7—
Segmentation of the Labor Market and Information on the Skill of Workers

The Fundamental Uncertainty of the Labor Contract

No firm that hires a new person knows quite what it is getting, or in particular what it will get on the average over the eight or so years it can expect to employ the new person. Even while the person is employed the firm does not know exactly what it has got. Let us start with the measurement of performance after the worker gets the job, since if that is impossible to obtain, predictions of that performance at the time of hiring cannot help but be yet more inexact (March and March 1978).

We will analyze first performances of athletes in professional athletics, for statistical measurement of performance is surely better developed there than in practically any other area (Kahn and Sherer 1986). Measurement is done by outsiders rather than by management, so that managers need not decide on a cost basis how carefully to measure performance. Athletes on baseball, basketball, football, or soccer teams work under standard conditions, conditions very nearly identical to those of people who play the same positions on other teams. The objectives on all teams are identical: to score in a well defined set of ways in a standard number of games against a set of teams that are nearly identical in composition. The performance appropriate for different positions is also measured in standard ways, so that in baseball, for example, batting averages, earned run averages, or fielding averages span performances appropriate (in different mixtures) to outfielders, pitchers, and infielders.

The difficulties with even this highly developed measurement system for the work of athletes are well known. The first, most obvious difficulty is that the measurements are statistical, and (given that half of all the teams lose and the other half win, except in games like soccer where ties are common) they are statistical under conditions where certain kinds of uncertainty are maximized as far as the opposing teams and the overall

managers of the competitive system can manage (for a description of the informal management of such a system of competition, Balinese cockfighting, see Geertz 1973a). This means that every measurement is only as statistically reliable as the sample size permitted by a season of games (or the season so far) makes it. For example, scoring is a very unreliable measure in soccer, because the expected number of scores per person in a season is very low. Pitchers who pitch few games, or passing by backs other than the quarterback, present comparable problems. But these are only extreme forms of the common problem of random measurement error in all performance statistics.

In addition, different sports depend to different degrees on teamwork, and different positions or roles in any one sport matter more or less to the performance of others. Thus, individual performance statistics cannot adequately summarize performance, because the qualities reflected in one person's scoring (or other performance) include qualities of teammates. This fact has shown up historically in American sports in the slower racial integration of the positions of pitcher and catcher in baseball or of quarterback in football, where the importance of the position's contribution to teamwork makes measurement of performance difficult. Some measures of performance (e.g. the number of double plays in baseball) are strongly affected by inseparability of teamwork performances.

Finally, some star performers are very popular with the fans, while some are a good deal less popular. Except for the factor of race (where it is fairly clear that in basketball a white star is more valuable to a team than a black star of equal performance, because there are more white fans than black fans, and whites prefer a star of their own race; see Kahn and Sherer 1986), the ability to get favorable notice in the news and bring fans to the stadium is not easily measurable.

So even in athletics, under the best of conditions, employers do not know quite what value they are buying for what they pay a worker.

Further, the very excellence of the measurement demonstrates a second point, that the performance of athletes over the course of their playing life and the length of that playing life are not well predicted by scouts or by recruitment contract rankings. Sports recruiters find out more surely than other personnel departments when they have made a mistake in judgment (in fact, the mistake rate of personnel departments may be the worst-measured performance characteristic of the business world, and surely is in the civil service). But of course, many sports have high injury rates as well. The long-run performance of an American football running back, even with a long series of knee operations, is not very well predicted by whether the knees held up through college.

This difficulty of predicting an athlete's performance at the time of recruitment occurs even though the situation in which the measurement of the talents and training of the recruit takes place—namely, college or amateur sports or farm clubs—simulates the actual work situation fairly well. This differs from the situation encountered by a recruiter of, say, engineers, who cannot examine many bridges built by a budding civil engineer, not even bridges built in a farm club that is not in the same league as a major civil engineering firm.

Consequently the contract of an athlete in any given year of his or her career depends intimately on current measurements of productivity, rather than on certificates consisting, perhaps, of batting averages from his or her college or farm club days. The bonuses for signing are of course a partial estimate of how much of the athlete's total value can be reliably predicted at recruitment time. In other parts of the economy, the labor contract has seniority provisions against layoffs and for moving up to better jobs by bumping less senior workers or for tenure after a six-year trial period, types of evidence of employer and employee unwillingness to make continuation and reward depend on current performance. These structures of tenure are wiped out in professional athletics by the possibility of good measurement of current productivity.

If seniority or tenure in the labor contract indicates poor measurement of current productivity, the employer—and even more, perhaps, the employee—must be uncertain about what the employer has in fact bought. This uncertainty is lower in athletics than in other sorts of work.

Part of the reason the possibility of measurement and prediction is so much worse in most jobs than in athletics is that the employer outside athletics will not be playing the same game as now in the future: the market will change, the technology will be improved, a different sort of teamwork will be needed because different groups will be working together, and so on. The artificial rigidity of the rules of sports makes it possible to predict what one will want in eight years in a third baseman. In eight years in the automobile industry, General Motors may be competing with the Japanese on quality rather than with Ford on style and price, and so may be wanting workers who will suggest how to avoid mistakes rather than workers who can keep the line going whether the carburetor gasket fits or not. An employer outside athletics wants to recruit now someone who will be good at doing whatever the firm decides to do in eight years, rather than someone who will still be good at what a third baseman has always done.

Part of the reason the labor market as a whole is less good than athletics at predicting productivity is that schools are not in general good

simulations of work life; the measurements at career beginnings are of what makes potential recruits good at schoolwork, with an overemphasis on intelligence and an underemphasis on qualities of character. Insofar as school recommendations do comment on qualities of character, we would all be better off to ignore them, because ordinarily the teachers who write the letters have very little opportunity to observe those parts of life in which such qualities play a great role.

Part of the reason the labor market as a whole is worse than athletics at measuring productivity and rewarding it is that employees will not stand for rewards based on performance unless they get a lot out of it. One can hire professional athletes even though one is going to fire them at roughly thirty-five years old on the basis of declining performance statistics, to withhold seniority privileges, to make their future income uncertain depending on such things as the success or failure of knee operations, only because one is willing to pay a lot for the years they work.

Not knowing what one is buying, or even what one has bought, makes the labor contract a perennial information problem for management. It might be imagined that at least the worker knows what he or she produces, but this also is not true. What workers conceive as valuable to the firm is strongly determined by the jobs they occupy. Manual workers typically believe that what really holds a firm together is physical production; clerical workers believe that if the flow of paper does not go right, everything will come apart; managers believe that the bottleneck factor in company success is management; financial officials of the company believe that the proper structuring of debt and equities is the fundamental determinant of the mixture of risks and profits of stockholders and debt holders, and so determines whether or not the firm will continue to exist. Thus, at least some of them must overestimate their productivity for the firm. And employees very generally have virtually no idea what their talents would bring on the open labor market.

Institutional Substitutes for Measurement of Productivity

Mutiple personnel information systems in organizations and in the labor market are directed at filling in the gaps in this picture of poor measurement and poor prediction of the productivity of labor. These systems of information include certificates from the educational system and sometimes from the professional system as well (e.g., specialty boards in medicine); union membership in the crafts; promotion decisions in internal

labor markets; informal respect in the work group; whatever experience firms have about whether men or women work out better under the salary, promotion, and incentive systems they typically set up for clerical workers; job evaluation schemes; recommendations by workers in the plant about new hires (a remarkable proportion of workers are recruited this way); footnote citations that distribute honor in the scientific community and so indirectly determine appointments, promotions, and salaries in universities; and so on.

The information that leads employers to prefer one kind of workers to another (e.g., that leads large manufacturing corporations to prefer people already employed there but that leads universities below the very first rank to prefer not to hire their own graduates as assistant professors) means that different segments of the labor force will have different opportunities: different incomes, different levels of employment security, different chances of promotion, and the like. Consequently, labor market segmentation is fundamentally a solution to a problem of the information structure about people and their performance that confronts organizations and employees. But since people's rewards depend on these flows of information, they have an interest in having the flows contain information that says they are better than others, whether that information is trustworthy or not (Edwards, Reich, and Gordon 1975).

The labor market part of the problem of organizations is shot through with agency problems, problems of the two parties to a contract having different amounts and kinds of information and wanting to use those information differentials to their own advantage. The example of an agency problem we gave in Chapter 1—whether a secondary school teacher's certificate of audiovisual competence was information about whether that teacher is likely to select good instructional films—demonstrates the deep problem of getting unbiased information about worker productivities. And it is this impossibility of gaining good information about employee performance that creates many opportunities for employees to monopolize opportunities, by monopolizing the chance to generate good information about their performance. We find relatively little monopolization in professional athletics (for instance, the residual racism in athletic teams' hiring and promotion is not at all comparable to the racism or sexism that still exists for jobs paying comparable salaries in corporations or universities), because in spite of the cautions we have outlined above, one can measure the productivity of athletes very well indeed. The purpose of this chapter is to examine the interplay of information on performance, agency problems biasing the information, and monopolization of opportunities by segments of the labor market.

Types of Information about Work Performance

An organization needs information about the immediate past work performance of a worker so as to allocate rewards fairly, and it needs information about future work performance in decisions about hiring, promotion, and retention. In order to be used as a basis for rewards, information needs to be, to some degree, "auditable," so that people do not simply get to choose their own reward levels. In order to be useful for predicting future performance, an employer needs information about response to a range of conditions that fairly represent the expected future of the work the person will do, and again the information needs to be auditable, so that workers do not unilaterally decide on their retention and promotion and potential recruits do not determine who should be hired by providing false information.

The general problem with information systems about employees is that the more accurate and wide-ranging a person's observation of the worker is, the more likely it is that the person will be a fellow worker (or a fellow student). This brings up two problems: the fellow worker may not be competent to render a judgment that would be accepted by a reward-giving or hiring executive; and the fellow worker may have his or her own fate intimately intertwined with the information given, and so be motivated to withhold or distort it. Informal pressures among fellow workers tend to discourage giving accurate information to management, especially if that information is negative (see, e.g., Halle 1984, 121–122, 138–144, 161–162). Information from a competitor for the same promotion might, however, be shaded in a negative direction for reasons of rational interest, while information from a fellow worker might be negatively shaded for vengeance if the worker had been too attentive to the evaluator's wife.

Information from hierarchical superiors tends to be based on less detailed observation but to have different, generally smaller, agency problems (but see the analysis below of why union firms are more productive, for contrary evidence). The information that a superior gives about the worker's performance gives the superior partial control over a reward for that worker. The superior, then, may trade the benefit of giving more favorable information to superiors for cooperation. While the organization generally benefits, in the short run, from the cooperation, the degraded information about the worker is a cost that must be paid in return. Of course, the cooperation required of a supervisor does not always contribute to organizational objectives, as sexual harassment cases demonstrate (cf. Charlton 1983).

Information is therefore often based on measures of performance that are easier to audit, such as units produced, hours recorded by a time clock, sales made, or (higher up the hierarchy) costs reduced, profits increased, delivery times reduced, scrap figures down, and the like. There are four broad types of difficulties with such measurements. The first is that they too can be manipulated by interested workers on the scene; for example, bargaining between workers and first-line supervisors about how much time should be allowed against a piece rate for down time of a machine is common in machine shops, because management cannot get accurate measurements of that down time from anyone but the workers and supervisors involved (Burawoy 1979).

The second difficulty with formal performance measurement is that the measures used may not be good estimates of true performance. In shoe factories in the Soviet Union where performance was measured by numbers of shoes produced under conditions of leather shortage, for example, factories produced only shoes of the smallest size (Granick 1954). When responsibility in dealing with extraordinary circumstances is a core function of a job, routine measurements of productivity tend to be worst exactly when the workers are doing the extraordinary parts of their jobs the best. For example, more soldiers under one's command are killed while one is successfully attacking or defending, and productivity of a chemical plant is lowest when the workers are putting out a fire.

The third difficulty is that such auditable measures tend to rigidify the measurement of performance, so that technical change or reorganization undermines the measurement system. With rigid performance measurement, the interests of workers embedded in the measurement system tend to lead to sabotage of technical and organizational change.

Fourth, such specialized measurement is generally expensive. One still needs the pyramid of supervision even if productivity is measured separately. One needs cooperation and the transmission of informal skills within the work group (and one has to pay the costs of cooperation) to deal with the ordinary contingencies of work, even if peer judgments are not used in the personnel process (cf. Blau, [1955] 1963). Because, for example, one cannot substitute the records of productivity, produced by an inspection department, for supervision by the foreman of a machine shop, measurement of performance by inspectors is an extra expense. Special performance measurement therefore tends to be narrowly focused on a few aspects of performance and to be done at a high level of abstraction, since it would be too expensive if it were not radically simplified.

For hiring and promotion decisions one needs to measure performances that will predict what kind of worker a person will be in the fu-

ture. Even if one or the other of the performance measurement types outlined above has been carried out before the hiring or promotion, predictive information has four difficulties additional to those associated with the measurement of current and past performance.

First, the performance that the hirer or promoter tries to measure may not predict the sorts of performances that will be required in the future job. Ordinarily the person doing the hiring cannot predict exactly what the performance requirements will be for the job being filled, because of technical and organizational change or because personnel officers do not know much about the job themselves. A crucial factor in the prediction may be how well a person is accepted by the future fellow workers and supervisors who can teach him or her to do it well; personnel people can usually predict this only by using crude categories such as ethnicity and sex, and those are not very good predictors.

A second additional trouble is that the performances measured previously may be systematically different from those required in the job a person is being recruited for or promoted to. Often much of the information for recruitment is about schooling, and the performances measured there (typically tests in courses or general ability tests), while fairly well audited, are not good samples of the tasks required for success in jobs. Work done in a lower position is often not representative of what the worker will need to do in a higher position (particularly in terms of "responsibility," management of people, and management of information flows at a higher level of abstraction, all of which tend to be more important in higher positions), so that excellent performance lower down may not predict good performance higher up (this uncertainty can be partly controlled by letting people who might be promoted pinch hit for superiors when those are sick or on vacation).

A third broad additional difficulty with recuitment information is that people providing the information often have no commitment to the organization they are supplying information to, so that letters of recommendation all rank people as among the top 10 percent. This is one reason why the more auditable parts of people's previous experience (grades rather than recommendations, and grades in calculus rather than grades in sociology) tend to overemphasized.

A fourth broad difficulty with recruitment and promotion information is lack of comparability. It is well known that it is harder to get good grades at Swarthmore or Reed or some other leading places than at less distinguished colleges and universities, but there is no general formula to translate grades from one school into their equivalents at another. Similarly, within a company some supervisors will be well known to give gen-

erally very high ratings to their subordinates, others to be somewhat tougher, but no general translation to a standard degree of softness or toughness is likely to exist.

Even without such variations in the standards of the judges, there would be differences in what is being measured. An engineer who has been doing an excellent job selling airplanes produced by a particular company does not show the same virtues in his or her work as an engineer who has been doing an excellent job designing efficient heat transfer from the exhaust to the incoming air and fuel of the engines of those airplanes. Even if both workers have substituted successfully for their superiors, their superiors have different jobs as well. Thus, prediction of future performance is even more difficult than measurement of past and current performance.

A General Theory of Certification

By a "certificate" we will mean a summary of information about some performances of a worker or a candidate for a job that has been subjected to some sort of special procedures that are thought to make it more trustworthy (the general approach of this section is taken from Heimer 1984; see also Shapiro 1987). The first thing we have to do, then, is to distinguish different certificates according to the degree of formalization of both the sampling of performances and the protection of the trustworthiness of the judgments. At the extreme low end of formality is the "certificate" of seniority in a particular job. The universe sampled in the knowledge that a person has held a job for some time is the correct universe for assessing performance in that job (insofar as the job is not predicted to change much), but the sampling process tells us only that no negative information has come to light that was sufficiently serious to fire the worker. No particular efforts have ordinarily been made to check and protect the trustworthiness of the information, to make sure that everyone who should be fired has been fired.

At the high end of the variable of formality is the degree (with a transcript) from a particular college or other school, which ordinarily specifies in what courses performance was sampled and certifies that the judgments were made by people relatively uninterested in providing spurious information. One can be reasonably confident that much of the information was derived from pencil and paper tests or from writing assignments, and that the subjects of those tests would be recognizable by someone who had taken a similar course at another school.

Certificates can be produced as an offshoot of all the types of informa-

tion systems described above, by fellow workers or bodies of peers, by superiors giving an overall judgment, by systematic performance measurement systems of a more or less numerical and auditable sort. Those providing the information may have sampled performances highly predictive or less predictive of job performance, may have more or less commitment to providing the hiring or promoting organization with accurate information, and may be providing information in a common metric (such as Graduate Record Examination scores) or in noncomparable form. The central sociological point of all this is that all information that is taken into account in recruitment, retention, or promotion tends to need legitimation. The main reasons for this are practical: performance measurement requires observation of a worker or recruit in performances that are relevant to the work he or she is to do; judgment of that information requires that the judge too be competent in the same performances; and transmission of the information requires honesty and discretion on the part of the transmitter.

Second, it is not only performances in school that become certificates. Seniority is almost always a part of the formal record, and is roughly of equal importance to education in determining job level, job security, and pay. Seniority is a crucial certificate of qualification, particularly for the job a worker now holds, because what it lacks in systematic measurement and trustworthiness of the judgments it makes up in relevance. The performances of which it is a sort of record are those that are crucial for doing the job. If labor market segmentation is due to differences in various populations' opportunities to give acceptable certificates of their competence in a job, then a big source of segmentation is the tendency of most jobs to be awarded next year to the person who holds them this year.

Seniority is also one of the components in promotion systems or "internal labor markets" that characterize large bureaucratic firms, the various levels of government (especially in educational, health, and military branches), and certain large nonprofit organizations such as private universities. Since in general not all workers get the same number and size of promotions, and since educational qualifications determine which promotion ladder different people are on within the same organization, seniority in this case is more complex than in the case of a job qualifying a worker for that same job. Most promotions go to whatever groups are now employed in large bureaucratic organizations, especially to those who entered from college. We have suggested in Chapter 2 some of what seniority in positions to which one is recruited from higher education may be certifying.

The other main source of certification in internal labor markets, again especially in those career ladders that start with recuitment from higher education, is the formal opinion of superiors. As one gets nearer to the top of the hierarchy, more of such opinions are formed in committee rather than by individual superiors, which means that to some degree the top ranks of the "internal labor markets" of large organizations must be treated as a subtype of craft and professional certification. At the top people are certified by a body of peers.

The third main type of certification is based on observation of work by people already certified as skilled in that work, and it is protected from the most extreme forms of interpersonal logrolling by a "committee of peers"–type structure. As Amitai Etzioni has pointed out (1975), committees of peers are actually made up of one's superiors, not one's peers. (The original notion presumably was that the peers are peers of one another, like the House of Lords, not peers of the candidate.) What the committee of peers structure means is that no one person is a particular hierarchical superior who has the certification within his or her own power. Instead, for example, the associate and full professors of a department meet as a committee to recommend to the dean whether a given assistant professor should be promoted, or the body of craftsmen meet to certify someone as a journeyman. Internal labor markets in corporations and large bureaucracies tend to be more craftlike toward the top than they are in the middle, because as one advances influence over promotion chances begins to pass away from a single hierarchical superior toward a group, the "management committee" or, in the extreme, the board of directors.

The basic idea of peer certification systems is that the true requirements of a role can be taught and judged only by those who can do the role themselves, who have teaching and supervisory responsibility for the candidate for the role, and who have multiple informal contacts with the candidate. Quite generally a strong seniority element is built into these certifications, as in the fixed term of apprenticeship in many traditional crafts or the fixed term of internship or residency of physician trainees. The expectation is often that anyone admitted to the peer tutelary period can learn to perform the roles involved, so years spent in the apprenticeship are in effect the only measures of competence in the certificate.

Quite often, however, there are additional examinations or performances (e.g., the traditional "masterwork" of some artistic crafts) or additional examination of the work (e.g., the assessment of published work by a committee on a promotion in a university). So while assistant professors rarely get promoted without a respectable five years' seniority in

the job, they may not get promoted even with the seniority because their work is not judged to be up to the standard of the profession.

We will expect the features of the certification system for particular positions to vary with the character of the information problems involved in those positions. In general, the more the crucial uncertainties of the system in which a role is placed are dealt with in the role itself (as, for example, medical diagnosis in hospitals is dealt with by physicians, not managers), and the more the information to deal with those uncertainties tends to be concentrated at the lowest level, the more one has to rely on role holders (peers) for judgments. Thus in general, the higher the skill level of a role, and the more important the uncertainties dealt with by workers' skills in the organization as a whole, the more the organization must tap the information held by peers, that is, by the senior people who perform the role themselves and who work with the candidate.

When it is difficult for peers to explain what it is they do and to judge whether someone else is doing it well, and when bureaucratic superiors cannot measure output well because of poor observation or incommensurability of outcomes, as in secondary school teaching or police patrol work, then very general certificates of appropriate character and background for the work are combined with seniority (for a general description of police and teaching promotion and salary systems, see Spilerman 1986, 54–65), but usually with relatively small reward for seniority. Where the skill itself is obscure, but the output of interest to the organization is readily measurable and auditable, as among wholesale or real estate salespeople or athletic stars, certificates of competence and seniority usually play a very small role and rewards are based on measured performance. Where crucial uncertainties have to be dealt with by teamwork, making individual performance difficult to measure, as in continuous process manufacturing, seniority mixed with hierarchical superiors' judgment tends to dominate.

These general considerations are combined in different ways into five main kinds of institutional systems for certifying people for jobs in modern economies, which we will outline in the next five sections; these systems are based on (1) continuity or tenure in the present job; (2) promotions from within in large organizations; (3) craft and professional certification by peers; (4) recruitment of a person from within the family or of oneself as owner; and (5) the odd case of certification through work in a union shop. After examining these main institutional systems for certification of competence, we will look at how the opportunities to provide information and to have it disseminated to the people doing the

hiring are distributed. If all people needed to get a job was to have the competences to do it, the labor market would be much more competitive than it now is, because competences can be learned in all sorts of ways. But one has to learn a competence in such a way that trustworthy information about that competence can reach a potential employer at a time when that employer is making a decision about whom to hire (for a fuller analysis, see Heimer 1984). Since one of the main forms of certification is being already in the job, and if someone satisfactory is in the job an employer does not make a decision, competition takes place mainly on the rare occasions of job vacancies.

But even in the case of job vacancies, only certain types of information are acceptable certificates to a potential employer for predicting a potential recruit's future performance. And since the information embedded in those certificates is generated in institutions that produce the opportunity for a person to perform, produce observations on that performance, and guarantee to some degree the integrity and comparability of the report of those observations, not everyone could conceivably provide certificates of competence because not all have access to those institutions. This situation produces a stratification of opportunities in the labor market. In general, those who cannot produce certificates of special competence occupy places in the "secondary" labor market, a market of open competition for bad jobs.

The Great Segmenting Factor Is Who Holds the Job Now

Around 1970 the status attainment literature, the literature about who got what jobs and how social origins influence what jobs they got, assumed that everyone got the highest occupational status they could, so it assumed uniformity of the values of workers choosing jobs (though these authors never discussed just what they were assuming). The "causes" in this literature, then, were amounts of people's resources, especially human capital resources, such as education, and ordinary capital or wealth, which facilitate or hinder that maximization (the clearest statement of this theory I have seen is Kelley 1972); in other words, the causes were not variations in subjective processes (e.g., in how much people value prestige versus money returns, or security versus higher average income), but rather variations in the situations that determine people's possibilities.

In the past decade, besides providing the main substantive material that has been used in the development of sociological methodology, the

status attainment literature has provided an environment in which variations in labor market demands by firms have become principal variables for explaining the distribution of labor market outcomes. And the new variables are provided by institutional features of the employers .

Overall, the central institutional feature of employers is that, by and large, they will employ the same workers in the same jobs this year as they did last year. For example, the average job occupied around 1970 by a male worker under fifty-five falls vacant in Yokohama about once every eight years, owing to either promotion within the firm or movement between firms; in Detroit such a job would fall vacant about once every four years (Cole and Siegel 1979, 70–71). So in Japan, about seven-eights of the jobs occupied last year would still be occupied this year by the same man; in the United States, about three-quarters would be.

In 1970, the typical male worker between thirty-five and fifty-five in the Detroit study (Cole and Siegel 1979), left the firm he worked for on the average of every fourteen to twenty years; in Yokohama in 1971, the typical male worker left the firm on the average of every thirty-three years. If these men did not move among firms, then chances are they would change jobs within the firm only once in sixteen to twenty years in Detroit, once in twenty years in Yokohama. On average, then, if a job is occupied by someone over thirty-five, nobody is going to move into it any more often than about once every ten to fifteen years (roughly half the jobs being vacated by promotion, the other half by the worker leaving the job), in either Japan or the United States.

Jobs occupied by young workers are more likely than those occupied by older workers to fall vacant because of both kinds of mobility. Jobs occupied by workers in large organizations are more likely than those in small organizations to fall vacant due to promotion , while those in small organizations are more likely to fall vacant because of mobility among firms .

But the overwhelming institutional fact about the labor market is that the boundary around the worker-job pair is the most impermeable of labor market boundaries—among markets, perhaps only the housing market for owner-occupied housing has a lower rate of movement (the housing market as a whole, at least in the United States, has very roughly the same rate of movement of people as does the labor market). High-seniority workers in middle age move out of their jobs about as often as homeowners move out of single-family houses that they own.

This means that the overall structure of the labor market reflects the mobility patterns of people now working which took place up to forty years ago when they were young. The tendency of people to retain their

jobs means that holding a job now does not represent an equilibrium created in the timeless instant of economic theory. About one-sixth of all lawyers in the United States are women, even though about one-third of all law school graduates in 1987 were women; past patterns of mobility into the legal profession will be cast in stone until the tight person-job ties between males and jobs as lawyers are dissolved by retirement or death (cf. Lieberson and Fuguitt 1967 on the comparable problem for the races; Caplette 1982). All sorts of firms have "internal labor markets" in the sense that the people who will occupy the jobs next year are mostly in the firm already.

What this comes down to is that the flexibility of the labor market in terms of mobility is mainly flexibility over what to do with the new cohort—people between twenty and thirty who are settling into the jobs that they will mostly then hold (except for promotions, if they are in the big firm–government sector). How this group settles in will heavily determine what the labor market looks like thirty to forty years from now. If 33 percent of law school graduates, but only 10 percent of engineering school graduates, are women, and if entry rates in those fields are about the same between 1986 and 1990, then in 2020–2025 the new recruits will be between sixty and sixty-five years old. A little over 33 percent of the lawyers and a little over 10 percent of the engineers aged sixty to sixty-five in 2020 (a few more of the men will have died than the women) will be women.

To put real numbers on this, in 1980 about 65,000 engineers out of 1,401,000, or 4.6 percent, and about 74,000 lawyers and judges out of 530,000, or 14.0 percent, were women (Statistical Abstract of the United States, 1984 , 416). In 1981, about 7,700 out of 75,000 engineering graduates were women, or about 10.3 percent (Digest of Education Statistics 1983–84 , 120); for law the figures are 11,768 out of 36,331, or about 32.4 percent (ibid., 126). By and large until 2020, the folks now in the profession will shape the sex composition of the legal and engineering elites. At around that time present law and engineering graduates will have moved into the elite to roughly double the respective percentages of women. So the fact that feminism has had a lot less effect on engineering training than on legal training recently will still be a fact set in stone in the elite of the engineering and the law elite labor force in 2020.

This means, however, that when the labor market aggregates do move because of some outside cause, such as a decrease in the number of elementary school–aged children, this movement has enormous implications for the only people who are mobile, namely the people just entering the labor market. For instance, the total number of teachers did not

change much between 1975 and 1981, as a percentage of what it was in 1975 (see ibid., 49; Digest of Education Statistics 1976 , 53). But because almost all the teachers already in service would hold onto their jobs, the adjustment of young people was tremendous. During that same period, the number of teachers graduating went down from 167,000 (Digest 1976 , 118) to 108,000 (Digest 1983–84 , 120), or by over a third. In the same years, the number of engineering graduates went up by over 50 percent, though surely the total number of engineers rose by only 3–5 percent or so; similarly, business administration graduates increased by over 75 percent, and computer science graduates more than tripled in just six years (Digest 1976 , 119; Digest 1983–84 , 120). These increases did not occur because things were really booming in this period, by factors of 1.5 to 6, in management, engineering, and computer work; rather if the dog of labor demand moves its hindquarters a little, the tail of demand for new graduates has to wag a lot.

The total size of the labor force has grown some lately: from about 95 million to about 114 million, or by about a fifth, in the eight years from 1975 to 1983 (Statistical Abstract, 1985 , 390). But jobs did not grow that much, so unemployment went from about 8 million to about 11 million (ibid., 390, 406). The unemployment rate , then, was about steady, at about 6 to 10 percent. But young people sixteen to nineteen had a rate of unemployment that varied from 16 percent to 23 percent, and for people twenty to twenty-four the rate varied from 10 percent to 16 percent (Statistical Abstract, 1987 , 390); that is, the total range of variation for teenagers was about twice that of the labor force as a whole. Roughly speaking, if unemployment overall goes up 1 percent, the unemployment rate for people out of school between sixteen and nineteen goes up 2 percent, and that for people out of school between twenty and twenty-four goes up 1.5 percent. In other words, the fluctuations of total employment hit the new cohort harder.

This tendency also shows up in wages. In the big firm sector, about a million unionized workers (miscellaneous journalistic sources, Spring 1985) have contracts that have separate scales for recently hired workers. Because the new cohort is so big, firms and other organizations can get good young workers more cheaply than good old workers. So companies not wanting to fight their unions agree with the older workers represented in unions that for the first ten or so years on the job new workers will get lower wages than the old workers in the same jobs, and they will progress up the seniority scale at a certain distance behind the older workers. So the adjustment in the wage rates responding to the increase in the total supply of workers is being disproportionately borne by young

workers. (My advice to the reader is to be born in 1933, when the birth cohort was the smallest relative to the total labor force; one will have opportunities far beyond what one's merit would otherwise earn.)

In addition, the new cohorts take up the uncertainty disproportionately. New businesses are a lot more likely to fail than old ones, especially in new industries. For instance, the death rate of firms in the high-tech electronics industry is a lot higher than in the old-line industry of newspapers (Freeman, Carroll, and Hannan 1983). Newspapers by and large employ old people, electronics firms young people. So when firms fail, it is more often young people who have to scramble.

If employers looked only at comparative wage rates, they would in general prefer to lay off workers now holding the jobs they have and to hire instead teenagers at the minimum wage or, by moving the factory, to hire Koreans or Taiwanese. But clearly most employers prefer the people they have now to the people they might get by laying all the old workers off and starting anew in the cheapest possible labor market. Employers believe they know something valuable about the workers who now occupy jobs in their organizations that they do not know about other workers. Let us briefly speculate about what an employer might know about workers with seniority.

First, first-line supervisors would presumably complain about a worker's misbehavior if that behavior were critical to production. The silence of first-line supervisors that is indicated by seniority, perhaps supplemented by a vague investigation of whether the worker has "kept his or her nose clean," then, means that there is nothing wrong with a worker's performance in any crucial respect.

Second, those close to production, at least, know that a worker has to learn a lot informally from other workers about the quirks of the machines or procedures or clients and how to deal with them, and seniority in a particular job usually measures at least a minimal fit into the production work group. That informal learning also provides workers with clues to at least the minimal requirements of teamwork that make a particular operation go. Production lines or other "unskilled" systems run a good deal better than engineers alone can guarantee because workers usually cooperate to remedy defects in their machinery by babying it (Kusterer 1978), by helping one another deal with technical interdependencies, where trouble for one creates troubles for others, and so forth. The workers often know their local part of the production or service process better than their supervisors could teach them. Seniority in the job certifies that workers know the technical and social competences that are learned in the work group.

Third, some of the routines that workers are required to use on their jobs are used very often, and some of those may create troubles if done badly. For instance, tellers count money many times a day, and it is important that they should get this task right (Kusterer 1978, 79–80). Whatever those frequently used routines are, it is exactly those routines which workers presently in the job are highly practiced at. Thus, they may be more productive than workers with the same repertoire of skills but different amounts of practice in the skills most used in their jobs, because the others have practiced on other jobs. This is less true of course, the more rapidly the task composition of a particular job changes (for instance, it is less true of maintenance electricians whose tasks change very frequently than of construction electricians who may work for days on a single task). But in many organizations the task composition of particular roles is quite stable, and in such organizations the people in the job are certified to be better fitted for those jobs by practice than substitutes with the same repertoires of skills.

Fourth, workers who hold the job now and have held it for some time show, by continuing to hold it, that the incentive system for that job is adequate to keep them working at it. One of the best predictors of being willing to do a job is having been willing to do it for some time. Seniority in a job is thus a measure of character and motivation, which is uniquely a measure of how the workers will likely respond to the incentives that in fact they will be confronted with in the job. New hires have much higher turnover than old lags.

There are many rational reasons for an employer to prefer the workers it has, because their holding the jobs they do is information about what sorts of jobs they can hold. Many of the costs of turnover are of course the sheer bureaucratic trouble of hiring people or in the expense of training them. But many of the costs of turnover come from the fact that the information employers have about workers and their fit to the work system, to the routines and incentives of the work role, is worse in conditions of high turnover.

What makes these facts institutionally crucial is that people who could just as well have been hired in the first place but were not can never show that they hold the job, with all that that implies about whether they will have the job next year. The largest cause of labor market rigidity is the preference of employers to stick with the evils they know rather than fly to others they know not of.

Of course, the whole situation would be most unstable if it were not for the fact that workers also prefer employers they know they can satisfy, prefer informal systems on the shop floor to keep production going that

they have learned to manage, prefer work that they have thoroughly mastered so they can, when they choose, perform it semiconsciously, and of course, prefer work where the incentives seem to them adequate to make it all worthwhile. The fact that workers have held jobs for a while shows them, too, that these are jobs they can hold.

Seniority on the job, then, is moderately reliable information about whether people can do the job, and perhaps the best available under conditions (which are normal) in which the management itself could not do the job very well. The stories of early union organizing days of managers stuffing tar paper into the furnaces to make smoke, to tell striking workers that they were managing fine, illustrates the difficulty of managers or staff workers learning the thousands of details involved in how the line really runs. Being an airplane engineer does not make one a fighter pilot, nor does managing a production line make one a semiskilled operative. Yet management is in a better situation to know how to do a semiskilled production job than a random worker picked from the street, so the workers with seniority will exceed such replacement workers even more than they exceed managers.

Thus, the central certificate in the labor market is the implicit certificate of already holding the job, a certificate that is especially valuable for certifying one for that particular job.

Segmentation by Internal Labor Markets: Promotions Go to Those Now Employed by Big Firms and Government

The second big institutional fact on certification is that big firms and government organizations have mobility within them among positions, presumably mostly promotions, while small firms in the classical small firm competitive sector have almost no promotions. Roughly three-eighths of all movements among jobs in both Japan (a low labor mobility society) and the United States (a high labor mobility society) are between positions within firms (Cole and Siegel 1979, 70–71), five-eighths between firms. These within-firm movements represent "internal labor markets" in the strong sense that even people who dissolve the person-job tie tend to move within the boundaries of the firm.

Such mobility structures are characteristic of bureaucratic organizations in the engineering-based industries (manufacturing of machines and generally metals fabrication, chemicals, oil, electricity generation and distribution, telephone and telegraph services), in finance, and in government (Stinchcombe 1979b, 231–233). The effect of the size of the

firm on the promotion rate is illustrated in Table 3, which shows that promotions are much more frequent in large firms than small ones. This is often called the creation by large firms of an "internal labor market," but it is crucial to recall that even people who do not change jobs are in the labor market: they sell their labor under conditions in which the employer could hire someone else, and it is just as much an internal labor market in the small firm sector when very few change either firms or jobs. The point here is not that a young woman who gets a nonclerical job in a big firm will be promoted more slowly than the men; rather very probably a forty-year-old woman does not now work for a large firm or the government in a job that leads to promotion, and so she is a lot less likely to get a promotion over the next twenty-five years than is a man of forty. Since there are very few promotions in the small firm sector employing a lot of women and blacks, and since promotions will go to people employed in firms where promotions take place, it is already a foregone conclusion that even with perfectly fair policies within firms on promotions, most promotions over the next couple of decades will go to white males. (It is important to note that the difference between large and small firms in promotion rates may be lower for women; in large firms women are disproportionately in dead-end jobs.)

But what is crucial from our point of view is that the information used to predict performance in the new job consists in large measure of observations of performances in the old job. In many jobs covered by a collective contract, that performance is certified by seniority alone. As we have argued above, this is not an information-free certification, and a person promoted to a higher-yielding machine in a machine shop because he or she has worked another machine for years is in fact a good bet.

Furthermore, jobs are ordinarily connected to other jobs in such a way that experience in one is thought to be relevant for experience in the

other; there are "career ladders." This is clearest where the higher job is really the same as the lower job, as in the case of full professors and assistant professors. Where the higher job is as first-line supervisor of the work of people doing the same job, it is generally conceded that experience in the supervised job is relevant. Engineering supervisors are recruited from among engineers, software development supervisors from among programmers, floorwalkers from among salespeople, foremen in the crafts from craftsmen, and so on. This means that many promotions will be restricted not only to people within the larger firms and governments, but also to people on the first rungs of career ladders.

Where the information is judged by a committee of peers (again, this means a committee of superiors) in an internal labor market, the work of the candidate is to some degree observable by that committee. It is more observable in universities when the significant work is published writing; teaching work is not usually very observable in American universities. But in businesses, the higher one goes in the organization, the more everyday work is done in committee and by long telephone conversations among executives in different departments, which means that a variety of people who might promote a candidate for higher management have had dealings with the candidate. Thus, a candidate's work is more observable by the promotion committees toward the top of the managerial hierarchy than it is for first-line supervisors toward the bottom.

Promotion systems, or internal labor markets in the strong sense, make much use of the same seniority information that is used in retaining the same people in the same jobs. But in addition, if information about job performance is available to superiors by direct observation, and if the position being filled is of considerable importance to the productivity of the organization as a whole, performance evaluations are used as a supplement to seniority information (Rosenbaum 1981).

This in turn means that the certification of job performance for higher and more difficult jobs tends to be judged by a "committee of peers," though in most pyramidal organizations the hierarchical superiority of the peers is obvious. At any rate, the dynamics of recruitment to higher management and to supervisory positions in high-status staff departments tends toward the professional model, to be discussed in the next section. Generally in systems of promotion from within, promotions to lower positions are based primarily on seniority in the next lower job, promotions to middle positions are based primarily on evaluations by hierarchical superiors, and promotions to top positions are more similar to professional or craft recruitment based on evaluations by a committee of peers.

Worker-Controlled Recruitment in Professional and Craft Occupations

The third big institutional fact in job certification is that some employers hire certified skilled or professional workers from other employers in the same line of work, producing a pattern of mobility in which people get the same sorts of jobs they had before when they move to a new employer. In this case, the central boundaries in the labor market are around occupations as well as around firms (though sometimes rather than around firms, as in highly unstable industries such as building construction). Occupational associations such as craft trade unions or professional schools govern mobility into the occupation.

The fundamental generalization about entry into craft and professional jobs (and, as we have argued above, into higher management and top staff jobs in bureaucratic organizations) is the more democratic control over recruitment into a set of jobs is —that is, the more entry and training are controlled by workers in that set of jobs—the fewer women, blacks, Mexican-Americans, or immigrant workers are employed in the group (Heimer 1986b, 766; see also Cohn 1985; Glenn and Feldbert 1977).

There are three main places in the labor market where "co-optation" is a common method of recruitment—that is, where members of a body of workers help train, select, and find jobs for young potential members of the body. These are (1) the professions, especially the highest-status professions in a given area of work—for example, professors in prestige colleges and universities; lawyers, but not paralegals; dentists, but not dental hygienists; physicians, but not nurses or lab assistants in a hospital (but chemists who work for chemical or oil companies who do the same sort of laboratory work as lab assistants in hospitals are included here, because while lab assistants in hospitals are recruited and trained by pathologists, professional chemists do recruit their own followers [Schroeder and Finlay 1986]); (2) craftsmen (or craftspeople, if the generalization I am arguing for were not true), such as craftsmen in the construction trades; machinists of various kinds (e.g., tool and die makers, pattern makers); molders in foundries; pressmen and typographers in printing, at least before computer typesetting; cooks and chefs in fancy restaurants; long-haul truck drivers driving big rigs; and riverboat pilots in Mark Twain's day—such workers are usually recruited, trained, and certified by those who already occupy positions as craftsmen; and (3) top management, executives high enough that they are likely to be on the board of directors as well as to hold an executive position.

In each of these cases an older tradition of organizing recruitment

applies, in which the training occurred only when someone who knew the trade allowed a novice to come into his office or shop and learn that trade (Jackson 1984, passim; Carr-Saunders and Wilson 1933; Lockwood 1958). Now recruitment in these crafts or professions is often by schools or by admission to practice in a hospital or some other organized and quasi-bureaucratic structure, but there is still "peer group" control over who gets in. A few other work settings are marked by recruitment with a peer group character, but these do not have the same labor-market-wide "occupational" form: in mining, new underground miners have classically been recruited by the working miners and so have had ethnic and sex continuity with the miners, while aboveground workers are hired by more bureaucratic methods and very often are of a different ethnicity than the underground miners. For example, in Arizona around the turn of the twentieth century the aboveground workers at a mine would be Mexicans, the belowground workers first Cornish tin miners, then later Anglos more generally (Boswell and Bush 1984, 141–142; see also Boswell 1986).

While professors in high-prestige places have a lot of control over recruitment, in junior colleges or in what are now called "state universities" the administration has generally more control over recruitment. The prediction, then, is that affirmative action with respect to race, gender, and ethnicity will have gone further in junior colleges and state universities than in prestige universities or tony liberal arts colleges.

How, then, does it come about that when workers control recruitment to their own ranks (when recruitment is "democratic"), the recruitment pattern is so discriminatory? It reminds one of the observation that to maintain the subjugation of a majority of another ethnicity stably in the long run, as is done in South Africa or Israel, there must be democracy within the ruling ethnic group. Only if, say, oriental Jews have a way to express their indignation over the treatment they get from the Ashkenazim, say, by supporting the Likud, can the more oppressed part of the Jews be prevented from joining their interests to those of the Arabs and undermining the solidarity of the ruling ethnicity. Similarly, the Boers in South Africa used an English-type parliamentary system to become the ruling ethnicity, then crowded out the English through the democratic system (democratic without blacks, needless to say) rather than allying with, say, the colored. Recruitment is one of the several cases in which democracy has undemocratic effects, and we would like to know why.

Let us consider professors aiming to hire another professor in their department. How often should one expect them to propose to the administration to hire the cheapest labor the university can get? The people

who do the hiring are likely to be either only professors themselves (usually functioning in a recruitment committee) or (in the United States) a professor who is also a part-time administrator who gets practically no more money for being chair of a department and who will go back to his or her professorial role after a term as chair. The recruiters, then, have no interest in beating down the wages of the professorate, and quite often have the reverse interest. (Someone trying to hire me told me that he wanted to go to the administration proposing the highest salary we could support with arguments, because then we [it would only be "we" after I came, of course] could use that as a benchmark for what the others of comparable merit ought to be getting.)

Now contrast this situation with that of a dean of a junior college, say in Gary, Indiana. He or she is likely to be hiring quite a few teachers, and will soon learn that the college can hire University of Chicago graduate students to teach a course for a few thousand dollars; they will be pretty smart, quite likely female, but will accept low wages partly because they are willing to work part-time so they can work on their dissertation the rest of the time. For the vocational (and high school makeup) classes, the junior college needs to hire teachers full time, and if it is willing to hire women it can get exceptionally good people who thought they were going to be stuck all their lives teaching in a high school. The junior college can get women a lot cheaper than it can get men, for it would have to hire men away from a machine shop or an auto repair shop or an engineering job. And of course, if a junior college hires black people, since by and large they still cannot get really good jobs, it can probably get them cheaper, too. A dean's own salary will not go down in the future because he or she has recruited a lot of cheaper labor; the dean is not hiring his or her own replacement, just substitutes for more expensive full-time college teachers with advanced degrees. And economizing on faculty salaries will surely increase the chances that a dean will be promoted or will increase his or her salary.

So the fundamental dynamic is set up. If an executive's future depends on minimizing the costs in a department or institution, then he or she will tend to hire less qualified workers, and will tend to prefer cheaper blacks, women, and immigrants, especially illegal immigrants, if the organization can use them. Pathologists, even though they are professionals, are not hiring their future replacement when they hire a lab assistant. Their future replacement will come out of a medical school (in 1981 that meant that three out of four of the replacements will be male, Statistical Abstract, 1984 , 170), so a pathologist will not be beating down his or her own wages by hiring cheaper women lab technicians (with four

years of training) instead of more expensive men chemists (with four years of training). On the other hand, in a chemical company if executives hire chemists to work in the lab they will very likely be hiring their own replacement, and the new hires' wages will shortly be competing with their own (Schroeder and Finlay 1986; Fitzgibbon 1988).

If the hirers are members of a group that co-opts and trains its own members, then if they beat down the wages of people coming in, they beat down their own wages as well. If hirers are executives who hire a lot of people who are not going to be promoted into their own positions, then if they beat down the wages of the people they hire, they themselves will get promoted faster. One of the easiest ways to beat down the wages of new hires is to hire disproportionately from among women and minorities. We thus predict that the more democratic recruitment is, the fewer women and minorities will be recruited; the more recruitment is in the hands of bureaucratic or entrepreneurial officials, the more women and blacks should be recruited. But for those minorities and women who do get hired by a body of fellow workers, the hirers have an interest in seeing that they get the same privileges and are not used by the organization to beat down the general level of wages. So those few women and minorities hired by a body of peers are less likely to be exploited (Heimer 1986b; Shack-Marquez and Berg 1982).

Gary Becker ([1957] 1971) argues that those who hire with the objective of minimizing costs will hire more blacks and women, on the assumption that it is entrepreneurs (who will never be replaced by the people they hire) who do the hiring; and in the competitive industries, where Becker finds most recruitment of women and blacks, entrepreneurs more often do in fact do the hiring.

The pressure for executives to hire women and minorities is intensified if there is a lot of competitive pressure on the organization. In a relatively labor intensive small firm sector such as apparel or contract construction, a firm may be in one of two basic situations. The first is where the cost of labor is standardized throughout a competitive market by craft unions, so that all employers are faced with the same wage rate for skilled workers (sometimes there is a nonunion sector that is less efficient, quite often with semiskilled workers that need a lot more supervision; that market situation will be analyzed below). In such conditions the employers do not mind too much turning over the recruitment of workers to craft trade unions, because then they get the same sort of workers as their competitors for the same wages. The second condition is that the competition over labor costs reigns unrestrained, and employers have to get the cheapest labor they can in order to stay competitive. In

competitive conditions with no shielding monopolies in the labor market, a rapid switch to women, blacks, immigrants, and illegals tends to take place (G. Becker [1957] 1971; Cohn 1985). So we get the four conditions described in Table 4.

Family Recruitment in Small Firm Sectors

A big institutional fact in some industries is that employers hire themselves and parts of their families to do some of the work of the firm. This is most characteristic of small-firm sectors of the economy, especially in farming and the service industries. People who have careers as owner of a firm are not governed by the same sort of relation between the supply of labor and the demand for labor that characterize other employment relationships, and these sectors tend to "oversupply" labor during downturns of the business cycle.

People do not seem to be very good predictors of their own work performance as entrepreneurs, as is evidenced by the large failure rate of small business enterprises. Roughly half of all new starts of small businesses in the urban economy go out of business in the first couple of years, indicating that the entrepreneur did not predict very well his or her own ability to make the enterprise go (or did not predict the other resources needed very well, which amounts to the same thing). In addition, the small businesses in the single-family residential and subcontracting parts of the construction industry tend to be destroyed on a

large scale in business recessions and depressions, because construction activity is very sensitive to the business cycle. The same is true of architectural firms (Blau 1984, 115–132). Physicians, lawyers, dentists, and a few other kinds of professionals are ordinarily successful selling their own labor, but they have some government help in certifying their abilities in that work and in support of their monopoly.

For reasons that have nothing to do with entrepreneurial talent, about half of all small farming businesses disappear in each generation in modern countries whose urban economy is developing fast enough to absorb them. But all studies of, for example, adoption of innovations by farmers show large differences in productivity. Corporate farms and partnerships get roughly 50 percent more sales per acre than individual family farms (computed from Statistical Abstract, 1987 , tables 1102–1103), which probably indicates poor judgment by many family farmers about whether they should be entrepreneurs in that business.

Thus, while overall the position of the owner of a business hiring himself or herself is a privileged one in the labor market, people do not show a great deal of ability in judging or predicting their own work performance. The certificate that a person gets from having owned a firm is to a considerable degree a certificate of foolhardiness.

Union Membership As a Certificate of Productivity

Union workers are about 10 to 30 percent more efficient than other workers (Freeman and Medoff 1984) and so do not run into hard resistance in bargaining up their wages until they exceed 30 percent above nonunion workers. In January 1979, the median male union member made about 28 percent more than the median male nonunion worker; unionized females made about 50 percent more, but there were very few of them (Statistical Abstract, 1984 , 435).

Unionized workers are more often blue-collar workers in manufacturing or transportation (41.4 percent of male blue-collar workers were union members, 29.1 percent of female; figures in this paragraph are from Statistical Abstract, 1984 , 435) as opposed to (a) service workers (males 24.7 percent, females 10.8 percent); (b) clerical and kindred workers (males 32.9 percent [many being post office letter carriers], females 12.2 percent); (c) professional, technical, and kindred workers (males 19.4 percent, females 26.5 percent [most of the unionized professional workers are teachers, and teaching is a heavily female profession, which is why more female professionals are organized than male ones]); (d) man-

agers and administrators (males 8.4 percent, females 5.3 percent), and (e) salespeople (males 3.3 percent, females 5.1 percent).

In general, unionized workers work in industries where there is quite a bit of competition, though not as much perhaps as in a lot of small firm competitive manufacturing. Freeman and Medoff (1984) show that unionization of the work force only pushes down profits in the less competitive sector. That is, only when there are excess profits in an industry, because of oligopolistic advantages, do unions manage to get part of what would otherwise be profits. In competitive industries, union and nonunion firms have about the same profit rates. So the question is, how do union employers stay in business if they have to pay roughly 30 percent more for their workers?

The basic answer seems to be that unionized workers work harder or with more skill than nonunion workers. Apparently they get more work out per day than workers in nonunion shops or on nonunion crews. There are a number of different possible explanations for this.

In some craft trades like the printing industry, people evidently start in the nonunion firms (Jackson 1984, 263). After they get thoroughly trained, so that they can work faster and more accurately than somebody still in training or with only a few years' experience, they get themselves jobs in the unionized sector and join the union. When nonunion firms that pay about 30 percent less than union firms sell in basically the same market (though there are some kinds of printing that until recently were done only on a union basis), the only way the union firms can compete is to produce the same products using about 70 percent of the worker hours that the nonunion firms use.

A similar thing seems to happen in the construction industry, though there the market is perhaps more segregated. In metropolitan areas, most big jobs (most commercial and industrial construction, and almost all government construction) are run on a union basis; repair jobs on houses in most trades are run on a nonunion basis; and the building of houses is run both ways. In the competitive branches, especially housing construction, where union and nonunion overlap, the union construction firms have to get by while paying 30 percent more than other firms. Union firms usually do not get jobs for which the client pays by the hour, because it seems to a noncraft client so much more expensive to hire union labor, and they have no way of knowing this labor is more efficient. Perhaps they have also heard about union featherbedding, and so imaging that union work will be more expensive because of work rules. Almost all repair work is done by nonunion firms, because it is almost always done on an hourly basis.

But where firms bid for the jobs, the union firms can do all right. This

means that the union firms must be about 30 percent more efficient than the nonunion firms in construction. Here again, I suppose the greater efficiency of union workers reflects the nonunion sector serving as an apprenticeship system for the union sector.

In factory work we find the same general picture, that union firms paying more than the nonunion firms are competitive, though they can't go too high above the nonunion wage and still make it. Here several different things seem to be going on. The first is merely that a union contract stimulates management to institute a speedup that they could have done with nonunion workers but somehow have not got around to. So part of the explanation is that union firms have more efficient managements, managements that drive the workers harder because they have to sell at the same price as their competition while paying higher wages. Incentive systems and industrial engineers are often part of the system introduced to speed up the work after the company goes union.

Second, because the system of industrial relations that is instituted under a union contract seems fairer to the workers, they may be willing to work harder and take orders more cheerfully. Grievances go into the grievance system rather than causing sullen workers to institute a secret slowdown because they think they are not being treated fairly.

Third, it may be that union firms can be more selective in hiring their workers than nonunion firms can. While most personnel recruitment processes cannot distinguish a good worker from a bad one, there may be some added fussiness in union firms about who is hired, and since the firm is paying more than other firms, it can get the better workers. For example, it seems that a lot of blue-collar jobs are filled through the recommendations of the firms' own workers (Granovetter 1973, 1974). It may be that in union firms managers that keep their ear to the ground and hear about good prospects among friends and relatives of their good workers can, because they are paying higher wages, generally get those friends and relatives. So union firms recruit from hard-working families and hard-working friendship circles, by following out the recommendations of their good workers and not listening to their shirkers.

The overall result, however, is that one way or another working for a union firm certifies that one can hold a job in a union firm, which in turn means that one can hold a job in which the whole system (whatever its origin) makes one considerably more efficient than workers in nonunion firms. Thus, the fact that union firms tend to offer their union jobs next year to their own workers, so that they too have the large majority of their jobs filled by the same person who had the job last year, results in the certificate of greater productivity being perpetuated in the same so-

cial groups who were more unionized last year. As the above data show, the more unionized groups are disproportionately the male working class.

Let us pick up the thread of the argument again. Certificates that give employers reason to pay above the going rate for unskilled labor are generated in five main ways: through holding the job now (seniority); through data collection within large firms and governments, leading to promotions; through certification of professionals and craftsmen by bodies of peers; through certification by owning a business that one is a manager; and through occupying a job in a union shop. Each of these conditions carries different amounts and kinds of information. Depending on what information is most problematic for particular roles in the organization, we would expect more than one certification system to be in place. For example, we have argued that in management the system for certifying competence for top roles is different from that for certifying competence for roles at the middle and bottom.

The central overall difficulty to which these various systems of certification respond is that the measurement, and more especially the prediction, of the quality of performance is problematic. About all we know about why union members are more productive is that they work in more productive organizations; but from a brute economic point of view, this means that the economy should (and does) pay union members more. While the preference for union workers is ordinarily imposed on firms from the outside and against their wills, that preference apparently allows them to pay their workers more. Similarly, much of the information that comes from the fact that a given worker now occupies a given job is inchoate, and has been hard to analyze systematically here. But since so many employers use that information to decide whom to employ next year, it serves as a certificate in the labor market. Seniority critieria are especially prominent when the qualities that constitute qualification for the job are also inchoate, as in teaching and police work. All such certificates, generated in different ways, and measuring different kinds of qualification for work, tend to preserve positions of privilege in the labor market.

The Secondary Labor Market

These big institutional facts (stability in jobs, internal promotions in large firms, occupational continuity in some occupations, family recruitment in some sectors, and the association of unionization with productivity) taken together distinguish, first of all, between people inside the boundary marking off the employed from those not employed. People

now employed have a much better chance of being employed next year than those not now employed.

But a second distinction has preoccupied recent labor market sociology. This is the distinction between those inside any boundary that protects privilege among those employed (such as being employed by a large firm that gives promotions or higher pay, being qualified as a craftsman or professional, owning a firm that gives the owners employment, being employed in a unionized shop or crew), generally called the "primary labor market," and those employed but not protected by such a boundary, generally called the "secondary labor market." Roughly speaking, those outside the circles of privilege amount to about 25 percent of those employed. About one out of four jobs (and about one out of four jobholders) have high rates of interfirm mobility and high rate of mobility into the status of unemployment, and have low wages, low unionization, and other features showing lack of privilege.

The people who occupy these unprivileged jobs in most modern economies are disproportionately ethnic minorities, women reentering the labor market in middle age, teenagers, and people recently unemployed (Doeringer and Piore 1971). In underdeveloped economies they are very often recent migrants from rural areas living in favelas or bidonvilles and working on the fringes of the urban service sector.

From the point of view of this chapter, what is distinctive about such populations is that they have very little opportunity to provide certification that will satisfy an employer with a good job on offer that they can be predicted to do well in that job. The main kind of certification that satisfies employers is that one already occupies such a job—and that is the hardest kind of certification for members of the secondary labor market to give. Those sorts of formalized information that justify promotion in internal labor markets are even more inaccessible to members of the secondary labor market, and the bodies of peers that govern self-recruited status groups have no reason to regard such people as potential peers, and much interest in keeping out low-wage labor.

Such lack of certificates no doubt reflects in part a true lack of competence, but this does not answer the sociological question of how such lack of certifiable competence (and perhaps of true competence) is systematically distributed. After all, people who end up with the good jobs were also born incompetent.

Conclusion

Almost all modern organizations must face a great deal of uncertainty in their predictions about worker productivity over the periods in which

those workers will be employed. There is a great deal of uncertainty even when those workers are already employed, because all the forms of uncertainty analyzed elsewhere in the book make it uncertain exactly what one will be asking workers to do, exactly what incentives they will get if they do that, and exactly what the economic or other consequences for the organization will be if they do it. Organizations have even more uncertainty when they are recruiting new workers.

This uncertainty matters more for some organizations than for others. Generally speaking, if an organization is confronted with uncertainty that requires skilled variation of routines at the worker level (as, for an extreme example, is required in basic scientific research), then the problem of picking workers who can deal successfully with the uncertainty is crucial to organizational success. In those cases will expect organizations to develop personnel systems that make heavy use of information from fellow workers (peer review systems), with all the problems of motivated distortion of personnel information that this involves. Such peer systems are likely to apply only to the professional or craft subsection of the organization; thus, for example, the business office of a university will probably use little peer review.

Organizations that require a great deal of teamwork, and in which supervisors contribute strongly to teamwork, will be expected to use a good deal of hierarchical information from superiors and superiors of superiors in predicting workers' future performance. The ideal-typical "internal labor market," in which promotions are governed by hierarchical superiors, should be a normal outcome of uncertainties that revolve mainly about the amount of teamwork.

Most work that is neither too difficult nor too involved in teamwork but that is poorly measured by reliable managerial information, such as teaching or police work, should result in certification systems organized around seniority. The main certificate indicating that person can manage the job (whatever the job really involves) is that he or she has been managing it already. In particular, the higher apparent productivity of union firms, while likely as mysterious to union firm managers as it is to economists and sociologists, is apparently, whatever its cause, adequately preserved by keeping union workers on their jobs.

Work that can be easily assessed by auditable measures in an environment that is sufficiently stable that the work is repetitive and requires the same talents (and the same measures of talent) should be paid according to its marginal productivity, just as economists assuming perfect information predict. Traveling salespeople, athletes, and lettuce pickers seem to be the main examples.

The general result, then, is that information about worker perfor-

mance is ordinarily very bad, and predictions of future worker performance even worse. People apparently cannot predict their own future performance as workers in businesses they start for themselves well enough to keep from going bust; and they can presumably be expected to do worse at predicting the performance of others. That is why the labor market institutions in most parts of the economy look so little like the institutions of commissions on sales, or superstar contracts in athletics, or subcontracting by piece of lettuce picked, which would be the ideal-typical outcome if employers actually measured marginal productivity in any direct way.

What one needs to know for analysis of institutions collecting and using information on workers is what information about the future performance of the work is contained in the main kinds of certificates employers use. We have tried to analyze what sort of information seniority, education, certified or formalized performance measurements from previous jobs, production outcome measurements, and the like contain, so as to give a functional analysis of labor market institutions.

This functional analysis gives something of a Panglossian view of labor market discrimination: the institutions that result in racial, ethnic, and gender discrimination are really all for the best in this best of all possible worlds. Of course, they are not best for all purposes, and especially not for the purposes of building a fair society. The political function of this analysis, then, is to try to suggest why discriminatory labor market institutions are so resistant to change, and to provide a basis for proposing ways to change them. The purposes these institutions best suit vary with the type of information system, and we have used this fact to explain where in the economy the different types of system are used. Knowing why institutions with a discriminatory effect are found where they are may be politically useful in undermining them.

The scientific function of this analysis is to explain why some kinds of labor market segmentation are found in some parts of the economy, different kinds in other parts. Given that explanatory purpose, it is not surprising that we do not succeed very well in explaining why in all parts of the economy, the general tendency is for all such institutions to discriminate against women, against racially oppressed groups, against teenagers, and against recent immigrants or guest workers.

The connection between information, judgment, and reward is central to the structure of status systems (Stinchcombe 1964, 25–40). Whenever information about work performance is central to the overall functioning of an organization, status systems will be shaped by the requirements of that information system. In colleges and secondary schools, the informa-

tion requirements for the grading of students are easily satisfied, and the preservation of the reward value of the resulting certificates is of greater functional importance. For example, the proportion of students flunking first-year chemistry varies hardly at all with the average SAT scores of entering freshmen, because the reward value of a grade in chemistry is primarily the relative ranking it provides a student as compared with other premedical students, not the absolute grade level. Conversely, reliable performance of maintenance tasks in complex, and frequently modified, machinery will be hard to measure and predict, but crucial for a chemical plant. The status system of maintenance people in such plants will therefore be shaped by the problem of collecting meaningful information on craft workers' performance.

Thus, the requirements of grading chemistry students and qualifying maintenance workers are different, even though they both depend ultimately on chemical knowledge. Hence, we find grading by standardized tests by chemists who themselves were promoted by committees of peers in colleges, and no grading, but instead apprenticeship and craft control, for maintenance workers.

We therefore predict different kinds of stratification systems in organizations with different information requirements, and different status systems in different parts of the organization depending on the particular information required. We will expect promotion in universities to depend not on hierarchical information from the business offices, but on faculty committees of peers, with the accompanying difference in the motivation of high-status people in the business offices to hire women and minorities, those in the faculty to hire white males and other high-priced labor. But we will expect the very top ranks of both to have the same low proportions of women and minorities, because as one approaches the top of the business office hierarchy people are recruited more by committees of peers, rather than recommended for promotion by their own superiors—just as in the case of faculty. The variations between and within organizations in the stratification system are likely to be explained by variations in the information needed about people for different productive tasks.