It should be noted that the age variable used in these analyses, age of entry into the firm, is very similar to the variable that economists customarily use to reflect previous experience (that gained before joining the current employer). The calculation of the two variables is highly similar (entry age is age minus tenure; previous experience is age minus tenure minus years of education minus 5), and the two are likely to be highly-though not perfectly-correlated in most cases. The main difference here is a matter of context. This variable means something different in a single organizational hierarchy than it does in the open labor markets studied by survey data. As we shall see, the present findings clearly indicate a different pattern of results.
College Selectivity Effects
College effects have attracted the interest of social researchers, but they have been regarded in very different ways. According to human capital theory, graduates of "better" colleges attain greater socioeconomic success because these colleges are more selective and provide a better education. According to sociologists, "better" colleges have more successful graduates because these colleges confer greater status to their graduates and these status properties increase their graduates' chances of success (Kamens 1977; Mayer 1977; Collins 1979). Mayer (1977) contends that colleges actually have implicit "social charters" which sometimes give highly specific social statuses and career niches to their graduates.
Operationally, economists have measured college quality in terms of rankings of the selectivity of colleges, using Astin's (1965) scale, which is based on the average Scholastic Aptitude Test (SAT) score of enrolled undergraduates.2 Economists interpret this ranking as a rough indicator of individuals' ability, assuming that individuals who attend more selective colleges are generally more able (see Weisbrod and Karpoff 1968; Wise 1975a). However, since more selective colleges are also It should be noted that the Astin scale was formulated in the early 1960s, while most of my sample graduated from college in the 1950s. It is possible that some colleges changed in student body composition after some of my sample members graduated. Although I have no systematic way to adjust for this possibility, my subjective impression is that the colleges which are most heavily represented in the sample would have ranked similarity over the previous decade.
However, empirical analyses can discern whether colleges affect job statuses and earnings, and, if the analyses are longitudinal, they can reveal the timing of these effects. As previously suggested, human capital theory suggests that ability affects earnings gradually over the first decade, so it would predict a gradually emerging effect of college selectivity. Social structural theory suggests that individual status attributes (like college status) affect attainments from the outset. To the extent that longitudinal analyses support one or the other prediction, they shall offer support to that theory and its interpretation of college selectivity.
These issues are not merely of theoretical interest; they have important practical implications. For instance, if occupational social structures are directly responsible for a large amount of the earnings variance, then the earnings distribution may be relatively unresponsive to changes in market factors or to changes in individuals' educational attainments unless the social structure is also altered. The issue of whether social structure precedes, and in some sense influences, earnings differences have crucial policy implications.
In order to compare the human capital and social structural theories, the present study compares the determinants of earnings and hierarchical level attainments for an entry cohort. Since the theories differ in their primary dependent variable, separate analyses with each dependent variable are done. Because the theories differ in their predictions about the timing of the impact of factors, these analyses are conducted longitudinally over a 13-year period so that changes in the impact of factors can be investigated.
In the present analysis, the structurally determined component of salaries is conceived as being the result of the level hierarchy of the organization. Of course, this is a simplifying assumption: It ignores other selection systems which may also contribute to the salary structure and it ignores antecedent selection and evaluation processes which may contribute to the nature and operation of the level hierarchy. However, for present purposes, an understanding of the two-stage model is attempted through a description of the operation of the level selection system as a distinct system, the ways that the level selection system contributes to the salary structure, and the ways that individual characteristics influence salaries independent of the level system.
The analyses required for the first stage in examining the influences of personal characteristics on levels have already been conducted. The second stage calls for analysis of the influence of levels and personal characteristics on earnings. The analysis is conducted by adding the level variable to the previous regression of personal characteristics on earnings. The level coefficient is quite large, and it brings the explained variation of the previous earnings regressions up to 74-89%. Indeed, the level co-efficient increases over time so that it explains increasing amounts of the variance, while the individual characteristics alone explain increasingly less. The level to which an employee is allocated has a large influence on earnings, even after controlling for individual characteristics; and this influence increases over time.
The influences of most personal characteristics on earnings are markedly reduced after controlling for level. Although entry age retains much of its influence on earnings when level is controlled, tenure and non-B.A. have much smaller influence, particularly in the later periods. By 1969 and 1972, the greater part of their influence on earnings is eliminated when level is controlled. Specifically, over four-fifths of the influence of having a college degree and over three-fifths of the influence of tenure on 1972 earnings are eliminated when level is con-trolled. Presumably, the primary influences of these factors would seem to be on level allocations in these later periods, an inference supported by the previous finding that they do have large influences on levels.
The progression of individuals from level to level seems to operate by some clear processes. Individuals' level attainments at any period are strongly influenced by their previous level attainments. This is not merely a matter of non-mobility, for the actual attainments in this cohort increase dramatically over these intervals. Despite large amounts of mobility, the employees who had attained the highest levels in one period tend to be at the highest levels in the next period. The individual characteristics that influence early level allocations are built into the level system and preserved into the future. For instance, the favored colleges affect initial levels and give an additional benefit between 1962 and 1965. Thereafter, their effect is carried forward into later periods without any further additional impact. Thus individuals are systematically allocated to these level classifications on the basis of some criteria and not others, these selections systematically operate at some times and not at later times, and individuals are systematically transferred between level classifications from one time to the next. These patterns suggest that the level classifications are indicators of a real system.
This level system seems to operate somewhat differently than the earnings system, and in some cases to make selections which are only subsequently manifest in earnings differences. In particular, the preferred colleges affect levels earlier than they affect earnings. If the level system is conceived as having a mediating effect, it is not a simple simultaneous process, but a lagged process. The level system does not immediately pass along the college effect in 1962. Rather it responds to the favored colleges very early, and it transfers these college effects to earnings at a later time.
Human capital theory does relatively well at explaining the gradually emerging and gradually increasing influence of many of the individual characteristics on earnings in a one-stage model. However, the analyses also suggest some reasons for doubting the one-stage model, and human capital theory fares much less well in the two-stage model. Human capital theory has only minimal value in explaining the determinants of level attainment. Its prediction of gradually emerging and gradually increasing influences does not hold for levels, which show initially large college effects and a sudden discontinuation of additional college effects. Since levels explain the greatest portion of earnings variance, the human capital theory has minimal explanatory value if one accepts the two-stage model.
