John C. Brown
Clark University


An important chapter in the history of industrialization is the development of a factory labor force accustomed to factory discipline and regular hours of work. Most commentators argue that high turnover and limited attachment to the job was typical of a large share of the labor force during industrialization. Factory labor forces were often recruited from the ranks of agricultural or proto-industrial workers, most of whom were unable or unwilling to adapt to the long hours, routine, and regular attendance required of factory work. The apparent contrast of nineteenth-century "nomadism" with the stability of post- World War II employment relationships has posed the question of what has explained the emergence of an industrial labor force with long-term job attachment. Since the 1970s, labor economists have stressed the efficiency gains from reducing turnover among workers placed in jobs where productivity rises with the acquisition of firm-specific human capital. The longer a worker trained by a firm remains with it, the more complete the amortization of the firm's and individual's investment in training. Ceteris paribus, a higher degree of security in the employment relationship would diminish the risk associated with employer or employee investments in firm-specific human capital. Saxonhouse has likewise found a strong relationship between productivity growth and average levels of experience in Japanese cotton textiles. The emergence of a labor force with a high degree of job attachment may have important implications for productivity growth associated with this form of human capital accumulation. The key questions in the debate as it has emerged are whether a shift to long-term employment occurred after all, and what caused it.

In the United States, historians such as (Jacoby [1985]) argue that the changeover from high to low mobility began in the 1930s. Before World War I, a small number of employers attempted to use home ownership schemes, company housing, and amenities such as lunch rooms in an attempt to stem high turnover. Loose labor markets after the war dampened employer interest in many of these personnel and employment policies designed to assist worker retention, although a minority did experiment with profit-sharing or stock plans, health insurance, pensions, and vacation plans. McHugh[1988: 24-25] suggests that although they may have been unsuccessful elsewhere, paternalistic policies of southern textile mills found in isolated rural locations contributed to the rapid stabilization of the labor force by the 'teens. Elsewhere in the economy, the rise of unionism during the Great Depression and World War II revolutionized labor-management relations, leading to the long-term employment typical of the postwar period.

Accounts of textiles, machine-making, and mining in Germany emphasize as well the high rates of fluctuation prevailing among workers before World War I (see Borscheid [1977: 276-282] among others). German firms invested heavily in efforts to stabilize the labor force with the direct provision of services such as company housing, health services, sickness and accident insurance, nurseries,etc. The mining colonies of the Ruhr coal district, industrial settlements such as those set up by Siemens outside of Berlin, and the famous Krupp settlements in the Essen area are the most complete expression of these efforts to develop a stable core of workers(Schulz [1990]). These efforts far outpaced comparable initiatives made much later in the United States. By the 1870s, before the enactment of compulsory insurance laws, 60 percent of German cotton textile firms offered employees sickness insurance and another 56 percent offered accident insurance. By comparison, an American survey in 1929 revealed that only one-sixth offered such insurance. Opinion is mixed, however, on how effective these efforts were in stimulating job attachment.(See Saxonhouse and Wright [1984], who argue that patterns of entrance and exit exogenous to firm actions can explain rising job attachment in the U.S. South or Japan).

Sparked by an interest in the macroeconomic consequences of sticky wages and long-term employment, Carter and Savocca [1990] question whether a changeover in job attachment took place at all. Hazard analysis of data from San Francisco manufacturing in 1892 prompts their conclusion that "most employment was concentrated in lengthy spells", of which the length for the average worker was 13 years. (The estimated distribution of spells for the average worker is found in column two of Table 1). The implication is, of course, that labor markets by 1890 looked very much like labor markets of the 1980s, notwithstanding the lows levels of unionization and the virtual absence of paternalistic


If the wage was basically set a low real wage floor (which only gradually drifted upward in response to improvements in agricultural labor productivity) we would expect the wage to be highly responsive to changes in the consumer price index but not to changes in productivity. In this section I present time series regression evidence using figures on nominal textile wages, labor productivity and the consumer price index, consistent with this interpretation.


Because the Japanese textile industpolicies typical of German labor markets.

Table 1
Distributions of Spells of Employment

Spell  (in Years)	United States	Germany			
	Interrupted Spells	Interrupted
Spells	Completed Spells		
	Urban Mfg.
1892	Cotton Workers in rural South 
Jan. 1, 1900	 Cotton Workers in rural South 
1895-1905	Cotton Workers in rural North 1905-1910	Skilled Workers in Berlin
<1	0	9.4	40.2	38.4
6.44	11.1	13.9	15.0	15.0
2-3		5.1	10.8	15.0	11.1
<3	6.44	25.6	64.9	68.4	79.0
3-5	26.9	8.5	13.9	21.7	11.0
5-10	39.7	11.1	9.7	9.9	5.0
>10	26.9	54.7	9.8		3.0
Avg.	13.0	20.56	4.1	na	na

Sources: Carter and Savocca [1990, Table 4]; Heiž [1910: 142-149]; Baden-Wčrttemberg Wirtschaftsarchiv, Bestand B39, Books 30-33; Stiftung Westfaelisches Wirtschaftsarchiv, Bestand F61, Nos. 88-101.

How long were job spells before 1914?

A recent contribution to the debate has suggested that these conclusions may rest on a faulty statistical foundation (Jacoby and Sharma [1992]). Analyses of duration data from surveys taken at a point of time (interrupted spells) will undersample short spells of the population at interest (employees at a firm) and oversample long spells. This sampling error may seriously overstate the average duration of employment spells at the firm: the key variable at issue in the discussions of turnover. An alternative sampling strategy is to use data on the experience of all workers exiting or entering employment during a certain period time; this strategy can be applied either to firm level or individual-level data. The data on distributions of employment spells in Table 1 illustrate the implications of using alternative measures. The German data on urban workers in the final column are from a survey of 110 skilled machinists, fitters, and lathe operators working in the precision engineering industry. Rather than a snapshot of workers employed in a particular firm, these retrospective data are based on the shortest and longest spell experienced by this group of workers. A remarkably large share of job spells (almost 80 percent) was less than three years. The short durations are particularly puzzling, given the general view that skilled workers tended to be less mobile than other groups. The data from the rural north German spinning mill compiled from payroll records imply as well that a substantial majority of workers were employed for relatively short durations. A third data source--the employee registers--from the records of the Suddeutschbaumwollindustrie, an integrated mill in Kuchen, Wčrttemberg (southern Germany) allows comparison of the sample distributions of interrupted spells on January 1, 1900 and the spells completed during the period bracketing that date. Before World War I, the Kuchen mill experienced low monthly rates of turnover of 2-3 percent, close to rates reported for the Krupp steel workers in 1910 and roughly equivalent to rates reported for American manufacturing during the 1960s. While only one-quarter of the interrupted job spells at the mill were under three years, almost two-thirds of the completed job spells during these ten years were under three years. The average spell of all workers in the sample leaving employment in the period is one-fifth of the estimated completed spell of those employed at the midpoint of the period.

These conclusions reflect only the sample distribution and may suffer from one important drawback: they are not corrected for the age or sex distribution. As is well known, job separation rates today tend to be much higher for younger than for older workers. Saxonhouse and Wright [1984: 30] suggest that quit rates for young men in southern textiles were much higher than for young women. Evidence on housing mobility ca. 1900 in Munich arguably points towards the same conclusion: the young moved frequently, but by the early 30s most families had settled in to long-term residence. If the mobile share of Kuchen millworkers were concentrated among young men, then the data for the Kuchen mill would not be comparable with the American data.

Although only available for the eight years prior to World War I, additional data on the age and job assignment of new entrants into the mill are available to estimate a simple proportional hazard model of the determinants of the length of employment in the Kuchen mill. The model includes the age, age squared, origin (whether from the village of Kuchen, the surrounding area, or further away), and potential for skill acquisition noted by Bernays[1910]. At the top of the hierarchy were the white collar workers (office clerks and sales agents). Given the heterogeneity of German firms' product mix, it may be expected that this group would be most likely to acquire knowledge of markets (and accounting practices?) particular to the Kuchen firm. Those traditional craftsmen working in the machine shop may have also held firm-specific knowledge. The skilled category includes weavers and spinners (both mule and ring). In contrast to the semi-skilled workers working in winding or the preparatory processes, a skilled worker typically required work for a period as an assistant before advancing to a full position as a weaver or spinner. The base case includes unskilled workers employed in the card and mixing room, cotton storage, and other day work. While the data do not distinguish marital status or the number of children, estimation of a similar model using data from a textile mill in northern Germany suggests that these variables would only insignificantly increase the likelihood of job separation for male workers.

Results of Estimation

Table 2 presents the results of this estimation. Chi-squared tests supported partitioning the sample into subsamples of males and females. Recalling that the coefficients reflect the impact of an independent variable on the likelihood of exiting the labor force, the time paths predicted by the age variables suggest a much stronger influence for women than for men. The probability of remaining on the job declined significantly for women until the late thirties, when it began a modest rise. The probability that males would exit rose gradually from ages 14 to 21; by age 30, the probability was equivalent to age 14. Consistent with the historical literature, rising skill levels--particularly the skills of a traditional craft and white collar employment--reduced the likelihood of quitting by 40 to 60 percent at a point in time for males, while the semi-skilled were least likely to quit among females. Finally, those from the area of the mill were fifty percent less likely to quit than workers from more distant locations.

Table 2
Results of Hazard Function Estimation
Variable	Males	Females
Age	.0146
(0.44)	.082
Age2	-.0005
(.982)	-.0011
Semi-skilled	-.668 
(1.13)	-.903 
Skilled	-.431 
(3.17)	-.430 
Traditional Craft	-.856 
White-collar	-.902 
(2.13)	-1.20 
From locality	-.432 
(1.79)	-.623 
From district	-.564 
(2.42)	-.686 

The specification of the proportional hazard model allows a separate estimation of the impact of the regressors (age, skill level, etc.) and underlying distribution of employment spells using the partial likelihood Cox method(Kiefer [1988]). The "survivor function" offers a convenient summary of this distribution, expressing the probability of reaching or exceeding a spell of a particular length. Table 3 offers the values of the survivor functions estimated for the mean age of each group of German millworkers (28 for men and 23 for women) as well as the function implied by the duration data for San Francisco workers(mean age of 29.5).

Table 3
Predicted Probabilities of Remaining on the Job at Various Tenures
Group	Length of Employment Spell			
	1/2 Year	1 Year	2 Years	5 Years
Male Cotton Workers: Germany	.67	.57	.44	.23
Female Cotton Workers: Germany	.72	.55	.39	.16
Male Manufacturing: USA	.94	.89	.79	.56

The survivor functions reveal two meaningful patterns. First, probabilities of remaining on the job were signficantly lower for the Kuchen workers, despite the higher average job tenure when compared with San Francisco. Second, differences in job retention over the life cycle do not fully account for male-female differences. After slightly higher reliability on the job during the first year, females experienced higher exit rates from employment. The strong downward impact of age noted in Table 2 would only further diminish the job attachment of females as they grew older.


This paper offers evidence that workers in German textiles experienced rates of job separation that were very high by today's standards. The results are at variance with estimation using American data from a similar period and contrast sharply with patterns observed in the U.S. South. Differences in the sampling procedures--the German data includes completed spells from several years--may help account for the divergent conclusions. Could German textile firms influence the pattern of turnover? Separate analysis of data from a north German mill suggest that company housing and use of the job ladder could heighten employee attachment. Whether these efforts were sufficient to overcome the competition of other employers (in heavy industry, for example) is an issue that requires further analysis.