Employment Duration and Industrial Labor Mobility in the United States, 1880–1980

Abstract

Recent studies of job tenure raise the question of the appropriate duration statistic to use in historical research. This article compares duration measures and examines their empirical and theoretical implications for historical research on employment tenure. Using a variety of data from the late nineteenth and early twentieth centuries, we find that although there existed a sector of stable jobs, most industrial jobs were brief. Since World War I, however, there has been a sharp shift in the relative size and importance of the short- and long-term job sectors.

Full text

Economic History Association
Employment Duration and Industrial Labor Mobility in the United States 1880-1980
Author(s): Sanford M. Jacoby and Sunil Sharma
Source:
The Journal of Economic History,
Vol. 52, No. 1 (Mar., 1992), pp. 161-179
Published by: Cambridge University Press on behalf of the Economic History Association
Stable URL: http://www.jstor.org/stable/2123349
Accessed: 01-09-2017 16:30 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
Cambridge University Press, Economic History Association
are collaborating with JSTOR
to digitize, preserve and extend access to
The Journal of Economic History
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

Employment Duration and Industrial
Labor Mobility in the United States,
1880-1980
SANFORD M. JACOBY AND SUNIL SHARMA
Recent studies of job tenure raise the question of the appropriate duration statistic
to use in historical research. This article compares duration measures and
examines their empirical and theoretical implications for historical research on
employment tenure. Using a variety of data from the late nineteenth and early
twentieth centuries, we find that although there existed a sector of stable jobs,
most industrial jobs were brief. Since World War 1, however, there has been a
sharp shift in the relative size and importance of the short- and long-term job
sectors.
Economists and statisticians have developed sophisticated tech-
niques for estimating the duration of unemployment, employment,
and other phenomena. These methods should be useful to economic
historians, especially those interested in the historical development of
labor markets. The new approach to duration modeling incorporates
data from incomplete (censored) spells and permits more precise
comparisons of past and present. With appropriate data, one can study
changes over time in the prevalence of long-term unemployment and
other issues related to labor market efficiency, or in the prevalence of
long-term jobs and issues related to internal labor markets. This article
focuses on the latter and asks how, if at all, has the distribution of
employment tenure in the United States changed since the late nine-
teenth century?
To answer that question, we discuss different measures of job
duration and their relevance to historical research. Next, we explore the
modeling of job duration and the assumptions that must sometimes be
made when using historical data. Those assumptions can be rather
restrictive, even implausible-a point we illustrate by a critical exami-
nation of recent historical studies of job duration. Finally, we present a
variety of evidence-some of it based on the new approach to duration
modeling-showing that most industrial jobs at the turn of the century
were less stable than those of more recent years.
The Journal of Economic History, Vol. 52, No. 1 (Mar. 1992). C The Economic History
Association. All rights reserved. ISSN 0022-0507.
Sanford Jacoby is Professor, Anderson Graduate School of Management, and Sunil Sharma is
Assistant Professor of Economics, both at the University of California, Los Angeles, CA 90024.
We would like to thank Janet Currie, William Issel, Keunkwan Ryu, Jules Tygiel, Mike
Waldman, and two anonymous referees for helpful comments. The usual disclaimer applies.
161
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

162 Jacoby and Sharma
I. DURATION STATISTICS
The standard published measure of job duration, taken from cross-
sectional surveys, is the mean or median duration ofjob spells currently
in progress (so-called right-censored spells, because their termination
date is not observed). In a steady state, the average duration of spells in
progress is roughly one-half their completed duration. This latter
statistic-the average completed duration of spells in progress-is
analogous to the average life span of those alive at a point in time when
ages are surveyed. Because it weights spells by their contribution to
total duration, one may call the statistic experience-weighted duration
(SEW).1
But there is a different statistic that weights equally the duration of all
spells, including those not currently in progress. This statistic-termi-
nation-weighted duration (STW)-measures the average length of all
spells that terminate during a given period. It is analogous to the average
age at death of those who died during some period (such as a year); in
a steady state it is equivalent to life expectancy at birth. The SEW
statistic is larger than the STW one because longer spells are more likely
to be in progress at any point in time. Akerlof and Main, from whom our
terminology is derived, have pointed out that "long-lived persons are
more apt to be seen in any given census; thus the person who dies at
eighty is visited by eight decennial censuses; the child who dies at ten is
seen by only one decennial census." Akerlof and Main estimated that
males in manufacturing jobs in 1973 had an SEW of 18 years, but an STW
of only about 4.5 years. Others have found differences of similar
magnitude
Which is the appropriate duration statistic? Each has its vices and
virtues. The SEW statistic suffers from length-biased sampling. It over-
samples spells that are long and undersamples short ones, many of
which ended prior to the survey. While some who held these latter jobs
are re-employed by the time of the survey, others remain unemployed or
exit the labor force. That is, SEW does not adequately incorporate
information on short-duration job holders.3
The other measure, STW, has the virtue of including information on
the work experiences of all workers whose spells terminate during a
1 The intuition for this is simple. Under steady-state or stable conditions, if a point is picked at
random to observe spells in progress, a captured spell is truncated with uniform probability over
its length. Hence, on average, an observed spell is half its completed length and, for a large
population (by the law of large numbers), SEW is twice the mean of observed (censored) spells. See
Kaitz, "Analyzing the Length of Spells."
2 Akerlof and Main, "An Experience-Weighted Measure," p. 1004. Also see Abraham and
Farber, "Job Duration," p. 284.
3 This is not a trivial issue. It has been argued that even in a quasi-panel data set like that of the
Current Population Survey, with its four-week point-in-time sampling scheme, there may be
substantial undercounting of short spells. Kiefer et al., "How Long is a Spell," p. 126. Also see
Clark and Summers, "Labor Market Dynamics," pp. 24-33.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 163
given period of time; not only those currently employed. The problem,
however, is that it contains repeated observations of workers who held
more than one job during the period.
When comparing duration distributions in different historical periods,
STW potentially is a superior statistic. If there has been a major change
in the distribution's lower tail, STW would pick it up, whereas SEW
would understate it. Demography provides a useful analogy. If a
country has a major change in its infant mortality rate, time-series
statistics on average life expectancy at birth will reflect this change,
whereas statistics on average age will not fully register it. In Mexico, for
example, male infant mortality declined by half between 1921 and 1960.
While the average age of the male population in 1921 and 1960 stayed the
same (about 22 years), male life expectancy at birth rose from 34 years
in 1921 to 45 years in 1960.4
The two statistics are likely to vary across different demographic,
occupational, and industrial subgroups. Recent advances in economet-
ric analysis of duration, and the availability of panel or longitudinal data,
allow one to model the entire duration distribution and to assess the
impact of different covariates on the distribution.5 Both in terms of
statistical estimation and the interpretation of coefficients, it is generally
convenient to specify models in terms of the hazard function. In the
continuous case, the hazard function is the conditional density at time t,
given survival until time t; in the discrete case, it is the conditional
probability of failure at time t, given survival to at least time t. The
distribution, density, and hazard functions are mathematically equiva-
lent in the sense that specifying any one of them implies specifying the
other two.6
An advantage of these econometric methods is that information
contained in censored observations can easily be incorporated in the
estimation procedure. As the proportion of censored duration spells
increases, however, the parameters of any model are less precisely
estimated because of the loss of information due to censoring.
The absence of information from completed spells can be a particular
problem for historical researchers, who often have available only data
composed entirely of censored spells (for example, from a point-in-time
survey). Consider a sample in which all observations are censored. For
simplicity and ease of exposition, we abstract from covariates, the
argument being true more generally. Let ti, i = 1,2, . .. , N, be the
4Arriaga, New Life Tables, pp. 196-206. Stable life tables illustrate the same phenomenon.
Take a population with a gross reproduction rate of two, in which the infant mortality rate drops
from 306 per thousand births to 32 per thousand. The population's average age will decline slightly
(from 34 to 28); average age at death (roughly equivalent to SEW) will increase 85 percent (from 34
to 63), while life expectancy at birth (STW) will increase nearly 300 percent (from 20 to 78). Coale
and Demeny, Regional Model Life Tables, pp. 656, 679, 728, 774.
5 See Kiefer, "Economic Duration Data," pp. 646-79.
6 See, for example, Cox and Oakes, Analysis of Survival Data, pp. 14-15.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

164 Jacoby and Sharma
censored observations for the ith individual. Suppose we fit a two-
parameter Weibull model (with parameters a, A).7 The likelihood
function for the data is the following equation:
N
L= H exp[ - Atia]
i = 1
The maximization of the above likelihood yields the corner solution A
equals 0. This makes intuitive sense. When all observations are censored
the maximum likelihood estimator makes the mean as large as possible.
Further, the duration dependence parameter a is not identified.8
With employment data consisting entirely of spells in progress at a
point in time, longer spells are more likely to be sampled than shorter
ones (we have length-biased sampling).9 To evaluate the contribution of
the observed censored spells under such a sampling scheme, we need to
make the steady-state assumption of constant entry. The combination of
length-biased sampling and the steady-state assumption identifies a and
gives an interior solution for A. Keep in mind, however, that the
steady-state assumption is a very serious simplification. The likelihood
for the data, then, is the following:
N -exp( - A tia)
L2= I A I I( 1
Note that the only difference between likelihood L1 and L2 is the
denominator, which represents the "correction" for length-biased sam-
pling under the steady-state assumption. The estimated parameters
should be interpreted as conditional on the assumptions that the
probability of getting a job is independent of calendar time and that the
7 The hazard function of a distribution F(t) with density flt) is h(t) = flt)/[l - F(t)]. The density
and distribution functions can be written in terms of the hazard function:
f (t) = h(t) exp[ - fO'H(u)du]
F(t) = 1 - exp[ - fo'h(u)du]
When we have covariates, the hazard conditional on covariates is commonly modeled using the
proportional hazard framework:
h(t,X,a,13) = h(t,a)- (X,13)
where X is a row-vector of regressors and 13 the associated column-vector of coefficients. A natural
choice for 4(X,f) [which is widely used] is exp(Xf3), because besides being reasonably flexible, it
guarantees non-negativity of the hazard without adding to the complexity of computation.
8 The duration dependence parameter captures the effect of time on the hazard function. For the
Weibull model, 0 < a < 1 implies negative duration dependence; a > 1 implies positive
dependence; and a = 1 gives the exponential case with no duration dependence.
9 Salant, "Search Theory and Duration Data," pp. 39-57; and Frank, "How Long is a Spell,"
pp. 285-302.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 165
composition of individuals becoming employed is stable over time. 10It
is clear that these assumptions are not tenable when a labor market is in
flux.
When estimating duration models, problems with the regressors can
arise. These are included in hazard specifications to control for heter-
ogeneity among individuals who, because of their various characteris-
tics and circumstances, have different duration distributions. In histor-
ical research, it may happen that key regressors affecting duration are
not included, perhaps because data are inadequate or unavailable. This
will give biased estimates of both the duration dependence parameter
and the coefficients of the included regressors. The estimate of the
duration dependence parameter is generally biased downward. The
effect on the coefficients of the included covariates is more complicated.
The estimates are biased and inconsistent, but a general result on the
direction of the bias is not known. Further, so-called neglected hetero-
geneity induces dependence between the included covariates and the
duration, so that the proportional hazard assumption, if assumed, is
violated because of misspecification.11
Neglected heterogeneity can be handled by specifying a baseline
hazard, a model for the effect of covariates, and a "mixing" distribution
for the heterogeneity.12 Recent research indicates that specification of
the mixing distribution is less important than an appropriate choice for
the baseline hazard, which is crucial. Parameter estimates are more
sensitive to the latter. 13
Cox's partial likelihood approach can be used to make inferences
about f8 (the covariates' coefficient vector) under the proportional
hazard assumption without specifying the form of the baseline hazard
function.14 Although the information contained in censored observa-
tions is used, the key to this approach is utilizing the information
conveyed by the ordering or ranking of the completed observed
durations. Unfortunately, when data consist only of censored spells,
one cannot take advantage of this semiparametric method to estimate
the impact of the regressors.
Finally, it should be pointed out that one must be clear as to what
exactly the duration variable measures are-job duration or employ-
ment duration. If the variable of interest is employment duration, but
the data do not distinguish between job and employment durations, then
10 Nickell made this point in his analysis of unemployment spells, noting that these assumptions
constitute a serious simplification. "Estimating the Probability," p. 1255. See also Frank, "How
Long is a Spell," pp. 285-302.
" Lancaster, The Econometric Analysis, chap. 4; and Sharma, "On Specification Diagnostics."
12 Heckman and Singer, "A Method," pp. 271-320.
13 Manton et al., "Alternative Models," pp. 635-44; and Ridder, "The Sensitivity of Duration
Models. "
14 See Cox and Oakes, Analysis of Survival Data, chaps. 7 and 8, for an exposition of the
partial-likelihood approach.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

166 Jacoby and Sharma
there is an error-in-variables problem. Even in large samples, this is
liable to bias the estimates of the regressor coefficients.15 In 1983 the
Bureau of Labor Statistics changed its duration measure from the
former to the latter. Because workers can change jobs without changing
employers, the new measure is a more accurate gauge of employment
patterns (and gives larger duration figures). Unfortunately, historical
surveys of employment duration do not usually distinguish between
these two measures, as for example, in the 1892 survey discussed
below. 16
II. HISTORICAL STUDIES
In a recent article in this JOURNAL, Susan Carter and Elizabeth
Savoca estimated job duration using historical data from an 1892 survey
of San Francisco workers. They found that the average completed job
tenure (SEW) in 1892 was 8.5 years for non-union males in San Francisco
and 13 years for nonfarm workers throughout the United States. The
latter is almost three-fourths as large as Akerlof and Main's SEW
estimate of 18 years for the 1970s, a degree of similarity, said Carter and
Savoca, "one would hardly expect." They concluded that at the turn of
the century as today, "most employment was concentrated in lengthy
spells." While Carter and Savoca admitted that there were "far fewer
lifetime jobs" in the nineteenth century, they nevertheless thought their
research called "into question the 'spot market' characterization of
labor markets in that era. " 9 17
Carter and Savoca's study is an innovative attempt to apply modern
duration modeling techniques to historical data. It is careful and precise.
But for the very reason that the study is important and likely to influence
future research, we think it necessary to point out some of its short-
comings. Based on our preceding discussion, we question both the
interpretation and reliability of their findings.
As should be clear by now, their use of an SEW measure masks
changes in duration patterns over time (especially for short-term jobs)
and undersamples spells of short duration. The first problem recalls our
discussion of a decline in infant mortality, which an SEW statistic may
not detect. The second problem is an echo of the 1970s debate over
unemployment-whether most jobless spells are frictional and brief or
whether most unemployment is spent in lengthy spells. As it turned out,
each side was right. Differing policy purposes were served by empha-
sizing either the majority of spells found in the lower tail or the fact that
most time spent in unemployment was concentrated in upper-tail spells.
15 Lancaster, The Econometric Analysis, p. 61.
16 Jacoby and Mitchell, "Sticky Stories," pp. 33-37.
17 Carter and Savoca, "Labor Mobility and Lengthy Jobs." Also see Carter and Savoca,
"Learning and Earning."
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 167
Yet both phenomena need to be acknowledged whether one is con-
cerned with unemployment in the 1990s or job tenure in the 1890s.'8
Another difficulty with Carter and Savoca's study arises from their
data being composed entirely of interrupted spells. As in the case of the
Weibull model discussed above, identification in their model is achieved
through a combination of length-biased sampling and the steady-state
assumption. Hence their results-estimates of regressor coefficients and
the finding that a equals 1, that is, no duration dependence-may simply
be an artifact of this combination. Moreover their empirical analysis
depends crucially on the steady-state conditions that the probability of
getting ajob is independent of calendar time and that the composition of
individuals becoming employed is stable over time.
These conditions are not valid in the case of the San Francisco labor
market in 1892. When the San Francisco survey was conducted in 1892,
the U.S. economy was still in an expansionary phase, as Carter and
Savoca noted. San Francisco, however, already was depressed. As
early as March 1892, the Coast Seamen's Journal reported that "not for
over twenty-five years has San Francisco witnessed such destitution,
misery, and suffering." It was reported elsewhere that San Francisco
was plagued by "substantial unemployment" in 1892. This depression
no doubt boosted job duration (especially of spells in progress), as
short-term workers lost their jobs and remaining workers were loath to
quit. Thus, the onset of the depression casts doubt on the steady-state
assumption underlying Carter and Savoca's estimation procedure.19
Third, Carter and Savoca have potential problems of neglected
heterogeneity due to the omission of some key variables, including
occupational status and skill. Previous historical studies of mobility
have focused on blue-collar workers, the group whose employment
patterns are thought to have changed the most during the past century.
Over time, blue-collar jobs came to have some of the stability that had
previously been the hallmark of white-collar work (although Table 1
shows that occupational status differences still remain). Yet Carter and
Savoca lump together blue- and white-collar wage earners: their sample
includes manual workers as well as clerks, foremen, managers, sales-
men, and pharmacists.20
18 Akerlof, "The Case Against Conservative Macroeconomics," pp. 219-37.
19 Cross, A History of the Labor Movement, p. 217; and Knight, Industrial Relations, p. 31.
Demography again provides a way of seeing the relationship between SEW (a population's average
life span) and employment growth (reproductive rates). At a given mortality level (job separation
rate), the reproduction rate inversely determines average life span. For numerical examples, see
Coale and Demeny, Regional Model Life Tables, passim.
20 Carter and Savoca also lumped together blue- and white-collar workers (and urban and rural
workers) in their table 1, which compares tenure of jobs in progress-from which SEW is
derived-in San Francisco versus other nineteenth-century areas. On blue- versus white-collar
work, see Kocka, White Collar Workers; and Jacoby, "Progressive Discipline," pp. 213-60. As
with personnel practices, unemployment rates for blue- and white-collar workers sharply differed
at the turn of the century: white-collar workers were much less likely to experience joblessness.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

168 Jacoby and Sharma
Their failure to take account of skill is bothersome. It was a primary
determinant of job tenure at the turn of the century, when skilled
industrial workers had lower rates of geographic and labor mobility than
did operatives and laborers. Indeed, in earlier work on the San
Francisco survey, Carter found that 93 percent of the long-term male job
holders (over 20 years' tenure) were skilled workers. Although Carter
and Savoca included schooling in their model, it is at best an imperfect
proxy for occupational skill, especially in the late nineteenth century.
Skilled blue-collar workers are over-represented in the 1892 San Fran-
cisco sample, both union and non-union. This explains why one-third of
the male workers surveyed in 1892 were union members, which is
roughly three times greater than San Francisco's unionization rate at the
time.2'
Finally, it is doubtful that Carter and Savoca's results are repre-
sentative of the San Francisco or nonfarm U.S. labor markets of the
early 1890s. This issue always arises when research is based on
fragmentary, local data. It is particularly a problem for historians, since
often these are the only kind of data available. The 1892 San Francisco
survey is fraught with "sampling bias." During the late nineteenth
century, San Francisco had a very large number of Asians (mostly
Chinese males) in its population. In 1890 Chinese workers made up 17
percent of the city's male labor force (or about 21 percent of its
non-union work force). Because of severe discrimination, the Chinese
were often relegated to jobs of low pay and status; presumably these
jobs were unlikely to offer career opportunities. Yet the researchers
who collected the San Francisco data interviewed "whites" only; not a
single Chinese man was surveyed! This is hardly surprising, given racial
attitudes in California in the 1880s and 1890s. Yet it does not inspire
confidence in Carter and Savoca's results to discover that their sample
nonrandomly omits one of five male workers. To say the least, there is
a sample selection problem that leads to undercounting of short em-
ployment durations.22
But the historical evidence on geographic mobility and occupational status is ambiguous, with
some studies finding higher rates of mobility for manual workers and others finding the opposite.
Keyssar, Out of Work, pp. 54-58; Thernstrom, The Other Bostonians, p. 230; and Guest, "Notes
From the National Panel Study," pp. 63-77. Also see Decker, Fortunes and Failures.
21 Carter, "The Changing Importance," pp. 291-92. On skill and mobility, see Thernstrom, The
Other Bostonians, p. 230; and Slichter, The Turnover, pp. 57-74. The San Francisco survey is
described in California Bureau of Labor Statistics, Fifth Biennial Report. Another variable that
would have been useful to include is establishment size (in employees) by industry. Both in the past
and today, large firms have lower turnover rates, perhaps because of higher pay (efficiency wages?)
or better management. Capital-labor ratios, which Carter and Savocca included, do not fully
capture the size effect. Brissenden and Frankel, Labor Turnover in Industry, pp. 54-55; and
Osterman, "Permanent Turnover Rates."
22 U.S. Bureau of the Census, Report on the Population, vol. 1, pt. 2, pp. 728-29; U.S. Bureau
of the Census, Population of the United States, vol. 1, pt. 1, pp. 531, 565; and Saxton, The
Indispensable Enemy, pp. 7, 210.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 169
Carter and Savoca, however, did recognize that San Francisco
differed from other American cities and attempted to correct for this.
They derived an estimate of completed spells in the United States by
taking 1890 nonfarm U.S. means for three variables (homeownership,
dependents, and marital status) and "plugging" them into their San
Francisco model. Even if one accepts the reliability of their model and
of this procedure (which we do not), the resulting national duration
estimate is unreliable because it does not control for other important
differences between San Francisco and the nonfarm United States at
this time.
Duration-related peculiarities of San Francisco include its relatively
sluggish employment growth during the late nineteenth century. In the
1880s the increase in the number of San Francisco's industrial wage
earners lagged slightly behind the United States (45 versus 47 percent),
but during the 1890s industrial employment rose by only 0.5 percent in
San Francisco versus 25 percent nationwide. As we have seen, even the
timing of the 1893 depression was different in San Francisco.23
Other idiosyncrasies include the maturity of San Francisco's male
labor force, which in 1890 had proportionately more men between the
ages of 25 and 64 than did the nation as a whole.4 San Francisco's
unionization rate in 1890 (12.4 percent) was more than double that of the
nonfarm United States, which fits with the city's reputation as a "labor
town." While Carter and Savoca properly excluded union members
from their sample, they were still left with a proportionately smaller
non-union work force than would be found in most other cities at this
time.25 San Francisco's industrial base was composed of relatively small
artisanal shops (14 wage earners per manufacturing establishment in
1890), very different from a city like Chicago, which had 30 workers per
establishment, or from the 1890 average of 22 workers for 164 U.S.
cities.26 Because it was a port city, San Francisco in 1900 had only 25
percent of its male labor force employed in manufacturing, less than in
any major American city except Washington, DC, and New Orleans.
Labor market dynamics would be affected by the smaller number of
23 U.S. Bureau of the Census, Report on Manufactures, pt. 1, p. 3; and pt. 2, pp. 1002-3.
24 Men between the ages of 25 and 64 accounted for 74 percent of San Francisco's labor force
versus 68 percent of the entire U.S. labor force. U.S. Bureau of the Census, Report on the
Population, vol. 1, pt. 2, pp. 728, 744.
25 The unionization rate is derived from Knight's estimate of 15,000 San Francisco union
members in 1888. The rate for the entire United States, using Wolman's estimate of union
membership in 1890, was 5.1 percent. The denominator in each case is the nonagricultural male
labor force. See Knight, Industrial Relations, p. 25; Wolman, The Growth of American Trade
Unions, p. 32; and U.S. Bureau of the Census, Report on the Population, vol. 1, pt. 2, pp. 304, 628.
26 Of course, these city differences also reflect industry mix. See U.S. Bureau of the Census,
Report on Manufactures, pt. 1, p. ccxlviii.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

170 Jacoby and Sharma
alternative employment opportunities available to San Francisco' s
non-union industrial workers.27
Finally, San Francisco had an unusual ethnic structure. During the
1880s and 1890s it remained relatively untouched by the "new"
immigration wave that originated in southern and eastern Europe. In
1900, 71 percent of the city's population was made up of European
immigrants and their children, nearly all of whom (94 percent) came
from northern Europe (Scandinavia or the British Isles). Thus, of the
foreign-born workers in the 1892 sample, 82 percent were born in
northern Europe or Canada. Previous studies have shown that workers
from the "new" immigrant wave were more mobile than earlier
immigrants (for example, they had higher rates of return migration), a
factor that may account for Carter and Savoca's finding that nativity had
no effect on their duration estimates.28
In light of these various omissions and the differences between San
Francisco and the nation that Carter and Savoca did not control for,
their projected estimate (13 years) of SEW for the nonfarm United States
is difficult to interpret. There is no reason to believe that it constitutes
even a rough approximation. Moreover, given their data and statistical
methodology, we believe their figure of 8.5 years for San Francisco must
be taken as tentative. Even if it were correctly estimated, it could be
telling us more about San Francisco in 1892 than about job duration and
employment stability in late nineteenth-century American cities. In
short, we have reason to doubt the claim that "jobs were not 'brief in
the nineteenth century."29
III. A DIFFERENT PERSPECTIVE
Our own view of urban, industrial labor markets is best expressed in
dual labor market terms: in the late nineteenth and early twentieth
centuries, a majority of blue-collar workers and jobs were unstable, but
there existed a small sector where workers held quasi-permanent jobs.
Since that time, the sectors reversed their relative size so that by the
1970s most industrial jobs were found in the stable sector.30 This can
27 U.S. Bureau of the Census, Population of the United States, vol. 2, pt. 2, pp. 505-7, 590; and
Issel and Cherny, San Francisco, pp. 54-57.
28 But nativity should be considered when comparing San Francisco and the United States.
Kazin, Barons of Labor, p. 35; California Bureau of Labor Statistics, Fifth Biennial Report, p. 218.
On the "new" immigration, see Hourwich, Immigration and Labor.
29 It should be pointed out that a national projection from San Francisco data should also control
for regional differences in variable interrelationships (covariance). For example, national home-
ownership rates were affected by ethnicity, age, and occupational status; there is also evidence of
these effects in San Francisco. The question is, were the national effects the same as in San
Francisco? See, for example, Monkkonen, America Becomes Urban, pp. 199-203; and Tygiel,
"Workingmen in San Francisco," pp. 273-86.
30 There is evidence, however, that the size of the primary sector has been shrinking since 1980.
Jacoby and Mitchell, "Sticky Stories," passim.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 171
TABLE I
DISTRIBUTION OF COMPLETED (PREDICTED) JOB DURATIONS
United States, 1968-1981
Proportion with Managerial
Completed San Francisco, and Blue
Duration 1892 Professional Collar
Less than 1 year 0.0 0.17 1.1
1-3 years 6.4 0.84 8.3
3-10 years 66.7 16.7 41.9
Over 10 years 26.9 82.3 48.8
Note: All figures are for non-union males and give the predicted completed duration of censored
spells.
Sources: The data in column 1 are from Carter and Savoca, "Labor Mobility and Lengthy Jobs,"
p. 13. The data in columns 2 and 3 are from Abraham and Farber, "Job Duration," p. 287.
best be seen by examining the shape-then and now-of the upper and
lower tails of the tenure distribution.
The upper tail is the locus of so-called primary or lifetime jobs. At the
turn of the century, stable manual jobs were the preserve of skilled
workers, typically employed in medium-to-large sized firms that oper-
ated on a year-round basis producing standardized commodities. Em-
ployers in these firms offered stable employment to their skilled workers
partly to retain scarce, skilled labor (skill differentials were compara-
tively wide in the United States at this time) and partly to pre-empt
unionization. In industries such as meat packing, printing, and rail
transportation, skilled workers were union members who exerted "craft
control" but combined it with employment rules adapted to the em-
ployer. Out of this grew practices such as promotional job ladders
governed by seniority as well as seniority-based layoff systems. In
pursuing these policies, trade unions adhered to the concept that
"employment was a permanent relationship between the union (a set of
workers) and the employer (a set of jobs)." As a result, unions came
"very close to creating a bureaucratic employment system for their
mostly skilled members."31
Putting aside the issue of whether Carter and Savoca's estimates are
reliable, a comparison of their upper tail with similar statistics from a
contemporary study shows that employment was less concentrated in
lengthy spells in the 1890s than in the 1970s. Table 1 shows the rapid
growth of lifetime jobs between the earlier and later periods, a change
that is obscured by simply comparing estimated means from the two
periods. The proportion of workers in lengthy jobs (over 20 years)
doubled or even tripled between the 1890s and the 1970s, going from 27
percent in 1892 to somewhere between 49 and 82 percent in the 1970s.
This is consistent with Carter's earlier work on San Francisco, which
31 Jacoby, Employing Bureaucracy, p. 30. See also Licht, Working for the Railroad; Jackson,
The Formation of Craft Labor Markets; and Elbaum, "Making and Shaping," pp. 71-107.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

172 Jacoby and Sharma
TABLE 2
PERCENT OF (CENSORED) JOB SPELLS OVER TEN YEARS
1913-1914 1928 1973
Males 15 33
All 16 30
Note: 1973 figures are for manufacturing workers.
Sources: Column 1 is from Brissenden and Frankel, Labor Turnover in Industry, p. 126. Column
2 is from Woytinsky, Three Aspects of Labor Dynamics, p. 40. Column 3 is from U.S. Dept. of
Labor, Job Tenure of Workers, pp. A13-A14.
showed the proportion of 1892 workers in jobs with current or eventual
tenure of 20-plus years to be between a tenth and a half of modern
levels. Today, the majority of male workers over the age of 40 hold jobs
that have lasted or will last 20-plus years; less than one-fourth held such
jobs in 1892. In other words, the relative size of the primary and
secondary sectors has been reversed.32
Tables 2 and 3, which are derived from censored spells in progress,
show similar results (even when the upper tail is cut at five years, as in
Table 3). Moreover, when comparing upper tails from the two periods it
should be kept in mind that labor force participation rates for older
males (over 60) fell from 66 percent in 1900 to 32 percent in 1980.33 All
other things equal, this would have reduced the relative size of the upper
tail in the 1970s and caused the reported historical changes to be
understated.
The lower tail of the tenure distribution, on the other hand, was
sizable.34 That most jobs and workers were unstable at the turn of the
century is expressed repeatedly in contemporary accounts of tramping,
"floating," reverse migration, and high quit and dismissal rates in
industrial firms. Instability was concentrated among the mass of un-
skilled and immigrant workers, although some craft workers still had a
tradition of itinerance. Job shopping was common at all ages, not just
among young workers as today. On the employer side, work was
seasonally unstable and the modal firm offered few incentives for
unskilled workers to sink roots.
Admittedly, labor turnover can be high but average duration lengthy
32 Carter, "The Changing Importance," p. 291; and Hall, "Importance of Lifetime Jobs." For
those wondering about the paucity of observations in the "under one year" cells of Table 1, note
two things; first, that these figures are derived from spells in progress, a measure that undersamples
spells of short duration (29 percent of spells in progress in the 1970s were under one year versus
64 percent of completed spells); second, that in a steady state spells in progress are observed on
average halfway through their completed duration. Among spells in progress for less than a year,
the majority have a duration greater than six months (this is our first point), so most will have a
predicted completed duration greater than one year. Also see Kiefer et al., "How Long is a Spell,"
p. 126.
3 Ransom and Sutch, "The Labor of Older Americans," p. 14.
34 Carter, "The Changing Importance," pp. 295-96; Raff, "Ford Welfare Capitalism," pp.
90-105; Jacoby, Employing Bureaucracy, passim; and Monkkonen, ed., Walking to Work.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 173
TABLE 3
PERCENT OF (CENSORED) JOB SPELLS OVER FIVE YEARS
Textiles
and
Manufacturing Automobiles Chemicals Machinery Metals Apparel
1913-1914 31 21 10 32 32 51
1917-1918 25 21 10 32 24 29
1973 49 60 55 48 55 44
Sources: 1913-1914 (n = 49,970 workers in 28 manufacturing establishments) and 1917-1918 (n =
45,791 workers in 40 manufacturing establishments) are from Brissenden and Frankel, Labor
Turnover in Industry, pp. 118-19. 1973 is from U.S. Dept. of Labor, Job Tenure of Workers, p.
A13.
if turnover is concentrated within an unstable minority of the labor
force. In a classic study of the 1910s, Sumner Slichter claimed that
turnover was "due to a few men changing rapidly while the great
majority of the force is stable." For Slichter, stable workers were those
with average durations (spells in progress) of a year or more.35 But was
Slichter right? Slichter's data were made up entirely of censored spells,
from which it is not possible to derive the size of the unstable labor force
(many of whom are between jobs at any point in time). Also, as
previously noted, spells in progress are longer, on average, than
completed spells. Among today's non-union blue-collar workers, 64
percent of completed spells are under one year, whereas 29 percent of
spells in progress are under a year. In other words, censored spells in
progress at a point in time do not accurately gauge the duration pattern
or size of the unstable labor force.36
Next we may ask, did the lower tail change its shape since the 1910s?
Recall that spells in progress surveyed at points over time will not fully
register a change in lower-tail volatility, just as periodic censuses of
average age can mask a large drop in infant mortality. What we need but
do not have are historical data on STW. Even so, the data on spells in
progress show sharp changes over time, with the proportion of workers
in short-term jobs (under one year, which was Slichter's criterion)
dropping by about 50 percent in the manufacturing sector since the
1910s (see Table 4). The changes are especially notable in the automo-
bile and chemical industries, where the proportion of workers in
short-term jobs today is one-third the level of the 1910s. Change was
less dramatic in the textile and apparel industries, which may be due to
peculiarities of those industries or to measurement error. Firms that
3 Slichter, The Turnover, pp. 43, 45.
36 Abraham and Farber, "Job Duration," p. 284. The combination of stability and mobility is
reconciled by, among others, Zunz, in his study of turn-of-the-century Detroit. He found the city's
blue-collar population to consist of a core of stable homeowners who served-as landlords and
Landsleute-as way stations for a much larger population of unstable and transient immigrants.
The Changing Face of Inequality, pp. 178-86.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

174 Jacoby and Sharma
TABLE 4
PERCENT OF (CENSORED) JOB SPELLS OVER ONE YEAR
Textiles
and
Manufacturing Automobiles Chemicals Machinery Metals Apparel
1913-1914 38 42 64 37 24 15
1917-1918 42 52 65 37 44 31
1973 22 15 16 22 18 25
Sources: See Table 3.
maintained seniority records in the 1910s were a select group, presum-
ably with job duration levels above the "true" mean.37
Although we lack historical data on STW, there are other statistics that
track changes in the unstable labor force. Each shows the same result as
Table 4-that the unstable labor force was large at the turn of the
century and has declined sharply over the years. These statistics include
measures of unemployment and labor turnover (which were concen-
trated among short-term job holders and had the effect of chopping up
job spells into shorter segments) and of spatial mobility (a source of job
mobility).
Mean levels of cyclical and seasonal unemployment were higher prior
to the 1920s than today. Seasonality at the turn of the century was
especially sharp. Industrial employment levels in 1909 fluctuated by 14
percent over the year, rising to 45 percent in the automobile industry,
whereas the industrial average in 1989 was 1 percent.38 Unemployment
need not have severed the employment relation if employers had
regularly rehired laid-off workers, but rehiring was less common at the
turn of the century than it is today.39 The incidence of unemployment-
the proportion of workers affected by job loss each year-summarizes
these changes. Incidence rates averaged about 24 percent in the late
nineteenth century; for trade union members, the average between 1908
and 1922 was 26 percent and was nearly 40 percent in 1908 and 1921. By
contrast, mean incidence for all workers between 1958 and 1987 was 15
37 Concerning their data, Brissenden and Frankel warned that "the establishments from which
the Bureau of Labor Statistics has secured labor mobility figures have necessarily been concerns
which had the figures to give, that is to say, concerns which had given more attention than most
firms to their force-maintenance problems.... In such establishments the instability is not likely
to be so serious as in the general run of American concerns, which as a rule pay little or no attention
to the flow of labor in and out and give little attention to its control." Brissenden and Frankel,
"Mobility of Labor," p. 40.
38 Keyssar, Out of Work, pp. 65-67; Jacoby, Employing Bureaucracy, p. 22; U.S. Dept. of
Labor, Employment and Earnings, Table B-2, 1989-1990; and Romer, "Spurious Volatility," p. 3.
39 Statistics from a large metal-working plant reveal that only 8 percent of all new hires after the
depression of 1907-1908 were rehires versus a recall rate in the 1970s of 60 to 65 percent. Slichter,
The Turnover, p. 126; Clark and Summers, "Labor Market Dynamics," p. 49.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 175
percent and only once was it above 20 percent-during the 1982
40
recession.
Data on labor turnover point in the same direction. Average turnover
rates were very high in the 1900s and 1910s, with monthly separation
rates commonly in excess of 10 percent, even during recessions. A
survey of 14 Detroit industrial concerns found an average monthly
separation rate of 15.3 percent in 1913-1914. Turnover rates dropped
sharply between the 1920s and the 1950s. Although a spike occurred
during World War II, it was considerably smaller than the previous
wartime peak. By the 1960s separation rates were running about 2 to 4
percent; the highest monthly rate in manufacturing during the 1960s was
4.9 percent in 1963.41
Spatial mobility is another way of tracking job mobility. Although the
measure is imperfect, since some separations involve changes between
jobs in the same area and some moves are by persons outside the labor
force, it is a useful gauge of trends in job mobility.42 Studies by Stephan
Thernstrom and other urban historians have found that spatial mobility
(and by implication job mobility) was greater in the late nineteenth
century than today. It was not unusual for decennial persistence rates in
late nineteenth-century American cities to be as low as 35 percent,
meaning that 65 percent of the population residing in a city at a point in
time was no longer living there 10 years later. Carter and Savoca argued
that a low geographic persistence rate need not be inconsistent with
lengthy residential spells. They estimate that, even with a persistence
rate of only 35 percent, the average resident stayed for over 10 years.
But this is an SEW estimate, based on the completed tenure of the
population residing in a city at the start of a decade. It ignores the
ceaseless movement into and out of a city over the course of a decade,
which can cause the mobile population to turn over several times. That
is precisely what happened in American cities of the late nineteenth and
early twentieth centuries.43
For example, Boston's population grew moderately during the 1880s,
from 362,839 to 448,477. The 1880s saw Boston's highest decennial
persistence rate of the nineteenth century-64 percent-which would
imply an average residential tenure (SEW) of 22 years. But the appear-
ance of stability is misleading. While net migration during the 1880s was
only 65,000 individuals, gross flows into and out of Boston during those
years were enormous, possibly as high as 1.5 million individuals (a
figure that dwarfs Boston's stable population). What was residential
40 Keyssar, Out of Work, pp. 51-53; and U.S. Dept. of Labor, Handbook of Labor Statistics, pp.
220-23.
41 Jacoby, "Industrial Labor Mobility," pp. 262, 268.
42 Currently 40 percent of separations involve a move and over 60 percent of the moves entail a
separation. Bartel, "The Migration Decision," pp. 775-86.
43 Thernstrom, The Other Bostonians, pp. 222-23.
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

176 Jacoby and Sharma
STW during the 1880s? While we cannot give a precise figure, we
estimate it somewhere between 1.3 years (or less) and 9.1 years, much
lower than the SEW estimate of 22 years. This is consistent with data
showing SEW to be three to five times as large as STW in recent years.44
Critics have found various biases in Thernstrom's estimates. Yet
even the critics have admitted the biases can go in either direction. Eric
Monkonnen, for example, argued that Thernstrom's figures are biased
downward, while David Galenson and Daniel Levy noted, "in practice
measured persistence rates for a population will rarely be greater than
true ones," although, they added, "the relationship between the two
often might be difficult to bound with confidence."45 One way to
"bound with confidence" is to examine other data sets. Estimates of
employment persistence derived from longitudinal payroll records are
less fraught with potential bias than are persistence data derived from
city directories. Yet the findings on nineteenth-century employment
persistence-that it was low relative to levels of recent years-mirror
the evidence on spatial mobility. Thus, data on geographic mobility are
a reliable guide to job mobility and suggest that a majority of the
population was residentially and occupationally unstable at the turn of
the century.46
IV. CONCLUSIONS
Information drawn from a variety of labor market statistics paint a
consistent picture. Most industrial jobs and workers were unstable
during the decades around the turn of the century, but there was also a
small, stable sector whose relative size and importance increased during
the decades following World War I. Accounting for the historical shift in
sectoral dominance is beyond the scope of this article. Suffice it to say
that the change resulted from a complex interaction between labor
supply factors (such as declining immigration), demand factors (in-
44 During the 1880s, 138,572 households left Boston. We do not know the average size of these
households: it could have been slightly greater than one (if mostly single males) or slightly larger
than five (which was the average size of resident Boston households). A reasonable assumption is
that departing households contained 3.5 persons, meaning that 485,000 residents left Boston during
the 1880s. The persistence rate indicates that 130,000 of these outmigrants resided in Boston in
1880; the remaining 355,000 arrived after 1880. Average tenure for the first group was somewhere
between two and twenty years; average tenure for the second group probably fell between one and
five years. This gives sufficient information to calculate tenure as a weighted average of the two
groups. It should be emphasized, however, that these figures are suggestive rather than definitive.
Thernstrom, The Other Bostonians, pp. 11-21; and Akerlof and Main, "An Experience-Weighted
Measure," p. 1007.
45 Monkkonen, American Becomes Urban, p. 195; and Galenson and Levy, "A Note on
Biases," p. 176.
46 Ginger, "Labor in a Massachusetts Cotton Mill," p. 84; Gitelman, "The Waltham System,"
pp. 227-53; and Navin, The Whitin Machine Works, pp. 160-61. Early twentieth-century payroll
records show similar instability; see Whatley and Wright, "Getting Started."
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 177
creased size and bureaucratization of firms), and institutional forces
(mass unionism and public regulation of the labor market).
Future historical studies of job duration must carefully consider the
statistical assumptions that allow derivations of completed spell dura-
tion when data are only available on censored spells. Attention should
also be given to the properties of various duration statistics. Ideally, one
would like to have panel data, akin to the modern Panel Study on
Income Dynamics, for late nineteenth- and early twentieth-century
workers. These data could conceivably be gleaned from censuses and
city directories. Alternatively, more systematic mining of corporate
payroll records would yield a sharper historical picture of job duration.
Yet care must be taken when analyzing samples from a particular firm
or region to place the data in historical context and to control for
sampling bias. Finally, it need hardly be emphasized that no single
statistic can capture all facets of a distribution's shape, much less
changes in its shape over time.
REFERENCES
Abraham, Katherine G., and Henry S. Farber, "Job Duration, Seniority, and Earn-
ings," American Economic Review, 77 (June 1987), pp. 278-97.
Akerlof, George A., "The Case Against Conservative Macroeconomics: An Inaugural
Lecture," Economica, 47 (Aug. 1979), pp. 219-37.
Akerlof, George A., and Brian G. M. Main, "An Experience-Weighted Measure of
Employment and Unemployment Durations," American Economic Review, 72
(Dec. 1981), pp. 1003-11.
Arriaga, Eduardo E., New Life Tables for Latin American Populations in the Nine-
teenth and Twentieth Centuries (Berkeley, 1968).
Bartel, Ann P., "The Migration Decision: What Role Does Job Mobility Play?"
American Economic Review, 68 (Dec. 1979), pp. 775-86.
Brissenden, Paul, and Emil J. Frankel, "Mobility of Labor in American Industries,"
Monthly Labor Review, 10 (June 1920), pp. 38-52.
Brissenden, Paul, and Emil J. Frankel, Labor Turnover in Industry (New York, 1922).
California Bureau of Labor Statistics, Fifth Biennial Report (Sacramento, 1893).
Carter, Susan B., "The Changing Importance of Lifetime Jobs," Industrial Relations,
27 (Fall 1988), pp. 287-300.
Carter, Susan B., and Elizabeth Savoca, "Labor Mobility and Lengthy Jobs in
Nineteenth-Century America," this JOURNAL, 50 (Mar. 1990), pp. 1-16.
Carter, Susan B., and Elizabeth Savoca, "Learning and Earning in Late Nineteenth
Century America: Gender Differences and the Role of Expected Job and Career
Attachment" (Smith Collge Working Paper 90-1, Department of Economics, May
1990).
Clark, Kim B., and Lawrence H. Summers, "Labor Market Dynamics and Unemploy-
ment: A Reconsideration," Brookings Papers on Economic Activity, 1 (1979), pp.
13-60.
Coale, Ansley J., and Paul Demeny, Regional Model Life Tables and Stable Popula-
tions (Princeton, 1966).
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

178 Jacoby and Sharma
Cox, David R., and David Oakes, Analysis of Survival Data (London, 1984).
Cross, Ira B., A History of the Labor Movement in California (Berkeley, 1935).
Decker, Peter R., Fortunes and Failures: White Collar Mobility in 19th Century San
Francisco (Cambridge, MA, 1978).
Elbaum, Bernard, "The Making and Shaping of Job and Pay Structures in the Iron and
Steel Industry," in Paul Osterman, ed., Internal Labor Markets (Cambridge, MA,
1984), pp. 71-107.
Frank, Robert H., "How Long is a Spell of Unemployment? Econometrica, 46 (1978),
pp. 285-302.
Galenson, David W., and Daniel S. Levy, "A Note on Biases in the Measurement of
Geographic Persistence Rates," Historical Methods, 19 (Fall 1986), pp. 171-79.
Ginger, Ray, "Labor in a Massachusetts Cotton Mill: 1853-1860," Business History
Review, 28 (Mar. 1954), pp. 67-91.
Gitelman, Howard M., "The Waltham System and the Coming of the Irish," Labor
History, 8 (Fall 1967), pp. 227-53.
Guest, Avery M., "Notes From the National Panel Study: Linkage and Migration in the
Late Nineteenth Century," Historical Methods, 20 (Spring 1987), pp. 63-77.
Hall, Robert E., "The Importance of Lifetime Jobs in the U.S. Economy," American
Economic Review, 72 (Sept. 1972), pp. 716-24.
Heckman, James, and Burton Singer, "A Method for Minimizing the Impact of
Distributional Assumptions in Econometric Models for Duration Data," Econo-
metrica, 52 (1984), pp. 271-320.
Hourwich, Isaac, Immigration and Labor: The Economic Aspects of European Immi-
gration (New York, 1912).
Issel, William, and Robert W. Cherny, San Francisco, 1865-1932: Politics, Power, and
Urban Development (Berkeley, 1986).
Jackson, Robert, The Formation of Craft Labor Markets (New York, 1984).
Jacoby, Sanford M., "Industrial Labor Mobility in Historical Perspective," Industrial
Relations, 22 (Spring 1983), pp. 261-82.
Jacoby, Sanford M., Employing Bureaucracy: Managers, Unions, and the Transforma-
tion of Work in American Industry, 1900-1945 (New York, 1985).
Jacoby, Sanford M., "Progressive Discipline in American Industry: Origins, Develop-
ment, and Consequences," Advances in Industrial and Labor Relations, 3 (1986),
pp. 213-60.
Jacoby, Sanford M., and Daniel J. B. Mitchell, "Sticky Stories: Economic Explanations
of Employment and Wage Rigidity," American Economic Review, 80 (May 1990),
pp. 33-37.
Kaitz, Hyman, "Analyzing the Length of Spells of Unemployment," Monthly Labor
Review, 93 (Nov. 1970), pp. 11-20.
Kazin, Michael, Barons of Labor: The San Francisco Building Trades and Union Power
in the Progressive Era (Urbana, 1987).
Keyssar, Alexander, Out of Work: The First Century of Unemployment in Massachu-
setts (Cambridge, 1986).
Kiefer, Nicholas M., "Economic Duration Data and Hazard Functions," Journal of
Economic Literature, 26 (June 1988), pp. 646-73.
Kiefer, Nicholas M., Shelly Lundberg, and G. R. Neumann, "How Long is a Spell of
Unemployment? Illusions and Biases in the Use of CPS Data," Journal of Business
and Economic Statistics, 3 (Apr. 1985), pp. 118-28.
Knight, Robert E. L., Industrial Relations in the San Francisco Bay Areas, 1900-1918
(Berkeley, 1960).
Kocka, Jurgen, White.Collar Workers in America, 1880-1940 (Beverly Hills, 1980).
Lancaster, Tony, The Econometric Analysis of Transition Data (Cambridge, 1990).
Licht, Walter, Working for the Railroad (Princeton, 1984).
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms

U.S. Industrial Labor Mobility 179
Manton, Kenneth G., Eric Stallard, and James W. Vaupel, "Alternative Models for the
Heterogeneity of Mortality Risks Among the Aged," Journal of the American
Statistical Association, 81 (Sept. 1986), pp. 635-44.
Monkkonen, Eric, ed., Walking to Work: Tramps in America, 1790-1935 (Lincoln,
1984).
Monkkonen, Eric, America Becomes Urban: The Development of U.S. Cities and
Towns, 1780-1980 (Berkeley, 1988).
Navin, Thomas R., The Whitin Machine Works: A Textile Machinery Company in an
Industrial Village (Cambridge, MA, 1950).
Nickell, Stephen, "Estimating the Probability of Leaving Unemployment," Economet-
rica, 47 (Sept. 1979), pp. 1249-66.
Osterman, Paul, " 'Permanent' Turnover Rates and Employment Practices" (Sloan
School of Management Working Paper, M.I.T., Sept. 1990).
Raff, Daniel M. G., "Ford Welfare Capitalism in Its Economic Context," in Sanford M.
Jacoby, ed., Masters to Managers: Historical and Comparative Perspectives on
American Employers (New York, 1991), pp. 90-105.
Ransom, Roger L., and Richard Sutch, "The Labor of Older Americans: Retirement of
Men On and Off the Job, 1870-1937," this Journal, 46 (Mar. 1986), pp. 1-30.
Ridder, Geert, "The Sensitivity of Duration Models to Misspecified Unobserved
Heterogeneity and Duration Dependence" (Unpublished manuscript, University of
Amsterdam, 1986).
Romer, Christina, "Spurious Volatility in Historical Unemployment Data," Journal of
Political Economy, 94 (Feb. 1986), pp. 1-37.
Salant, Stephen W., "Search Theory and Duration Data: A Theory of Sorts," Quarterly
Journal of Economics, 91 (Feb. 1977), pp. 39-57.
Saxton, Alexander, The Indispensable Enemy: Labor and the Anti-Chinese Movement
in California (Berkeley, 1971).
Sharma, Sunil, "On Specification Diagnostics for Econometric Models of Durations,"
Journal of Quantitative Economics, forthcoming.
Slichter, Sumner H., The Turnover of Factory Labor (New York, 1921).
Thernstrom, Stephan, The Other Bostonians: Poverty and Progress in the American
Metropolis, 1880-1970 (Cambridge, MA, 1973).
Tygiel, Jules, "Workingmen in San Francisco, 1880-1901" (Ph.D. diss. UCLA, 1977).
U.S. Bureau of the Census, Report on the Population of the United States at the
Eleventh Census: 1890 (Washington, DC, 1897).
U.S. Bureau of the Census, Population of the United States at the Twelfth Census: 1900
(Washington, DC, 1901-1902).
U.S. Bureau of the Census, Report on Manufactures at the Twelfth Census: 1900
(Washington, DC, 1902).
U.S. Department of Labor, Bureau of Labor Statistics, Job Tenure of Workers: Janu-
ary 1973, Special Labor Force Report No. 173 (Washington, DC, 1974).
U.S. Department of Labor, Bureau of Labor Statistics, Handbook of Labor Statistics,
Bulletin 2340 (Washington, DC, 1990).
U.S. Department of Labor, Bureau of Labor Statistics, Employment and Earnings
(Washington, DC, various issues).
Whatley, Warren, and Gavin Wright, "Getting Started in the Auto Industry: Black
Workers at the Ford Motor Company, 1918-1947" (Paper presented to the
American Economic Association, Washington, DC, Dec. 1990).
Wolman, Leo, The Growth of American Trade Unions, 1880-1923 (New York, 1924).
Woytinsky, W. S., Three Aspects of Labor Dynamics (Washington, DC, 1942).
Zunz, Olivier, The Changing Face of Inequality: Urbanization, Industrial Develop-
ment, and Immigrants in Detroit, 1880-1920 (Chicago, 1982).
This content downloaded from 128.97.27.20 on Fri, 01 Sep 2017 16:30:19 UTC
All use subject to http://about.jstor.org/terms