Technological and Organizational Change and the Careers of Workers *

Transcription

1 Technological and Organizational Change and the Careers of Workers * Michele Battisti, Christian Dustmann, and Uta Schönberg This version: December 2017 Abstract: This paper addresses what happens to workers who hold jobs that disappear due to technological and organizational change (T&O). We show that although T&O reduces firm demand for mainly routine-task based jobs, affected workers face no higher probability of non-employment or lower wage growth than unaffected workers. Rather, firms that adopt T&O play an important active role in curtailing its potentially harmful effects by offering affected workers re-training opportunities to upgrade to more abstract jobs, with the exception of older workers. Overall, our findings paint a more positive picture of T&O s welfare effects than is commonly suggested. * We gratefully acknowledge funding by Norface s Welfare State Future program. Ifo Institute for Economic Research at the University of Munich and CESifo Research Network. Contact information: battisti@ifo.de (Michele Battisti). University College London (UCL) and Centre for Research and Analysis of Migration (CReAM). Contact information: c.dustmann@ucl.ac.uk (Christian Dustmann), u.schoenberg@ucl.ac.uk (Uta Schömberg).

2 Technological and organizational change (T&O) is seen as the main reason for the decline in the employment share of occupations with routine task content. 1 The vast majority of the existing literature analyzing the consequences of T&O focuses on its impact on firms and their demand for routine tasks. But what are the consequences of T&O for the workers who held routine jobs (i.e., worked in occupations with a predominantly routine task content) before changes were implemented (hereafter, routine jobholders)? Although hardly any work investigates this question, this aspect is of critical importance. If routine jobholders lose employment permanently or for a prolonged period of time, or if they relocate to inferior and lower paying jobs, T&O may have severe negative welfare implications. This scenario dominates the popular public debate at least in the US context. 2 But if routine jobholders retrain and are able to move to jobs with more abstract skill content, either in the same firm or in other firms, T&O offers tangible benefits. Identifying which of these two scenarios applies is crucial not only for determining T&O s welfare effects on workers but also to provide more information in a political debate which is strongly influenced by populists who foment anxieties about T&O-induced job destruction specifically and globalization more generally. To assess which of the aforementioned scenarios dominates we match a survey panel data set on German firms which covers 15 years and contains detailed information on T&O with registry data providing the entire work histories of all workers ever employed in these firms. This allows us to follow workers over time irrespective of whether they stay in the firm that carried out T&O or not. We begin our investigation by showing that in firms that implement T&O the employment share of routine jobs declines and that of abstract jobs increases relative to firms that 1 See, for example, Autor, Levy, and Murnane, 2003; Goos and Manning, 2007; the recent survey by Acemoglu and Autor, 2011, and the recent theoretical contribution of Feng and Graetz, See, for example, the Economist ( and MIT Technology Review ( 1

3 do not implement T&O. This result holds even within industries and local labor markets. Firm-level T&O is also associated with a rise in the firm level employment share of medium aged workers (30 49) and a decline in that of workers over 50. The results of our placebo tests suggest that these shifts in firm workforce composition are caused by T&O rather than by different pre-existing trends in the demand for specific worker types. In our main empirical analysis, we shift our focus from firms to workers to investigate how workers particularly, routine jobholders and older workers adjust to T&O measured at the firm level. A possible confounder in our analysis could be that companies engaged in T&O may be high-wage firms with low turnover rates employing high-wage workers who are more likely to upgrade to abstract jobs than transition into under- or non-employment. In order to address this concern, we control for worker and firm wage fixed effects in our regressions. We further use event study methodology to corroborate our results. We find that although T&O reduces the firm s employment share of routine jobs, those holding routine jobs at T&O implementation on average suffer neither employment losses nor reduced wage growth. Rather, these workers are more likely to move to more abstract jobs. We provide indirect evidence that these upward movements are facilitated by an increase in firm-provided training opportunities, which can actively dampen the potential harmful effects of T&O on routine jobholders. Although those holding routine jobs at T&O implementation do not lose out on average, those aged 55 and above do not succeed in upgrading to abstract jobs either within or outside the firm. Rather, if affected by T&O, they are more likely to permanently transition into non-employment. In fact, T&O increases such transitions not only for older routine jobholders but for all older workers, including those in abstract jobs and those with a university degree. Not only are these findings consistent with the assumption that older workers lack the workplace skills necessary to deal with T&O, but they also suggest that when workers are close to 2

4 exiting the labor market and therefore do not have much time to enjoy the return to training, neither they nor the company find it worthwhile to invest in new skills. We next ask whether there are particular characteristics of firms that facilitate the re-training of workers after T&O. One facilitating factor may be the degree to which firms have training personnel at hand, or have experience with training workers in-house. We test this hypothesis by investigating whether firms that train more young workers within Germany s apprenticeship training system are more likely to upgrade the skills of their other workers when they carry out T&O. We find strong support for this hypothesis: routine jobholders are more likely to upgrade to abstract jobs following T&O if the share of apprentices in the firm prior to T&O implementation is higher. An additional reason for firms to retrain or further train workers may be the presence of unions that support training activities. Although their influence has dwindled over the past two decades, unions continue to be more powerful in Germany than in countries such as the US and the UK. Moreover, compared to other countries with powerful unions such as France and Italy, unions in Germany play a more active role in advising and supporting firms and workers in all matters of apprenticeship and further training. 3 Our findings indeed suggest that unions may mitigate T&O s effects on routine jobholders: movements from routine to abstract jobs in response to T&O occur more frequently in firms that recognize a union (and are thus bound by collectively negotiated wages, but also benefit from services such as training support that unions offer) than in firms that do not. These findings highlight that firms in-house training capacity and institutions such as unions may alleviate some of the negative welfare implications of T&O, by facilitating upward movements from routine to abstract task-based jobs. 3 See for example Bahnmüller (2009) and Seitz (1997) for an overview. Nearly all unions in Germany, including the three largest IG Metall, Verdi, and IG Bergbau, Chemie und Energie, provide further training programs at often highly subsidized rates. Dustmann and Schönberg (2012) also show that unions increase apprenticeship training in Germany. 3

5 Our paper contributes to three main strands of the literature research on the link between routine-biased T&O and job polarization, on T&O adjustment mechanisms of both workers and firms, and on age-biased T&O. 4 Three recent papers aim to establish direct evidence for the polarization hypothesis by exploiting T&O variation either across industries and countries (Michaels, Natraj, and Van Reenen, 2013) or across local labor markets (Autor and Dorn, 2013; Akerman, Gaarder, and Mogstad 2015). 5 One innovative strength of our analysis is that our 15 years of linked survey-registry data allows us to exploit T&O variation across firms (rather than industries or local labor markets) and detailed worker characteristics to identify a direct link between T&O and firms demand for certain tasks. Our results not only corroborate the key findings of these three studies, but also demonstrate that T&O substitutes for routine jobholders within both industries and local labor markets. Several older papers draw on firm-level T&O data as we do, but because they tend to have no longitudinal information on worker skill composition, they are unable to relate T&O to changes in the firm s workforce composition (see for example Bresnahan, Brynjolfsson, and Hitt (2002)). Caroli and van Reenen (2001) use longitudinal data on firm skill composition, but are prevented from directly testing the polarization hypothesis since they cannot observe workers education and the types of task they perform. Altogether, none of these papers tracks the careers of workers affected by T&O, which is the focus of our paper. In investigating what happens to routine jobholders after their jobs are eliminated, our analysis is closely related to that of Cortes (2016), who uses PSID 4 A related recent literature investigates the labor market effects of a specific technology industrial robots exploiting variation in robot exposure across industries (Graetz and Michaels, 2015, 2017) or local labor markets (see Acemoglu and Restrepo, 2017 for the US and Dauth et al., 2017 for Germany) as opposed to variation in T&O across firms, as we do. In contrast to our paper and the literature summarized below, this literature emphasizes the effects of industrial robots on overall (rather than skill- or age-specific) employment and wages. 5 Gaggl and Wright (2017) use a discontinuity design where firms below a size threshold qualify for ICT subsidies to investigate firms labor demand responses, and find evidence for complementarities between ICT and non-routine cognitive-intensive work. 4

6 data to explore the effects of routine-biased technical change on the patterns of occupational transition out of routine occupations and the wage changes of individual workers. 6 In contrast to Cortes (2016), our approach does not rely on structural assumptions to infer the effects of T&O on task transitions. Rather, we exploit direct information on T&O at the firm level to investigate routine jobholder s movements downward (to non-employment and manual jobs) and upward (to abstract jobs) in response to T&O. By drawing on detailed firm data, we can also shed light on whether firms invest in training when introducing T&O, and which types of firms are particularly successful at upgrading workers from routine to abstract jobs. We are, to the best of our knowledge, the first to document that upward movement to abstract jobs within and outside the firm facilitated by firm training activities is a dominant T&O adjustment mechanism of routine jobholders. 7 Our paper likewise adds to the literature on age-biased technological change by analyzing the T&O s impact on changes in a firm s employment share of older workers and identifying these workers heterogeneous adjustment mechanisms to T&O. Existing work typically concentrates on T&O s impact on workers close to retirement. Bartel and Sicherman (1993), for example, use industry level data to show that unexpected implementation of T&O induces workers to retire earlier. 8. Conversely, Aubert, Caroli, and Roger (2006) demonstrate that firms using innovative workplace practices employ fewer older workers. 9 We go beyond these papers by analyzing in detail how different types of older workers respond to T&O. 6 Autor and Dorn (2009) also touch at this question, by showing that routine occupations are getting old, and that young college workers but not older college workers or non-college workers reallocate to high-skill non-routine employment. 7 Several previous papers also establish a positive correlation between T&O and firm training activities (Bresnahan, Brynjolfsson, and Hitt, 2002; Behaghel, Caroli, and Roger, 2014; Lynch and Black, 1998; Behaghel, Caroli, and Walkowiak, 2012; Behaghel and Greenan, 2012). 8 See also Peracchi and Welch, 1994, for earlier evidence on older workers labor force participation and Hægeland, Rønningen, and Salvanes, 2007 for more recent work. 9 Beckmann (2007) uses cross-sectional IAB Establishment Panel data from 1995 to assess the effect of self-reported technological state and IT investment on age specific labor demands, while Rønningen (2007) uses Norwegian panel data to investigate the relation between the introduction of 5

7 1 Background, Data, and Descriptives 1.1 Background We conduct our analysis for Germany, which, similar to the U.S., UK, and other OECD countries, witnessed an erosion of jobs in the middle of the wage distribution in the 1990s and 2000s (see Spitz-Oener, 2006; Dustmann, Ludsteck, and Schönberg, 2009; Black and Spitz-Oener 2010). Such employment polarization is commonly attributed to routine-biased technological change, which lowers firm demand for routine tasks typically located in the middle of the wage distribution, while increasing demand for the abstract tasks that dominate its upper end (Autor, Levy, and Murnane 2003; Goos and Manning, 2007). 10 We illustrate this polarization in Figure 1a by plotting decadal changes in employment shares of occupations ranked according to their median wage in Figure 1a shows that employment shares of occupations in the middle of the wage distribution decreased, particularly through the 1990s. Consistent with the polarization hypothesis and in line with findings for the U.S. (e.g., Acemoglu and Autor, 2011; Autor and Dorn, 2013), the share of workers employed in routine occupations declined in Germany by about 10 percentage points between 1990 and 2010 (from about 42% to 32%), whilst the share of workers in abstract occupations increased by roughly the same amount and the share workers in manual occupations remained roughly constant (Figure 1b). 11 In this paper, we first explore whether and to what extent firm-level T&O is indeed new or improved products, processes, and innovation activity at the firm level on (levels of) wage bill shares for workers in different age groups. Likewise, Behaghel, Caroli, and Roger (2014) use cross-sectional information on ICT and innovative work practices to study changes in the wage bill shares of older workers in France. All these papers tend to find that the employment of older workers is negatively affected by technological innovation. 10 Barany and Siegel (2014) emphasize that at least part of the observed polarization may stem from structural changes rather than responses to technological change per se. 11 See Section for the classification of occupations into manual, routine and abstract. 6

8 responsible for the decline in middle wage routine jobs. We then investigate how T&O affects workers careers. 1.2 Data and Descriptives Our empirical analysis combines three main data sources: the IAB Establishment Panel (IABEP), registry records for all firms and workers covered by social security, and the Qualification and Career Survey. The IABEP, administered from 1993 to 2012, is an annual representative survey of up to 16,000 firms that provides detailed information on organizational change, innovation, and training activities. We match these data with registry records of the work histories of all workers ever employed at a surveyed firm and are thus able to trace out workers careers even after they have left the company. To assess occupational task content we rely on the Qualification and Career Survey The IAB Establishment Panel The IABEP survey was first administered in 1993 to 4,265 West German firms, and extended to East German firms in By 2010, the number of surveyed firms had increased to over 16,000. From this database, we select all West German firms that participated in the IABEP in any of the years for which information on organizational change is available. Furthermore, we require that these firms employed at least 10 employees, and were not active in agriculture. In addition to data on such variables as IT investments, product and process innovation and training activities, the survey contains a series of questions on different types of technological and organizational changes occurring in the company over the previous 3 years. Questions on organizational change were asked first in 1995 and then at 3-year intervals (i.e., 1998, 2001, 2004, 2007, and 2010). We focus on four questions asked consistently in all these waves to which the firm answers either yes or no: whether it has shifted responsibilities and delegated decisions; whether it has introduced team work or working groups with their own responsibilities; whether it 7

9 has introduced profit centers (that is, units or departments that carry out their own cost and result calculations); and whether it has internally restructured or merged departments or areas of activities. Our baseline measure of T&O intensity simply adds up the number of organizational changes over the 3-year period. Each change not only implies important alterations in how the firm operates but commonly requires changes in the technology being used (e.g., the introduction of a profit center typically implies the adoption of a new computer system). More generally, following Bresnahan, Brynjolfsson, and Hitt (2002), we view our organizational change measures as part of a clustered three-way complementarity between IT investments, reorganization, and product innovation. In such a complementary relation, rapidly falling IT prices motivate firm investment in IT, which in turn changes the organization s optimal structure, and both the IT investments and reorganization ultimately help firms offer better services and products. We present relevant summary statistics for our T&O measures in Figure 2. First, by pooling over all 3-year periods between 1992 and 2010 and weighting by baseline employment to make results representative across workers, Figure 2a shows that nearly 45% workers work in firms that did not implement any measure of T&O in the previous 3 years, while about 12% of workers are employed in firms that adopted at least three changes. According to Figure 2b, departmental mergers were the most common type of T&O. More than 40% of the firms implemented one over the previous 3-year period. For comparison, about 28% of the firms shifted responsibilities across departments, 18% introduced team work, and 17% introduced a profit center. In Panel A of Table 1, we provide evidence that, as hypothesized for example by Bresnahan, Brynjolfsson, and Hitt (2002), our T&O measures are indeed complementary to IT investment and product innovation. Relative to firms that do not carry out any T&O, firms that adopt at least three such measures over a 8

10 3-year period are more likely to undertake IT investments, and invest more than twice as much in IT per employee. Firms that heavily reorganize are also nearly three times more likely to implement product or process innovations. Overall, this evidence suggests that our T&O measure is indeed capturing fundamental changes in how firms are operating. In Panel B of Table 1, we show that companies that carried out at least one T&O change over a three years period are larger, pay higher wages, and exhibit a higher revenue per worker at baseline (i.e., at the beginning of the 3-year measurement period) than those that implemented no changes. The differences between firms that introduced one or two changes and those that implemented at least three, however, are small. As the panel shows, at baseline, restructuring firms tend to have a higher fixed firm effect and employ workers with a higher worker fixed effect than non-restructuring firms. 12 Restructuring firms also employ a larger share of workers in abstract jobs. These differences, however, are relatively small and largely disappear when comparing firms that introduced one to two versus three to four organizational changes Registry Data Our second data source is the registry of social security records in Germany over the 1975 to 2010 period, which, in addition to unique worker and establishment identifiers, contains detailed information on factors such as workers wages, education, age, occupation, industry, and place of work. From this database, we select all individuals employed at least once in an IABEP firm (as of June 30 in a given year) together with all their employment spells before joining and after leaving the firm, thereby tracing out their pre- and post-t&o careers. As is typical in registry data, wages are censored at the social security limit, so, following Dustmann, Ludsteck, and Schönberg (2009), Card, Heining, and Kline (2013), and 12 We employ the full registry data to compute a fixed firm and fixed worker effect for each firm and worker in our sample estimated over rolling 7-year windows prior to T&O adoption, building on Abowd, Kramarz, and Margolis (1999) and Card, Heining, and Kline (2013) (as detailed in the next section). 9

11 others, we impute the censored wages by assuming that wages are drawn from a normal distribution with a variance that varies by age, gender, and education (please see our Appendix for details). As detailed below, we use the occupation variable in the registry data, combined with detailed information on task usage from the Qualification and Career Survey to classify workers into those that hold manual, routine, or abstract jobs. We also distinguish three educational groups: low-skilled workers without any postsecondary education (9.6% in our sample), medium-skilled workers who completed an apprenticeship or equivalent (75.2%), and high-skilled workers who completed a university or college degree (15.2%). We further differentiate between three main age groups: individuals aged below 30, between 30 and 49, and above 50. In the firm level regressions, we restrict the sample to regular employees not in training, aged between 16 and 65; in the worker level regressions, we additionally exclude workers aged over 62 at the beginning of the observation period (who are thus over 65 at the time of observation). We use the full registry data to estimate wage regressions that include both a fixed firm effect and a fixed worker effect (as in Abowd, Kramarz, and Margolis, 1999, and Card, Heining, and Kline, 2013), in addition to controls for age, age squared, and year fixed effects. We estimate these regressions in 7-year rolling windows, starting in Then, for each worker and firm in our sample, we merge in the pre-estimated worker and establishment fixed effects that refer to a 7- year period before T&O implementation The Qualification and Career Survey To categorize the occupations in the registry data, we use detailed information from the 1991/1992 wave of the German BIBB/IAB Qualification and Career Survey (BIBB) on 19 activities performed at work, which we classify into routine, abstract, and manual tasks (see Appendix Table A1). Then, following Antonczyk, Fitzenberger, and Leuschner (2009), for each survey individual i, we proxy the time 10

12 spent on each task type as the number of tasks performed of type j divided by the total number of tasks performed: Task ij = Number of tasks of type j performed by individual i Total number of tasks performed by i Thus, if an individual carries out 6 tasks in total, 3 of which are routine tasks, the routine index is 0.5. We then aggregate the individual task indices at the 3-digit occupational level, use the maximum mean task index to classify the occupation as routine, manual, or abstract, and finally merge this information with the registry data, again at the 3-digit occupational level. Table 2 summarizes differences between the three task types along several analytically relevant dimensions. The table shows that routine, abstract, and manual occupations across all years represent around 37.1, 45.2, and 17.7 percent, respectively, of employment in our sample. The observation that the own-task index on the main diagonal is substantially larger than cross-task index off the diagonal in Panel A highlights that our classification of occupations into manual, routine, and abstract works well. Panel B indicates that abstract occupations enjoy a 33% wage premium over routine occupations, while wages in routine and manual occupations are roughly similar. There is a strong correlation between occupational task content and worker educational attainment, with a 30% share of highly educated workers in abstract occupations, but only 3% and 2.7% in routine and manual occupations, respectively. These figures suggest that high-skilledworkers work mainly in abstract occupations, while medium-skilled workers are more likely to work in routine and manual occupations, although the share of medium-skilled workers in abstract occupations can still be as high as 66%. Age, in contrast, is roughly evenly distributed across occupational categories. 2 Empirical Specification and Identification Strategy We first explain how we estimate the extent to which a firm s T&O affects its employment shares of task and age based worker groups (e.g., workers in routine 11

13 vs. abstract jobs, younger vs. older workers). We then describe how we trace out the effect of T&O on the careers of the workers whose job it affects. This methodology will allow us to address questions such as: does a firm s introduction of T&O result in routine jobholders or older workers exiting into non-employment, or does it result in them transitioning to abstract jobs? 2.1 Estimation at the Firm Level Consider the following relationship between the employment share of workers of type g, y jt g, (e.g., workers in routine jobs or older workers) and technological and organizational capital (TO jt ) in firm j at time t: y jt g = γ g TO jt + τ t g + f j g + e jt g where τ t g denotes worker type-specific year fixed effects, and f j g designates firm fixed effects that are allowed to vary across worker types. We estimate this regression in first differences, relating 3-year changes in the employment share of type g workers to changes in technological or organizational capital (ΔTO jt ): Δy g jt = γ g ΔTO jt + τ g g t + Δe jt We measure ΔTO jt as the number of organizational changes implemented by a firm over a 3-year period, between t-3 and t. By first differencing, we eliminate any worker type-specific firm fixed effect that may affect the firm s demand for organizational capital, on the one hand, and for routine, abstract, or older jobholders, on the other. Hence, in equation (1), a positive (negative) coefficient γ g > 0 (γ g < 0) is suggestive of complementarity (substitutability) between technological and organizational capital and group g workers. We also adopt an additional tighter specification that controls for worker type and industry specific fixed effects at the one-digit level, I j g, and commuting zone fixed effects (R j g ), as well as firm size (in logs) at baseline (FS jt 3 ). This results in the following first-difference equation: (1) 12

14 Δy g jt = γ g ΔTO jt + τ g t + I g j + R g j + β g g FS jt 3 + u jt (2) This specification allows firms in certain industries (e.g. those particularly affected by increased trade) or larger firms to be systematically more likely to implement T&O, while simultaneously exhibiting differential trends in their demand for certain worker groups. At the same time, using commuting zone fixed effects to compare workforce composition shifts in firms that do and do not implement T&O within the same local labor market addresses concerns of reverse causality whereby firms experiencing an increased supply of workers who typically perform abstract (routine) tasks are more (less) likely to adopt T&O. When estimating both these equations, we weight firm level observations by employment at baseline and cluster our standard errors at the firm level. We additionally present results of placebo tests to demonstrate that any shifts in the employment shares of certain worker types are indeed likely to be caused by T&O. Although we focus throughout the paper on T&O s effects on employment share changes rather than wage bill share changes (which some view as a more direct measure of firm demand for certain worker types), results using wage bill shares are very similar to those reported below Estimation at the Worker Level To investigate T&O s effects on the careers of the workers it affects, we estimate specifications of the following type, where j denotes the firm at which worker i was employed in t-3 (the baseline firm), g designates the group to which worker i g belonged at baseline (e.g., employed in a routine job) and y ij t measures the 13 Michaels, Natraj, and Van Reenen (2013), Caroli and van Reenen (2001), and Behaghel, Caroli, and Roger (2014) focus on wage bill share specifications. Caroli and van Reenen (2001), for example, discuss how wage bill share equations emerge naturally from a setup in which the firm s cost function is assumed to be a restricted translog and firms minimize costs under given factor prices and fixed capital stock so that the only variable inputs are different types of labor (i.e., workers with different types of skills). 13

15 outcome at time t (e.g., employment) of worker i who was employed in firm j and belonged to group g in the baseline period t 3: g y ij t = γ g TO j t + I j g + R j g + τ t g + δ 1 g FS j t 3 + x it 3 δ 2 g + δ 3 g f j + δ g 4 θ it 3 + e g ij t, (3) Analogous to our firm level regressions, TO j t denotes the number of organizational changes implemented over a 3-year period (t-3 to t) by the firm at which individual i was employed at baseline (in t-3). In addition to worker typespecific year fixed effects τ t g, we control for the same set of baseline firm characteristics as in the firm-level regressions (i.e., industry and commuting zone fixed effects I j g and R j g, and firm size in t-3, FS j t 3 ). We also condition on a set of individual characteristics (x it 3 ) at baseline (gender, foreign citizenship, and age). To account for the possibility that restructuring firms are high-wage firms that offer differential career prospects to their employees irrespective of organizational change, we additionally control for (baseline) firm wage fixed effects (f j ), preestimated jointly with worker fixed effects on the full registry wage data or the t-9 to t-3 period (see Section 2.2.2). Likewise, to allow for the possibility that these restructuring firms employ high-wage workers unlikely to transit into nonemployment and prone to upgrade to higher paying jobs, we further condition on worker wage fixed effects (θ it 3 ), also pre-estimated jointly with the firm fixed effects on the full registry data for the t-9 to t-3 period. To paint a detailed picture of how T&O affects worker careers, we consider the following main outcome variables (y g ij t ): the individual is working; the individual is working in the same occupational group; the individual is working in a manual, routine, or abstract occupation; the individual is working in a different occupational group but in the same firm; or the individual has left the firm. In our baseline specifications, we measure these outcomes up to 3 years after the 14

16 implementation of T&O. We also adopt a more long term perspective and analyze the impact on workers career outcomes over a 6-year period. We cluster standard errors at the baseline firm level. 3 Results 3.1 Effects of T&O on Jobs T&O and Workforce Composition In Table 3, we present results based on equations (1) and (2) to determine whether and to what extent firm T&O affects workforce composition in terms of tasks, education, and age. In the first set of columns (columns (1)), we control only for year fixed effects (as in equation (1)); in the second (columns (2)), we add controls for firm size at baseline, commuting zones, and industry (as in equation (2)). As Panel A shows, T&O is associated with a reduction in the employment share of routine jobs, with each organizational change over the previous 3 years reducing this share by between and percentage points. Given the average 1.2 percentage point overall reduction in the routine employment share over a 3-year period (last row of Panel A), this response is substantial, as the estimated reduction in column (2) implies that each additional organizational change can account for about 19% of the overall decline in routine employment over that period. Given that firms may adopt up to four organizational changes over a 3-year period, multiple changes have an even larger impact on routine employment shares. Panel A also underscores that the reduction in the routine employment share is accompanied by an increase in the abstract employment share of roughly the same magnitude, while the manual employment share remains largely unchanged. Panel B reports the outcomes of performing the same analyses with workers classified according to education rather than main task type. The results in the first set of columns, which are conditional only on year fixed effects, indicate that, in 15

17 line with Michaels, Natraj, and Van Reenen (2013), T&O significantly reduces the employment share of medium-skilled workers, while significantly increasing that of abstract workers and leaving that of low-skilled workers unchanged. However, these estimates become smaller and statistically insignificant once we control for firm size at baseline, and industry and commuting zone fixed effects (second set of columns). We should expect a weaker association between T&O and changes in the employment shares of medium- and highly-skilled workers (relative to changes in those of routine and abstract jobholders), due to the particularities of the German education system where about 45% workers who have completed apprenticeship training (our medium-skilled category) are employed in abstract occupations that would require college training in the U.S. and UK (see also Panel B of Table 2). Panel C of Table 3 clearly reveals an age bias of T&O by showing the relation between T&O and changes in the employment shares of younger (aged below 30), medium-aged (aged between 30 and 49), and older workers (aged 50 and above). That is, although T&O does not alter the employment share of younger workers, it sharply increases that of medium-aged workers while sharply lowering that of older workers. Lastly, we perform a placebo test in which we regress changes in the taskage specific employment shares on both contemporaneous and future T&O with the sample restricted to firms who participated in the IABEP over two consecutive 3- year periods (Table 3 Panel D). The estimated effects of contemporaneous T&O for this restricted sample are similar to those for the full sample (Panel A). Future T&O, in contrast, has little impact on firm workforce composition, suggesting that our baseline results are not simply picking up differential firm specific trends or reverse causality T&O, Firm Growth, and Reshuffling A reduction in the routine employment share in response to T&O does not necessarily imply the destruction of routine jobs in restructuring firms; if such firms 16

18 grow in size, for example, their total employment of routine jobholders may well increase even though their routine employment share declines. A reduction in the latter, accompanied by an increase in the abstract employment share, could also be driven by routine jobholders switching to abstract occupations within the firm (internal reshuffling). In Table 4, therefore, we provide initial evidence on the relation between T&O and firm growth, hiring, internal reshuffling, and worker separations. As the table shows, conditional on year, commuting zone, industry fixed effects, and baseline firm size, each additional adopted organizational change is associated with a slight (statistically insignificant) increase in mean firm wages and firm size, a slight (statistically insignificant) decrease in separation rate (the number of workers leaving the firm over the 3-year period divided by employment at baseline), and a modest (statistically significant) increase in the external hiring rate (the number of workers joining the firm over the 3-year period divided by employment at baseline). Overall, there is therefore little evidence that firms that implement T&O grow differentially from firms that do not. On the other hand, the estimates on internal reshuffling (number of continuing employees who changed task over the 3-year period divided by baseline employment) provide clear evidence that T&O is associated with more internal task reshuffling. 3.2 Effects of T&O on Jobholders Our analysis so far revealed that T&O reduces firm demand for routine-based jobs. But how are the workers who held routine jobs before T&O implementation affected by T&O? Do they exit into under- or non-employment or do they successfully upgrade to abstract occupations? If the latter holds true, does this upgrading take place within the firm and how do these transition rates differ by workers age? We provide the first overview of T&O s effects on worker career paths over the 3-year period using regressions based on equation (3) that relate 17

19 organizational changes in the baseline firm between t-3 and t to the career outcomes of the workers at time t employed in these firms at baseline (in t-3). All regressions control for year fixed effects, commuting zone, and the industry fixed effects of the baseline (t-3) firm, baseline firm size, gender, foreign status, baseline age, a preestimated firm wage fixed effect for the baseline firm and a pre-estimated worker wage fixed effect, the latter two jointly estimated over a 7-year period prior to T&O adoption Transitions by Task Usage at Baseline For routine jobholders at baseline (Table 5, Panel A, row (2)), the results show that, in line with our findings that T&O reduces the firm s routine employment share (Table 3, Panel A), T&O also decreases the probability that routine jobholders will continue to be employed in routine occupations (column (2)). Yet T&O appears to neither affect the probability that these workers will transit into non-employment (column (1)) nor increase the probability that they will separate from the firm (column (4)). Rather, and in line with the evidence in Table 4, T&O increases the probability of a routine jobholder being assigned to a different (manual or abstract) occupational group within the same firm (column (3)). This effect is not only highly statistically significant but also large in magnitude since the overall probability of switching tasks within the same firm over a 3-year period of 1.61%. Moreover, conditional on workers being employed in both periods, T&O is associated with increased wage growth of those workers holding routine jobs at baseline (column (5)) 14. Overall, the estimates provide little support for the contention that T&O harms the careers of routine jobholders either in absolute terms or relative to manual or abstract jobholders, in large part because it increases the probability of their undertaking different tasks within the same firm. 14 In the wage growth regression, we do not condition on worker and firm wage fixed effects. 18

20 3.2.2 Transitions by Age at Baseline To measure transitions by age, we repeat the analysis in Panel B of Table 5, now distinguishing between three age groups: younger worker (aged below 30), medium aged workers (aged between 30 and 49), and older workers (aged 50 and above). In line with our conclusion that T&O decreases the employment share of older workers (Table 3, Panel C), we find that T&O reduces the probability that these workers will continue to be employed in the same occupational group. The results further suggest that most older workers who move away from their baseline occupation in response to T&O transit into non-employment. More specifically, each additional organizational change lowers the probability of an older worker remaining employed between t-3 and t by percentage points, an effect roughly equal to that of one additional organizational change on the probability of older workers leaving the task performed at baseline. T&O also increases the probability that older workers will move to a different occupational group within the same firm; however, this effect is much smaller than that for routine jobholders. For medium aged workers, in contrast, T&O not only improves employment probabilities but strongly increases the probability of occupational switching within the firm, while reducing the probability of exit. To better understand which age groups among older workers are particularly affected by T&O, we further distinguish between those aged 50 54, 55 59, and >59 (Table 5, Panel C). Our estimates indicate that T&O has little impact on the employment probabilities of workers in the age group, but rather increases their probability of moving to a different occupational group within the same firm. This finding contrasts sharply with T&O greatly increasing the transitions into non-employment and separation of workers over 54 while inducing no increase in their transitions into other occupational groups within the same firm. The above observations highlight that, although occupational switching within the firm helps to cushion T&O s possible adverse effects on routine jobholders, it only 19

21 benefits workers younger than 55. Those over 54 not only respond to T&O primarily by separating from the firm and moving into non-employment, but (according to our longitudinal data) most such movements are permanent and likely to represent transitions into early retirement Transitions of Routine Jobholders by Age To explore such worker transitions in more detail, we focus in Table 6 on workers employed in routine occupations at baseline. A remarkable 75% (0.183/0.242) of routine jobholders affected by T&O respond to T&O by upgrading to an abstract occupation, about two-thirds of them (0.122/0.183) within the same firm (Panel A). T&O s effect on transitions into non-employment or manual work, in contrast, is statistically insignificant. This pattern holds for both 3-year and 6-year transitions. Splitting up these routine jobholders into the same three age groups as before (Table 6, Panel B) further reveals that T&O increases upward movement from routine to abstract occupations for all age groups, but particularly for workers below 50. Whereas for workers under 30 upgrades occur to a similar extent both within and between firms, for workers aged 50 and above, almost all upgrades into abstract occupations take place within the firm. In this older age group, T&O s effect on upgrading from routine to abstract tasks varies widely by age (Table 6, Panel C), with nearly all upward movement driven by workers under 55. Routine jobholders aged 55 and above, in contrast, are more likely to permanently move into nonemployment following T&O Firm Training Activities All our results so far imply that firms can play an important role in actively curtailing T&O s harmful effects by offering upgrading opportunities to routine jobholders. This interpretation is further supported by the results in Table 7, documenting that T&O is associated with a rise in firms internal and external training activities (conditional on year, commuting zone, and industry fixed effects 20

22 and (log) baseline firm size). Not all workers benefit equally, however. Training activities increase for medium- and high-skilled workers much more than for lowskilled workers. Unfortunately, our data do not allow us to distinguish firm training activities by age or task Labor Market Transitions of Older Workers The findings in both Tables 5 and 6 underscore that T&O results in routine jobholders 55 and older transitioning predominantly into non-employment while benefiting little from a larger number of firm training opportunities. These observations suggest that for firms and workers to undertake investment in new skill acquisition, the period over which this investment amortizes must be sufficiently long, which makes it not worthwhile for routine jobholders aged 55 and over who are near retirement age. Older workers might also find it more difficult to learn the new skills required after T&O implementation. For a better understanding of how older workers more generally respond to T&O, we next explore whether T&O also increases non-employment rates for older abstract jobholders or whether a university education shields older workers from T&O s possible adverse effects. According to Table 8, T&O increases the probability that workers over 55 will separate from the firm and move to nonemployment regardless of whether they were previously employed in a manual, routine, or abstract occupation. These effects are as strong for high-skilled workers as for low- or medium-skilled workers. Hence, neither employment in a (previously) abstract occupation nor a university education appears to cushion T&O s adverse career effects on older workers. Rather, our results support the notion that older workers do not generally possess the workplace skills necessary to deal with such changes, and investing in new skill acquisition seems not worthwhile, for firms or workers, when workers are close to retirement age. 21

23 3.2.6 Robustness Checks Our estimates in Tables 5, 6 and 8 condition on (pre-estimated) firm and worker wage fixed effects (f j and θ it 3 in equation (3)), in addition to firm and worker characteristics at baseline such as the firm s industry affiliation and size and the worker s age. The former two control variables eliminate two important sources of potential biases: First, restructuring firms may be high wage firms offering more upgrading opportunities to their employees irrespective of T&O. Second, restructuring firms may employ high wage workers who are likely to move up to higher paying abstract jobs irrespective of T&O. To provide additional support for the hypothesis that the associations between T&O and routine jobholders upward movement from routine to abstract occupations and T&O and older workers movements into non-employment reflect a causal relation, we perform an event study as a robustness check. We summarize these findings in Figure 3. The figure contrasts firms that do not introduce T&O between t-3 and t but implement at least two changes between t and t+3 (treated firms) with those that carry out no changes over the entire t-3 to t+3 period (control firms). Figure 3A plots the number of organizational changes implemented by the treatment and control firms in each 3-year period, between t-6 and t+6. By construction, neither treatment nor control firms implement any T&O between t-3 and t (the pretreatment period). In the treatment period (between t and t+3), treatment firms implement 2.6 changes on average, compared to no changes in control firms. Treatment firms also introduce more organizational changes than control firms in the t+3 to t+6 post-treatment period (1.3 vs 0.4 on average). Figure 3B shows that upgrading from routine to abstract occupations in treatment relative to control firms predominantly occurs in the t to t+3 treatment period, during which treatment firms adopt at least two organizational changes while control firms do not adopt any. Figure 3C illustrates that the share differences between treatment and control firms for transition into non-employment by older 22

24 workers are small in the t-6 vs. t-3 and t-3 to t pretreatment periods, but open up in the treatment t vs. t+3 period, which supports the hypothesis that transitions into abstract occupations and non-employment are indeed caused by T&O. 3.3 Task Upgrading, Education and Labor Market Institutions Overall, our findings paint a more positive picture of T&O s welfare effects than usually discussed. Facilitated by firms training opportunities, routine jobholders aged below 55 affected by T&O respond to T&O by moving from routine to abstract occupations, rather than to non-employment or to inferior jobs. An important question is whether this upgrading is facilitated by particular institutional and firm characteristics. Facilities inside the firm that support retraining of workers may be a first factor that promotes upgrading following T&O. For example, firms that routinely train young workers may find it easier and less costly to provide the necessary support for more experienced workers to upgrade to an occupation with more abstract skill content. Germany has a well-established youth training scheme, often referred to as the apprenticeship system that educates nearly 65% of school leavers and equips them with both practical skills through hands-on training inside a firm and academic skills through classroom-based learning in vocational schools. To qualify as a training firm, firms need to fulfill certain conditions, such as employing qualified training personal (see Dustmann and Schönberg, 2012, for details). We might therefore expect that firms that train a larger share of young workers through the apprenticeship scheme are also more likely to retrain adult workers when they carry out T&O. To investigate this hypothesis, we exploit variation in the employment share of apprentices in the firm at baseline (i.e., the number of apprentices divided by all employees in year t-3). We regress workers career outcomes at time t on T&O implemented by the baseline firm between t-3 and t, the share of apprentices at 23