4

Synthetic Turnover

One of the most important concerns in the sociology of gender has been to explain why women are concentrated in occupations with low-status and low-wage attainments. Many writers have suggested what shall be referred to here as the “natural turnover hypothesis.” The natural turnover hypothesis argues that women’s domestic pressures and sex-role socialization lower their commitment to labor force participation. This increases their turnover relative to men, making them less attractive for jobs involving long preparatory tenure. Since highly paid jobs usually require extensive training and experience, conservation of training costs would mandate giving these jobs to men (Caplow 1954; Oppenheimer 1970). There are a large number of arguments relating turnover differences to hiring practices. Writers usually invoke one or the other of them as a general explanation of occupational sex inequality.¹

This chapter suggests that there may be limits to the applicability of the hypothesis. The most important critique is that the female turnover differential has been overstated. When appropriate controls are applied, sex differences in turnover often disappear (Stoikov and Raimon 1968; Armknecht and Early 1972; Fry 1973; Laws 1976; Viscusi 1980; Begnoche Smith 1979; Blau and Kahn 1981). A few studies do show strong net sex effects (Parnes 1954; Burton and Parker 1969; Pencavel 1970); however, the bulk of the literature casts doubt on the premise that, net of job characteristics and age, gender has an independent effect on quitting.²

In the face of such evidence, some critics have attempted to salvage the theory by modifying it. One such variant is the theory of statistical discrimination (Phelps 1972). This claims that the mean labor force commitment of men and women may be similar but that the unexplained variance in female tenures is greater than that of males. Since employers have difficulty predicting female turnover from standard pre-employment indices, they err conservatively by barring females from superior positions. Statistical discrimination models do not provide an alternative to the natural turnover hypothesis. They, in fact, restate it. In the absence of precise information on turnover, statistical discrimination would only be effective in lowering costs if the true female turnover rates were on average higher than the true male rates. This, however, is the questionable premise from which the theory claimed to be independent (Aigner and Cain 1977).³

There is a second, more damaging and heretofore unexamined, ground for challenging the natural turnover hypothesis. The natural turnover hypothesis assumes that it is economically rational for employers to minimize turnover within their ranks, especially in jobs involving a great deal of employer-subsidized training. Keeping women out of occupations involving large amounts of firm-specific skill acquisition would protect employers against economic losses should such positions be vacated. Turnover is not always economically unprofitable, however, and can in many circumstances even be economically advantageous. The profitability or unprofitability of turnover is situation-specific, as will be discussed below. In the cases where turnover is economically desirable, employers will cease trying to minimize quit rates of their staff and may, in fact, seek to increase them artificially. Therefore, a job segregation model based on the presumption of turnover avoidance will be valid only when turnover is costly. When turnover is beneficial, a different explanation must be invoked.

This chapter proposes an explanation of job segregation for the case involving beneficial turnover. To distinguish it from previous arguments, it will be called the “synthetic turnover hypothesis.” A series of cases are described where the synthetic turnover hypothesis should have predictive power superior to that of the natural turnover hypothesis. Data are then presented that show the goodness of fit of the synthetic turnover model. It is not claimed that the synthetic turnover model is superior in all cases to the classical explanations but rather that it better explains occupational segregation in a particular well-defined domain, though this domain is probably fairly large.

The Economics of Turnover

Most sociologists tend to overestimate the costs of turnover, probably because they consider primarily the costs of replacing human capital. It is true that many job holders in the course of their careers acquire skills not possessed by a new entrant and that, when an employee resigns, his or her employer may lose productivity until the replacement has acquired these skills. Because these human capital losses are never totally recovered, it is often assumed that employers are inevitably averse to turnover.

However, there are compensatory savings associated with turnover that can override these costs. A new employee can be paid an entry-level wage, while the pay of veteran employees may have been augmented with seniority-related increments. Many organizations use tenure-based salary systems to motivate steady performance. Good work is rewarded with regular year-end raises and by promotion to jobs with higher maximum salaries (Thurow 1975; Williamson 1975; Edwards 1979). While time-based rewards may be effective in motivating personnel, they have the drawback of steadily increasing labor costs. A certain amount of salary increase can be tolerated when it is balanced by an increase in employee productivity. In the early part of his or her career, when an employee learns new skills, raises may be economically justifiable. A some point for most jobs, however, the learning curve flattens. Returns to experience are not unlimited, and promotions and raises eventually translate into higher costs for what is an unchanging amount of employee product. Once learning has stopped, in a time-based salary regime, an employer faces a difficult choice. He may continue to give raises, which will produce higher per-unit labor costs. Alternatively, he can eliminate tenure-based raises. This undercuts incentives to perform and reduces employee output. In such a case, it might be advantageous to allow the employee to resign. A replacement could be brought in at a lower rate of pay, with potential salary increases once again compensating high levels of effort.

The desirability of turnover to an employer is a trade-off between the desire to avoid retraining and a preference for recirculating staff into entry-level positions. The relative importance of these factors is situation-specific. When labor markets are tight and replacements hard to find, employers will prefer to conserve manpower and minimize turnover. The same would hold if the savings obtainable from replacement were small, as in the cases where wages are paid at a flat rate or by the piece. Flat-rate wages are independent of tenure lengths while piece rates are directly linked to productivity. Replacement savings would also be low when learning curves are long and steep. A large percentage of the labor force would still be at early stages of their careers when salary increases are economic. In contrast, an employer would seek to maximize turnover when the opposite conditions hold: when replacements are plentiful, the incentive system is tenure-based and learning curves are short with early crests. These predictions are summarized in Figure 4.1.

Synthetic Turnover and Gender

When turnover is profitable, one would expect that employers would act to increase quit rates beyond their natural level. When an employer proactively induces resignations in his staff, this can be considered synthetic turnover. There are a variety of strategies of synthetic turnover, some of them involving gender.

Some employers withhold organizational rewards from individuals as they reach the crest of their productivity. When successful, such withholding demoralizes the job holder and leads to an early quit. However, there are some individuals and some organizations for which this will not be an effective strategy. If a job holder has little opportunity for comparable employment elsewhere, he or she may remain despite diminished organizational incentives. Non-quitters might include employees of firms with relatively high wages, employees who are poor performers, and those with idiosyncratic, nontransferable skills. Particularistic exclusion is not sex-based, but it is not always practical.

A second tactic is selective recruitment. An employer can intentionally hire employees who are likely to quit early voluntarily. Selective recruitment, however, requires good ability to predict voluntary quits. The employer needs to identify a population who will stay long enough to avoid taxing the firm’s retraining capacity but not so long as to create stagnation problems. The natural attributes of workers may not fit a business’s ideal tenure plans.

A third tactic is to place maximum tenure limits on positions with short learning curves. After a given number of years, the employee is forced to resign. Tenure bars provide effective limitations on career lengths, but they may create fundamental problems of legitimation. Most organizational incentive systems would be undercut if employees perceived the ultimate reward of good service as being dismissal. Tenure bars can only work when the moment of forced resignation coincides with the arrival of some exogenously determined plausible excuse for dismissal. Mandatory retirement age is an example of an effective tenure bar because of ideologies associating old age with decreased performance.

The different mechanisms of synthetic turnover vary in their relevance to the sorting of women into low-status jobs. Particularistic sanctioning is irrelevant to sex discrimination. Selective recruitment has a weak and transitory effect. In contrast, tenure bars can create systematic, lasting sex discrimination.

Particularistic sanctioning should have no gender consequences because both hiring and the withholding of organizational rewards are based on objective measures of productivity. Since men and women are likely to be equally productive, they would receive punishments at equal rates, resulting in equal numbers of men and women at all organizational ranks.

A selective recruitment effect would only be noticeable if employers hired women because they expected women to choose short careers voluntarily. Many employers genuinely believe that women are marginal labor force participants (National Manpower Council 1957; A. Hunt 1975). The literature on sex differences in turnover shows that the magnitude of independent sex effects depends on the precise number and identity of the control variables that are applied. Employers’ perceived sex differentials will likewise depend on the sophistication of their analysis. With no experience, an employer might staff a new position with women in the hope of obtaining shorter tenures. Experience gathered over the long term would most likely disconfirm this prediction. However, the original sex composition could stay intact if either the employer failed to perceive these non-differentials or bureaucratic inertia or other rigidities did not permit a change in sex type once such non-differentials were discovered. Selective recruitment for short tenures could then produce sex discrimination; however, such discrimination could persist only under a fragile set of extenuating circumstances.

Tenure bars provide the greatest potential for determining sex-type. This is because the most common tenure bar historically has been the marriage bar. In the past, many employers hired young girls on the condition that they retire at marriage. A girl who started working at sixteen and married at twenty-one would have had a career of only five years. Marriage was one of the very few demographic transitions that could ideologically be held to be grounds for dismissal. Traditional sex-role ideology stressed the incompatibility of work and family; dismissal would be accepted and understood by the women themselves. There were few other occasions for imposing a bar that would have been as easy to legitimate.

The consequences of marriage bars for women’s status should be clear. In firms where bars were operative, jobs with short learning curves would have been reserved for young single girls. Jobs with long learning curves for which employers wanted longer tenures would have been given to men. This would tend to confine women to low-status jobs because learning curves are shorter for those positions. Turnover differences would have indeed occurred. However, these would have been due not to the female labor supply but rather to the proactive policies of management.

Several types of data are available to support these claims. First, an examination of the marital policies of a number of firms will show the extent of marriage bars historically. Second, an in-depth analysis of the Great Western Railway and the General Post Office, which had explicit marriage bars, will show the inapplicability of natural turnover models to these settings. Third, data will be presented to show the special relevance of synthetic turnover to clerical work.

The Extent of Discrimination against Married Women

There have been a number of systematic attempts to measure the degree to which employers have explicitly restricted female career lengths with marriage bars. One of the most important studies was done in England by the British government in 1946. The Whitley Council Report on Marriage Bars investigated the marriage policies of several large clerical employers (National Whitley Council 1946). Their sample consisted of a set of private firms, some British governmental agencies, and the civil services of a number of foreign countries.

The study found a remarkably high level of discrimination against married women. Virtually every British private employer interviewed had a marriage bar (see Figure 4.2). Marriage bars were also universal among British public employers. Lower levels of discrimination were found overseas, with only half of the foreign civil services having marriage bars. The sampling of organizations in this study was idiosyncratic: the sample size was small and biased towards highly visible bureaucracies. Nevertheless, the findings were probably fairly accurate. Nowhere in the records of the commission is there any indication that any of the members disputed the empirical accuracy of the presented findings. This is significant because the commissioners were selected for their expertise in current personnel management practices and for their divergent views on marriage bars.

Marriage bars also existed in the United States and were especially prominent in elementary school teaching. The National Education Association monitored the use of marriage bars in the public school systems of cities of 2,500 or more from 1928 to 1956. Before World War II, it was the general practice neither to hire married women as teachers nor to permit them to continue teaching after marriage. In 1928, over 60 percent of all urban school districts did not hire married women and over 50 percent required resignation soon after marriage. In 1941, the comparable figures were 87 percent and 61 percent (Oppenheimer 1970).

Studies of marriage bars have also been conducted using a wider range of industries. In 1936, Purdue University surveyed 250 American firms on their policies concerning married women (see Table 4.1). The survey found that 50 percent of the factories and 61 percent of the offices had some sort of restriction on the hiring of married women. Forced resignation was somewhat rarer. Twenty percent of the factories and 33 percent of the offices had forcible marriage bars (Best 1938). A comparable survey was performed by the National Federation of Business and Professional Women’s Clubs in 1940. The study found that 43 percent of public utilities, 29 percent of large manufacturing concerns, 23 percent of small private businesses, and 13 percent of department stores had some sort of restriction on the employment of married women (Shallcross 1940; Oppenheimer 1970). The methodology of these studies is not reported. If these surveys are similar to others of this period, they probably contain serious methodological errors. Taken at face value, however, they suggest that marriage bars were not uncommon among American employers. Bars seem to have been confined to a minority of employers, and the extent of bar use varied by industry and occupation. Nevertheless, the absolute number of employers practicing discrimination would have been very large.

In the early 1960’s, the International Labor Office commissioned an extensive study of discrimination against married women. Labor experts from all over the world submitted reports on the use of marriage bars in their home countries. No attempts at quantitative verification were made. Instead, the ILO relied upon the impressionistic accounts of the local observers. The ILO report is striking, however, in the breadth of discrimination that was suggested. Reporters from Japan, the Netherlands, Italy, and Ireland noted that, in their countries, it was standard practice in most industries for employers to hire young women with the explicit provision that they resign upon marriage. Such practices were also found in other countries, although the correspondents only referred to specific industries. Married women were not employed as nurses in Australia and Belgium. They were barred from working as bank clerks in England, Belgium, Greece, Australia, and South Africa. There were comparable restrictions in the textile factories in Portugal, metallurgical factories in Italy, and chemical factories in both Italy and Belgium. A very wide range of countries reported marriage bars for both teachers and airline stewardesses. The United States was comparatively infrequently mentioned in these discussions, suggesting a low level of bar usage. Nevertheless, it would seem that marriage bars were a phenomenon of worldwide importance (International Labor Office 1962).

The above evidence is based on historical data. However, marriage bars can also be found in more contemporary settings. Robert Cole notes that in the 1970’s married women are generally expected to resign from employment in Japan. This policy is not explicitly written into the employment contract, in fact, explicit discrimination against married women was made illegal in the 1970’s. The practice persisted, however, despite the legal strictures, by unwritten informal understandings (Hanami 1965; Cole 1979; Cook and Hayashi 1980). The Japanese case also suggests that the extent of discrimination against married women is poorly measured by the presence of formal restrictions. As such, much of the survey evidence given above is likely to underestimate the amount of discrimination.

Turnover and Job Allocation in the Great Western Railway and the General Post Office

In this section we look in depth at the turnover and job allocation dynamics of the two main firms of this study provide further support for the model. Both of these firms used synthetic turnover, because they both had severe over-annuation problems with their white-collar staff. The stagnation problems were so severe as to require sustained attempts to augment employee quit rates artificially. Both firms had incentive systems involving time-based salary increments. Employees were paid on graduated salary scales. All promotion was internal, and employees could expect several major promotions over the course of their careers (GPO Establishment Books 1857–1937). The firms’ problems were exacerbated by a policy of guaranteeing continuous employment up through retirement age. Both firms hired employees only at the age of sixteen. Once hired, an employee had lifetime job security and only extreme malfeasance would warrant dismissal (Gladden 1967; Kingsford 1970). Postal jobs were protected by civil service regulations. Railway jobs were safeguarded by past company promises, made in an era of labor scarcity. The time-based salary-promotion system produced rising labor costs that could outstrip worker productivity. Once the stagnation point occurred, however, the only source of relief was voluntary employee exits.

Particularistic sanctions were incapable of raising the level of employee turnover. This was because postal and railway employees had limited employment opportunities in the outside world. Postal skills tended to be non-transferable. The Post Office had a monopoly on mail delivery and telephone and telegraph service. Workers with training in mail-sorting or telecommunications had no other employer to turn to. The clerical force had some salable skills. However, the civil service paid at well above market rates. An accounts clerk was unlikely to duplicate his salary in the private sector (Gladden 1967; Klingender 1935).

Railway immobility was created by the rigidity of railway employers. These employers only hired sixteen-year-olds. Even exceptional performers had difficulty transferring after the age of twenty (Staff Statistics 1870; Clerical Appointments Committee 1924). Much of their work was railway-specific and had few analogues in other firms. Some skills, such as basic bookkeeping, would have been transferable. However, the idiosyncratic nature of railway accounting and paperwork would have induced other private sector employers to have limited interest in former railway clerks (Jenkinson 1914). As a result, marginal civil servants and railway clerks tended to persist, even when deprived of raises and promotions. The firms needed stronger devices to ensure turnover in the lower ranks.

Both firms had absolute marriage bars. Women were hired at the age of sixteen and allowed to work until they married, at which point they were contractually obligated to surrender their positions. Neither willingness to continue work nor superior performance of duty could be invoked to waive this requirement. However, the danger remained that women might delay marriage to continue working. The bar could be avoided by the willing celibate. To counter this, both firms instituted the payment of marriage dowries. After six years of employment, the firm would pay a woman a substantial cash bonus on the occasion of her marriage. The size of the dowry varied, but it was often the equivalent of a month and a half’s salary. In present-day terms, this meant a payment of between $1,300 and $1,800. The firms paid these dowries to most female employees, indicating an enormous financial commitment to creating short female careers.

It should be noted that the employers were not absolute turnover maximizers. A woman had to remain at least six years in order to qualify for her dowry. This is consistent with the hypothesis that employers sought to keep women during the early stages of their careers, during which the salary system is financed by increased employee productivity. The dowry insured against overly short tenures, but it also kept those tenures from becoming too long.

There was a second, more minor, mechanism of exclusion. Many firms used long-term incentives such as pension plans to create loyalty among their employees (Edwards 1979). Women generally had less access to these programs. In both firms there was a substantial period of time during which men were offered superannuation systems and women were not. When pensions were finally offered to women, they were given at lower retirement ages. The male retirement age was sixty-five, while the female age was fifty-five. This is actuarially inappropriate since females generally have lower mortality rates than males. This too suggests that employers were willing to see women’s jobs vacated, even if it meant paying early benefits.

There is evidence that the employers actually perceived the problem this way themselves. Many contemporary discussions of the advantages of female labor laid heavy emphasis on the need to keep stagnation low and the promotion process fluid. Consider the following testimony given by Francis Scudamore, controller of the Postal Telegraph Services, to a committee of Parliament in defense of women’s employment:

Permanently established civil servants invariably expect their remuneration to increase with their years of service . . . even in the cases necessarily very numerous in which from the nature of their employment, they can be of no more use or value in the twentieth than in the fifth year of service. There must always be in the Post Office an immense number of duties which can be just as well performed by a lad of eighteen as a man of forty; but when the same person continues to perform the same duty from his eighteenth to his fortieth year, it is* impossible to permanently resist his claim for additional remuneration, and when he continues to perform to his sixtieth year, it becomes equally impossible to resist his claim for a retiring allowance.

Nor would it be possible to maintain a rule under which persons employed on certain duties should perforce retire after a short time say five to seven years.

Women however will solve these difficulties for the Department by retiring for the purpose of getting married when they get the chance. If we place an equal number of women on the same ascending scale of pay, the aggregate pay to the females will always be less than the aggregate pay to the males . . . and further there will always be fewer females than males on the pension list [Scudamore 1871, p. 78].

Scudamore’s observations closely parallel the theoretical basis of the synthetic turnover model. He points to the gap between tenure-based pay increases and observed productivity curves and notes the difficulty of imposing unlegitimated tenure bars. His hope was that, by hiring women, he would have a workforce that would leave voluntarily. Scudamore was speaking at the date of the very inception of female hiring, at which time the Post Office still had no marriage bar.

The marriage bar was instituted in the late 1870’s. The timing, seven or eight years after the introduction of hiring women, suggests that women were not voluntarily leaving in large numbers upon marriage and thus required additional exit incentives. The similarity between the timing of the introduction of the marriage bar and the desired six-year length of female careers consistent with a marriage bar policy is interesting although possibly coincidental.

After the introduction of the bar, managers continued to defend its use in terms consistent with the synthetic turnover theory. In 1945, over sixty years later, the British civil service authorized a commission of government administrators, labor representatives, and community leaders to consider the future of the marriage bar. The committee’s role was advisory, and it was specifically prohibited from making any concrete policy recommendation. Nevertheless, there was one argument the commission found so compelling that it was given special emphasis in the body of the report and was reiterated forcefully in the conclusion:

Where there is a large amount of routine work to be done it is a real advantage to employ women who stay only a few years and retire on marriage. The amount of routine work is so great that the Service can not provide adequate careers for all the young persons recruited to the lower grades. It is essential to have fairly rapid turnover, and marriage wastage is a highly important factor in this [National Whitley Council 1945–46, p. 12, item xi].

And further on:

The amount of routine work is so great . . . that the Civil Service Staff pyramid is bound to have a very broad base, and promotion to the higher ranks can not come to many.

. . . [The retention of women after marriage] would be bound to slow down the rate of promotion. . . . This would not be welcomed by their colleagues whose prospects thereby would be worsened (National Whitely Council, 1945-46, p. 15, item xxxix].

These statements are clearly quite similar to those made by the telegraph controller in 1870. Thus it would seem that, for the entire late nineteenth and early twentieth centuries, the Post Office was formally committed to working towards short female careers. This policy had a visible impact in reducing the opportunities for promotion and advancement that were open to female employees.

Women tended to be concentrated in those jobs for which short tenures were advantageous to the firm. As predicted by both the natural and synthetic turnover models, these were jobs with flat learning curves and short promotion chains. One index of such placement is the extent to which women are placed in non-entry-level jobs.

Table 4.2 clearly shows that women were less likely to hold non-entry-level positions. At the minimum, men had a 3 to 2 advantage in these jobs. At the extreme, on the Great Western Railway, men were nearly nine times as likely to be in jobs with long learning curves. This is consistent with the sorting by sex expected if job allocations were made on the basis of anticipated tenure.

The confinement of women to low-status jobs in these settings cannot be explained by the natural turnover hypothesis. The natural turnover hypothesis is inconsistent with the patterns of quit rates associated with these firms. The standard argument is that women leave employment before they can remunerate their employers for the cost of advanced training. Women either quit during the training period or leave so soon afterwards that the employer loses the anticipated benefits of his investment in human capital. However, if women’s job tenures were shown considerably to exceed the period of skill acquisition, this would be inconsistent with the natural turnover premise of high absolute female instability.

To test this proposition, one must compare female job tenure with the learning curves of both male and female jobs. Information on job tenures can be found in actuarial records that employers used to calculate the details of their pension plans. Pensions require accurate yearly turnover figures for the adjustment of premiums and benefits. Both organizations therefore endeavored to collect accurate data on career lengths.

The data consist of several series of age- and sex-specific exit rates. Using these rates, one can perform a standard life-table analysis using the period rates to simulate the experiences of a synthetic cohort. This analysis assumes that the entering radix of sixteen-year-olds are subject to the sixteen-year-old rates. Those who survive to age seventeen were subjected to the seventeen-year-old rates. Each year the cohort was reduced by the next sequential rate. The median career length is calculated by finding the age at which 50 percent of the cohort had disappeared. A prospective recruit has a 50 percent chance of surviving up through the year at which half of his or her cohort has disappeared.⁴

There are further analytic possibilities for these data. The female exit rates are divided into marital and non-marital exits. These allow simulation of the effects on turnover of manipulations of the marriage bar policy. Since the bar and dowry affected only maritally based exits, the marital exit rate can be altered in various ways to reflect possible outcomes of eliminating the marriage bar. The precise results of an imaginary bar removal are conjectural. However, estimates of the maximum and minimum effects can help define the range of possibilities. The minimum result would occur if the bar was actually irrelevant to the determination of career lengths; in this case eliminating it would have no effect. The minimum effect of bar removal is adequately represented by the observed career lengths. The maximum effect would occur if, in the absence of the bar, no one would have left for marital reasons. In this case, the female quits would be totally represented by the non-marital rates. To simulate this, the tenure lengths are recalculated assuming the marital exit rates to be zero. An intermediate simulation is also included, one that assumes that, in the absence of the bar, the marital exit rate would have been halved.

The analysis indicates that female careers were indeed shorter than male careers (Table 4.3). The observed rates show that men worked about forty years to women’s ten to thirteen, a very substantial difference. Furthermore, this gap was not wholly a function of the marriage bar. Even under the most generous assumption of bar removal, that in which female marital exits are eliminated entirely, women worked only thirty-one years. This is a long time, but is nevertheless nine years shorter than the average male rates.

However, the significance of this differential is diminished by the other major finding in the table. Women’s careers in absolute terms were quite long. Observed women’s careers averaged about thirteen years. One could qualify for a dowry after six years of service. Female career lengths substantially exceeded this. The median age of marriage for English women in this period was twenty-six, which would allow a ten-year career (Wrigley 1969). The careers shown here are longer than those typical of British women of the period. Both the Post Office and the GWR provided unusually high-paying, attractive careers for women. This suggests that when offered better-than-average employment, women were willing to defy both organizational incentives and social convention in order to keep it. Career lengths of thirteen years are inconsistent with the prevailing stereotype of women as transitory employees.

These career lengths are substantially longer than the training times for clerical positions in these firms. Data on contemporary human capital requirements are availible from the Hobhouse Commission on Postal Labor that sat in the House of Parliament in 1906. Both management representatives and union activists testified before this committee on the equities and inequities in civil service pay scales, with both sides providing extensive data on the training required to perform various duties in order to substantiate their pay claims. Not surprisingly, the human capital estimates vary depending on which side is testifying and are often only sketchily supported with empirical data. Nevertheless, by averaging the conflicting claims and identifying consistencies between the rival accounts, some general conclusions may be drawn. The longest learning curves were those for telegraph signalling supervisors and engineers. These were all trained in-house. Telegraph overseers required six to eight years of training, and engineers required even more time (Hobhouse 1906). At the opposite extreme were mail-sorting overseers. A filing supervisor could be trained in two years. A very common intermediate case was the accounting supervisor. Their skills were comparable to those of the modern certified public accountant, who can be trained in about five years (Montagna 1975). Training times were thus distributed from two to eight years, with a peak at the five-year mark.

Observed female careers were over twice as long as these estimates. Removing the marriage bar would only increase this discrepancy. Female careers could theoretically have been extended further by judicious promotion. During apprenticeship, an employer would have been able to identify women with particularly high levels of labor force commitment and, instead of promoting at random from all women, he could promote from the top half of the anticipated tenures. An employer promoting a woman at the seventy-fifth percentile of anticipated longevity could have obtained an expected career length of over twenty-six years, the equivalent of twenty-one years of service after the completion of training. This is inconsistent with the expectations of short female job tenures expected from natural turnover theory. However, long tenures are precisely the problem that necessitates synthetic turnover. Given forty-year male careers, women with low propensities to quit, and a tenure-based salary system, these employers had a strong incentive to increase employee quit rates. The policies of marriage bars and the confinement of women to entry-level positions helped to achieve this end.

The Location of Synthetic Turnover

Synthetic turnover offers an important alternative to natural turnover explanations of low female occupational status. However, it cannot provide a general explanation of this phenomenon. A major objection is that overt discrimination against older or married women has not been a universal aspect of female employment. Marriage bars are virtually non-existent in the United States today. Surveys of American employers before World War II found such overt discrimination only in a subset of employers. The Purdue sample found that 40 to 50 percent of all employers did not formally bar the hiring of married women. Since the theory only applies to particular employers in particular periods, it is necessary to specify the precise circumstances under which a synthetic turnover regime can appear.

As argued above, synthetic turnover regimes are most likely to appear when all of the following are present: renumeration based on tenure-based salary scales; learning curves that peak early; a substantial supply of replacement labor; and a normative climate that will tolerate overt sex discrimination. This last condition rules out the likelihood of synthetic turnover regimes existing in contemporary America. The presence of vigorously enforced equal opportunity legislation makes implementing such a regime unfeasible today. However, overt discrimination would have been consistent with the sex-role ideologies of earlier periods. In those periods, the presence of marriage bars would have been determined largely by the first three economic considerations.

The first factor helps to explain why discrimination against married women is so prevalent in Japan. Japanese employment has long been characterized by tenure-based salary scales, even among blue-collar workers. A common system is one in which a substantial percentage of pay is determined by the age and seniority of the worker. This system, called “nenko,” is associated with the extremely well-developed internal labor markets that are typical of Japanese firms. Age-based payment systems are particularly prone to automatic inflation. In Japan, they have been financed by continuing organizational growth and economic profitability. In the 1970’s, as Japanese economic growth began to stagnate, the cost burden of the nenko system became particularly intense (Cole 1979). The pervasive use of seniority pay in Japan would have been a strong incentive to introduce synthetic turnover. The 1962 International Labour Office report characterized Japan as an economy-wide user of marriage bars.⁵ In the 1970’s the need for such a system would have increased. Japan may be one of the only countries in the developed world that still uses bars extensively.

Among occupations, the combination of tenure-based salary scales and short learning curves is most likely to appear in clerical work. Unfortunately, no satisfactory data exist on the precise distribution of systems of remuneration or training requirements in the labor force; this is especially true for historical discussions. The following argument is put forward as a conjecture only; the data used in its support are illustrative, but not necessarily conclusive.

White-collar work—professional, managerial, clerical, and to a lesser degree sales—is particularly likely to pay employees on tenure-based salary scales, while blue-collar workers are more like to be paid with flat rates or piece rates. One indicator of this is the relationship between earnings and age. White-collar workers have pay profiles that increase with age over the entire course of the life cycle, whereas blue-collar profiles are relatively flat, with a slight increase in early youth and a decline in old age (Phelps Brown 1977).

Human capital theorists might argue that the steep white-collar pay profile reflects a long period of learning on the job, because in the human capital model, pay is related to productivity (Mincer and Polachek 1974). However, there is evidence that the increased remuneration of white-collar workers over time is due less to productivity increase than to automatic tenure-based pay raises. James Medoff and Katherine Abraham examined the relationship between tenure, productivity, and salary among three samples of executives. They found that for these executives salaries routinely increased over time. These raises were only partially related to productivity. Productivity did correlate with pay raises due to promotion. However, non-promoted employees also received substantial pay increases. The correlation between these within-grade pay increases and productivity was close to zero. The fact that salaries increased over time with no clear relation to performance strongly suggests the presence of automatic, tenure-based scales (Medoff and Abraham 1980, 1981).

Managers and clerks tend to be paid under similar regimes. In the Great Western Railway and the General Post Office, both managers and clerks were paid using salary systems that incorporated automatic raises. The only feature that distinguished the two groups was the mean level of remuneration (GPO Establishment Books 1857–1937; GWR Staff Census 1933). However, clerks and managers do differ on one critical aspect; clerks are much less skilled than managers. Most clerical workers enter employment pre-trained. They learn typing, dictation, and bookkeeping in secretarial courses, offered privately or in high school (Oppenheimer 1970; Rotella 1981). Clerical learning curves should flatten out before managerial curves do. If both workers are paid on the same system, clerical work will create much more severe problems of overannuation.

Manual occupations with flat salary scales have no relation to tenure. Levels of remuneration are often set by collective bargaining, with the resulting rates applied to particular jobs, irrespective of the individual attributes of the job holder. Futhermore, manual work is often paid at piece rates. Since piecework relates remuneration to actual output, the contradictions of tenure-based systems are thus avoided. Therefore, in manual occupations there is no need to engage in synthetic turnover.

The service occupations and sales work are intermediate cases. Service occupations are too heterogeneous to allow for simple characterization. Sales work does utilize time-based salary scales. However, in many circumstances these scales are augmented by commissions. These commissions are productivity-based and reduce the threat of superannuation.⁶ If these considerations are valid, clerical work should be more likely than other occupations to use synthetic turnover because of its combination of tenure-based salary scales and short learning curves.

Of all the surveys of employer behavior reviewed in this study, that which found the highest extent of discrimination against married women was the British National Whitley Council Study of 1946. This survey found that every private employer in its sample used marriage bars. The Whitley Council sample was the only one that consisted entirely of large bureaucratic clerical employers. Large bureaucracies should be especially prone among clerical employers to use synthetic turnover because they are especially likely to utilize internal labor markets (Edwards 1979). The Whitley sample thus had a very high propensity to use tenure-based salary scales.

Studies of marriage bars in multiple industries tend to support the correlation between clerical work and discrimination. The Purdue study found that offices were more likely than factories to restrict the employment of married women. Sixty-one percent of the offices had some form of bar on the hiring of married women, as opposed to 50 percent of the factories. Thirty-three percent of the offices required resignation upon marriage, while only 20 percent of the factories did so. The Internal Labor Office reporters also noted a predisposition of clerical employers towards discrimination. The ILO accounts were largely impressionistic. No systematic attempts were made to compare industries; rather when an observer found a particularly striking example of the use of bars, it was included in his nation’s report. Nevertheless, there are systematic tendencies as to which industries are cited and which are not. The reporters from western Australia, Belgium, Colombia, El Salvador, and Italy all explicitly noted that discrimination against married women was more common among salaried employees than among wage workers; no one reported the opposite trend (International Labor Office 1962).

A more systematic measure of discrimination against older women is available: the actual age structure of occupations. Very few older women will show up in an occupation having a marriage bar in effect. Young age distributions are also a product of selective hiring in favor of juveniles. A preference for female juveniles is a preference for women who may self-select out of employment on the occasion of marriage. Table 4.4 comes from a survey of American employers in seven cities in 1955.⁷ It shows the age structure of female employment in the eight-code census category occupations. In most occupations 36 percent to 40 percent of the women were older than forty-five. A marked exception was clerical work. Clerical work had by far the youngest age distribution among women, with only 22.8 percent of its employees being older than forty-five.

There are several alternative explanations of why clerical work might have a young female age structure. It could be that an absence of older women represents a high female propensity to quit because of domestic obligations. But it is hard to see why women in clerical professions would feel more domestic than those in other occupations. Women may leave in reaction to the monotony and low pay of clerical work; unskilled labor, which is also poorly paid, has the second youngest age distribution in the table. However, consider the service occupations, which are made up in large part of private domestic workers, waitresses, and the like. These are unremunerative, deadend jobs with an age distribution that is relatively old.

A young age structure could alternatively be the result of employers’ preferences for young, attractive employees. Certainly such a case could be made for clerical work, where receptionists and secretaries are often hired to work with the public. However, other occupations requiring female contact with male customers do not have particularly young age structures. Service occupations are in the middle of the female age distribution, and female sales workers are the second oldest in the table. The previous arguments suggest that neither high natural female turnover, low pay per se, nor bias towards physical attractiveness adequately explains the unusually young age distribution of female clerks. Such a unique position is completely consistent with the synthetic turnover model.⁸

Labor scarcity also has an impact on synthetic turnover. When replacement workers are difficult to obtain, increasing turnover is no longer rational. Marriage bars were fairly common during the Great Depression, but they became much less common during the labor scarcity of the 1950’s. Valerie Oppenheimer presents data on the extent of discrimination against married women as school teachers in American urban school districts. In 1928, nearly 60 percent of all school districts discriminated against married women. This figure rose to 77 percent in 1930 and 87 percent by 1941. After the war, the figure dropped to 18 percent and has declined to negligible levels since then (Oppenheimer 1970). What is true for American schoolteachers may have been true universally. The 1962 ILO study still found substantial evidence of discrimination. Nevertheless, most reporters remarked that the extent of such discrimination was decreasing. Many referred to particular discriminatory industries as isolated cases of policies that were steadily disappearing from general practice.

Conclusion

This study has shown that there is a clear domain of employment in which the natural turnover theory is inappropriate. These are the firms with explicit policies of synthetic turnover, fims that confine women to low-status positions as a strategy to reduce the need for incentive wages. While this practice has never been universally followed, it was implemented by a relatively large percentage of employers in earlier times, particularly for salaried positions. Where such practices can be clearly documented, the natural turnover hypothesis must be rejected as an explanation of low female status.

Women’s confinement to subordinate economic positions has been a general phenomenon. Synthetic turnover regimes have been much more restricted in their application. The question arises as to how one could explain women’s low status in these industries that have no apparent marriage bars in operation.

It is plausible that policies of synthetic turnover that have existed in the past could have a lagged effect on sex-typing, even if the actual discriminatory policies themselves have been abolished. Many occupations may have had marriage bars at the time of their inception. Sex types are remarkably stable: most occupations do not change sex composition by as much as 5 percent per decade (Snyder et al. 1978). Norms prohibiting women from access to certain positions could have been created in the era of marriage bars. A marriage bar is likely to be buttressed by the creation of an elaborate ideology “explaining” why women cannot be considered for certain superior jobs. After the marriage bar disappears, this ideology could remain, creating a lasting barrier to female advancement.

The natural turnover hypothesis remains questionable. Its major premise, the intrinsically higher turnover of women as compared to men, has never been convincingly demonstrated. Synthetic turnover is probably only part of the explanation of the overall occupational subordination of women. New strategies of explanation are clearly required if we wish to understand the phenomenon in economic terms.

Notes

1. Most writers claim that the mean labor force commitment of women is lower than that of men, although they also admit that there is variance among women such that some have preferred job tenures equal to or longer than those of men. There is disagreement on the specification of which jobs become less likely to be female. One argument states that higher mean female turnover disqualifies women from jobs with large amounts of employer-financed human capital acquisition (National Manpower Council 1957; Oppenheimer 1970; Davies 1975; Lyle and Ross 1973; Thurow 1975; Rotella 1981). A second version argues that women are barred from jobs involving high skill rather than long learning curves per se (Caplow 1954; Baker 1964; Holcombe 1973; Madden 1973,1975; Grimm 1978). Others have suggested that women are allocated to jobs with flat wage time curves (Zellner 1975), jobs with no power (Wolf and Fligstein 1979a, 1979&), or jobs in the secondary labor market (Piore and Doeringer 1971; Barron and Norris 1976; Blau and Jusenius 1976).

2. See the extensive literature reviews of B. W. Anderson (1974) and Price (1977).

3. Aigner and Cain (1977) have developed a statistical discrimination model that does not depend on unequal mean attributes between majority and minority. They claim that black occupational performance is harder to predict than white performance from measures of educational attainment. However, its relevance to gender is unclear. No one has developed a concrete analogue of such uneven predictability for gender that does not reintroduce the premise of lower mean female job commitment.

4. The analysis is subject to the methodological dangers inherent in estimating cohort parameters from period statistics. The younger rates are based on the experience of a set of younger workers. The older rates are based on the experience of a different set of older workers. If there are cohort-specific differences, then the younger group will not replicate the older rates when they age. Thus the synthesized career will not reflect the experience of any particular historical cohort (Shyrock and Siegel 1975). Despite these problems, these tables are still useful. Management used these data in actual policymaking. At a minimum, they represent prevailing managerial perceptions of gender differences. If the cohort differences are minor, they also reflect real worldly experience.

5. The other three society-wide users of marriage bars, according to the ILO, were the Netherlands, Italy, and Ireland. Further research is needed to identify the factors facilitating discrimination in these countries.

6. Department store clerks provide a significant exception to this discussion. They are usually paid on straight tenure-based salary. Failure to perform, however, is often accompanied by dismissal. By firing non-performers, management keeps salaries and productivity mutually consistent. Productivity in sales work is easily measurable by sales. Productivity in clerical work is harder to measure. Furthermore, individual sales are likely to fluctuate from period to period. Clerical productivity is likely to be fairly constant over time. The lack of readily observable short-term slumps in productivity may have kept clerical employers from using particularistic firing to keep costs in line.

7. The seven cities are Detroit, Los Angeles, Miami, Seattle, Philadelphia, Min neapolis-Saint Paul, and Worcester.

8. The relative rankings of age distributions by occupation for men are roughly similar. The percentage of all male employees who were forty-five or over for the various occupations was as follows: clerical, 32.1; unskilled, 33.2; semi-skilled, 30.4; service, 54.1; skilled, 40.3; sales, 35.4; professional-managerial, 36.5. For both sexes, clerical, unskilled, and semi-skilled work were the youngest occupations, while sales, professional, skilled, and services were much older. The correlation between an occupation’s percentage over forty-five for men and women is .41. The relatively young age of male clerks undoubtedly was not caused by marriage bars. Here the primary mechanism was the selective recruitment of juveniles. Upward mobility could also play a role; male clerks, for example, may move into managerial occupations.

Male clerical workers are only the second youngest occupation in the table because the male clerical workers tended to be older than the female clerical workers. Male clerical workers might be expected to hold more skilled positions than female clerks. If this is the case, male clerical workers are not obvious examples of salaried workers with short learning curves. Tenure-based salaries with medium learning curves have effects similar to low skills with flat rates and produce a male clerical age structure similar to that of unskilled workers.