Appendix A The Postal and Railway Samples
General Post Office
Most of the data from the Post Office comes from the Post Office Establishment Books. These are annual listings of the staff of every department of the Post Office along with statistical breakdowns by sex and job title. The Establishment Books also provide information on the pay scale associated with each position and provide both the names and career summaries of the top job holders in each department. Although the data is available on an annual basis, for economy’s sake analysis was confined to the following years: 1871, 1876, 1881, 1886, 1891, 1896, 1901, 1906, 1911, 1914, 1918, 1921, 1926, 1931, and 1936. In general, every effort was made to preserve even five-year panels except during World War I, when the dates were selected to represent the last pre-war observation, the middle of the war, and the first year after the immediate post-war adjustments.
The data set is not a neat rectangular one of N departments times M years. The number of departments in the Post Office fluctuated over time, with new departments being created as the Post Office diversified and old departments being eliminated through merger or loss. In 1921, for example, all the Dublin offices were lost due to Irish independence. Furthermore, the percent female in very small offices is subject to dramatic and somewhat random fluctuations due to the addition or subtraction of one individual. As a result, the analysis is confined to all departments with a staff of at least ten, a fact that produces some mild fluctuations in the number of cases from year to year.
The departments used in the analysis are the Secretary’s Office, the regional secretaries in Dublin and Edinburgh, the Registry, the Receiver and Accountant General’s Office (RAGO), the Money Order Department, the Savings Bank, the regional accounts offices in Dublin and Edinburgh, Central Telegraph Control, the Central Telegraph Signalling Gallery, the Central Telegraph Cable Section, the Central Telegraph Intelligence Section, the London East Central (EC) Telegraph Office, the West London Telegraph Office, the telegraph departments of Dublin and Edinburgh, the Telephone clerical section, the Telephone operating force, both types of telephone offices in Dublin and Edinburgh, the Engineering Department, the Factory Department, the Stores Department, the Returned Letter Offices in London and Edinburgh, London Postal Control, the main London East Central (EC) sorting office, the West London Sorting Office, and the sorting offices of Dublin and Edinburgh.
The West London Telegraph and Sorting offices are used as proxies for the policies of the London postal service for staffing the district offices as a whole. The correlation between the levels of variables calculated for the West London office alone and those for the entire Metropolitan District are above .90. Using the West London office to represent trends in the Metropolitan District provided extremely substantial economies in data-collecting and processing. The Money Order Department includes data from the Postal Order section. In some periods, the Postal Order section was under the supervision of the Money Order Department, but in other years it was under the control of the Receiver and Accountant General’s Office. In order to prevent department scores from fluctuating randomly due to changes in the possession of a division that was in itself rather constant, it was decided to allocate the Postal Order section to the Money Order Department for all dates. Excluded from the analysis are the Medical Department and the Surveyor’s clerks, due to their small size and extremely irregular reporting procedures in the data. Furthermore, the only postal employees included in the analysis are those employed in London, Dublin, or Edinburgh. Data are only available for clerks in other locations for selected dates in the twentieth century, and the scanty quantitative data is matched by an even greater paucity of supporting archival materials. A crude examination of the provincial data for those years for which they are available suggests patterns very similar to those reported for London, Dublin, and Edinburgh.
There are actually three sources of personnel statistics for the railway. The first is the series of staff censuses discussed in the text. Data are provided for every department, and within departments they are disaggregated where appropriate into three categories: headquarters staff, divisional office staff, and depot staff. In this study, each of these divisions within the department is treated as a separate unit. The second source of data is the annual series of organizational charts of the salaried staff of the railway. These charts are lists of the clerical staff of every department by name, sex, and salary grade. These are clustered by individual office within each department; for example, within the Goods Department Headquarters one can find listings for the wages office, the scheduling office, and the rates office among others. The office staff lists are accompanied by brief descriptions of the work done in each office, allowing for crude analysis of the relation between job duties and staffing patterns. The last source of data is a set of strike statistics compiled after the General Strike of 1926 for the Chief Mechanical Engineer’s Department. For every office within the Chief Mechanical Engineer’s Department, there are data on the sex composition of the office and its strike rate. One particular advantage of these data is that they include material on the wage clerks of the shop floor. Wage clerks are included in the GWR census but are excluded from the organizational charts of salaried staff.
The departments used in the analysis are as follows:
The statistical records for the Post Office and the Great Western Railway contain data on a wide variety of employees, of whom some were clerical, some non-clerical, and some borderline. The following criteria were used to determine who would be within the sample and who would be excluded. All employees whose job title was “clerk” were included. Unfortunately, the title “clerk” also includes some personnel who were clearly managerial. It is desirable to exclude top executives from the study, and the only systematic basis for defining executives was by salary. Therefore all employees who in 1870 were earning £500 a year or more were excluded from the study. The £500 cut line usually identified a clear occupational border between clerks and non-clerks, which was then used in subsequent years. This definition is reasonably generous in allowing upper-level employees into the sample; thus the study clearly includes many workers who would be called administrative assistants or first-line managers today. Workers who had been given a job title with “clerk” in the 1870’s but were renamed in later years were kept in the sample. Furthermore, workers with obviously clerical job duties, such as bookkeepers, paper sorters, or writers, were included. Many workers at the bottom of the office hierarchy were excluded. All janitorial and security personnel were removed, as well as messengers and office boys. Messengers and office boys performed substantial amounts of clerical work in the nineteenth century, but their primary duties tended to be non-clerical. Itinerant workers were also kept out of the analysis. This includes canvassers, inspectors, surveyors, traveling sales personnel, and members of the Travelling Post Office. The normal conception of office work involves sedentary personnel. A variety of blue-collar clerks were excluded as well. These were workers hired within a manual chain of command to do primarily clerical duties. An example would be a worker on the Great Western Railway called a “caller-off,” whose job is to identify all of the packages entering or leaving a freight car and check them off against a bill of lading. Ideally such workers would have been included and their work settings considered as additional units of analysis. The problem is that the job titles of most of these workers are somewhat arcane and are not easily identified by the modern-day observer. Because these workers are easy to miss, there is a very real danger that only some of these workers would be included in the sample and the resulting description of this population would be seriously distorted. Stationmasters were generally excluded as being managerial. Technical and scientific personnel were removed whenever possible. Because of the ambiguities of nineteenth-century terminology, a high number of technicians were included in the engineering sample. Repetition of the main analyses of the book excluding all the disputed cases provides findings similar to those presented in the text, suggesting that the effect of including the hidden technical personnel is not substantial.
The gender wage differential correction is designed to eliminate that component of the lowering of wages that is strictly due to the fact that women get paid less than men for doing identical work. The overall goal is to calculate a set of rates that reflect how much would have been paid had all the work been done by men. This entails calculating an average salary for an office that uses the observed male salary rates and a set of female rates that have been inflated by a factor measuring the differences in pay for men and women doing jobs of comparable worth.
To quantify the sex-discrimination factor, I searched the job classifications to locate jobs where men and women did work that was close to identical. Such assessments inevitably involve subjective judgments. Five pairs of jobs were selected for the sex-discrimination index. These are male and female shorthand typist, male paperkeeper and female sorting assistant, male and female counter clerk/telegraphist, male and female executive officer, and male and female high-grade clerk. Of these five, the closeness of fit of the first three pairs is easier to establish than is the case with the last two.
The meaning of the term “shorthand typist” was the same in the Post Office as it is today. Both men and women did the same sorts of duties; men, however, were employed in settings such as the engineering office that preferentially hired male employees. The paperkeepers and sorting assistants were essentially filing clerks. “Paperkeeper” is merely an older term for the job. To some extent, the female sorting assistants were more skilled. They were expected to rotate from office to office doing whatever putting away was required. Thus, they had to learn the filing systems of many different departments. The paper-keepers were limited to one department and could confine themselves to a single system. Counter clerks/telegraphists were the clerks who served at postal windows. They were also expected to transmit local telegraph messages using Morse code. Since local post offices were either all male or all female, and the work of each post office was fundamentally similar, the work of these two classes can be viewed as identical.
The situation for the two pairs of supervisory grades is somewhat more ambiguous. The executive clerks were management trainees. They were started at a very low salary and given broad pre-management training. Their scales allowed them to by-pass the bulk of the clerical force and advance quickly to responsible positions within the bureaucracy. It is unclear whether female executives got to the same levels of authority as their male counterparts. The high-grade clerks were supervisors who were promoted up from the ranks. Their primary responsibility was the supervision of large bodies of single-sex subordinates. The nature of the advanced substantive tasks involved with their jobs is unclear.
The ratio of male to female pay in these job categories was as follows:
All of these figures are within .12 of each other, and the three most similar jobs are all within .05 of each other. This suggests a relatively consistent pattern of bias in pay scales.
Note also that the discrimination coefficients for the supervisory jobs is lower than those for the rank-and-file positions. One would normally expect higher-level jobs to show more discrimination; had these jobs been “truly” dissimilar, the reasonable bias one would expect would be that men would have jobs having significant responsibilities and remuneration, while women would have had more limited positions with a lower ceiling on pay. This would have produced a greater gap between male and female earnings. The reverse bias that we observe suggests that the jobs were in fact “truly” identical and that the lower-grade positions overestimate the global degree of discrimination involving jobs of comparable worth.
The rhetorical goal of Chapter 3 is to argue that real wages increased in the Post Office. Findings of high levels of sex discrimination allow for the calculation of higher estimated “true” wages. To allow a conservative bias into my calculations, I have included the supervisory categories, for a low sex-discrimination correction factor and a comparatively modest estimate of real trends in postal wages.
Under Assumption B, the percentage of non-entry-level jobs is defined as all clerks with a status equal to or lower than the highest grade 5 male clerk. In 1933, there was a standard salary that represented the cutting line between a grade 5 and a grade 4 clerk: £210 per annum. The strategy used here is to calculate the percentage of clerks in 1870 who had the same relative salary or less. An equivalent 1870 salary was calculated by estimating the range of salaries in 1933 from minimum to maximum. Most salary distributions are highly skewed at the upper end. To reduce the skew, the top 2 percent of salaries were removed from the distribution. It was then determined that £210 was on the 47.8th percentile of this range of salaries. Note that this range is not the actual distribution of observed salaries; the range is the distance between the minimum and maximum salaries. The 47.8th percentile was then calculated for the 1870 salary range (after first removing the top 2 percent as before). This produced a cut point for 1870, £150 per annum. The top 2 percent was then added back into the distribution, and the observed percentage of all salaries greater than £150 was the measure of the percentage of non-entry-level jobs.
Center, Stella, and Herzberg, Max
|1929||Secretarial Procedure. New York: Ronald.|
|1942||Office Management. Rev. ed. New York: Ronald.|
Dicksee, Lawrence, and Blain, Herbert
|1906||Office Organization and Management Including Secretarial Work. London: Pitman.|
|1929||Practical Office Supervision. New York: McGraw-Hill|
|1917||Scientific Office Management. Chicago: Shaw.|
|1918||Making the Office Pay. Chicago: Shaw.|
|1925||Office Management: Principles and Practice. Chicago: Shaw.|
|1926||Office Appliance Manual. N.p.: National Association of Office Appliance Manufacturers.|
Leffingwell, William, and Robinson, Edwin
|1950||Textbook of Office Management. New York: McGraw-Hill.|
|1920||Office Training and Standards. Chicago: Shaw.|
|1923||Textbook of Filing. New York: Appleton.|
|1934||Personal Secretary: Differentiating Duties and Essential Personal Traits. Cambridge, Mass.: Harvard University Press.|
Schulze, J. William
|1913||The American Office: Its Organization, Management, and Records. New York: Key.|
|1919||Office Administration. New York: McGraw-Hill.|
The calculation of the growth rates that the GWR and GPO would have needed to absorb their graduating junior clerks can be easily calculated using stable population techniques. Each organization was assumed to hire incoming recruits as junior clerks for the first two years of their careers. Subsequently, such clerks would graduate to senior clerkships, where they would work for thirty-eight years. It was assumed that between hiring and the fortieth year of employment, there would be no resignations or exits. Furthermore, at the fortieth year, all clerks were assumed to retire. These simplifying assumptions are comparatively similar to the turnover behavior suggested by the actuarial data in Chapter 4. There was a very slight amount of turnover in the first few years of employment, and some mild deviations from retiring after the fortieth year. However, the absolute differences between the observed rates and the simplifying assumptions is not large. The actuarial data come from the 1910-25 period and may not be typical of what was experienced in earlier periods. Given this room for ambiguity, the errors in using the observed actuarial data to represent the experience of the whole 1870–1930 period are probably greater than those caused by a cosmetic simplification of the mathematical model.
For the present discussion, an adult is defined as anyone with more than two years of experience, while a junior has two or fewer years of experience. Then, using stable population techniques, one can estimate what rate of population growth is necessary to keep a fixed percentage of the population with two years of experience or less. The formula for the growth rate is
where r is the rate of population growth; g(r) is the ratio of adults to juniors that is to be maintained by management; and g(0) is the rate of adults to juniors that would cause no population growth. Put differently, g(0) is the ratio of individuals above the age of two to people below the age of two in a hypothetical stationary population. In this case, a completely rectangular age distribution will produce zero population growth, so g(0) is 38/2, or 19. MJ(0) is the mean age of juniors in the stationary population, while MA(0) is the mean age of adults in the stationary population. These figures are 1 and 21 respectively.
I am extremely grateful to Alberto Palloni of the University of Wisconsin Sociology Department for generating these equations and explaining their use.
Labor cost buffering theory would suggest that those firms that have a lot of organizational slack should be less likely to attempt to cut costs. Relatively prosperous firms should be less concerned with minimizing wages, and thus more tolerant of hiring adult male workers. Thus Francine Blau and Carol Jusenius (1976) and R. D. Barron and G. H. Norris (1976) are probably correct in arguing that women are more likely to be concentrated in the secondary sector and less likely to appear in the primary sector.
This is a controversial claim because statistical tests of this hypothesis have generally been unsuccessful. Bridges found that out of six measures of primary sector status only capital intensity had any consistent predictive power in explaining the percentage of females in American industries. Wallace and Kalleberg, using slightly different measures of industrial structure, were only able to explain 9 percent of the variance among industries in the use of women using dual sector-related variables. Most of their measures of primary sector status had no significant relation with sex. While these initial findings are discouraging, they should not be viewed as definitive. Bridges is probably correct that some of the variables used in dual sector models have little relation to occupational sex-typing. However the effect of others has been seriously underestimated, due to methodological limitations of the data.
Most of the quantitative studies of both occupational sex-typing and dual labor markets have used industries as units of analysis. This strategy has been necessitated by the practical consideration that the economic data upon which these studies depend is available in industrial aggregates. As is often noted even in these very studies, however, most dual labor market theories operate at the level of the individual firm. Most of the propositions in dual labor market theory are designed to contrast relatively monopolistic firms with more marginal firms. The contrasts between firms in a single industry are likely to be far more dramatic than the contrast across industries. General Motors and Weyerhauser have far more in common with each other than they do with Red Star Auto Parts and Tuscaloosa Plywood. Nevertheless, General Motors and Red Star Auto Parts are defined as being in transportation equipment and are thus primary sector, while Weyerhauser and Tuscaloosa Plywood are defined as lumber companies and are thus secondary sector. It is true that there should be compositional effects that allow industry attributes to be inferred from the average attributes of its firms. Nevertheless, most industries are somewhat heterogeneous, so the potential for measurement error is extremely non-trivial.
The unit of analysis problem does not mean that all of the findings of these studies should be discarded. Some of the relationships under consideration should be robust to whether firms or industries are being considered; others should be more sensitive. In general, the effects of economic concentration and economic scale are likely to be misrepresented by industry-level variables, while the firm size, state intervention, and capital intensity can tolerate the higher level of aggregation. Let us consider each of these measures in turn.
Economic concentration is the aggregate measure of whether firms in a given industry have oligopolistic power through their domination of market share. This variable is, by its very nature, zero-sum. If four firms control 70 percent of an industry’s market, the other one hundred firms must be competing for the remaining 30 percent. Thus, in a primary sector industry, the peripheral firms should be more marginal than those in a secondary sector industry, and one would expect substantial improvements in the performance of this variable if one had access to good-quality firm-level data.
Economic scale is a measure of financial resources; it is based on such measures as the average amount of assets per firm and reported levels of profitability. In some respects, this variable can be viewed as a measure of financial “size.” Economic scale is quite likely to be distorted by the use of grossly aggregated units of analysis. Within any industry, there should be fairly substantial differences in absolute wealth and profitability between the major and minor producers.
The number of employees per firm is also likely to be distorted by industry-level measures. However, attempts to remeasure firm size at a smaller level should not improve its poor ability to predict sex-type. The number of employees per firm has always been a questionable indicator of primary sector status. If a firm is underfinanced, competing for market share, and burdened with a labor-intensive technology, it is hard to see why its employment policies would be any different for a labor force of a hundred workers or a labor force of ten thousand workers. In the latter case, it is managing more employees but does not have any additional resources for paying them. Bridges is undoubtedly right in suggesting that firm size will be a poor predictor of female employment, but the behavior of size net of economic scale, concentration, and capital intensity should be irrelevant to a dual sector model of female employment.
The extent to which the state serves as a market for a firm’s goods is likely to vary significantly from firm to firm within an industry. This factor is less likely to be a problem with state regulation, since this tends to apply across the board within industries. As is the case with firm size, however, it is not clear that re-estimating the variable at a lower level of aggregation would improve its predictive power. This is because state intervention has a secondary effect that is likely to counteract the exclusionary effects of governmentally induced financial security. In recent years, the government has vigorously enforced affirmative action laws. One of the most common strategies for doing this has been to put strict minority hiring requirements on employers holding government contracts. Firms that are regulated by the state have also been vulnerable to government pressure and have had incentives to cooperate with affirmative action. If the economic security that comes from government support has had any exclusionary effect on women, this would only have occurred before 1965, when equal opportunity programs were not vigorously enforced.
Capital intensity, like state regulation, is not seriously mismeasured by using data at the industry level. This is because capital intensity is to a large extent determined by technological considerations that are common to all firms within the industry. Most factories in the lumber industry are sawmills with a fairly standard ratio of capital to labor. Most steel mills are capital-intensive, regardless of their modernity or level of technological sophistication. There are some industries where firms do differ in the extent of automation; core automobile factories use assembly lines, robots, and other forms of mechanized production, while the peripheral firms are traditional machine shops. Nevertheless, the within-industry variance in labor intensity is probably far less than the between-industry variance, a statement that is not true for monopoly power or economic scale.
This long methodological discussion has important implications for interpreting the results of Bridges, and Wallace and Kalleberg. Capital intensity is likely to remain as a significant and important predictor of the percentage of females in a firm. Firm size is likely to be unrelated to sex while the relationship between sex and levels of state intervention is likely to be extremely complex and unstable over time. Market share and economic scale should correlate with the use of white adult males. These findings would support the claims that buffering from labor costs reduces the use of women. Those firms that are relatively wealthy and have financial surpluses can afford to hire an expensive labor force.