Institutions, Technical Change, and Diverging Life Chances: Earnings Mobility in the United States and Germany

by Thomas A. DiPrete, Patricia A. McManus
Citation
Title:
Institutions, Technical Change, and Diverging Life Chances: Earnings Mobility in the United States and Germany
Author:
Thomas A. DiPrete, Patricia A. McManus
Year: 
1996
Publication: 
The American Journal of Sociology
Volume: 
102
Issue: 
1
Start Page: 
34
End Page: 
79
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

Institutions, Technical Change, and Diverging Life Chances: Earnings Mobility in the United States and ~erman~'

Thomas A. DiPrete Duke University

Patricia A. McManus

Indiana University

Some scholars argue that growing wage inequality stems primarily from technical rather than institutional factors. However, this con- clusion assumes that institutional differences operate chiefly at the level of individual industries. This article argues in contrast that important institutional effects are countrywide and demonstrates the effect of country-level institutional differences by comparing recent earnings dynamics in the United States and Germany. The recent trend in real earnings has been steeper in Germany, while the variance in earnings mobility has been greater in the United States. This is partly due to higher rates of U.S. job mobility, but cross-national differences in earnings trajectories are evident even for workers who did not change jobs.

INTRODUCTION

It is now well known that earnings trajectories in the United States have been diverging during the past 15 or 20 years after a 30-year period of stability following the Second World War (e.g., Karoly 1993). This divergence in cross-sectional, snapshot views of the American workforce is also associated with substantial levels of turbulence in the career trajec- tories of individual American workers, as documented in the numerous studies of restructuring and downsizing at the firm level (e.g., Harrison

' Earlier versions of this article were presented at the 1993 summer meeting of the research committee on stratification and mobility of the International Sociological Association, at the University of Mannheim in December 1993, at the World Congress of Sociology in Bielefeld, July 21, 1994, and at the American Sociological Association annual convention in Los Angeles, August 5, 1994. The research has been supported partly by a National Science Foundation grant (SES 92-09159), partly by a grant from Duke University's Arts and Science Council, and partly by a grant from the Trent Foundation. This research was made possible by a contract with the Deutsches Institut fur Wirtschaftsforschung to use the English-language version of the German Socio- Economic Panel. We would like to acknowledge the helpful comments of three AJS reviewers. Direct comments to Thomas A. DiPrete, Department of Sociology, Duke University, Box 90088, Durham, North Carolina 27708-0088.

O 1996 by The University of Chicago. All rights reserved.

0002-9602/97/10201-0002$01.50

34 AJS Volume 102 Number 1 (July 1996): 34-79

and Bluestone 1988) and of worker displacement at the individual level

(e.g., Hammermesh 1989; Seitchik and Zornitsky 1989; Jacobsen, La-

Londe, and Sullivan 1993) and in trend studies of the covariance struc-

ture of earnings (Gottschalk and Moffitt 1994). While there is hardly

universal agreement on the causes of this trend reversal, many-

especially American-scholars (Bound and Johnson 1992; Murphy and

Welch 1993) see this trend as a response to fundamental technological

and market forces.

Some of these forces may be primarily American in character. Blue- stone and Harrison's (1982) theory of deindustrialization argues that global competition caused the American manufacturing base to shrink and manufacturing wages to fall. While the globalization of markets presumably has an impact on all countries, it might be argued that Amer- ican manufacturing industries were particularly vulnerable to this trend for three reasons: (1) the American domestic market was for a long time relatively free of foreign competition, (2) this domestic market was so huge that most American firms did not feel compelled to invade foreign markets, and (3) success in the huge American domestic market went to firms that used Fordist production principles (Piore and Sabel 1984). In contrast, according to Piore and Sabel, foreign manufacturing firms used more flexible production systems that were better suited for competition in their smaller markets. Ironically, the technological and market changes that are creating the global marketplace give an advantage to flexibly organized firms. Consequently, American firms were at a particular and to some extent unique disadvantage and thus have had to make particular and to some extent unique adjustments (see also Appelbaum and Schett- kat 1990b).

A related, and to some extent uniquely American, argument attributes the turbulence in American labor markets to the decline in the size and market power of labor unions (Freeman and Medoff 1984; Cornfield 1991). According to this argument, weakened labor unions could not withstand company demands for concession bargaining and were not capable of offsetting membership losses due to decertification and to firm or industry shrinkage via successful union organizing in nonunion firms. This argument is related to the deindustrialization argument, of course; the decline of the labor movement was certainly accelerated by the de- cline in manufacturing industries.

While these two arguments are in many respects compelling, any argu- ment that the American situation is somehow exceptional is undermined by two deficiencies. First, while certain aspects of the American situation may be unique, it does not follow that overall competitive pressures were substantially more intense on American firms than they were on foreign firms, which faced their own problems and have also been forced to restructure their operations in order to stay competitive (Bosch 1990; Streeck 1993). For example, the sharp rise in unemployment rates in Europe since the early 1970s suggests a high rate of structural change there (Phelps 1970; Schettkat 1992), and the relatively high rate of pro- ductivity growth in Europe points to a similar conclusion (Wohlers and Weinhert 1988; Baumol, Blackman, and Wolff 1989). Second, the dein- dustrialization thesis and the related deunionization thesis do not appear to be the main explanations for trends toward greater inequality in Amer- ican labor markets. Bound and Johnson (1992) used CPS data to test four competing explanations for changes in the structure of U.S. wages in the 1980s: (1) declines in the relative size of the manufacturing sector,

(2)
declines in the wage premiums paid to male blue-collar workers, (3) lags in supply responses related to the small size of the baby bust cohorts,
(4)
changes in the technical efficiency of high skill versus lower skill workers. They concluded that the fourth of these explanations was domi- nant: in their view, technical change was increasing the value of skilled labor relative to less skilled labor. They reached this conclusion primarily because the wage changes favoring highly educated workers were so pervasive, occurring in all industries.

Technical change has, of course, been occurring at a rapid rate throughout the 20th century, so the reason why it would be the cause of the pronounced trends of the past two decades is not immediately self-evident. Bound and Johnson attribute the change to the computer revolution (Mincer 199 1; Krueger 199 1). This particular explanation has not been completely accepted since inequality has been growing in indus- tries that make minimal use of computers (Danziger and Gottschalk 1993). However, other research (e.g., Murphy and Welch 1993) has gen- erally supported Bound and Johnson's findings that the trend toward inequality is pervasive and has led many scholars to favor a technological explanation for the trend.' Because the aggregate trend in earnings mobil- ity is clearly linked with career turbulence (i.e., individual-level changes in earnings, hours of work, job, occupation, and employer) technical change becomes an obvious explanation for this latter phenomenon as well.

The rejection of institutional explanations as primary causes for recent trends stems from the putative inability of institution-based explanations

The approach of Bound and Johnson (1992) was to estimate wage regressions for 32 age-gender-education groups with CPS data for three different times covering the 1973-88 period and then to search for explanations for the changing wage structure revealed in their equations. Murphy and Welch (1993) similarly analyzed time-series data drawn from the CPS for experience-gender-education groups.

to account for the pervasiveness of change. However, we see two signifi- cant problems with too quick a rejection of institutional in favor of tech- nical explanations. The first of these problems concerns the way that institutions have been treated by this recent literature on inequality trends. Simply put, in this literature, the common identification of institu- tional features with unions is excessively narrow. The extensive research on different forms of internal labor markets makes quite clear that labor markets have an institutional structure that goes beyond and exists in the absence of labor unions or collective bargaining agreements. Further- more, explanations that equate institutions with labor unions, by their nature, can only account for that portion of change that occurs between sectors (e.g., the union vs, the nonunion sector), even though there is significant variation both in institutional arrangements (e.g., in the struc- ture of labor markets) and in earnings trends within these sectors. Fi- nally, and most pertinent for the present article, by focusing on one aspect of institutional structure that exhibits internal variability within a country, this approach has avoided the many institutional features that tend to characterize a nation's employment system (e.g., legal structure, the existence of employer associations, systems of education and training, and personnel practices, management and worker patterns of behavior that might generally be described as "cultural") and therefore are not adequately captured by models such as the unionlnonunion model that emphasize within-nation variation.

The second major problem with the technical-change thesis is that the fundamental explanatory force put forward by these writers is unob- served. It is assumed that an unmeasured force producing pervasive change is technological in nature, perhaps because this seems to the au- thors of these studies to be the only force sufficiently universal to account for the trend. An assumption, however, cannot prove a thesis: the valid- ity of this inference obviously depends upon the underlying assumption that pervasive trends within nations must be technologically driven.

Technological arguments do not deny the influence of institutional fac- tors on labor market outcomes. The United States, after all, is character- ized by extremely heterogeneous labor markets offering various levels of firm-based and occupation-based protection, and this heterogeneity could play an important role in the observed trends. However, the implication of the technical-change thesis is that the institutional structures character- izing American labor markets are either passing the impact of technologi- cal and market forces directly to workers or that the institutional struc- tures themselves are changing in ways that cause American workers to be directly exposed to the broader technological and market forces.

The evidence for institutional change is readily apparent. Many seem

ingly stable internal labor markets continue to be dismantled as one large company after another submits to reengineering.3 Even workers whose jobs continue to enjoy a level of protection are not truly shielded from broader trends because of the decoupling of internal labor markets in the United States. While relationships among jobs within an internal labor market may be determined through political and administrative mecha- nisms, the benchmark wage typically is strongly affected by external market forces. For the quarter century following the Second World War, collective bargaining agreements enjoyed a certain degree of linkage. Certain unions in effect would establish a pattern of wage increases that would then be followed to a greater or lesser extent across the industrial spectrum (Piore and Sabel 1984; Kochan, Katz, and McKersie 1986). The pattern-setting agreements were themselves linked to productivity trends. While white-collar labor markets were not linked through the same collective bargaining mechanism, cultural norms concerning legiti- mate pay differences (Alves and Rossi 1978) and the implications of these norms for fairness judgments (Dornstein 1988), worker morale, and union efforts to control pay gaps between white-collar and blue-collar work (Freeman and Medoff 1984) created an informal linkage. To be sure, the pay relationships across white-collar occupations were not fixed, but they changed slowly and against an institutional drag.

More recently, however, while relationships among jobs within mar- kets remain a strong force in structuring earnings, the institutional links that connect internal labor markets (ILMs) have weakened. Market forces have always had an important influence on wage relationships across ILMs. The fragmentation of markets (i.e., increased variance in market demand for occupations of a given "social" distance) has decon- structed many ILMs and seriously weakened nonmarket institutional links among surviving ILMs. As a consequence, the variance of earnings across jobs is increasing, and career variance in earnings is increasing through a combination of voluntary job mobility, involuntary job mobil- ity, and the evolution of earnings within jobs (Gottschalk and Moffitt 1994)~~

Seen from this perspective, institutions have played an important role

"Firms that provided secure jobs in the past can, and often do, alter course" (Oster- man 1993).

Whether the breakdown of ILMs increases or decreases earnings variance depends, of course, on whether these ILMs were creating or suppressing variance before their breakdown. Freeman and Medoff (1984) concluded that union-based ILMs suppressed variance because the reduced inequality between blue-collar and white-collar workers more than offset the increased inequality between blue-collar union and nonunion workers. In either case, this breakdown increases individual-level turbulence, at least in the short term.

in the trend toward growing inequality, but the technical-change thesis assigns them the role of mediating technological and market forces and thus having no independent effect on trends. The real test of such a thesis must be comparative. Technological explanations for recent trends are by their nature universal explanations that in principle apply to all advanced industrialized countries. An adequate test must involve cross- national comparisons. Such comparisons are important for another rea- son; they provide an additional, cross-national source of institutional variation. The evidence for the technical-change thesis primarily rests upon the conclusion that a common attribute of American labor markets is at work in generating these trends.' To distinguish between common technical and common institutional forces requires cross-national com- parison.

As already noted above, the absence of legal or other institutional linkages is a significant common, institutional feature of American labor markets. This feature contrasts strongly with institutional arrangements in many other industrialized societies. In Europe, linkages among labor markets frequently occur through state coordination or through negotia- tion between centralized employer and labor associations. These institu- tional arrangements-often described as corporatist-can vary substan- tially in form, with the United States at one extreme and the Sweden of the 1960s or 1970s perhaps at the other (Lange, Wallerstein, and Golden 1995). Relative to the United States, however, all countries that have been described as corporatist are distinguished by their relatively central- ized wage-setting institutions.

While centralized wage-setting institutions might be expected to reduce wage dispersion, these institutional arrangements do not sever the link between market forces and career outcomes. Indeed, some scholars sug- gest that they only postpone the market adjustment that must inevitably be made (Lindbeck et al. 1994). In this view, trade-offs are necessitated by economic forces. Swenson (1989), for example, argued That employ- ment, average wage levels, wage equality, and price stability are often inconsistent goals and that unions and business in countries with strong labor unions make explicit choices about these trade-offs. When trade- offs are postponed or only incompletely made, market realities may force more drastic adjustments on business, labor, and the state. These adjust- ments can occur at the level of individual outcomes, or they can so thoroughly change the institutions that a situation of "corporatist break- down" occurs (Lange et al. 1995; see also Streeck 1993).

Similarly, Bound and Johnson (1992) found that the industry-specific (i.e., industry- varying) component of technical change was relatively unimportant in comparison to the common industry component.

The evidence for institutional change demonstrates clearly that institu- tions do not exist in a vacuum; they are subject to the influence of political and economic forces. At the same time, they also have the potential to mediate economic and technological trends and to affect both the range of permissible trade-offs and the outcome of these trade-offs. This article uses Germany as a test case to introduce institutional variation of a cross-country rather than within-country form to examine the adequacy of a technical-change argument for trends in earnings. Germany is often described as quasi-corporatist even though its wage-setting institutions differ from those of the "classically" corporatist Scandinavian countries; instead, it has "functional equivalents" to corporatist institutions that have generally produced similar collective bargaining outcomes (Thelen 1991; Cameron 1984; Lange et al. 1995). German labor market institu- tions differ from those of the United States in two key respects: the process of wage determination and the structure of job mobility.

The Process of Wage Determination

Germany has a strong institutional system of wage determination, even though only 40% of German workers are members of (both white- and blue-collar) labor unions. The IG Metal1 union plays a leadership role in setting a benchmark for other collective bargaining agreements, which are generally concluded at the regional-industrial level (Soskice and Schettkat 1993). Employers who are members of the main employer asso- ciation (Bundesvereinigung der Deutschen Arbeitgeberverbande [BDA]) apply these agreements to their nonunion as well as union members. In addition, the federal or state German labor minister can extend the agreements to cover firms that are not members of the BDA. These collective bargaining agreements specify minimum rates of pay: individ- ual works councils may participate in informal negotiations over the level of the "wage drift" that separates the effective wage from the minimum that is guaranteed in collective bargaining agreements (Swenson 1989; Thelen 1991; Abraham and Houseman 1993).

The institutional actors that make up quasi-corporatist societies such as Germany do not, of course, work toward identical outcomes; business, labor, and state interests can instead be divided with respect to goals (Swenson 1989). However, there are three clear reasons to expect that the net effect of institutionalist forces will be in the direction of greater equality. First, despite internal conflict, wage leveling is often an impor- tant goal of labor unions, whether in the United States (Freeman and Medoff 1984) or in Europe (Swenson 1989). Second, equality of outcomes is generally furthered by pattern bargaining, which is an important part of the German process of wage determination. Third, it has at least been argued (Bosch 1990; Mahnkopf 1992) that properly motivated strong unions find ways to channel economic forces to mitigate their wage im- pact on manual labor, for example, by supporting training and retraining efforts that increase the level of useful skills among workers.

The Structure of Job Mobility

Germany has well-developed occupational internal labor markets with tight linkages between educational or training credentials and occupation (Haller 1987; Blossfeld and Meyer 1988; Marsden 1990). This system contrasts strongly with the United States, where relatively pure occupa- tional labor markets are limited to construction crafts and certain profes- sions. Largely because of these tight linkages between training and job, Germany has low rates of occupational mobility and job mobility (Carroll and Mayer 1986; Konig and Miiller 1986; Kappelhoff and Teckenberg 1987; Blossfeld, Giannelli, and Mayer 1993). Germany's low rates of job mobility also derive from the strong legal mechanisms in place to reduce involuntary job mobility that derives from economic conditions (as op- posed to individual performance). A series of legislative acts make it difficult for firms to dismiss German workers and in fact provide employ- ers with incentives to avoid dismissals (Bosch and Sengenberger 1989; Abraham and Houseman 1993; Buechtemann 1993). German law does not allow dismissals for economic reasons unless the company has ex- hausted other means of avoiding layoffs and has consulted with the plant- level works c~uncil.~

When companies contemplate mass dismissals, Ger- man law requires companies to work out a social plan with works councils that includes prior notification, compensation, and retraining. Finally, German law allows employees to collect unemployment compen- sation when their hours have been reduced as part of a plan to prevent layoffs.

It is fairly obvious that differences in the structure of job mobility can contribute to cross-national differences in the structure of earnings mobility. In the United States, a major force behind downward earnings mobility is worker displacement. Roughly half of the workers displaced from their jobs fail to find a new job that pays as well as their old one (Herz 1991), though it is still something of an open question about how much of these losses are permanent (e.g., Ruhm 1987; Hammermesh 1989). Constraints on involuntary job mobility in Germany should reduce the impact of job mobility on earnings.

Works councils represent about 70% of German workers (Miiller-Jentsch 1995).

In the two respects of wage determination and job mobility, Ger- many is quite different from the United States. At the same time, Germany's economy appears to be exposed to similar technological and global economic forces (Dertouzos et al. 1989; Appelbaum and Schettkat 1990a; Schettkat 1992)~~

Germany's economy is heavily reli- ant on manufacturing and is more export driven than that of the United States. While its location within the European Common Market may provide a certain level of insulation, its high-wage labor force creates even stronger pressures for technological upgrading than those found in America, where many employers have the option of keeping costs down by hiring low-wage workers (Wegner 1985; Baumol et al. 1989; Schettkat and Wagner 1990; Thurow 1992). If the technical change that engenders "a rising demand for skill" is the main determi- nant of the trends toward greater inequality and career turbulence in the United States, one would expect to see the same impact in Germany. If, however, earnings trends are influenced by an "institu- tionalist" logic as well as by technical forces, cross-national similarity in earnings trajectories does not necessarily follow from the similar technological position of the two countries. This hypothesis postulates as a key variable a common institutional factor operating within each country as a key determinant of individual earnings trajectories. Just as Bound and Johnson (1992) argued that the common component of technical change in the United States dominates the industry-specific components, so the common institutional component may dominate the variable institutional component that fails to account for most of the rising earnings inequality in within-country econometric analyses.8

This article examines differences in earnings trajectories for Germany and the United States with data for the relatively recent past, the period during which the trends discussed earlier are most evident in the United States. First, it addresses the question of whether there is similarity in the patterns of earnings change in Germany and the United States. Second, it addresses the question of whether class differences in earnings mobility are more pronounced in the United States than in Germany. This second question is of interest because institutional forces, like technological

'In fact, it has been argued that the availability of low-wage labor in the United States serves to slow down the rate of technological progress in those industries for which low-wage workers are available relative to countries (such as Germany) where the option to hire low-wage labor is absent (Wegner 1985).

Some studies in fact find substantial institutional effects on earnings inequality within the context of the United States, but even Freeman's study (1993) finds that declining unionization accounts for only 20% of increased earnings inequality.

forces, can be biased toward or against class inequality even as they have strong within-class effects. Third, it addresses the mechanisms that create intercountry differences. Two mechanisms are fundamental: (1)the evo- lution of earnings in a given job and (2) the earnings change that arises from job mobility. Earnings change in the same job occurs through a variety of market and institutional processes: through collective bar- gaining in the case of union jobs, through external market forces where they apply, through internal market forces that work to keep stable rela- tionships between positions located in the same market. A portion of earnings change in the same job can occur through changes in hours worked, which is sometimes though not always reversible, while another portion can occur through changes in the rate of pay, which tend to be more durable. Earnings change, of course, also results from job mobility; indeed, it is this component of change (and particularly that component connected with occupational mobility) that has received primary atten- tion in most of the sociological literature (see Hannan, Schijmann, and Blossfeld [I9901 for a notable exception), even though it is generally not the dominant source of earnings rn~bility.~

In summary, two important issues in stratification can be addressed by a comparison of German and American earnings mobility. The first concerns the ability of institutions to withstand apparently strong eco- nomic tendencies toward divergence in life chances on the basis of skill that arise from rapid technological change coupled with a globalization of markets. Radically different outcomes in the two countries suggest either that institutional mechanisms can counterbalance economic forces -at least in the short term-or that they can channel these forces in ways that mitigate their impact on earnings inequality. The second con- cerns the importance of job mobility as a source of earnings mobility and of between-class earnings inequality, and of the importance of this feature of labor market structure in generating cross-national differences in the structure of earnings change.

DATA SOURCES AND METHODOLOGICAL ISSUES

Analyses in this article involve tabular and multivariate analyses of earn- ings change designed to answer the questions posed above. In this section we describe the data sources and address measurement issues concerning

Job change is certainly an important source of earnings change, even if it does not necessarily dominate. Topel and Ward (1992), using data for the 1950s through the early 1970s, found that job change accounted for one-third of earnings growth during the first 10 years of the career (see also Topel 1986).

three critical variables: position change, earnings change, and occupation.

Data Sources

German data are from the English-language version of the German Socio- Economic Panel (GSOEP) for the six-year period from 1985 to 1991.1° Given the recency of the reunification, we restrict our attention to the sample drawn from the area that was West Germany (which we refer to as Western Germany), though we take account of reunification in the interpretation of the Western German results." American data for 1985-91 are taken from the Panel Study of Income Dynamics (PSID). We divided this six-year period into two three-year segments and analyzed three-year earnings mobility from 1985 to 1988 and from 1988 to 1991. Results in this article are computed for men only between the ages of 18 and 61 who were employed in 1985 and were present in the survey from

1985 to 1988, or men who were 18-61 years of age, employed in 1988, and present in the survey from 1988 to 1991. The males in the German sample, when weighted, constitute a representative sample of the West- ern German states and include the foreign population resident in Western Germany. The males in the PSID, when weighted, are representative of American male household heads (exclusive of immigration since 1968). While the two samples are not perfectly comparable, they offer the most accurate possible comparison of earnings mobility, and, as we shall see, the comparison provides the basis for drawing rather solid conclusions about cross-national similarities and differences.'' Because much of our interest is in population comparisons, we use weighted PSID samples for our cross-national comparisons of earnings mobility. l3 In our multivariate analyses designed to interpret cross-national differences, we use the full (unweighted) PSID sample.

lo The first wave of the GSOEP was collected in 1984. This article's use of recent panel data and its inclusion of foreign workers who reside in Germany (Auslhder) distinguish the sample used in this article from the

samples used in other analyses of earnings mobility in Germany (e.g., Carroll and Mayer 1986; Hannan et al. 1990). l2 Male household heads probably have more stable earnings than do all males in the

same age range. Therefore, if the results had shown that the American sample had more stable earnings trajectories than the German sample, it would have been difficult to tell whether this difference was real or due to the (unavoidable) American sample restriction. However, if the results had shown the American sample to have less stable earnings trajectories than the German sample, generalizing to the respective populations would have been unproblematic, though our estimate of the country difference probably understates the true difference.

l3 Heads who are not descendants of original 1968 PSID sample families are given "0" weights in the PSID.

Measures of Position Change

We measure position change in the two countries with survey questions that ask respondents if their position or employer has changed since the previous year survey. If a respondent had made at least one employer change during the three-year period, we coded him as having changed employers. If a respondent had changed positions within the same firm without changing employers, we coded him as having changed positions internally. l4

Measures of Earnings and Earnings Change

Perfectly comparable measures of earnings-and hence of earnings mo- bility-for Germany and the United States are difficult to obtain because of the different ways that earnings are measured in the PSID and the GSOEP. The PSID contains measures of earnings and other income and of hours worked for the calendar year that precedes the interview. The PSID interviewers also ask the respondent about his current hourly wage. However, no information about current hours worked or about current weekly or monthly pay is solicited. GSOEP interviewers ask the respon- dent about earnings in the previous calendar year and also about earnings in the previous month. They also obtain information about usual and actual hours worked. However, they do not directly solicit information about hourly wages.

Not surprisingly, alternative measures of earnings give somewhat dif- ferent results for a respondent. Furthermore, each measure doubtless includes measurement error. In a recent validation study (Rodgers, Brown, and Duncan 1993), the PSID instrument was administered to respondents of a large manufacturing firm and the reports on earnings and hours were checked against company records. The results demon- strated that measures of annual earnings, of earnings in the previous pay period, and of hours worked all contain measurement error. In particu- lar, efforts to reconstruct hourly earnings from reports of last pay period or "usual" hours and earnings led to very unreliable result^.'^

The presence of measurement error is a matter of particular concern when studying earnings mobility because of the possibility that indepen-

l4 It is thus possible that someone coded as an employer changer also changed positions internally. In 1988, the PSID changed the questions it used to probe for position changes. The more detailed questions appear to have led to an increase in reported internal job changes.

l5 Our checks of the GSOEP led us to similar conclusions. Regressions of gross monthly earnings on measures of skill, demographic variables, occupation, and indus- try produced notably weak effects of hours worked for blue-collar workers.

dent errors in earnings reports will create large amounts of error in mea- sures of change computed from two points in time.16 Measurement error in earnings creates three problems for an analysis such as this one. First, it overstates the level of earnings mobility within a single country. Sec- ond, it could bias comparisons between countries. Differences in the level of measurement error across countries could overwhelm differences in the underlying true rates when naive comparisons of measured earnings mobility for the two countries are used. Third, measurement error com- plicates the analysis of individual-level earnings mobility over time. That portion of the measurement error that was independent over time would generally create a sawtooth pattern in earnings trajectories: the portion of measured earnings growth and decline between times t, and t, that was an artifact of measurement error would disappear when earnings change between t, and t, was examined. Consequently, individuals would appear to rebound from earnings growth and decline even if no rebound actually occurred.

In order to reduce the impact of measurement error on our results, we adopted the following strategy: (1)to estimate a measurement model for reported earnings, (2) to use this measurement model to impute the latent true earnings, and (3) to analyze change in terms of the imputed measure. In light of the results of Rodgers et al. (1993) and our own experience with the GSOEP, we chose to avoid creating a measure of hourly earnings and instead make comparisons of monthly earnings in the two countries. Since our earnings indicators refer to monthly earnings, our latent earn- ings measures will also apply to monthly earnings. In our multivariate models, we separate the effect of hours worked on earnings mobility from the effects of other variables by controlling for hours worked in the previous time period.

We begin with a fairly standard measurement model that draws on covariance structure modeling, a standard approach in sociology (Bollen 1989). A latent, ''true" earnings level is assumed to combine with sto- chastic disturbances to generate two measures of reported earnings at each time point in both the German and the American surveys. It is assumed that the covariance in measures of reported earnings is due to their common dependence on true earnings." While distinct measures of

l6 Rodgers et al. (1993) report rather weak (0.09-0.23) correlations for the errors in

reported earnings at two times spaced four years apart. l7 This model also contains the common (though probably incorrect) assumption that true scores and measurement error are uncorrelated. Rodgers et al. (1993) found that the correlation between true earnings in the previous pay period and measurement error is around -0.3. The relaxation of the assumption of independence creates identi- fication problems and is in any case difficult to implement with available statistical software. We leave it as a future research problem.

previous monthly earnings are not available in a single survey, it is possi-

ble to construct two such measures by using adjacent waves of the

GSOEP and PSID. We used the following two measures of monthly

earnings for the GSOEP sample.

The retrospective report from the following year survey of gross yearly earnings was divided by 12 to convert it to a monthly scale (GEARNl). Respondents are asked to report earnings including sick pay and compen- sation payments (but excluding bonuses such as holiday money), pay- ments for training, self-employed earnings, and earnings from second jobs. To these earnings we added respondent reports of bonus earnings, including "thirteenth month" payments. The second measure of monthly earnings was the respondent's answer to the question, "How high were your [gross] earnings last month?" It includes overtime but not holiday or back pay (GEARN2).

For the PSID sample, we used the following two measures. The total labor earnings in a calendar year as reported in the following PSID survey were divided by 12 (PEARNl). The projected main job earnings for a calendar year were computed from the wage per hour on the main job as reported in that year, the number of weeks worked on main jobs that year as reported in the following year's survey, and the average hours per week on main jobs in the calendar year as reported in the following year's survey (PEARN2).

We constructed the measures for each country to make them as compa- rable as possible. In each case the first measure includes earnings varia- tion that arises from variation in weeks worked throughout the year in the primary and any secondary jobs, while the second measure focuses on earnings based either on reports of current monthly earnings (in Ger- many) or the current wage per hour in the primary job (in the United States). We corrected the earnings data in both countries for inflation. We used CPI-U (the Consumer Price Index U.S. city average for all urban consumers) for the United States, and Preise und Preisindizes fur die Lebenshaltung for Germany. In both the German and the American measurement model, we specified the metric of the latent earnings vari- able to be the same as the first of the two earnings measures (GEARN1 or PEARN1).

For some of our substantive analyses, we focus on changes in the log of mean earnings. However, because of our interest in the divergence of earnings over time, we focus much of our attention on explicit measures of earnings mobility, defined as the proportion of respondents whose (real, inflation-adjusted) monthly earnings rise or fall by at least 5%, IS%, or 25% over a three-year period. Workers who exit employment between the starting and ending time are assumed to have an earnings decline of more than 25%. While these benchmarks are arbitrary, we believe they do a good job of characterizing levels of change that have a direct substantive interpretation.

Distributional Assumptions

As is well known (e.g., Bartholomew 1987), the distribution of latent variables in a measurement model is, in principle, arbitrary and, in prac- tice, typically assumed to be multivariate normal. While such an assump- tion is not necessarily problematic, it does create problems of interpreta- tion in the case of earnings. The observed distribution of earnings is manifestly not normal. A latent variable whose distribution is normal obviously may not closely correspond to the nonnormal earnings variable that would be obtained if measurement error could somehow be elimi- nated.

In order to allow for a more straightforward approach to the problem of interpretation, we transformed our data. Our transformation is some- what novel but appears to work rather well in practice. The use of log transformations to reduce skew in earnings is well known. However, this transformation still leaves earnings data far from normal. When earnings differences (measured in the log scale) are computed, the distribution of these change scores is quite spiked. If we were to assume that this distri- bution was normal, our predictions of the proportion of a population whose earnings increased or decreased by more than 15% or 25% would be much too high. To counter this problem, we modified the usual log transformation, which takes the form

where x, is the measure of earnings at time t, and f, is the transformed earnings. Instead, we used the following strategy to transform earnings. First, since our analysis focuses on earnings change with respect to bench- marks in a -25%to 25%range, we placed a floor and ceiling on reported earnings change for respondents who were working at both start and end points so that extreme values would not have an unduly large effect on our results: our floor was a 63% earnings drop, while our ceiling was a 272% earnings rise.'' We then used the following more flexible form to transform earnings

b

f, = log x,,

where b is chosen so that if we assume that f,has a normal distribution, the predicted proportions of people whose 5, changes by more than 5%,

l8 These points correspond to a (-1,l)range for differences in the natural logarithm of earnings.

15%, or 25% will match the empirical data as closely as possible. To accomplish this, we used nonlinear least squares estimation to estimate

b. We let pi be the (weighted) proportion of respondents who were work- ing at a starting and an ending time in which

where ci = log(1.25), log(1. 15), log(1.05), log(0.95), log(0.85), and log(0.75) for i = 1, . . . , 6. In simpler terms, pi is the proportion of respondents working at two different times whose earnings increased by less than 25 %, by less than 15 %, by less than 5 %, and whose earnings decreased by more than 596, by more than 15 %, and by more than 25 %. We let this empirical distribution (pi) be the dependent variable in the following equation:

where F is the cumulative distribution function for the normal distribu- tion and where 2, is defined as in equation (1). Nonlinear least squares provides an estimate of b that causes the right side to most closely match (by the least squares criterion) the left side. In effect, this procedure finds the normally distributed variable that most closely reproduces the empirical distribution for the six points listed in ci.19

The results of this strategy can be seen in table 1for the United States and in table 2 for Germany. The first column shows the observed fre- quencies in each of the seven rows for respondents in the PSID (table 1) and GSOEP (table 2) who were employed in both 1985 and 1988 and in both 1988 and 1991. The measure reported in the first column is PEARNl (table 1) or GEARNl (table 2); appendix table A1 shows the distributions of PEARNl, PEARN2, GEARN1, and GEARN2, as well as other supplementary earnings measures for the United States. In tables 1and 2, the second column is the naive prediction for a random variable with a normal distribution and the same mean and variance as log[(xt+,)l (x,)], where t equals either 1985 or 1988. Clearly, this naive prediction has a strong positive bias for predictions of earnings change greater than 15%. In contrast, the prediction based on the transformed 5 is substan- tially more accurate than the naive prediction. We used these trans-

l9 If one transformed the original distribution for the log difference in such a way that the mean and standard deviation were different (suppose the relationship between the "new" means and standard deviations and the "old" ones was m, = d,m,, and s, = d,n,), but the empirical distribution at the six points was the same, the nonlinear least squares solution would give the same normalized distribution, with 6" = Pld,.

TABLE 1 PROBABILITYOF A CHANGEIN MONTHLYEARNINGSIN A THREE-YEARPERIOD.FOR U.S. MALE

NONFARMWORKERSAGES18-61 IN START YEAR

Observed
Probability Groups PEARNl
1985-88 (conditional on employment in
1988):    

> 25% rise ...................................... 25 > 15% rise ...................................... 36 > 5% rise ........................................ 54 -5% to 5% ..................................... 19 > -5% fall ..................................... 27 > -15% fall ................................... 16 > -25% fall ................................... 11

Rvmeasurement model) ....................... 1988-91 (conditional on employment in

1991): > 25% rise ...................................... 18 > 15% rise ...................................... 27 > 5% rise ........................................ 41 -5% to 5% ..................................... 21 > -5% fall ..................................... 39 > -15% fall ................................... 20 > -25% fall ................................... 12

Rvmeasurement model) .......................

Naive Prediction Based on PEARNl

Transformed Prediction Based on PEARNl

.25 .37 .52 .16 .32 .18 .07

Corrected Prediction from 5 Distribution

.19 .34 .55 .21 .24 .092 .021

Prediction from Average Imputations (Four-Measures Sample)

.17

.32

.54

.26

.20

.092

.043

Prediction from Average Imputations (Full Sample) Unconditional Corrected Prediction from 5 Distribution

.17 .31 .49 .19 .32 .18 .12

.86-. 91

NOTE.-Thii information is from the PSID (Consumer Price Index adjusted) .

PROBABILITY IN A THREE-YEAR PERIOD. FOR GERMANMALE

OF A CHANGEIN MONTHLY EARNINGS NONFARM WORKERS AGES18-61 IN START YEAR

Prediction Unconditional Naive Transformed Corrected from Average Prediction Corrected Prediction Prediction Prediction Imputations from Average Prediction Observed Based on Based on from (Four-Measures Imputations Probability Groups GEARNl GEARNl GEARNl Distribution Sample) (Full Sample) Distribution

1985-88 (conditional on employment in

1988):
> 25% rise .....................................
> 15% rise .....................................
> 5% rise .......................................

(n -5% to 5% ....................................

I-. > -5% fall ....................................
> .15% fall ..................................
> -25% fall ..................................
RZ(measurement model) .......................
1988-91 (conditional on employment in
1991):
> 25% rise .....................................
> 15% rise .....................................
> 5% rise .......................................
-5% to 5% ....................................
> -5% fall ....................................
> .15% fall ..................................
> -25% fall ..................................
R2 (measurement model) .......................

NOTE.-This information is from the GSOEP (Consumer Price Index adjusted)

.$ from

FIG. 1.-Measurement model for monthly earnings in years t and t + 3.

formed earnings measures as indicators in the measurement model shown in figure 1.

The fourth column of tables 1 and 2 gives the corrected prediction of earnings mobility, which was computed from the joint distribution of the latent earnings variables, and includes the range for R2 from the measurement model. Because our model consists of two latent factors, each of which predicts two observed measures, there are four equations for each range of years, and the R2measures the variance in the observed measure explained by the latent factor. The R2 is essentially a measure of reliability: in this context, our measures would be considered conge- neric (see, e.g., Bollen 1989, pp. 206-9). Our reliability estimates are higher than the 0.65 reported by Rodgers et al. (1993) for the manufactur- ing company they studied and are in the range of the 0.82 that Bound and Krueger (199 1) reported in their comparison of annual earnings mea- sures from matched CPS and Social Security Administration data. The Rodgers et al. result is probably too low as an estimate for national data, since their sample from one firm has restricted variance in true earnings relative to a national sample. In one respect, our results should overesti- mate reliability because measurement errors for two indicators of earn- ings in a given year (even when taken from different panel waves, as ours were) will generally be positively correlated (our measurement model must necessarily specify this correlation to be zero). On the other hand, our measures (which involve yearly averages, monthly measures, and measures based on wages) would not be identical even if there were no measurement error, and this fact would tend to reduce R2.

In all cases, our latent variable has a higher R2with earnings measures based on current conditions (monthly earnings in Germany, current earn- ings per hour in the United States) than with measures based on annual earnings. The R2 range for PEARN2 is 0.88-0.90, while the range for GEARNZ is 0.89-0.91 . The R2 tends to be lower for PEARNl (0.83-0.86) and for GEARNl (0.72-0.86). These results suggest that our latent variables are doing a rather good job of measuring current earnings (PEARNZ and GEARNZ), which is reassuring because our analysis of the impact of job change on earnings change uses job change measures that also apply to a worker's current employment situation. Our results also suggest that estimates of earnings mobility that were computed from noisy raw data (instead of latent factors) would be about 5-10 percentage points too high at the 15% and 25% benchmarks. As it turns out, we will show that the difference between the German and the American patterns of earnings mobility is considerably larger than the level of noise in the data.

Our transformation allows us to use the latent variables to compare levels of earnings mobility between the United States and Germany. Our descriptive results are computed from the estimated unconditional distributions of these latent variables. In order to perform multivariate analyses of earnings change, we used our measurement model to impute the latent earnings measures to the sample members. With appropriate corrections for standard errors (Rubin 1987), these imputed variables can be used in ordinary regressions. This strategy has two advantages over direct estimation of model parameters with covariance structure method- ology: (1) it makes the use of categorical covariates more straightforward, and (2) it allows us to estimate the effects of interactions, which are difficult to estimate with standard covariance structure models (Bollen 1989).

Multiple Imputation of Earnings and Mobility

Employing the usual notation (e.g., Bollen 1989), let

where we assume that

Under these assumptions, it follows that x and 5. have a joint multivariate normal distribution:

where

Since x and 5 have a joint multivariate normal distribution, it also follows that the conditional distribution of 5, given x, has the following form:

In reality, of course, the observed earnings variables do not have a nor- mal distribution. Instead, we use the transformed, normalized li: to esti- mate the distribution of 5. These estimates can then be used in equation

(2) to compute the conditional distribution of 5 for each case for which we have measures of x. From this conditional distribution, we can then make random draws in order to impute latent earnings." These latent earnings have the same scale as 2. When the transformation is reversed, the scale of x itself is recovered. The reverse-transformed 6 can then be used to compute empirical distributions and to perform multivariate analyses. A standard result from the theory of conditional distributions (e.g., Feller 1966) relates the conditional and unconditional variance of a variable as follows:

Because the reliability of our earnings indicators is relatively high and because the variation in measured earnings in our samples is substantial, almost all the variation in 6 comes from variation in the conditional expectations, not from the simulation. Nonetheless, the uncertainty from this simulation should be taken into account in calculations of standard errors and hypothesis tests. To do so, we imputed five values of 6 for each observati~n.~~

We then performed our multivariate analyses five times (once for each of the five imputations) and corrected the standard errors through the method discussed in Rubin (1987).

An additional advantage of using the multiple imputation approach is that it allows us to recover many of the cases that were necessarily miss- ing from our measurement analysis. A relatively large number of respon- dents in the two countries lack one of the four measures of earnings necessary to estimate the measurement model. For example, while some self-employed individuals in the United States report a wage rate, many do not. Other cases in both countries are lost because of the necessity for a sample member to be present in four different waves of the GSOEP or PSID (if the comparison is made between years t and t + 3, data must be available for years t, t + 1, t + 3, and t + 4) in order to have measures on all four indicators.

However, while the measurement model uses four measures of earn- ings, equation (2) does not require four measures of earnings for the computation of the conditional distribution of 6. So long as there is at least one measure of earnings for year t and one measure for year t + 3,

20 This is done by drawing the earnings at the first time (1985 or 1988 in our analysis) from the univariate normal distribution for the latent earnings at the first time. Then a difference between the second and first time is drawn from the univariate normal distribution for the difference between the second and the first latent earnings. This difference is then added to the imputed value for the latent earnings at the first time to get an imputed value for latent earnings at the second time. This method produces "paired" imputations. For 1988, the middle year in our analysis, we get separate imputations depending on whether we are pairing 1988 with 1985 or with 1991.

21 Our imputations were based on separate measurement models for each category of the job change variable, in order to improve their accuracy.

the conditional distribution can be calculated. Thus, while our mea-

surement model results are based on the subset of cases with two mea-

sures of earnings in each year, our multivariate results are based on all

respondents who were working in time t and who were either not work-

ing or were working with at least one valid measure of earnings in year

t + 3.

Columns 5 and 6 of tables 1 and 2 provide comparison between esti- mates of earnings mobility that are based on the distribution of 5 and estimates that are based on the imputations. Specifically, column 5 gives the distribution of the average of our five imputations for each sample member for the four-measure sample (i.e., the sample of workers with two valid earnings measures for each time point), while column 6 gives the distribution for the larger "full" sample that has at least one earnings measure at each time point. The distribution of the average individual- level imputations is generally similar to the distribution of 5 and shows compressed variation relative to the observed sample as expected. The differences in the distributions of 5 and the imputations within countries are small relative to the between-country differences, which is reassuring. In both the United States and Germany, the workers represented in the sample shown in column 6-those for whom at least one earnings mea- sure is available for the starting and ending time-are somewhat more mobile than are the workers in the four-measure sample shown in column

5. It is still true, however, that between-country differences are much larger than within-country differences across the different samples or methods used to estimate earnings mobility.

Measurement of Occupational Groups

Our analysis involves a comparison between manual and nonmanual workers in each country. We chose to operationalize the distinction be- tween manual and nonmanual workers through a version of the EGP (Erikson, Goldthorpe, and Portocarero 1979) occupational classes as adopted by Ganzeboom, Luijkx, and Treiman (1989). Occupation in the PSID is coded with the 1970 U.S. census classification system. We used the conversion system of Ganzeboom and colleagues (Ganzeboom et al. 1989; Ganzeboom, De Graaf, and Treiman 1992) to map 1970 U.S. cen- sus codes into 1968 three-digit International Standard Classification of Occupations (ISCO) codes. We further used the conversion system of Ganzeboom et al. to map 1970 U.S. census codes into EGP occupational categories. While the Ganzeboom conversion could easily be used with the three-digit ISCO codes that are contained in the German-language version of the GSOEP, it could not be implemented with the English- language version of the GSOEP, which, for data security reasons, does not contain the third digit of the ISCO codes." However, the German work-status (Stellung im Beruf) categories map into EGP codes without complications, and so we used them instead of the two-digit ISCO cate- goriesZ3

RESULTS: EARNINGS MOBILITY IN THE UNITED STATES

AND GERMANY

We report estimates of earnings mobility based on the estimated distribu- tion of the latent variable, 5.24Columns 4 and 7 of tables 1 and 2 show the overall comparison between Germany and the United States. Column 4 presents the probability of change in monthly earnings, conditional on working at the survey time in both 1985 and 1988. Column 7 presents unconditional probabilities of earnings change, where those who were not working at the survey time in 1988 are included as having experi- enced more than a 25% earnings decline.

The results in table 1 show quite clearly that there were a higher average earnings gain and a lower variance in Germany than the United States for the period 1985-88, during which the American economy grew at a faster rate than did the German ec~nomy.'~ The combination of relatively high mean and relatively low variance means that the probabil- ity of earnings gain in Germany was significantly higher than in the United States, while the probability of earnings decline was significantly lo~er.'~

During the 1988-91 period, an economic slowdown occurred in the United States. The German economy reached its peak growth rate in

22 The German language version of the data cannot be used outside Germany.
23 These categories, the Stellung im Beruf system, are essentially the same distinctions
made by Konig and Miiller (1986). To improve comparability with the U.S. data, we

combined the categories of semiskilled and unskilled workers into a single category,
as is found in the EGP scheme.
24 Column 1 of tables 1 and 2 reports the frequencies for PEARNl and GEARN1.

For comparison, we include the frequencies of several alternative indicators of earn-

ings mobility in table Al.
25 Real GNP in the United States grew at a 3.2%, 2.9%, 3. I%, and 3.9% rate for the
years 1985-88, while the Western German GNP grew at a rate of 1.8%, 2.2%, 1.590,
and 3.7% (International Monetary Fund 1994).

26 Table A2 provides an alternative method of comparing these distributions in which
the comparison is made against the average earnings change within each sample.
Thus while tables 1 and 2 contain information about mean mobility and about vari-
ance, table A2 provides information only about the variance of earnings mobility.

1990 but did not begin to experience a serious slowdown until 1992.~~ This second period also encompassed the reunification of Germany, which contributed to that country's economic slowdown. Columns 4 and 7 of table 2 show that, during this second period, there was a slowdown in the rate of earnings gain in Germany. However, the German variance was still compressed relative to the United States. Even during this pe- riod of economic slowdown, earnings in Germany rarely declined among those who continued to work, though in this second period earnings gains at the 15% level or higher were not as common as in the United States. However, while American workers had a somewhat higher probability of earnings gains, they paid for this with a sharply higher probability of earnings decline. We note that the earnings declines reported in the table are declines in real dollars, not in nominal (current) dollars. Workers whose nominal earnings do not change over time nonetheless experience a decline in their real earnings due to inflation. In many cases, however, the workers in question suffered declines in nominal earnings as well as in real earnings.

RESULTS: EARNINGS MOBILITY AND POSITION CHANGE

As we noted in the introduction, one source of the greater variance in earnings outcomes in the United States could be its higher rates of job mobility. Table 3 shows the rates of position change in Germany and the United States during these two periods of time. In both periods, Ameri- can workers were much more likely to change positions than were Ger- mans. The rates of within-firm changes were higher, as were the rates of between-firm change^.^' Since we expect that job mobility could be a major source of the higher earnings mobility variance in the United States, it is important to examine the impact of job mobility on earnings mobility.

The first column of table 4 reports the rate of earnings change for males who did not change position at any time between 1985 and 1988, while the first column of table 5 reports similar results for the 1988-91 period.29 The second column of tables 4 and 5 shows odds ratios of

27 Real GNP in the United States grew at a 3.9%, 2.5%, 0.8%, and -.1.2% rate for the years 1988-91, while in Western Germany the real GNP growth rate was 3.7%, 4.0%, 4.9%, and 3.6% (International Monetary Fund 1994).

28 Rates of within-firm change in the 1985-88 period are probably biased downward for American workers relative to rates in the second period. In 1988, the PSID changed the way it asks about internal position changes. Our results suggest that the more precise questions increased the reported rate of internal position change.

29 In computing the results of tables 4 and 5, we used separate scaling factors for each of the three job change categories, in order to improve the accuracy of our results.

TABLE 3

Within-Between-Total

Firm Firm Count Employment Same Job Change Change (Employed Exit (%I (%I (%) Both Times) (%)

1985-88: Manual workers: United States ..........

Germany ..................

Nonmanual workers: United States ............

Germany ..................

1988-91: Manual workers: United States ............

Germany ..................

Nonmanual workers: United States ............

Germany ..................

NOTE.-Unweighted counts are in parentheses. The first row for each entry is the total sample while the second row for each entry is the subsample with valid scores on all four earnings variables.

earnings change for those who changed positions relative to those who did not change positions.30 During 1985-88 internal job change improved the chances for earnings gain in both countries. In contrast, the conse- quences of employer change were variable in both countries: the probabil-

30 Thus, e.g., an odds ratio of 1.5 in the "P > 15% rise" row of col. 2 implies that the odds of at least a 15% increase in monthly earnings in the three-year period were 50% higher for those who changed positions with their employer than for those who were still in the same position. In parentheses, we report the estimated probability or earnings mobility. The SEs for these probabilities are obtained from the covariance matrix for the estimates of the means and variances of the latent earnings variables. We sampled 110 times from the distribution for the mean and variance of the latent variables and computed the probability of earnings change at our seven benchmark points. After discarding the five largest and five smallest values, we report the mean values and standard deviations of these 100 remaining values. The SEs in the third column concern only the portion of uncertainty in earnings mobility for those working at both times.

TABLE 4

PROBABILITY MOBILITYFOR MALES NOT CHANGING

OF EARNINGS JOBSAND ODDS RATIOS OF EARNINGSMOBILITYFOR JOB CHANGERS, RELATIVE TO JOB STAYERS, 1985-88

New Position, Left Employer Same Employer (unconditional) Odds Ratio vs. Odds Ratio vs.

Same Job "Same Job" "Same Job" Probability Groups P (SE) (PISE) (PISE)

United States:
>25%rise ........................ ,071 (.01) 3.2 (.20/.03) 5.2 (.28/.02)
> 15% rise ........................ .20 (.02) 2.5 (.38/.03) 2.5 (.38/.02)
> 5% rise ......................... .43 (.02) 2.1 (.61/.03) 1.3 (.50/.02)
-5% to 5% ....................... .29 (.02) .67 (.21/.01) .34 (.12/.007)
> -5% fall ....................... .28 (.02) .55 (.18/.03) 1.6 (.38/.02)
> -15% fall ..................... ,080 (.01) .62 (.051/.02) 4.3 (.27/.02)
>-25%fall ..................... ,010 (.004) .82(.0083/.006) 22 (.19/.02)

Germany:
> 25% rise ........................ ,052 (.02) 4.4 (.20/.04) 14 (.44/.04)
> 15% rise ........................ .26 (.03) 2.3 (.44/.04) 3.7 (.56/.04)
> 5% rise ......................... .66 (.02) 1.5 (.74/.04) 1.1 (.68/.03)
-5% to 5% ....................... .28 (.01) .60 (.19/.02) .29 (.10/.009)
>-5%fall ....................... ,057 (.02) 1.2 (.069/.02) 4.5 (.21/.03)
> -15% fall ..................... ,0025 (.001) 3.4 (.0084/.005) 66 (.14/.02)
> -25% fall ..................... ,001 1 117 (.10/.008)

NOTE.-Probabilities and SEs of job change for position and employer changers are in parentheses in cols. 2 and 3.

ity that employer change led to earnings gain was higher than the proba- bility of gain for those who did not change jobs, and the probability of earnings loss was also higher.31 In 1988-91, probabilities of earnings gain via external job change dropped in both countries (from .5 to .35 in the United States; from .68 to .61 in Germany), and the probabilities of earnings loss from external job change rose. Internal job change contin- ued to be the most reliable source of earnings gain in both countries in the 1988-91 period, while employer changing produced variable results, especially when the possibility of an employment to nonemployment tran- sition is included.

Clearly, the big difference between the two countries is in the experi- ence of workers who did not change jobs. In the 1985-88 period, German

31 In absolute terms, the probability of earnings loss from job change is higher in the United States than in Germany. However, the contrast between the odds of downward mobility for employer changers and for stayers is generally much greater in Germany (this is reflected in the odds ratios reported in tables 4 and 5).

TABLE 5

PROBABILITY MOBILITYFOR MALES NOT CHANGING OF EARNINGS JOBSAND ODDS RATIOSOF EARNINGSMOBILITY

FOR JOB CHANGERS, RELATIVETO JOB STAYERS, 1988-91

New Position, Left Employer
Same Employer (Unconditional)
Odds Ratio vs. Odds Ratio vs.
Same Job "Same Job" "Same Job"
Probability Groups P (SE) (PISE) (PISE)
United States:    
> 25% rise ......................    
> 15% rise ......................    
> 5% rise ........................    
-5% to 5% .....................    
> -5% fall .....................    

> -15% fall ....................

> -25% fall ....................
Germany:

> 25% rise ......................
> 15% rise ......................
> 5% rise ........................
-5% to 5% .....................
> -5% fall .....................
> -15% fall ....................
> -25% fall ....................

NOTE.-Probabilities and SEs of job change for position and employer changers are in parentheses in cols. 2 and 3.

workers almost never experienced earnings decline, and two of three experienced a real earnings gain of better than 5%. In the United States, however, real earnings often declined even for those who did not change jobs, while the probability of earnings gain was lower in the United States than in Germany. In the 1988-91 period, the big discrepancy in the two countries continued to be the much larger rate of downward earnings mobility in the United States than in Germany. Our analyses make clear that cross-national differences in job mobility rates are a relatively small source of the cross-national difference in earnings mobil- ity; the earnings trajectories of American workers diverged even for those who stayed in the same job while earnings trajectories for comparable German workers generally moved in the same direction. Appendix table A3 makes this point in a quantitative way. It reports estimates of earnings mobility for American workers after the American probabilities of job mobility are adjusted to match the German probabilities of job mobility. As can be seen, this adjustment accounts for only a small proportion of the cross-national difference in earnings mobility.

RESULTS: EARNINGS MOBILITY AND CLASS

Table 6 provides an answer to the question whether earnings outcomes varied by social class in the United States and in Germany. In both countries, within-class outcomes were much more variable than between- class outcomes. However, cross-national differences in the experience of the two classes are also apparent. In both 1985-88 and 1988-91, Ameri- can nonmanual workers experienced about six percentage points higher chance for earnings gain and six percentage points lower chance for earn- ings loss than did manual workers. Nonmanual German workers had a higher chance for a large earnings gain than did manual workers in the first three-year period. However, manual workers had the higher chance for a small gain, while the chances for earnings loss were roughly equal for the two groups. In the more recent three-year period, the situation changed slightly in favor of nonmanual workers. Overall, however, class differences in Germany were muted relative to the United States.

THE EVOLUTION OF INDIVIDUAL EARNINGS TRAJECTORIES

The analyses discussed above have focused on earnings mobility between two closely spaced points in time. The presumption of this analysis was that earnings declines would provide an informative measure of career turbulence since the usual life-cycle model of careers predicts that socio- economic outcomes rise to a plateau. A more direct measure of turbu- lence, however, examines outcomes at multiple points in the life course to determine how often trajectories reverse direction and the extent of recovery from losses suffered in a previous period. In order to examine the evolution of individual-level trajectories in the two countries, we examined earnings mobility at three points in time, which provides the opportunity to observe earnings rebounds.

We hypothesize that earnings rebounds are asymmetric: workers will on average hold on to their earnings gains, while they will recover at least partially from earnings declines. They hold on to their gains, first of all, because they want to and, second, because the "inertia" built into labor markets (even turbulent ones as in the United States) improves their chances for success in their attempt. Recovery from earnings losses can be accomplished either by increasing hours worked or by changing jobs, but this task is harder, since it is ultimately a function of the struc- ture of opportunity. Therefore, we expect it to be more difficult to reverse losses than to hold gains.

The question of interest is whether the more flexible American labor market-where earnings losses are more common than in Germany-also makes it easier for workers to recover from losses and whether the

more rigid German labor markets-which protect workers from losses- also make it more difficult for German workers to recover from losses when they do occur. The categorization of the two labor markets as "flexible" and "rigid" supports a prediction that the earnings rebounds will be larger in the United States. However, one might alternatively argue that in the German occupational labor markets, where such a high proportion of workers have occupational credentials, the worker has greater control of his future life changes. In the United States, in contrast, the proportion of earnings coming from firm-specific human capital (or perhaps more accurately, from returns to tenure with a given employer), gives workers less control over their destiny and hence makes it more difficult to recover from earnings declines that may be triggered by employer-initiated actions such as layoffs.

Our analytical strategy involves three regressions. The first (model A) specifies the log of imputed earnings in 1985 as a function of age, educa- tion, training, occupation, industry, and (in Germany) whether the worker was on a term or an indefinite ~ontract.~'

Education is measured with a set of dummy variables for various degree or certification levels. The "apprenticeship" category for Germany includes training in voca- tional school or other on the job training. Occupation was measured in EGP categories, with semiskilled/unskilled manual work as the omitted category. Industry was measured using the common four-industry sector distinction, though we further divided manufacturinglmining into two categories based on the Stinchcombe (1979) clas~ification.~~

The second regression (model B) models differences in the log of 1988 and 1985 earnings (recalling that the difference of logs is the same as the log of the ratio) as a function of the same set of variables with two additions. First, we include a measure of the log of hours worked on a monthly basis in 1985. The GSOEP question on hours asks respondents to report their average hours worked per week, including possible over- time during the previous month. We imposed a floor of 30 hours per week and a ceiling of 60 hours per week in Germany for full-time workers and multiplied the result by 4.5 to get a monthly measure.34 For the

32 We used age rather than labor market experience in these results because the differ- ences in the education and training institutions of the United States and Germany make it difficult to construct comparable imputed measures of experience.

33 That portion of manufacturing/mining that fit Stinchcombe's (1979) definition of "large scale engineering" (this includes mining) was grouped together as the omitted category. The sample was limited to those who had a job as of the survey date.

34 This procedure was necessary to eliminate outliers. When we regressed weekly earnings on variables in the variables included in model A plus our modified measure of hours worked, we get a larger effect and a smaller SE than when we use the unmodified measure of hours worked.

PSID, we took the report of hours worked on all jobs in the calendar year (this information was taken from the following year survey) and divided by 12. The second additional variable in model B was our im- puted measure of the log of monthly earnings in 1985. The sample for model B was limited to those working in both 1985 and 1988.

The third regression (model C) specifies the log of the ratio of 1991-88 earnings as the dependent variable. The independent variables in model C are similar to those of model B. In addition, we included a measure of imputed earnings change between 1985 and 1988, and we allowed the effect of earnings change between 1985 and 1988 to vary depending on whether the 1985-88 change was negative or positive. The sample for model C was limited to those working in 1985, 1988, and 1991 (later in this article, we compare patterns of exit and reentry to employment).

Table 7 shows both important similarities and differences between the United States and Germany. In both countries, earnings rise with age at a declining rate, earnings are higher for more educated workers, and earnings are higher for managers and professionals. There are also notice- able differences in the determinants of earnings. There is a greater vari- ance in the payoff to educational levels in the United States: the gaps between high school and less than high school or between college degree and high school diploma are larger than are the corresponding differences in Germany. Furthermore, the gap in earnings between skilled and semi- skilled/unskilled manual occupations (the latter is the omitted reference category in the regressions) is larger in the United States than in Ger- many. Third, the gap between earnings in light and heavy manufacturing is greater in the United States than in Germany, while the payoff to self-employment is significantly larger in Germany than in America. The higher returns to skill and to higher-class positions in the United States are consistent with our earlier findings of greater class divergence in earnings mobility in America than in Germany.

Table 8 contains our results for earnings mobility between 1985 and 1988. In both countries, the rates of earnings gain (net of earnings level in the starting period) are higher for the better-educated and for those in the higher-status occupations. Both countries also show a certain ten- dency for earnings to equalize over time in the sense that, net of other factors, workers who begin the period with higher- or lower-than-average earnings will generally move back toward the average earnings level. Every 10% higher level in 1985 earnings reduces earnings gains by roughly 1% in the United States and 2% in Germany, net of other factors, while every 10% lower level in 1985 earnings increases earnings gains by roughly 1% in the United States and 2% in Germany. Finally, in the United States though less so in Germany, earnings mobility appears to be driven by changes in hours worked as well as by changes in pay. The

TABLE 5 LOG EARNINGS. 1985 FOR THE UNITED STATES AND GERMANY

UNITED STATES GERMANY
Estimate t-Ratio Estimate t-Ratio
Age .........................................................
059
7.1 .05 8.7
Age2 ....................................................... -.001 -6.3 -.001 -7.9
Education and training, United States:      
High school .......................................23 7 7.2    
Some college ........................................ 30 8.0    
College degree .......................................      
Nonacademic certification ........................      
Education and training. Germany:      
HauptschuleiRealschule ...........................      
Abitur .................................................      
University/Fachhochschule ......................      
Apprenticeship ......................................      
Fachschule ...........................................      
Occupational class:      
Upper professional, manager ....................      
Lower professional. manager .....................42 1 12.4 .27 9.5
Routine nonmanual .................................165 3.9 .154 4.2
Self-employed ........................................ -.046 -.8 .388 8.4
Manual supervisory .................................
37 1
6.5 .143 3.4
Skilled manual .......................................25 7.8 .062 2.9
Employer tenure ........................................017 10.0 .002 2.1
Union member ..........................................175 6.6 .028 1.8
Term contract ..........................................   -.I41 -4.1
Industry sector:      
Light manufacturing ............................... -.182 -5.5 -.077 -4.0
Communications/transportationiutilities .......013 .3 -.I37 -4.7
Wholesalelretail trade ............................. -.193 -4.6 -.20 -5.8
Services ............................................ -.25 -7.6 -.199 -8.6
Constant ................................................. 10.361 66.2 5.348 48.5
R ............................................... .35 .32
N .......................................................... 2, 903 2, 649

NOTE.-Parameter estimates and t-ratios are based on five imputations . The omitted education cate- gory is less than high school/GED in the United States and less than middle school in Germany . The omitted occupation category is semiskilled/unskilled workers . The omitted industry sector is heavy manufacturing.

TABLE 8

EARNINGSMOBILITYFROM 1985 TO 1988. FOR THE UNITED STATES AND GERMANY*

UNITED STATES GERMANY
Estimate t-Ratio Estimate &Ratio
Age ...............................................        
Age2 ..............................................        
Education and training. United States:        
High school .................................        
Some college ..............................        
College degree ..............................        
Nonacademic certification ...............        
Education and training. Germany:        
HauptschulelRealschule ..................        
Abitur ........................................        
UniversitylFachhochschule .............        
Apprenticeship .........................        
Fachschule ..................................        
Occupational class in 1985:        
Upper professional. manager ...........        
Lower professional. manager ...........        
Routine nonmanual .......................        
Self-employed ...............................        
Manual supervisory .......................        
Skilled manual .............................        

Employer tenure

...................

.......

Union member ................................
Term contract .................................

......

Log hours 1985 ..................

.....

Industry sector: Light manufacturing ...................... Communications/transportationi

utilities ....................................
Wholesaleiretail trade ....................
Services .....................................

Log earnings 1985 * ..........................
Constant ........................................

NOTE.-Parameter estimates and t-ratios are based on five imputations . The omitted education cate- gory is less than high schooliGED in the United States and less than middle school in Germany . The omitted occupation category is semiskilled/unskilled workers . The omitted industry sector is heavy manufacturing.

* Log (earnings 1988iearnings 1985) is for men employed in both 1985 and 1988 .

more hours an American male worked in the previous period, the more likely he was to experience downward earnings between the first and second period, presumably because of a tendency for hours worked to regress to the mean. The reverse is also true: those who worked relatively few hours in the first period are more likely to experience earnings gain in the second period. Changes in hours worked appears to be a less important mechanism for earnings mobility in Germany.3s

Finally, table 9 displays results for earnings mobility from 1988 to 1991 as a function of earnings in 1985 and earnings mobility between 1985 and 1988 for those who were employed in 1985, 1988, and 1991. U.S. workers in upper-status occupations experienced more upward earnings mobility than did lower-status workers, net of other factors. Further- more, higher education produced a higher rate of earnings increase inde- pendent of class position and of previous earnings history in the United States. These variables had less of an impact in Germany. Third, the results continue to show an equalizing tendency in earnings mobility from 1985 levels, with the effect somewhat larger in Germany (-.094) than in the United States (-.067). These results add further support to the claim that earnings divergence-in particular by class-was stronger in the United States than in Germany.

The 1988-91 analysis also shows a clear tendency for earnings to re- bound in the sense that those who gained in the previous period tend to gain less in the present period and vice versa. Furthermore, these effects are asymmetric: recoveries from earnings losses are bigger in an absolute sense than are givebacks of gains. In the United States, 29% of losses between 1985 and 1988 (net of other factors) were made up between 1988 and 1991 despite the onset of the recession. In Germany the recovery was much larger. About 85% of losses between 1985 and 1988 were made up between 1988 and 1991. In both countries, workers who gained earnings in the first period tended to hold on to these gains in the second period. American workers on average lost only 14 percentage points (-0.288 f .152) of the previous period's gain, net of other factors, while German workers lost about 13 percentage points (-.843 f .709) of previous gains, net of other factors. Thus, the labor market character- istics of both countries allowed workers to hold on to gains and to recover a portion of their losses. The difference in the rate of recovery of losses favoring Germany suggests that the occupational labor markets of that country provide workers with greater control over their future than did the more heterogeneous labor markets of the United States.

The reason for this weaker relationship in Germany is not clear given the apparent willingness of German firms to adjust worker hours in response to changes in demand (Abraham and Houseman 1993) and bears further scrutiny in future research.

TABLE 9

EARNINGSMOBILITYFROM 1988 TO 1991. FOR THE UNITED STATES AND GERMANY*

UNITED STATES GERMANY
    Estimate t-Ratio Estimate t-Ratio
Age ...............................................        
Age2 .............................................  
Education and training. United States:  
High school ...............................  
Some college ..................................  
College degree ...............................  
Nonacademic certification ................  
Education and training. Germany:  
HauptschulelRealschule ...................  
Abitur ..........................................  
UniversitylFachhochschule ...............  
Apprenticeship ...............................  
Fachschule ....................................  
Occupational class in 1988:  
Upper professional. manager .............  
Lower professional. manager ............  
Routine nonmanual .........................  
Self-employed ................................  
Manual supervisory .......................  
Skilled manual ...............................  
Employer tenure ...............................  
Union member ..................................  
Term contract ........................... ........  
Log hours 1988 .................................  
Industry sector:  
Light manufacturing .......................  
Communicationsitransportation/  
utilities ......................................  
Wholesaleiretail trade ......................  
Services ......................................  
Log(earn88/earn85)* ........................  
Log(earn88learn85) if positive* ............
Log earnings 1985 * ............................
 
Constant ..........................................  
R .................................................  
N ...................................................  

NoTE.-Parameter estimates and t-ratios are based on five imputations . The omitted education cate- gory is less than high schooliGED in the United States and less than middle school in Germany . The omitted occupation category is semiskillediunskilled workers . The omitted industry sector is heavy manufacturing.

* Log (earnings 1988iearnings 1985) is for men employed in both 1985 and 1988 .

While the results in table 9 suggest that a greater portion of earnings loss among employed workers is more transitory in Germany than in the United States, the GSOEP and PSID data show that unemployment after employment exit tends to be more transitory in the United States than in Germany. Of those American workers who were employed in 1985 but no longer employed in 1988, we calculated that 55% were again employed in 1991. In contrast, only 36% of the German workers who were employed in 1985 but not in 1988 were again employed in 1991. Some of this difference is due to the higher levels of early retirement in Germany than in the United States (Naschold et al. 1993), and some of it is due to the established tendency of German workers who leave employment to remain unemployed for a longer time than is typical in the United States (Schettkat 1992). Furthermore, the consequences of returning were different in the two countries. As expected, (e.g., Ham- mermesh 1989; Seitchik and Zornitsky 1989; Jacobsen et al. 1993) Ameri- can workers who returned to employment did not achieve their previous earnings level (they made only 81% as much in 1991 as they had made in 1985). In contrast, the 36% of German workers who returned to em- ployment made on average 24% more money in 1991 than they did in 1985.

Only a fraction of this large cross-national difference can be accounted for by differences in the aggregate earnings trend for Germany and the United States. We suspect that sample selection accounts for much of this difference. The more generous welfare benefits in Germany make it less painful to stay out of the workforce, so that the workers who return are disproportionately those who can do well in the labor market. The difference between the countries can be put in further context by compar- ing the 1985-91 earnings change of those who were not working in 1988 with the earnings change for all men working in both 1985 and 1991. In the United States, the average increase in earnings between 1985 and 1991 for those working at both times was 4%. Therefore, the individual who worked at the time of the 1985 interview, who then was not working at the time of the 1988 interview, and who had returned to work by the 1991 interview did 22% worse than the average worker. In Germany, the overall average earnings increase was 22%, so those who left and returned did 2% better than the average worker. In other words, exit and reentry did not carry the same penalty in Germany as it did in the United States. Finally, we can use the data on reentry to compute the average labor earnings in 1991 for all workers who worked in 1985 but were not working in 1988, regardless of whether they were working in 1991. This calculation shows that American workers who were working in 1985 but not working in 1988 made 45% of their average earnings in 1985 (.81 of the 1985 average for those had returned to work by 1991 X

.55, the fraction who returned). In effect, this is a 45% earnings rebound

from the null 1988 monthly earnings. The comparable figure for Germany

is also 45% (1.24 X .36).

In summary, workers in both countries demonstrate a measure of re-

bound from previous earnings levels over the life course. Education and

skill differences lead to greater earnings divergence in the United States

than in Germany, and the tendency for earnings to equalize, net of skill

and other differences, was somewhat weaker in the United States than

in Germany. While both American and German workers tended to hold

on to earnings gains, German employed workers did better at recovering

earning losses.

DISCUSSION

The results of our analyses demonstrate that labor earnings are much less volatile in Germany than in the United States. In the recent past, the earnings of German males have been rising at a faster rate than have their American counterparts. At the same time, the volatility in earnings mobility has been much lower than in the United States. The difference in earnings trajectories can be seen both at the individual level and at the class level. We argued at the start of this article that a contrast between Germany and the United States should reveal the power of institutions to counteract, mediate, or otherwise channel the effects of structural forces on individuals. In the United States, market forces are passed through the institutional filters relatively unmodified, at least in

comparison to Germany. The most dramatic evidence for this is in the earnings changes for workers who do not change jobs. While these changes are naturally less dramatic than are the changes for workers who do change jobs, they are still pronounced. Some American workers gain substantially just by remaining where they are, while other workers lose out. To a certain extent, these losses occurred through the erosion of paychecks that did not rise over time, so the fluctuations may not have been as noticeable to the individual workers as they are in statistical tables. Nonetheless, the divergence is real and reflects the extent to which American labor markets respond to market developments.

While numerous authors (e.g., Rosenberg 1989; Stille 1990; Bosch 1990; Schettkat 1992; Streeck 1993) have argued that Germany and in- deed all of industrialized Europe is experiencing industrial t~rbulence,~~ one might argue that the structural forces operating on the German labor

36 Indeed, by Appelbaum and Schettkat's (1990~)measure of structural change, the German product market has actually shown higher rates of structural change in recent years than has the American market.

market actually differ substantially from those in the United States, where Fordist manufacturing was more entrenched than in Germany (which instead emphasized high-quality custom production) and where diversification away from labor-intensive consumer goods may have lagged behind a similar transition in Germany (Sengenberger 1990). While it is difficult to provide a definitive answer to this question with statistics, some comparisons are possible. As noted earlier in this article, average manufacturing productivity increases, which in the view of many economists are the primary driving force behind wage increases, have been generally similar. Other measures also suggest qualitative similarity in the economic situations of the United States and Germany, with some- what more turbulence in the United States. Estimates of the rate of growth in GNP during the 1980s shows a clear advantage for the United States, even when the recession of the early 1980s is taken into account (Abraham and Houseman 1993, table 3-1). However, the United States experienced a deeper recession than did Germany. Furthermore, data on manufacturing demand (operationalized as manufacturing shipments; see Abraham and Houseman 1993) in the United States and Germany show similar long-run trends. However, American industries show more vola- tility. American industrial response to macroshocks such as recessions or the OPEC oil crisis has been greater, and American industries show longer periods of above-trend or below-trend shipments. Unemployment rates in the two countries have been quite similar (Abraham and House- man 1993, table 3-2). Business failure rates in the United States during the period 1981-90 average 0.9% per year (Dun and Bradstreet 1994), while in Germany, the recent rate appears to be 0.7% per year (Mort- siefer, quoted in Harm 1992).

In summary, there may be some truth in the argument that the greater volatility of American earnings is a consequence of the more turbulent environment in which American firms find them~elves.~' However, this truth is only partial, and it begs the question of why the environment is more turbulent. There clearly is a different relationship between economic performance and labor markets in Germany than in the United States. Abraham and Houseman found that in the United States, "pro- duction employment and total prodaction move up and down together,"

37 One might also question whether country differences are due to differences in the segment of world production served by the manufacturing sectors of the two countries. While such differences may be important, they are unlikely to explain the cross-national difference. One might note, e.g., that German and American comparisons of earnings trajectories of nonmanual workers (the great majority of whom work outside the manufacturing sector in both countries) look similar to German and Ameri- can comparisons of manual workers (see table 6).

while in Germany it was total hours that varied with production, not employment. At a more macro level, total employment in the United States is more responsive to GDP growth than is total employment in Germany. From 1971 to 1980 employment in the United States increased by 0.24% for each 1% rise in the GDP, while in Germany the comparable figure was only 0.07% (Institut fiir Arbeitsmarkt und Berufsforschung 1994; Holst and Schupp 1994; see also Wohlers and Weinert 1988).

Germany, in other words, appears to create more earnings and job stability for individual workers at the expense of creating more jobs. In part, the slow growth in employment reflects the relatively low rate of female labor force participation, the low rate of growth of the population, and the relatively high rate of early retirement. Our results make clear that German earnings stability is enhanced by the high rate of job stabil- ity, but there would be a strong cross-national difference even if the job mobility rates were adjusted to the same level. At the same time, it is important not to understate the extent to which the historical growth in German unemployment rates and the long duration of this unemployment have created a polarized labor market in Germany, with stable earnings for employed workers and diminished job prospects for the unemployed.

Thus, while the analyses performed here have been reasonably exten- sive, we must keep in mind that individual earnings are only one compo- nent of life chances. A comprehensive cross-national comparison of the stability or instability of life chances requires considering total private compensation (including benefits), the experience of women workers, family income from private sources, and government transfers (i.e., the "social wage"). Such a comparison also requires a more systematic explo- ration of how institutional mechanisms mediate economic pressures on product markets, as well as a more systematic exploration of the links between macro-and microchange. We therefore see the analyses dis- cussed here as an important component of a larger story.

In a deeper sense, the analyses presented here can only be partial. Two arguments that follow naturally from the literature cannot be tested with existing data. Those who argue that technical change is fundamentally restructuring the demand and therefore the wages of different types of labor would predict that countries with "rigid" institutions like Germany will inevitably exhibit a pattern of turbulence more similar to that of the United States as the pressure for change becomes too great for German institutions to resist. On the other side, if American firms were especially unsuitable for competition in a global market based on flexible produc- tion, the level of turbulence in the American labor market may subside somewhat after the current round of corporate restructuring is concluded. Either position would predict a certain amount of convergence in the pattern of German and American earnings trajectories. Whether such convergence occurs, and the extent to which such a convergence occurs, will only become clear with the passing of time.

APPENDIX

TABLE A1

EARNINGS FROM VARIOUSUNCORRECTED

MOBILITY MEASURES IN THE PSID AND GSOEP (%)

PSID

Average GSOEP Current Annual PROBABILITY WageIHour PEARNl PEARN2 GEARNl

GROUPS WageIHour GEARN2

1985-88:
> 25% rise ...........
> 15% rise ...........
> 5% rise .............
-5% to 5% ..........
> -5% fall ..........
> -15% fall .........
> -25% fall .........

1988-91:
> 25% rise ...........
> 15% rise ...........
> 5% rise .............
-5% to 5% ..........
> -5% fall ..........
> -15% fall .........
> -25% fall .........

TABLE A2

EARNINGS DEFINEDRELATIVETO THE AVERAGEMOBILITYIN THE

MOBILITY, SAMPLE,LIMITEDTO THOSE WORKING AT BOTH TIMES (%)

> 1.25 X average change ........................... 10 11 1 1
> 1.15 X average change ........................... 2 1 22 7 7
> 1.05 X average change ........................... 39 3 9 3 1 30
Within 5% of average change ...................... 23 2 2 40 41
< .95 X average change ............................38 39 30 2 9
< .85 X average change ............................ 18 18 5 4
< .75 x average change ............................ 5 5 0 0

TABLE A3
EARNINGSMOBILITYESTIMATESFOR THE UNITED STATES, UNADJUSTED

AND

ADJUSTED JOB MOBILITY RATES, LIMITED

TO THE GERMAN TO THOSE EMPLOYED AND END DATES

AT BOTH START

---~ -~ --

UNITED STATES PROBABILITY Unadjusted Adjusted

GROUPS GERMANY 1985-88:

> 25% rise

.15 .ll .14

...................

......

> 15% rise .......................... .............. .28 .24 .36
> 5% rise ........................................... .50 .47 .66
-5% to 5% ...................................... .30 .29 .24
> -5% fall ......................................... .27 .28 .10
> -15% fall ..................................... .10 .09 ,013
> -25% fall ................... .................. .03 .02 ,001

1988-91:

............

> 25% rise

> 15% rise

.09 .09 ,047

.20 .19 .19

........................

.......
..............
......

...................

.....

> 5% rise ........................... ........ .40 .39 .49
-5% to 5% ....................................... .25 .2 5 .33

fall ...................

.......

> -15% fall ....................................... .08 .07 ,025
> -25% fall ...................................... .05 .05 ,001

NOTE.-Col. 1 of table A3 will not generally be identical to col. 4 of table 1. In computing the conditional probabilities of earnings mobility given job change, we used separately computed scaling factors for each conditional distribution, and our latent conditional distributions are derived from esti- mates of a covariance structure model fit to the separate subsamples (defined by no job change, a job change with the same employer, or a change of employer).

REFERENCES

Abraham, Katharine G., and Susan N. Houseman. 1993. Job Security in America: Lessons from Gemany. Washington, D.C.: Brookings.

Alves, Wayne M., and Peter H. Rossi. 1978. '(Who Should Get What? Fairness Judgments of the Distribution of Earnings." American Journal of Sociology 84: 541-64.

Appelbaum, Eileen, and Ronald Schettkat. 1990a. "Determinants of Employment Developments: A Comparison of the United States and the Federal German Econo- mies." Labour and Society 15:13-31.

. 1990b. Labor Market Adjustments to Structural Change and Technological Progress. New York: Praeger. Bartholomew, D. J. 1987. Latent Variable Models and Factor Analysis. New York: Oxford University Press. Baumol, William J., Sue Anne Batey Blackman, and Edward N. Wolff. 1989. Productivity and American Leadership: The Long View. Cambridge, Mass.: MIT Press.

Blossfeld, Hans-Peter, Gianna Giannelli,-and Karl Ulrich Mayer. 1993. "Is There a New Service Proletariat? The Tertiary Sector and Social Inequality in Germany." Pp. 109-35 in Changing Classes: Stratification and Mobility in Post-Industrial Societies, edited by Gdsta Esping-Andersen. Beverly Hills, Calif.: Sage.

> -5%

............ .35 .36 .18

Blossfeld, Hans-Peter, and Karl Ulrich Mayer. 1988. "Labor Market Segmentation in the Federal Republic of Germany: An Empirical Study of Segmentation Theories from a Life-Course Perspective." European Sociological Review 4: 123-40.

Bluestone, Barry, and Bennett Harrison. 1982. The Deindustrialization of America. New York: Basic. Bollen, Kenneth A. 1989. Structural Equation Models with Latent Variables. New York: Wiley.

Bosch, Gerhard. 1990. Retraining-Not Redundancy: Innovative Approaches to In- dustrial Restructuring in Germany and France. Geneva: International Labour Office.

Bosch, Gerhard, and Werner Sengenberger. 1989. iiEmployment Policy, the State, and the Unions in the Federal Republic of Germany." Pp. 87-88 in The State and the Labor Market, edited by Samuel Rosenberg. New York: Plenum.

Bound, John, and George Johnson. 1992. "Changes in the Structure of Wages in the 1980s: An Evaluation of Alternative Explanations." American Economic Review 82:371-92.

Bound, J., and A. B. Krueger. 1991. "The Extent of Measurement Error in Longitu- dinal Earnings Data: Do Two Wrongs Make a Right?" Journal of Labor Economics 9:l-24.

Buechtemann, Christoph F. 1993. Employment Security and Labor Market Behavior. Ithaca, N.Y.: Cornell University Press.

Cameron, David R. 1984. "Social Democracy, Corporatism, Labour Quiescence, and the Representation of Economic Interest in Advanced Capitalist Society." Pp. 143-78 in Order and Conjlict in Contemporary Capitalism: Studies in the Political Economy of Western European Nations, edited by John H. Goldthorpe. Oxford: Oxford University Press.

Carroll, Glenn R., and Karl Ulrich Mayer. 1986. "Job-Shift Patterns in the Federal Republic of Germany: The Effects of Social Class, Industrial Sector, and Organiza- tional Size." American Sociological Review 51:323-41.

Cornfield, Daniel. 1991. "The U.S. Labor Movement: Its Development and Impact on Social Inequality and Politics." Annual Review of Sociology 17:27-50. Danziger, Sheldon, and Peter Gottschalk. 1993. Uneven Tides: Rising Inequality in America. New York: Russell Sage.

Dertouzos, Michael L., Richard K. Lester, Robert M. Solow, and the MIT Commis- sion on Productivity. 1989. Made in America: Regaining the Productive Edge. Cambridge, Mass.: MIT Press.

Dornstein, Miriam. 1988. "Wage Reference Groups and Their Determinants: A Study of Blue-Collar and White-Collar Employees in Israel." Journal of Occupational P~ychology61:22 1-35.

Dun and Bradstreet. 1994. The Failure Record through 1994. New York: Dun & Bradstreet.

Erikson, Robert, John H. Goldthorpe, and Lucienne Portocarero. 1979. "Intergenera- tional Class Mobility in Three Western Societies: England, France, and Sweden." British Journal of Sociology 30:415-41.

Feller, William. 1966. An Introduction to Probability Theory and Its Applications, vol. 2. New York: Wiley.

Freeman, Richard B. 1993. "How Much Has De-Unionization Contributed to the Rise in Male Earnings Inequality?" Pp. 133-64 in Uneven Tides: Rising Inequality in America, edited by Sheldon Danziger and Peter Gottschalk. Beverly Hills, Calif.: Sage.

Freeman, Richard B., and James M. Medoff. 1984. What Do Unions Do? New York: Basic.

Ganzeboom, Harry B. G., Paul M. De Graaf, and Donald J. Treiman. 1992. "A Standard International Socio-Economic Index of Occupational Status." Social Sci- ence Research 21:l-56.

Ganzeboom, Harry B. G., Ruud Luijkx, and Donald J. Treiman. 1989. "Intergenera- tional Class Mobility in Comparative Perspective." Research in Social Stratijica- tion and Mobility 8:3-84.

Gottschalk, Peter, and Robert Moffitt. 1994. "The Growth in Earnings Stability in the U.S. Labor Market." Manuscript. Boston College.

Haller, Max. 1987. "Positional and Sectoral Differences in Income: The Federal Republic, France, and the United States." Pp. 172-90 in Comparative Studies of Social Structure, edited by Wolfgang Teckenberg. Armonk, N.Y.: M. E. Sharpe.

Hammermesh, Daniel S. 1989. "What Do We Know about Worker Displacement in the U.S.?" Industrial Relations 28:Sl-59.

Hannan, Michael T., Klaus Schomann, and Hans-Peter Blossfeld. 1990. "Sex and Sector Differences in the Dynamics of Wage Growth in the Federal Republic of Germany." American Sociological Review 55:694-713.

Harm, Christian. 1992. "The Financing of Small Firms in Germany." Working Paper no. 899. World Bank, Country Economics Department, May. Harrison, Bennett, and Barry Bluestone. 1988. The Great U-Turn: Corporate Restruc- turing and the Polarizing of America. New York: Basic. Herz, Diane E. 1991. "Worker Displacement in the 1980s." Monthly Labor Review 114 (May): 3-9.

Holst, Elke, and Jurgen Schupp. 1994. "Erwerbsbeteiligung und Erwerbsorientier- ung von Frauen in West- und Ostdeutschland 1990 bis 1993." Discussion Paper no. 90. German Institute for Economic Research, Berlin.

Institut fur Arbeitsmarkt und Berufsforschung. 1994. "Kurzbericht," no. 7, March

31. Nurenberg: Bundesanstalt fiir Arbeit. International Monetary Fund. 1994. International Financial Statistics Yearbook, vol.

47. Washington, D.C.: International Monetary Fund. Jacobson, Louis, Robert LaLonde, and Daniel Sullivan. 1993. The Costs of Worker Dislocation. Kalamazoo, Mich.: Upjohn Institute for Employment Research.

Kappelhoff, Peter, and Wolfgang Teckenberg. 1987. "Intergenerational and Career Mobility in the Federal Republic and the United States." Pp. 3-49 in Comparative Studies of Social Structure: Recent Research on France, the United States, and the Federal Republic of Germany, edited by Wolfgang Teckenberg. Armonk, N.Y.:

M. E. Sharpe.

Karoly, Lynn A. 1993. "The Trend in Inequality among Families, Individuals, and Workers in the United States: A Twenty-Five Year Perspective." Pp. 19-97 in Uneven Tides: Rising Inequality inAmerica, edited by Sheldon Danziger and Peter Gottschalk. New York: Russell Sage.

Kochan, Thomas A,, Harry C. Katz, and Robert B. McKersie. 1986. The Transforma- tion of American Industrial Relations. New York: Basic.

Konig, Wolfgang, and Walter Muller. 1986. "Educational Systems and Labor Mar- kets as Determinants of Worklife Mobility in France and West Germany: A Com- parison of Men's Career Mobility, 1965-1970." European Sociological Review 2 (2): 73-96.

Krueger, Alan B. 1991. "How Computers Have Changed the Wage Structure: Evi- dence from Microdata, 1984-89." Working paper. Princeton University.

Lange, Peter, Michael Wallerstein, and Miriam Golden. 1995. "The End of Corporat- ism? Wage Setting in the Nordic and Germanic Countries." In The Workers of Nations: Industrial Relations in a Global Economy, edited by Sanford Jacoby. New York: Oxford University Press.

Lindbeck, Assar, Per Molander, Torsten Persson, Olof Petersson, Agnar Sandmo, Birgitta Swedenborg, and Niels Thygesen. 1994. Turning Sweden Around. Cambridge, Mass.: MIT Press.

Mahnkopf, Birgit. 1992. "The 'Skill-Oriented' Strategies of German Trade Unions: Their Impact on Efficiency and Equality Objectives." British Journal of Industrial Relations 30:61-81.

Marsden, D. W. 1990. "Institutions and Labor Mobility: Occupational and Internal Labor Markets in Britain, France, Italy, and West Germany." Pp. 414-43 in Labour Relations and Economic Performance, edited by R. Brunetta and

C. Dell'Aringa. London: Macmillan.

Mincer, Jacob. 1991. "Human Capital, Technology, and the Wage Structure: What Do the Time Series Show?" Working paper. National Bureau of Economic Re- search, Cambridge, Mass.

Miiller-Jentsch, Walther. 1995. "Germany: From Collective Voice to Co-Management." Pp. 53-78 in Works Councils: Consultation, Representation, and Cooperation in Industrial Relations, edited by Joel Rogers and Wolfgang Streeck. Chicago: University of Chicago Press.

Murphy, Kevin M., and Finis Welch. 1993. "Industrial Change and the Rising Impor- tance of Skill." Pp. 99-165 in Uneven Tides: Rising Inequality in America, edited by Sheldon Danziger and Peter Gottschalk. New York: Russell Sage.

Naschold, Frieder, Maria Oppen, Holger Peinemann, and Joachim Rosenow. 1993. "Germany: The Concerted Transition from Work to Welfare." Pp. 117-82 in Regulating Employment and Welfare, edited by Frieder Naschold and Bert de Vroom. Berlin: Walter de Gruyter.

Osterman, Paul. 1993. "Pressures and Prospects for Employment Security in the United States." Pp. 228-46 in Employment Security and Labor Market Behavior, edited by Christoph F. Buechtemann. Ithaca, N.Y.: Cornell University Press.

Phelps, Edmund. 1970. "Money Wage Dynamics and Labor Market Equilibrium." Pp. 124-66 in Microeconomic Foundations of Employment and InBation Theory, edited by E. S. Phelps. New York: Norton.

Piore, Michael J., and Charles F. Sabel. 1984. The Second Industrial Divide: Possibil- ities for Prosperity. New York: Basic.

Rodgers, Willard, Charles Brown, and Greg J. Duncan. 1993. "Errors in Survey Reports of Earnings, Hours Worked, and Hourly Wages." Journal of the American Statistical Association 88:1208-18.

Rosenberg, Samuel. 1989. "Labor Market Restructuring in Europe and the United States: The Search for Flexibility." Pp. 3-22 in The State and the Labor Market, edited bv Samuel Rosenbern. New York: Plenum.

Rubin, r avid. 1987. ~ultipll Imputation for Nonresponse in Surveys. New York: Wilev. ~uhm,~~hristo~her

J. 1987. "The Economic Consequences of Labor Mobility." Industrial and Labor Relations Review 41:30-42. Schettkat, Ronald. 1992. The Labor Market Dynamics of Economic Restructuring: The United States and Germany in Transition. New York: Praeger. Schettkat, Ronald, and Michael Wagner, eds. 1990. Technological Change and Em- ployment: Innovation in the German Economy. Berlin: Walter de Gruyter.

Seitchik, Adam, and Jeffrey Zornitsky. 1989. From One Job to the Next: Worker Adjustment in a Changing Labor Market. Kalamazoo, Mich.: Upjohn Institute for Employment Research.

Sengenberger, Werner. 1990. "Flexibility in the Labor Market: Internal versus Exter- nal Adjustment in International Comparison." Pp. 144-62 in Labor Market Adjust- ments to Structural Change and Technological Progress, edited by E. Appelbaum and R. Schettkat. New York: Praeger.

Soskice, David, and Ronald Schettkat. 1993. "West German Labor and East German Transformation." Pp. 102-27 in Labor and an Integrated Europe, edited by Lloyd Ulman, Barry Eichengreen, and William T. Dickens. Washington, D.C.: Brookings.

Stille, Frank. 1990. "Structural Change in the Federal Republic of Germany: The Case of Services." Pp. 93-116 in Labor Market Adjustment to Structural Change and Technological Progress, edited by Eileen Appelbaum and Ronald Schettkat. New York: Praeger.

Stinchcombe, Arthur. 1979. "Social Mobility in Industrial Labor Markets." Acta Sociologica 22:2 17-45.

Streeck, Wolfgang. 1993. "The Rise and Decline of Neocorporatism." Pp. 80-101 in Labor and an Integrated Europe, edited by Lloyd Ulman, Barry Eichengreen, and William T. Dickens. Washington, D.C.: Brookings.

Swenson, Peter. 1989. Fair Shares: Unions, Pay, and Politics in Sweden and West Germany. Ithaca, N.Y.: Cornell University Press. Thelen, Kathleen A. 1991. Union of Parts: Labor Politics in Postwar Germany. Ithaca, N.Y .: Cornell University Press. Thurow, Lester. 1992. Head to Head: The Coming Economic Battle among Japan, Europe, and America. New York: Warner. Topel, Robert. 1986. "Job Mobility, Search and Earnings Growth." Research in Labor Economics 8: 199-233. Topel, Robert H., and Michael P. Ward. 1992. "Job Mobility and the Careers of Young Men." Quarterly Journal of Economics 107:439-79.

Wegner, M. 1985. "Die Schaffung von Arbeitsplatzen in Dienstleistungsbereich: Ein Vergleich zwischen den USA and der Bundesrepublik." IFO-Schnelldienst 85 (6): 3-13.

Wohlers, Eckhardt, and Giinter Weinert. 1988. Employment Trends in the United States, Japan, and the European Community: A Comparative Economic Study. New Brunswick, N.J.: Transaction Books.

Comments
  • Recommend Us