career outcomes (see Berlew and Hall 1966). It also posits that assessments occur repeatedly, and an employee must continue to pass these hurdles in order to advance. How-ever, the tournament model goes beyond most of the career literature in hypothesizing some further patterns: for example, the relationship of early promotions to specific career outcomes such as future promotion chances, level attainments, and career ceilings and floors. Although these hypotheses are consistent with findings in the career literature (Kanter 1977), the actual patterns of career mobility over long periods of time have rarely been described.
Hypotheses of this sort have rarely been tested because time-series data over a sufficiently long interval are difficult to obtain, and that, in fact, was a limitation in the tracking study. The data available in the present study are unusual in permitting analysis of career mobility over a 13-year period, thereby providing the first opportunity for testing these hypotheses in an organizational context.
Level Mobility in an Organizational Hierarchy
In the 1970s and 1980s, research on mobility in the United States has largely focused on occupational status changes. This is a particularly valuable approach to mobility between generations, but when applied to intragenerational (career) mobility, it overlooks the patterns of mobility within occupations or organizations that may be the most common form of mobility for some segments of the labor force.
In many occupations, change of levels (or status) in an organizational or professional hierarchy is the main kind of advancement. Most professionals, managers, and skilled workers tend to stay in the same occupation throughout their work lives (Reynolds 1951, pp. 19-36; Lipset and Bendix 1952; Blau and Duncan 1967). The aim is descriptive analysis, not causal inference. These analyses do not seek to assert that early career paths cause later career mobility. Rather they seek to ascertain whether early career paths are related to later career mobility. Regardless of whether such a relationship indicates a unique causal influence or a mediating influence for other causal factors (e.g., sex, ethnicity, and supervisors' ratings), the existence or nonexistence of such a relationship would have important implications.
In the first place, if such a relationship were discovered, it would identify highly visible social signaling cues about individuals' career futures (Spence 1974). Because of the diversity of selection criteria and irregularity of their application (Kanter 1977) and the invisibility and lack of clarity of selections (Goldner 1965), employees often have difficulty inferring their likely career futures. This unclarity is even truer in judging other employees-supervisors, peers, and subordinates-for whom the relevant attributes (education, supervisors' evaluation) are often unknown to observers (Berg 1971, p. 78). Regardless of whether previous career paths have a causal influence on later careers, if they are related to later careers, then they are clearly visible signals which may influence the way employees are regarded and treated by others. Even as a phenotypic phenomenon, this would illuminate an important stratification process in organizations which may have ramifications for social interaction.
Furthermore, the state of knowledge about organizational careers is such that the very basic descriptive knowledge is lacking. Are there patterns of career paths in organizations? Are early career paths related to employees' likely career futures? If so, what aspects of early career paths are related and how early does this phenomenon appear? Does such a relationship continue to hold, even after controlling for present position? These basic descriptive questions are fundamental to an understanding of organizational careers and, in particular, for understanding whether Markov or semi-Markov models, which require the path-independence assumption, can be applied to organizational careers.
Given the aim of this article, nonparametric statistical techniques are quite appropriate for the analyses. Although multivariate tests are commonly used, they are not necessary for these hypotheses, and they have the disadvantage of requiring assumptions about the shape of promotion probabilities (e.g., as linear, logarithmic, or logistic). At this stage of analysis, the choice was made not to constrain the relationship with such assumptions because neither theory nor empirical work suggested a particular functional form. Of course, the costs of these assumptions are also accompanied by certain advantages of multivariate analyses, particularly if one desired to model the complete status attainment process in an organization. However, the hypotheses to be tested here are straightforward, descriptive ones intended to test specific relationships in organizational career paths, and they can be tested very suitably by nonparametric tests (Goodman 1962).