Matching skills of individuals and firms along the career path

Matching skills of individuals and firms along the career path Abstract Research since Gary Becker equated specific human capital with firm-specific human capital. This paper divides firm human capital into a specific and a general component to investigate the relationships between firm- and occupation-specific human capital and job switches. Applying the task-based approach, the results show that the degree to which firm knowledge is portable depends on task similarities between the firms. Firm- and occupation-specific knowledge are negatively related to wages in a new job, but achieving a good occupational, instead of firm, match is more important for employees. The occupational intensity, reflecting the overall knowledge composition on the firm level, decreases with experience and can outweigh the costs of covering long task distances. As regards matching, a small distance between the firm and occupational tasks matters for medium-skilled employees. 1. Introduction It is well established that firm-specific knowledge increases with firm tenure and that it is lost when employees switch employers (Becker, 1964). Many studies use firm tenure as a proxy for accumulated firm-specific knowledge. The question remains, however, as to what exactly this knowledge is and whether all firm knowledge is specific and, thus, not transferable. The skill-weights approach (Lazear, 2009) models firm-specific knowledge by letting firms place different weights on general skills. This assumes that a certain amount of all knowledge is transferable across firms. Gathmann and Schönberg (G&S, 2010) test this approach for occupation-specific knowledge by investigating the relationship between occupational knowledge and wages. Following Lazear’s theoretical arguments, G&S show empirically that the amount of specific knowledge and, thus, the number of portable skills, varies between occupations and along the career path. To date, there has been no investigation of the degree to which firm knowledge is portable across establishments. Studies often argue that, in the case of a firm switch, all firm knowledge is lost; thus, the likelihood of firm switches decreases with the accumulation of firm knowledge as measured by increasing firm tenure. However, the importance of firm knowledge as an obstacle to switches would be overestimated if part of the firm knowledge was transferable. Therefore, the aim of this paper is to discover how the firm distance of moves (knowledge differences between firms) varies along the career path and how it relates to wages. In addition, the relative importance of firm and of occupational knowledge for wages is determined, providing an indication for switchers whether it is more important to find a good firm or a good occupational match. Regarding the incentives to switch, the analysis considers on-the-job learning possibilities as proxied by two new variables for the specific knowledge structure of firms: the share of occupational peers on the firm level (occupational intensity) and the match of the firm and occupational task sets (firm-occupation distance). This allows determining to what degree knowledge that is specific to firms can benefit workers who switch jobs. I begin by formally modelling the relationship between specific, non-transferable knowledge and labour mobility. A combination of the approaches of Lazear (2009) and G&S (2010) allows developing one that accounts for firm and occupational knowledge as well as occupational intensity. Both knowledge types can be decomposed into a transferable and a specific component. The model yields several predictions which can then be investigated in the empirical analysis. As stressed by Lazear (2009), firm knowledge is understood as a unique arrangement of skills in that the skills taken alone are not specific but become so through their combination.1 This variation is lost in empirical analyses that only look at the occupational level because, opposite to my analysis, individual-level data is often unavailable. The predictions are empirically tested with the task-based approach that analyses which tasks are performed on the job (cf. Autor et al., 2003; Poletaev and Robinson, 2008; G&S, 2010). The data used for the empirical analyses are from three sources. The Sample of Integrated Labour Market Biographies (SIAB) covers a representative sample of the working population in Germany. Information about firms is drawn from the Establishment History Panel (BHP). Details on occupational skill sets and tasks are provided by the BIBB/BAuA Employment Survey 2006. As a preparatory step, a factor analysis is applied to categorize tasks into groups. Firm knowledge is measured, first, as the task composition of firms which is determined via the occupational task composition of the workforce. Second, the weight of tasks’ combinations shows the importance of single occupations, instead of single tasks (occupational intensity). The first procedure disaggregates occupational information to the task level, while the second procedure allows accounting for the importance of having several tasks embodied in one occupation. Third, the differences between a worker’s and a firm’s task composition are identified to measure an overlap of knowledge (firm-occupation distance). As knowledge naturally differs by qualification levels, this is accounted for by analysing different qualification groups separately. The results are in line with the predictions, confirming that long-distance firm switches occur more often early in the career than later on. Which type of knowledge can be transferred and to what extent depends on the qualification level of workers. In the case of joint firm and occupation switches, firm- and occupation-specific knowledge both matter for wages. The present evidence indicates that firm-specific knowledge matters less than occupation-specific knowledge. Interestingly, it can further be shown that individuals start work in firms that have a relatively higher share of employees in the same occupational group, e.g. a high occupational intensity, and that this share decreases with increasing work experience. Taking this a step further in the OLS estimations reveals that a lower occupational intensity is associated with higher wages. However, workers might strategically select into firms with a certain occupational diversity. To control for this selection problem, occupational intensity is instrumented with occupational diversity on the industry level. Contrary to the OLS results, occupational intensity now shows a positive effect on wages for medium- and high-skilled workers and becomes insignificant for low-skilled employees. Surprisingly, the sizes of the coefficients, which are calculated using standardized variables, suggest that the benefits of occupational intensity can clearly outweigh the costs of both distance measures. This effect supports the notion that a high occupational intensity shows on average a higher demand for an occupational task set which is reflected in higher wages. The overall results generally hold for both men and women. In addition, firm-occupation distance starts at similar levels across qualification levels at the beginning of the employment biography. It then increases for low-skilled, remains stable for medium-skilled, and decreases for high-skilled workers. When estimating 2SLS regressions, medium-skilled workers are affected by higher costs for higher distances while low-skilled seem unaffected. The paper is embedded into the literature on skill-weights and on the task-based approach. Lazear (2009) suggests that all human capital is general; it becomes specific through weights that are firm specific. His approach is novel because, compared to earlier work (cf. Becker, 1962), there is no longer a clear, exogenously given distinction between general and specific human capital and instead they are defined endogenously with observable market parameters. First, empirical tests of Lazear’s model show that the investigated predictions of the model are borne out by the data, making the model a worthwhile basis for the present analysis (Backes-Gellner and Mure, 2005; Geel et al., 2010). The approach of this paper also relates to Neal (1999), who stresses the prevalence of complex job switches, involving a higher degree of changes as measured by simultaneous industrial/occupational (career) and firm moves at younger ages. Also, Neal suggests that careers are chosen first, followed by selection of firms. Although my analysis measures the moves simultaneously and may not make statements on the sequence of job decisions, the relative coefficient sizes confirm that, in terms of the final decision, career changes have a larger impact on wages than firm changes. To date, to estimate which types of switches are more costly, scholars analyse whether specific knowledge is more tied to occupations, firms, or industries. For instance, some scholars are more in favour of industry-specific human capital (Neal, 1995; Parent, 2000) while others prefer the idea of occupation-specific human capital (Poletaev and Robinson, 2008; Kambourov and Manovskii, 2009). Parent (2000) acknowledges that industry-specific human capital might measure something similar to occupation-specific human capital. Pavan (2011) criticizes that the importance of firm-specific human capital is regularly underestimated. With the exception of Poletaev and Robinson (2008), all studies mentioned above use tenure variables for industries, firms, and occupations, which prevents that knowledge can be divided into a sticky and a portable component. The advantage of task data is that they allow moving away from such a generic classification of specific or general skills and measuring instead the degree of similarities (general) or differences (specific) between portfolios. G&S (2010) are first to make use of this approach by combining task-specific human capital and a skill-weights approach to investigate job mobility. Although several of their wage regressions look at simultaneous firm and occupational switches, there is no explicit investigation of the differential impact of firm knowledge. The present paper fills this gap by being the first to conduct an analysis of specific versus general human capital on the firm level using the skill-weights approach, providing a new measure for firm knowledge. In addition, it moves beyond measures of specific knowledge to also capture the impact of a unique skill composition in firms. While specific human capital on the individual level is regarded as costly in the case of a job change, a specific knowledge structure on the firm level could motivate workers to cover long distances. Keeping in mind the transferability of knowledge, the question arises as to what degree careers might be structured to minimize the loss of knowledge and, simultaneously, optimize learning potential and wages. The majority of workers receive additional on-the-job training (Mincer, 1962; for an overview, 1989). Studies have estimated varying incidences of on-the-job training, but there is clear evidence for the importance of formal and also informal training (for an overview, Barron et al., 1997). Naturally, what can be learned depends on the job chosen and the workforce composition at the firm, as shown in organization research (e.g., Baron, 1984; Hannan, 1988), which has long emphasized that organizations shape jobs, e.g. through their size, growth, demography, technology, and unionization. Occupational intensity can be interpreted as a signal sent by the firm that shows either how large the firm’s market power and hence bargaining power is (reflected in lower wages) or how much a firm values an occupational group (as reflected in higher wages). Also, the match between firm and occupation tasks is directly related to occupational intensity but also takes into account other workers in the firm. It reflects the similarities of the own task profile with the complete task composition of the workforce in a firm. In more detail, from a supply-side perspective it can be argued that being around a larger peer group might be costly—because it reduces the uniqueness of own knowledge—but important at the beginning of an occupational career to develop own skills. Learning processes are time-consuming and can hence be costly, as individuals and firms cannot immediately reap the benefits of the investments. As wages decrease, occupational intensity increases (firm-occupation distance decreases). After some years, the importance of workers in the same group will decrease and potentially turn to zero because individuals have accumulated more knowledge, in sum providing fewer incentives to accept costs associated with a high occupational intensity (low firm-occupation distance) and potentially encouraging switches to firms with lower occupational intensities (higher firm-occupation distances). Occupational intensity (firm-occupation distance) should therefore decrease (increase) along the career path and wages should increase as the occupational intensity decreases. Opposite to this but in line with Bidwell and Briscoe (2010), a demand-side perspective suggests that organizations with a higher demand for a bundle of occupational skills also have a higher occupational intensity (lower firm-occupation distance). By providing more complex, challenging task combinations and being willing to also pay higher wages, the firms become very attractive for workers. Assuming that with increasing experience workers become even more attractive to firms would imply that occupational intensity increases (firm-occupation distances decreases) with experience. It further follows that wages increase as occupational intensity increases (firm-occupation distance decreases). Theoretically it appears plausible that both the supply and the demand effect are at work, but which one dominates is an empirical question. The remainder of this paper is organized as follows. Section 2 presents the conceptual framework. In Section 3, the data set and variables are introduced. Section 4 contains the results and a discussion of their implications. Section 5 concludes. 2. Conceptual framework To investigate the relationship between specific human capital and wages, I take as a starting point the conceptual framework by G&S (2010). The final framework allows disentangling the effects that occupational and firm knowledge have on worker productivity. It further helps to interpret the different roles that firm knowledge can play for switchers as regards costs and benefits, which are addressed in terms of firm-specific knowledge and knowledge composition of firms. Both firm and occupational knowledge are divided into a specific and a general component. In the following paragraphs, the focus is on changes to the framework of G&S that were implemented to incorporate firm knowledge.2 Ultimately, the idea is that the previous calculations with task data overestimate the importance of occupational knowledge by disregarding firm knowledge. As regards the terminology, Acemoglu and Autor (2011) suggest distinguishing skills from tasks because ‘a skill is a worker’s endowment of capabilities for performing various tasks’ (Acemoglu and Autor, 2011, p. 1045). They cannot, necessarily, be taken to be equivalent. Now, Lazear refers to skill weights, which in G&S’s approach are labelled task weights. Nonetheless, in light of the similarity of both approaches, it appears reasonable to assume that the idea behind both models is the same and, thus, does not change with labels. Preference is given to tasks because this is also in line with the methodology chosen in the empirical section. It is also acknowledged that the individual has one knowledge base and will not necessarily distinguish between occupational or firm knowledge although they may be of different value. However, in many data sets, information on the individual level is not available and instead researchers rely on information that is added on the occupational level from other sources. Nonetheless, individuals in the same occupation differ as regards the overall accumulated human capital because of different work experiences at different firms, introducing additional variation across individuals within the same occupation. As sketched above, firm knowledge is understood as the result of a unique variation of otherwise general skills, implying that both occupations and firms weigh skills in unique ways that result in a unique skill portfolio. Also, from a firm’s perspective it is of interest what the individual has learned through experience in another firm as regards organizational practices when recruiting personnel. It is hence assumed that firm knowledge can enter in a wage equation separately. In the end, there will be one price but two components contribute separately to it. To illustrate the basic idea, the two knowledge types will be handled separately in the following framework. Suppose that the output in a job is determined by fulfilling a variety of general tasks that become specific by the relative importance attached to them in an occupation and in a firm. Following Lazear and G&S, my approach uses two tasks j, which can be interpreted as analytical and manual tasks (=A, M). The productivity (S) of a worker (i) varies by jobs (g) and by the time spent in the labour market (t). The jobs are a combination of the occupation (o) and the firm (f) for each task (with, for instance, βg=βo+βf). The relative weight β shows the importance of tasks in an occupation o or firm f. G&S suggest that the importance corresponds to the time spent on that task. Worker i’s productivity (measured in log units) in occupation o at firm f and at time t is   ln Sifot=[(βo+βf)tioftA+(1−(βo+βf))tifotM]+αfoXfot. (1) G&S’s original equation ( ln Siot) is adapted to additionally capture the firm dimension ( ln Sifot) in terms of the weights attached to tasks. The task composition on the firm level results from the occupational task structure of the workforce, reflecting an interaction between the occupational and firm level. As mentioned before, this allows capturing the idea that all knowledge is general and that both firm and occupation can attach different weights to this knowledge. In other words, the analysis decomposes the βs into a firm and occupational component which jointly determine the worker’s productivity. It further allows moving a step closer to what the workers know on the individual level because different influences of the career path are taken into account. For instance, a worker might hold an occupation that is specialized in manual tasks but choose to work in a firm that places more weight on analytical tasks. The worker’s skill portfolio then changes while working at the firm when compared to a worker with the same occupation but whose employer puts a higher weight on manual tasks. Again, this variation would normally not be captured in the empirical analyses. It is now possible to calculate the absolute distances between current and previous firms (occupations) by comparing their task weights, for instance, |βf−βf'| ( |βo−βo'|). The more similar firms (occupations) are, the smaller the absolute difference. Although the empirical analyses consider multiple tasks, it is sufficient to consider only two at the moment to illustrate the logic behind the analytical setup. While the weights for task-specific human capital vary on the individual level, worker productivity can also be influenced by the knowledge composition of the firm, as argued above. In other words, task productivity not only arises from the weights placed on individual tasks but also from the weights of the tasks’ combinations. This implies that firms might view certain skills as complementary and prefer these to be used in combination on the individual level. A combination that is very relevant to workers is the one that corresponds to the own occupation and which is captured by Xfot as the share of workers in the same occupational group in a firm or the differences between the own and the firm’s task composition. As outlined above, the relative size of occupational peers or the degree of similar tasks at the current firm can but does not need to be of advantage; the direction of the relationship is determined by the parameter αfo. It takes on positive values ( αfo>0) when the share of firms prepared to pay higher wages for a unique skill composition is larger than the share of firms that pays lower wages, reflecting the trade-off that firms face between attracting suitable workers (higher wages) and making use of their own market power (lower wages). A corollary that can be tested in this context is that as experience increases, occupational intensity/firm-occupation distance can either decrease or increase. Xfot also influences the previously accumulated human capital, as outlined below in eq. (3). Next, the worker’s task productivity in a job tigtj (where go,f}) needs to be determined with   tigtj=tij+γgHigtj(j=A,M), (2) where tij describes the ability of worker i in a certain task j (initial endowment). Higtj includes all previously accumulated human capital of worker i in task j in different jobs. This equation illustrates the influence of previously acquired human capital on both the firm and occupational level, which is combined into one job. Hence, in contrast to G&S, I account for variation on the firm level and, correspondingly, on the occupation level, which is necessary if I want to investigate the difference between firm- and occupation-specific human capital. The equation incorporates the idea that workers gain more knowledge on the job. The degree to which this can be achieved in a certain task t depends on the importance of βg, which is assumed to be captured with the time spent on a task. The more experienced workers are, however, the lesser they can learn. This can be written as   HigtA=βg′Fig′tHigtM=(1−βg′)Fig′t︸ experiencein  priorjobs, (3) where Fig't is the experience of worker i in previous jobs, as indicated by the prime. Since all knowledge is general in this approach, acquired human capital is a function of the combination of task weights on the firm and occupational level. Combining the equations above gives   lnSifot=γfo[(β0+βf)HifotA+(1−(β0+βf))HifotM]︸Tifot+(β0+βf)tiA+(1−(β0+βf))tiM︸mifot+αfoXfot, (4) where γfo measures the returns to task-specific human capital of jobs. Tifot can be observed as a time-variant measure of task-specific human capital; and mifo is the unobservable match to the job that does not vary over time. The equation further includes Xfot, which represents occupational intensity. To investigate labour mobility, wages in different jobs need to be compared. These are determined by multiplying the productivity with the skill or tasks prices of jobs  Pfo that is wifot=Pfo*Sifot. Next, the equation is logarithmized and yields the following expression:   lnwifot=(pfo︸skill pricejob+γfoTifot +mifo︸ln Sifot)+lnαfoXfot︸occupational intensity, (5) where pfo=lnPfo ( pfo=lnPfo). Equation (5) can be used to investigate labour mobility of workers. Like Lazear, G&S suggest a two-period setup where the worker has to decide whether to stay or to switch jobs in the second period. A job switch occurs when   ln wifot︸wages in new job >ln wif'o't︸wage in previousjob . This equation can be rearranged as follows:   pfo-pf'o'+mifo-mif'o'+lnαfoXfot-lnαf'o'Xf'o't+γfoTifot>γf'o'Tif'o't, (6) which shows that what is paid for task-specific human capital in the previous job must be exceeded by the sum of the returns to task-specific human capital in the new job, the difference of skill prices, of the task match, and the improved learning environment. To illustrate the influence of the βs the equation can be rewritten as   pfo-pf'o'+γfo-γf'o'Tif'o't︸wage growth in job+mifo-mif'o'︸task match+lnαfoXfot-lnαf'o'Xf'o't︸changes in task environment>-γfo(βo+βf)-(βo'+βf')HifotA-HifotM︸-γfoTifot-Tif'o'tloss in human capital. (7) The right-hand-side term in eq. (7) shows the loss in task-specific human capital where one can again see the influence of the difference between the βs. The left-hand side is the sum of the difference of the firm task match, the wage growth attributable to an increase in skill prices, and the returns to previously acquired task-specific human capital. In addition, it shows the advantages resulting from a good match in terms of the task environment. When the weights of the knowledge composition are of disadvantage, the term for the task environment could be moved to the right-hand side of the equation, as it adds to the costs associated with switches. 2.1 Empirical predictions The analytical setup yields the following intuitive results, part of which were tested for the case of occupational human capital by G&S but, according to my argument, should simultaneously matter for human capital at the firm level. Although all these predictions should hold equally across qualification levels, the overall knowledge base is expected to differ by the type of knowledge base. It is to be expected that what determines switching behaviour, like the distance of a switch, also differs by qualification level. That is why the qualification categories are investigated separately in the empirical analysis. The first prediction states that less task-specific human capital is lost when the switch takes place between firms that are more similar with regard to their task composition. Therefore, switches occur more often between similar firms. Second, the distance covered in a switch will be the highest early in the career because the likelihood that the left-hand side of eq. (7) is greater than zero becomes less likely with experience. Specifically, during early years of employment, workers are still looking for their best possible match, which might include a certain amount of trial and error. After having spent a longer time in the labour market, workers are less likely to travel long distances because, possibly, they have already found a good match, and as they have accumulated more knowledge, switches become more costly. Third, wages at the source firm are expected to be a better predictor of wages in the target firm if both positions require similar tasks. This follows from the idea that with a higher number of transferable skills, a better match is achieved because distances are shorter. Fourth, both occupation and firm knowledge matter for wages. When investigating joint switches (that is, simultaneous firm and occupation switches), I thus take into account both knowledge types to compare their relative importance. Finally, the task match of the worker and the firm can affect wages negatively because of a time investment in learning by the employee. The theoretical counterargument suggests a positive relationship, showing that a set of occupational skills is more highly valued by firms. The direction of the effect depends on the parameter value of αfo, which will be determined in the empirical analysis. The analysis carried out in this paper is innovative in its methodological measurement of firm knowledge, providing a direct empirical measure for firm-specific knowledge, and it incorporates a unique test of potential (dis-)advantages in terms of learning opportunities arising from the occupational composition of firms. 3. Data 3.1 Data sources Three data sources are accessed for the analysis. The first data set is the weakly anonymous Sample of Integrated Labour Market Biographies (SIAB). Data access was provided via on-site use at the Research Data Centre of the German Federal Employment Agency at the Institute for Employment Research and subsequently remote data access. The SIAB contains a very long observation period (1975–2008) and information on labour market histories of 1.5 million individuals in Germany (Dorner et al., 2010). It is the most comprehensive administrative micro-level data set on employment histories currently available for Germany. In addition, it is possible to link the establishment information of the Establishment History Panel (BHP) to the SIAB. This combination of individual labour market histories (SIAB) and firm employment structure (BHP) makes the data perfectly suited for this analysis. The SIAB provides information on wages and occupations of individuals, and the BHP has information on the occupational categories of all employees in a firm. A detailed description of the final data set is included in the Appendix. In short, I restrict the analysis to men, to employees with an average daily wage of at least 10 euros, and to voluntary switches. The lower the wage is, the higher the likelihood that individuals may hold another job to make a living. For instance, self-employment is not reported to the Research Data Centre of the German Federal Employment Agency and can hence not be controlled for. To identify and later exclude involuntary switches, I start with job switches where simultaneously structural changes occurred in the firm, for instance, a change of ownership or the firm’s exit from the market. This group is augmented with other involuntary switchers, who are identified by receiving unemployment benefits immediately after leaving the firm. Note that in Germany, workers who give notice, in contrast to being given notice, may not receive unemployment benefits for three months. Voluntary switches are more likely to purposefully decide on new firms, while involuntary switchers are more likely to be forced into their new job and hence willing to accept any firm environment. The classification of individuals and firms according to their task sets requires, of course, information on tasks. The BIBB/BAuA Employment Survey 2006 (Hall and Tiemann, 2006; Rohrbach-Schmidt, 2009), which was undertaken in 2005 and 2006 by the Federal Institute for Vocational Education and Training (BIBB) and the Federal Institute for Occupational Safety and Health (BAuA), provides all necessary information. This wave consists of a random sample of 20,000 people who are active in the labour force in Germany. In addition to individual-specific data, the survey includes information on the task requirements of occupations. For further examples using this database, see Spitz-Oener (2006, 2008) and Borghans et al. (2014). The BIBB/BAuA data are merged by occupation (SIAB) or occupational groups (BHP). 3.2 Variables The dependent variable is the logarithm of wage, as proposed in the analytical setup. Wage is measured as gross daily income of employees and reported in euros. Occupational intensity—the share of occupational peers—is calculated by dividing the number of workers in the same occupational group, using Blossfeld categories, by the total number of workers in the firm. The Blossfeld classification, which is the only available unit for occupations on the firm level in the BHP, is based on the three-digit occupation of an individual as specified by the employer in the notification to the social security agencies. Blossfeld first distinguishes between three upper-level groups, namely, production, service, and administration, and second, ranks occupations according to the type of required skills. Accordingly, blue-collar workers who perform simple manual tasks and white-collar workers who provide simple services are regarded as unskilled; blue-collar workers engage in complicated tasks, white-collar workers perform qualified tasks, and semi-professionals are regarded as skilled workers. The third and most highly qualified group includes engineers, technicians, professionals, and managers. The Blossfeld classification thus assigns upper-level groups and then ranks individuals according to their skill requirements. To address a potential bias in the estimation, I further calculate the degree of occupational diversity on an industry level (28 industries) to create a Herfindahl index and use this variable as an instrument for occupational intensity. The idea is that firms located in diverse industries will exhibit lower shares of occupational intensity. There is, however, no empirical evidence that occupational diversity on the industry level has an effect on wages. A detailed description of the instrument variable Herfindahl Blossfeld can be found in Section 4.3. To measure general work experience, I calculate the number of years someone has worked since labour market entry by using information on the exact number of working days, excluding periods of unemployment. It is common practice in wage regressions to include a squared term for work experience because a concave relationship is in line with changes that occur later along the career path. This specification is more restrictive than suggested by the analytical setup but still in line with the general idea. I distinguish three levels of education in the regressions. Low-skilled workers are defined as those who did not pass the Abitur (German university entrance qualification) and have not completed apprenticeship training. This also includes unskilled workers. Medium-skilled workers passed the Abitur and/or have completed nothing above an apprenticeship. High-skilled workers hold a degree from a university or university of applied sciences. Incentives to switch firms can be driven by regional characteristics and, therefore, controls for region types are introduced. Additional controls include, as dummies, years, industry, occupational groups, and the logarithm of establishment size. The summary statistics as well as correlations for the most important variables can be found in Tables A.4 and A.5 in the Appendix. The analyses of employment biographies are carried out separately for men and women because the two groups are known to show significant differences in terms of wages and employment careers. For instance, due to fertility decisions, women are known to behave differently from men in their human capital accumulation over the life cycle. Note that I only report the results for men, leaving the results for women to the discussion at the end of the section. The measures for task distance between occupations, using the BIBB/BAuA data, include both men and women (see Section 4.1). In the regressions, all variables are standardized. 4. Analysis 4.1 A task-based measure for specific human capital of firms The main variable of interest is a measure of the firm- and occupation-specific human capital. G&S group tasks manually into three categories: analytical, manual, and interactive. This categorization makes it possible to combine tasks from different years. As they show, the task content of occupations changes only slightly. In contrast, this paper lets the data structure determine the task groups, which has the advantage of allowing me to take into account more tasks because they do not have to be included in every survey wave. The disadvantage is that this procedure cannot be carried out with every survey because tasks do, and therefore factors would, vary. Thus, I rely on G&S’s result that, over time, task variation in occupations is low, and I instead use a factor analysis. Here, a principal factor analysis shows whether certain tasks need to be clustered on the occupational level in latent variables. The first advantage of this procedure is an easier interpretation of the data due to condensed information and orthogonal factors. In addition, since task level is determined by executing a task regularly or by the degree of expert knowledge required, it takes more than a high value in one task to end up with a high value in a factor. Thus, the factor reflects the task level in a certain domain and the level can change through adjustments of different tasks. Indeed, using exploratory factor analysis on high-dimensional task data is in line with Green (2012), who also discusses the risks associated with classifying tasks by hand. In addition, my analysis does not need to reproduce existing categorizations nor interpret the results in the framework of routine versus non-routine tasks, as done by Autor et al. (2003). Indeed, Rohrbach-Schmidt and Tiemann (2013) show that significant changes in the questionnaire can bias analyses that combine data from all cross-sections of the German data. In sum, although using task data from different survey waves can be of advantage, it is ultimately a trade-off between (1) exogenously determined task categories to which survey questions from different years are assigned; and (2) endogenously determined task categories from one observation period. In the current context, applying factor analysis (option (2)) is considered to be the more appropriate procedure. A selection of 31 survey questions from the BIBB/BAuA Employment Survey 2006 gives information about tasks applied in the employee’s current job. The closest approximation to tasks of individuals in this context is achieved on the occupational level. It is acknowledged that this procedure only allows measuring variation on a more aggregate level, but this is still an improvement to previous tenure measures. The survey question asks respondents to assess the task level that they use in their current position. The calculations of the factor analysis return seven factor variables that explain around 91% of the total variation in 248 occupations (see Table A.2, for an overview, and Appendix A, for details on the data and computations). The factors are then labelled according to their content, which is the combination of certain tasks, placing most emphasis on the variables that load the highest. This is similar to what Poletaev and Robinson (2008) and Nedelkoska and Neffke (2011) do. The factor labels are: intellectual, technological, health, commercial, instruction, production, and protection. To make the occupational classification more transparent, Table A.3 reports the occupations with the highest and lowest values in each factor. The example occupations set out in the table make intuitive sense, thus confirming the plausibility of the principal factor analysis. For instance, the technological factor has a strong focus on the application of technological and manual knowledge, both of which are characteristics of occupations such as aircraft engine mechanic or optometrist. The health factor is most important for various types of medical practitioners and other occupations in the healthcare system. More routine tasks like producing and manufacturing goods, measuring, testing, and operating machines load highest in the production factor, which is where occupations such as machine operators for dairy and paper products are found. The task composition of the workforce is determined with information on the 12 occupational groups by Blossfeld (1985; see Table A.1) and task data. The occupational classification does not allow seeing whether the firm employs workers in the same three-digit occupation as held by the switcher. From an employee perspective, however, it is unlikely that they have detailed information as to all the occupations of prospective co-workers. Thus, the Blossfeld classification appears to be an adequate indicator of one aspect that is driving a (voluntary) job switcher’s decision. Task factors for each Blossfeld group are calculated as follows. First, the average factor value of each task is determined for all occupations that belong to one Blossfeld group ( tb). These task factors are then weighted by multiplying them with the corresponding number of workers in a firm in that Blossfeld group ( nfb). Since the focus is the structure of the workforce, this value is divided with the sum of all weighted task factors to calculate the relative importance of a task factor in a firm.   Task importance in firms= tb*nfb∑b=1ntb*nfb This procedure captures a firm’s unique task composition, which is translated into task weights on the firm level. As soon as larger differences between firms can be detected, it can be expected that workers’ skill portfolios—although they might be holding the same occupations—will differ according to variations on the firm level. The more similar firms are with regard to the task composition, the more firm knowledge can be reapplied by the worker after a switch. Job switchers are, thus, also included because they are part of the firm’s task structure. This procedure returns the relative importance of tasks in a firm to avoid the notion that firm size drives differences. Next, the distance of firms/occupations is determined by using the angular separation or uncentered correlation of two vectors representing two firms/occupations (for details on the computational method, see Jaffe, 1986; G&S, 2010). The equation is   AngSepgg'=1-∑j=1Jqjg*qjg'∑j=1Jqjg2*(∑k=1Jqkg'2)1/2  Distancegg'=1-AngSepgg', where q is the vector of all tasks in a firm/occupation. The measure is slightly adjusted so that a value of 1 (0) means that the firms/occupations are completely different (identical). This distance measure reflects the differences between firms or occupations with regard to their task-specific human capital. The measure for occupations is calculated on the basis of the original factors of occupations from the factor analysis (see Table A.2). For firms, the relative task importance is compared between origin and target firm. For firm-occupation distance, the factors from the firm and the occupation level are laid against each other. 4.2 Transferability of firm and occupational knowledge In what follows, the analysis always distinguishes between qualification levels of employees. This is important because the amount of human capital and, thereby, general and specific knowledge can be expected to differ between groups. First, the share of switches by different firm distance intervals is calculated. The results in Fig. 1 show that the majority of switches involve low firm distances. The largest share of joint occupational and firm switches occurs in the lowest firm distance interval, confirming that switches occur more often between similar firms.3 For the first two intervals, which cover around 90% of the sample, firm distance appears to decrease when qualification level increases. Possibly, workers with higher qualification levels can be more selective in choosing a suitable target firm or low-skilled employees are to a smaller degree affected by firm distance. The figure also reveals, as a control analysis, that the distribution differs for layoffs which cover slightly longer distances than voluntary switchers. As announced, layoffs are thus excluded from the following analysis. Fig. 1 View largeDownload slide Distribution of joint switches across firm distance intervals Fig. 1 View largeDownload slide Distribution of joint switches across firm distance intervals In Fig. 2, I investigate the relationship between firm distance and different years of work experience for all male employees. The graph shows the predicted margins with confidence intervals when firm distance is regressed on work experience. The more experienced workers are, the smaller firm distance becomes. Across qualification groups, the negative trend turns out to be very similar. In general, low-skilled have the largest average firm distance values, followed by medium- and high-skilled workers. The results could be interpreted as evidence that workers accumulate more specific knowledge or achieve a better match along the career path, providing less incentive to cover larger distances and pay associated costs. Fig. 2 View largeDownload slide The relationship between work experience and firm distance Fig. 2 View largeDownload slide The relationship between work experience and firm distance Following, the relationships between task-specific human capital at the occupational as well as firm level and wages are investigated. The analysis focuses on switchers and builds on the analytical framework by estimating the following equation:   lnwifot= γoDot+γfDft+αXoft+βZifot+εifot, where the dependent variable is the logarithm of wage ( lnwifot), D is the distance on the firm or occupational level, Xoft is occupational intensity (replaced with firm-occupation distance in Section 4.3.2), Zifot is a vector of control variables, and ϵ is the error term. The equation is first estimated using ordinary least squares in the baseline, and in the next section using two-stage least squares. All regressions report coefficients based on standardized variables which are needed to compare the relative contribution of occupational and firm knowledge in explaining the variation of the model.4 Past wages are included as a control measure for the reservation wage and, in addition, interactions between past wages and task distance follow the idea that wages at the source firm are expected to be a better predictor of wages in the target firm if both positions require similar tasks. The estimations further include work experience, work experience squared, firm size, as well as dummies for occupational groups, regions, industry, and years. It is acknowledged—for instance by not claiming causal relationships—that the estimation procedure cannot account for endogeneity in the decision to switch jobs, leading to a potential bias in the estimations. Hence, to at least increase homogeneity in the group, the focus remains on voluntary switchers. Nonetheless, the results continue to provide information on the relative importance of firm and occupational knowledge, which is the goal of this exercise and the preparation for the analysis of (dis)advantages of firm knowledge in the following section. Table 1 reports the OLS results by qualification level, following a stepwise inclusion of the variables. The baseline specification in column A shows that previous wage and current firm size contribute positively to the current wage. Work experience has a positive relation with the current wage, but the coefficient decreases over time. Occupational distance shows a negative sign. I continue by replicating the results by G&S, using only occupational distance (columns B–C).5 As mentioned earlier, their wage regressions usually look at simultaneous firm and occupational switches but they do not include a task-based measure for firm knowledge. Across qualification groups, most variables have the expected signs. Occupational distance decreases the current wage. Except for high-skilled workers (column 13), previous wage correlates positively with the current wage but the coefficient decreases with increasing task distance. Columns D and E complement the previous estimations by including the firm distance variables. With one exception for low-skilled workers (column 4), firm distance matters in addition to occupational distance. With the exception of high-skilled employees, the interactions between the distance measures and previous wages are significant (column 15). Whenever both firm and occupational distance measures are significant, the coefficient of occupational distance is roughly twice as large as the one of firm distance, reflecting that a good occupational match is relatively more important. So far, the results confirm that the newly constructed measure for firm knowledge plays a significant role in explaining wages in target firms. In all specifications, occupational intensity contributes negatively to wage, but further evidence is needed to corroborate this finding. From this I can conclude that the skill-weights approach as implemented by G&S, whose results are replicated in the regressions, can and should be extended to firm knowledge. An occupational measure for specific knowledge picks up part of the story but cannot fully capture the costs of losing human capital. On average, the sum of the coefficients suggests that the total costs of specific knowledge are one and a half times as large as when estimated only with occupation-specific variables. Hence, firm-specific human capital costs on average half of the sum of occupation-specific human capital. At this point, instead of proceeding the same way as G&S did, the focus will now shift to a more in-depth investigation of firm knowledge in terms of on-the-job learning. Table 1 Specific knowledge—distance of switches and the correlation of wages (OLS) Depvar: Current wage (log)  A  B  C  D  E  Low qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.111***  0.110***  0.155***  0.109***  0.167***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Occ intensity  –0.009**  –0.009**  –0.008**  –0.009**  –0.008**  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        –0.001  –0.009***        (0.002)  (0.002)  Firm dist * previous wage          –0.013***          (0.002)  Occ distance    –0.006***  –0.017***  –0.006***  –0.015***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.020***    –0.017***      (0.002)    (0.002)  Firm size (log)  0.095***  0.095***  0.094***  0.094***  0.093***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.800***  0.799***  0.794***  0.804***  0.796***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.600***  –0.600***  –0.598***  –0.602***  –0.597***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.014  –0.003  0.032  –0.220***  –0.179***  (0.082)  (0.081)  (0.082)  (0.063)  (0.063)  Observations  32,306  32,306  32,306  31,516  31,516  R-squared  0.324  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.176***  0.167***  0.233***  0.163***  0.242***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Occ intensity  –0.032***  –0.031***  –0.030***  –0.028***  –0.027***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.014***  –0.015***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Firm size (log)  0.068***  0.068***  0.067***  0.065***  0.063***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.476***  0.464***  0.455***  0.467***  0.456***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.292***  –0.282***  –0.278***  –0.285***  –0.281***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.112  –0.064  –0.029  –0.345***  –0.292***  (0.073)  (0.073)  (0.074)  (0.036)  (0.036)  Observations  100,935  100,935  100,935  98,363  98,363  R-squared  0.406  0.413  0.418  0.415  0.421  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.155***  0.145***  0.152***  0.142***  0.151***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Occ intensity  –0.022***  –0.024***  –0.024***  –0.022***  –0.022***  (0.005)  (0.005)  (0.004)  (0.005)  (0.005)  Firm distance        –0.012***  –0.011***        (0.002)  (0.002)  Firm dist * previous wage          –0.003          (0.003)  Occ distance    –0.035***  –0.035***  –0.033***  –0.034***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.004    –0.003      (0.003)    (0.003)  Firm size (log)  0.075***  0.075***  0.075***  0.072***  0.072***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.306***  0.304***  0.308***  0.306***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.201***  –0.200***  –0.205***  –0.203***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.610**  –0.522*  –0.523*  –0.209  –0.209  (0.289)  (0.282)  (0.282)  (0.138)  (0.139)  Observations  21,669  21,669  21,669  21,381  21,381  R-squared  0.437  0.444  0.444  0.445  0.445  Depvar: Current wage (log)  A  B  C  D  E  Low qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.111***  0.110***  0.155***  0.109***  0.167***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Occ intensity  –0.009**  –0.009**  –0.008**  –0.009**  –0.008**  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        –0.001  –0.009***        (0.002)  (0.002)  Firm dist * previous wage          –0.013***          (0.002)  Occ distance    –0.006***  –0.017***  –0.006***  –0.015***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.020***    –0.017***      (0.002)    (0.002)  Firm size (log)  0.095***  0.095***  0.094***  0.094***  0.093***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.800***  0.799***  0.794***  0.804***  0.796***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.600***  –0.600***  –0.598***  –0.602***  –0.597***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.014  –0.003  0.032  –0.220***  –0.179***  (0.082)  (0.081)  (0.082)  (0.063)  (0.063)  Observations  32,306  32,306  32,306  31,516  31,516  R-squared  0.324  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.176***  0.167***  0.233***  0.163***  0.242***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Occ intensity  –0.032***  –0.031***  –0.030***  –0.028***  –0.027***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.014***  –0.015***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Firm size (log)  0.068***  0.068***  0.067***  0.065***  0.063***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.476***  0.464***  0.455***  0.467***  0.456***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.292***  –0.282***  –0.278***  –0.285***  –0.281***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.112  –0.064  –0.029  –0.345***  –0.292***  (0.073)  (0.073)  (0.074)  (0.036)  (0.036)  Observations  100,935  100,935  100,935  98,363  98,363  R-squared  0.406  0.413  0.418  0.415  0.421  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.155***  0.145***  0.152***  0.142***  0.151***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Occ intensity  –0.022***  –0.024***  –0.024***  –0.022***  –0.022***  (0.005)  (0.005)  (0.004)  (0.005)  (0.005)  Firm distance        –0.012***  –0.011***        (0.002)  (0.002)  Firm dist * previous wage          –0.003          (0.003)  Occ distance    –0.035***  –0.035***  –0.033***  –0.034***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.004    –0.003      (0.003)    (0.003)  Firm size (log)  0.075***  0.075***  0.075***  0.072***  0.072***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.306***  0.304***  0.308***  0.306***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.201***  –0.200***  –0.205***  –0.203***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.610**  –0.522*  –0.523*  –0.209  –0.209  (0.289)  (0.282)  (0.282)  (0.138)  (0.139)  Observations  21,669  21,669  21,669  21,381  21,381  R-squared  0.437  0.444  0.444  0.445  0.445  Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of OLS regressions. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(5) are workers with low, (6)–(10) with medium, (11)–(15) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1 4.3 On-the-job learning The descriptive evidence documents the importance of learning in firms. The BIBB/BAuA survey (N = 15,796 with sample restrictions as defined above) shows that 78.2% of the workers need on the job either a longer training or instruction to carry out their current activities and 60% declare to need special courses or trainings. A total of 78.3% often receive support from colleagues and 58.3% from supervisors. In addition, 23.8% have acquired their skills primarily and 37.3% secondarily through experience. In the latter case, this is the answer that was most often chosen among all options. These results confirm that, for the majority of workers, their previously acquired knowledge did not match perfectly the job that they were carrying out. Instead, additional effort was required to learn, for instance, from colleagues while working on the job. Figure 3 shows the relationship between work experience and occupational intensity for all male employees, based on the margins from a regression. The lower the number of observations is—due to either the qualification level or the decreasing number of observations by work experience—the larger the standard errors become. Nonetheless, the first working years, which are of particular interest, show lower variation and provide an interesting indication of the development of occupational intensity. There is a downward trend across all qualification levels, with the exception of the first working years of low-skilled workers, where the margins increase. Fig. 3 View largeDownload slide The relationship between work experience and occupational intensity Fig. 3 View largeDownload slide The relationship between work experience and occupational intensity To address the relationship between wages and learning opportunities in more details, I estimate two-stage least square regressions. Learning opportunities are measured with occupational intensity or the firm-occupation distance. Both measures are closely related but tell different stories. Thus, a high occupational intensity (large share of co-workers in the employee’s occupation) should relate to a low firm-occupation task distance. This follows from the concepts that (1) firms resemble the occupational structure of industries; and (2) similar (different) firm and occupational tasks imply small (large) differences. Due to the negative relationship between occupational intensity and firm-occupation task distance, I would also expect the sign in front of the coefficient to change when compared to the occupational intensity. This new measure allows me to compare two aspects of on-the-job learning: (1) Do workers benefit from co-workers who show the same or a different task portfolio (occupation perspective, measured with occupational intensity)? (2) Do workers benefit from working in a firm that shows a similar or a different task portfolio (firm perspective, measured with firm-occupation distance)? Naturally, the two aspects are closely related because when all co-workers and the employee share the same occupation, the firm-occupation distance is zero. However, firm-occupation distance also captures similarities with the task portfolios of other occupations, not subsuming them into one group. It is thus likely to show smaller changes than occupational intensity when different occupations are added to a firm. Conceptually, it does not make sense to include both measures in a regression because they both measure knowledge concentration with occupational data (either aggregate tasks in occupations or pure task data), and thus they are investigated separately. Without instrumenting these variables, it is unclear whether the observed relationships result from a specific sorting into firms. As discussed before, workers may select firms that pay lower wages because they expect to profit via on-the-job learning. Alternatively, they could expect higher wages when working in more specialized firms. A priori it is unknown whether occupational intensity then serves as a reliable measure for different learning opportunities. For instance, a larger firm-occupation distance may capture other costs related to switches. A higher occupational intensity may reflect a firm’s specialization instead of pure learning opportunities. To ensure that occupational intensity does not proxy another firm characteristic, I use the Herfindahl Blossfeld index as an instrument for learning opportunities (occupational intensity or firm-occupation distance) to measure occupational diversity on an industry level, as opposed to the standard procedure of measuring industrial diversity on a regional level. A Herfindahl index can be understood as a concentration or diversity measure where the minimum value of 0 corresponds with an equal distribution of shares while the maximum value of 1 describes the concentration on one share. The occupational composition of industries should relate to the occupational composition of firms because industrial and firm labour demand should be highly correlated. In the data, the newly created Herfindahl Blossfeld index is significantly positively but moderately related to occupational intensity (r = 0.2943, see Table A.5) and significantly but only weakly related to firm-occupation distance (r = –0.1325). To my knowledge there is no evidence yet that being in an occupationally diverse industry has a direct impact on wages. I therefore start by investigating several potential connections, which one may theoretically expect, and report the results here. The Herfindahl Blossfeld index shows indeed negligible correlations with all the other variables in the analysis, most importantly with wages (r = 0.0436). Technically, an indirect relationship could exist since workers’ wages can be understood as a function of industry productivity which in turn is determined by firm size and the type of workers needed. The calculations show, however, that the correlation between firm size and the Herfindahl Blossfeld index is very low (r = –0.0295, see Table A.5). Also, since the Herfindahl Blossfeld index is calculated as a percentage measure, it should be independent of the number of workers. There is in addition no evidence for an important correlation between workers’ qualification levels and the Herfindahl Blossfeld index, ruling out that the index serves as a proxy for the positive effect of education on wages. In any case, the regressions are carried out separately for qualification level. To avoid that the final results are driven by unobserved, indirect connections or that the Herfindahl Blossfeld index picks up other industrial or firm characteristics, control dummies for the 28 industries of the Herfindahl Blossfeld index, for firm size and for occupational groups (as a measure for type of worker), are included. Therefore, as the Herfindahl Blossfeld varies on the industry and year level and since the regressions include industry controls, the Herfindahl Blossfeld then picks up variation arising from changes in the workforce composition across years. Looking at the development of the index across years and industries shows clear differences between industries, confirming that the variable captures additional industry-specific characteristics. 4.3.1 (Dis-)advantages of occupational intensity in firms Across all qualification levels, the OLS results suggest that occupational intensity and wages correlate negatively. To address potential endogeneity, Table 2 reports the 2SLS results. The odd-numbered columns (also columns A and C) show the first-, the even-numbered columns (also columns B and D) the second-stage regressions. The specifications are the same as the final regressions in columns D and E of Table 1, including all variables of interest.6 The Herfindahl Blossfeld index relates positively to occupational intensity in the first regression in all estimations. The F-statistic on the excluded instrument is always clearly above the threshold of 10. The results for occupational intensity in the second stage clearly differ from the OLS regressions. Occupational intensity becomes insignificant for low-skilled but positively significant for medium- and high-skilled workers. The pattern and signs of the other variables reflect closely the OLS results in Table 1. One important difference is that the coefficients of firm distance increase substantially for medium- and high-skilled employees but are insignificant for low-skilled employees. Nonetheless, occupational distance continues to show larger coefficients than firm distance. Interpreting the results against the construction of the variables shows that the change in the coefficients of occupational intensity is due to an important correlation between firm and industry patterns in workforce composition. In the OLS estimations, occupational intensity also captures labour demand in the form of workforce changes on the industry-year level, leading to biased results. Table 2 The (dis)advantages of occupational intensity (2SLS)   First stage   First stage   Depvar: Current wage (log)  A  B  C  D  Low Qualification  (1)  (2)  (3)  (4)    Occ intensity    –0.063    –0.063    (0.069)    (0.069)  Herfindahl Blossfeld  0.132***    0.132***    (0.014)    (0.014)    Occ distance  –0.019***  –0.007***  –0.011***  –0.016***  (0.003)  (0.002)  (0.004)  (0.002)  Occ distance * previous wage      0.014***  –0.016***      (0.004)  (0.003)  Firm distance  0.055***  0.002  0.055***  –0.006  (0.003)  (0.004)  (0.004)  (0.004)  Firm dist * previous wage      0  –0.013***      (0.004)  (0.002)  Previous wage (log)  –0.002  0.109***  –0.032***  0.165***  (0.006)  (0.004)  (0.010)  (0.007)  Firm size (log)  –0.231***  0.081***  –0.230***  0.080***  (0.007)  (0.016)  (0.007)  (0.016)  Work experience  0.096***  0.809***  0.099***  0.801***  (0.022)  (0.017)  (0.022)  (0.017)  Work experience 2  –0.104***  –0.608***  –0.106***  –0.603***  (0.028)  (0.023)  (0.028)  (0.023)  Constant  –0.672***  –0.163***  –0.695***  –0.129***  (0.096)  (0.038)  (0.096)  (0.038)  Observations  31,516  31,516  31,516  31,516  R-squared  0.208  0.321  0.209  0.324  F-statistic  87.044***    86.772***      Medium qualification  (5)  (6)  (7)  (8)    Occ intensity    0.096***    0.079***    (0.020)    (0.020)  Herfindahl Blossfeld  0.223***    0.223***    (0.007)    (0.007)    Occ distance  –0.001  –0.029***  0.002  –0.034***  (0.002)  (0.001)  (0.002)  (0.001)  Occ distance * previous wage      0.018***  –0.030***      (0.002)  (0.002)  Firm distance  0.051***  –0.020***  0.050***  –0.021***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      –0.007***  –0.012***      (0.002)  (0.001)  Previous wage (log)  –0.028***  0.167***  –0.053***  0.247***  (0.004)  (0.003)  (0.006)  (0.005)  Firm size (log)  –0.249***  0.096***  –0.249***  0.089***  (0.004)  (0.005)  (0.004)  (0.005)  Work experience  –0.088***  0.478***  –0.084***  0.466***  (0.013)  (0.009)  (0.013)  (0.009)  Work experience 2  0.031**  –0.289***  0.029**  –0.284***  (0.013)  (0.009)  (0.013)  (0.009)  Constant  0.374***  0.059**  0.359***  0.065***  (0.067)  (0.023)  (0.068)  (0.023)  Observations  98,363  98,363  98,363  98,363  R-squared  0.194  0.386  0.194  0.399  F-statistic  966.873***    972.649***    High qualification  (9)  (10)  (11)  (12)    Occ intensity    0.127***    0.128***    (0.034)    (0.034)  Herfindahl Blossfeld  0.215***    0.213***    (0.014)    (0.014)    Occ distance  –0.030***  –0.029***  –0.031***  –0.029***  (0.004)  (0.003)  (0.004)  (0.003)  Occ distance * previous wage      0.012***  –0.005      (0.004)  (0.004)  Firm distance  0.061***  –0.021***  0.063***  –0.021***  (0.004)  (0.003)  (0.004)  (0.003)  Firm dist * previous wage      –0.021***  0.001      (0.004)  (0.003)  Previous wage (log)  –0.006  0.143***  –0.001  0.150***  (0.006)  (0.005)  (0.010)  (0.008)  Firm size (log)  –0.149***  0.094***  –0.149***  0.094***  (0.007)  (0.007)  (0.007)  (0.007)  Work experience  –0.076***  0.319***  –0.078***  0.317***  (0.025)  (0.018)  (0.025)  (0.018)  Work experience 2  –0.02  –0.200***  –0.018  –0.199***  (0.027)  (0.019)  (0.027)  (0.019)  Constant  –0.371  –0.002  –0.383  –0.004  (0.292)  (0.111)  (0.290)  (0.111)  Observations  21,381  21,381  21,381  21,381  R-squared  0.316  0.409  0.317  0.409  F-statistic  234.185***     231.983***       First stage   First stage   Depvar: Current wage (log)  A  B  C  D  Low Qualification  (1)  (2)  (3)  (4)    Occ intensity    –0.063    –0.063    (0.069)    (0.069)  Herfindahl Blossfeld  0.132***    0.132***    (0.014)    (0.014)    Occ distance  –0.019***  –0.007***  –0.011***  –0.016***  (0.003)  (0.002)  (0.004)  (0.002)  Occ distance * previous wage      0.014***  –0.016***      (0.004)  (0.003)  Firm distance  0.055***  0.002  0.055***  –0.006  (0.003)  (0.004)  (0.004)  (0.004)  Firm dist * previous wage      0  –0.013***      (0.004)  (0.002)  Previous wage (log)  –0.002  0.109***  –0.032***  0.165***  (0.006)  (0.004)  (0.010)  (0.007)  Firm size (log)  –0.231***  0.081***  –0.230***  0.080***  (0.007)  (0.016)  (0.007)  (0.016)  Work experience  0.096***  0.809***  0.099***  0.801***  (0.022)  (0.017)  (0.022)  (0.017)  Work experience 2  –0.104***  –0.608***  –0.106***  –0.603***  (0.028)  (0.023)  (0.028)  (0.023)  Constant  –0.672***  –0.163***  –0.695***  –0.129***  (0.096)  (0.038)  (0.096)  (0.038)  Observations  31,516  31,516  31,516  31,516  R-squared  0.208  0.321  0.209  0.324  F-statistic  87.044***    86.772***      Medium qualification  (5)  (6)  (7)  (8)    Occ intensity    0.096***    0.079***    (0.020)    (0.020)  Herfindahl Blossfeld  0.223***    0.223***    (0.007)    (0.007)    Occ distance  –0.001  –0.029***  0.002  –0.034***  (0.002)  (0.001)  (0.002)  (0.001)  Occ distance * previous wage      0.018***  –0.030***      (0.002)  (0.002)  Firm distance  0.051***  –0.020***  0.050***  –0.021***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      –0.007***  –0.012***      (0.002)  (0.001)  Previous wage (log)  –0.028***  0.167***  –0.053***  0.247***  (0.004)  (0.003)  (0.006)  (0.005)  Firm size (log)  –0.249***  0.096***  –0.249***  0.089***  (0.004)  (0.005)  (0.004)  (0.005)  Work experience  –0.088***  0.478***  –0.084***  0.466***  (0.013)  (0.009)  (0.013)  (0.009)  Work experience 2  0.031**  –0.289***  0.029**  –0.284***  (0.013)  (0.009)  (0.013)  (0.009)  Constant  0.374***  0.059**  0.359***  0.065***  (0.067)  (0.023)  (0.068)  (0.023)  Observations  98,363  98,363  98,363  98,363  R-squared  0.194  0.386  0.194  0.399  F-statistic  966.873***    972.649***    High qualification  (9)  (10)  (11)  (12)    Occ intensity    0.127***    0.128***    (0.034)    (0.034)  Herfindahl Blossfeld  0.215***    0.213***    (0.014)    (0.014)    Occ distance  –0.030***  –0.029***  –0.031***  –0.029***  (0.004)  (0.003)  (0.004)  (0.003)  Occ distance * previous wage      0.012***  –0.005      (0.004)  (0.004)  Firm distance  0.061***  –0.021***  0.063***  –0.021***  (0.004)  (0.003)  (0.004)  (0.003)  Firm dist * previous wage      –0.021***  0.001      (0.004)  (0.003)  Previous wage (log)  –0.006  0.143***  –0.001  0.150***  (0.006)  (0.005)  (0.010)  (0.008)  Firm size (log)  –0.149***  0.094***  –0.149***  0.094***  (0.007)  (0.007)  (0.007)  (0.007)  Work experience  –0.076***  0.319***  –0.078***  0.317***  (0.025)  (0.018)  (0.025)  (0.018)  Work experience 2  –0.02  –0.200***  –0.018  –0.199***  (0.027)  (0.019)  (0.027)  (0.019)  Constant  –0.371  –0.002  –0.383  –0.004  (0.292)  (0.111)  (0.290)  (0.111)  Observations  21,381  21,381  21,381  21,381  R-squared  0.316  0.409  0.317  0.409  F-statistic  234.185***     231.983***     Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of two-stage least squares regressions. The odd-numbered columns (also, A and C) show the first-stage results. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(4) are workers with low, (5)–(8) with medium, (9)–(12) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1 The reduced-form estimates for this analysis are also in line with the expectations, showing significant, positive coefficients of the Herfindahl Blossfeld index for medium- and high-skilled employees but insignificant coefficients for low-skilled employees in the wage regressions (results are available upon request). Comparing the size of the distance variables and occupational intensity shows that the sum of the negative distance coefficients never amounts to the size of the positive coefficient for occupational intensity. In other words, although an individual might cover costly distances, these costs are outweighed by the benefits of working in an occupationally intensive firm. Going back to eq. (7), one could summarize that since in practice the distance measures and occupational intensity cancel each other out, a joint switch occurs when there is wage growth and/or an improvement in the task match. This is an interesting finding, as there may be various other reasons why individuals switch but nonetheless, on average, the (dis)advantages along knowledge dimension balance out. In sum, instrumenting occupational intensity confirms that the OLS results are biased and that for medium- and high-skilled employees wages in target firms increase with occupational intensity. Thus, in terms of wages, the 2SLS estimates for these groups confirm the demand-side hypothesis by Bidwell and Briscoe, according to which occupationally intensive firms are of advantage to workers. The decreasing share of occupational peers with experience is in line with the supply-side hypothesis. Taking into account that distance and occupational intensity decrease with increasing experience suggests that, when distances are shorter, there is less need to choose occupationally intensive firms as a compensation mechanism. However, this cannot be causally interpreted. Low-skilled workers are not affected by firm knowledge, neither by firm distance nor by occupational intensity—an even stronger result than in the previous models. This again might explain why they cover on average longer firm distances. To verify the robustness of the 2SLS estimations, the same equations were estimated using limited information maximum likelihood (LIML) and generalized method of moments (GMM). The results stay virtually the same. Including the squared term of occupational intensity provides no robust evidence for a non-linear relationship between occupational intensity and wages. All of the previous regressions focus on men. Additional analyses for women show the same patterns. In fact, the positive relationship between occupational intensity and wages can also be confirmed for low-skilled women. All results are available upon request. 4.3.2 Matching task profiles of workers and firms In a final step to better understand on-the-job learning, the focus shifts towards the newly developed measure for the match between occupational and firm tasks. It calculates the distance between the worker’s occupational tasks and the task composition of firms. The data show that for 45.6% (35.3%) of the sample, the most (least) important tasks of workers and firms are identical. As a cross-check, for both tasks the share where the important firm (occupational) task is not the least important occupational (firm) task, implying no match, is clearly above 90%. These results already suggest that employees work at firms with similar task portfolios to their own. Figure 4 depicts the development of the firm-occupation match over the career path, based on results when the firm-occupation distance is regressed on work experience. While, interestingly enough, all qualification levels start at similar levels, the distance increases for low-skilled, remains relatively stable for medium-skilled, and decreases for high-skilled workers. This is interesting, as it suggests that the lower the qualification level, the more likely individuals are to work for firms with a lower overlap of task profiles between firm and occupation when they gain more experience. Fig. 4 View largeDownload slide The relationship between work experience and firm-occupation distance Fig. 4 View largeDownload slide The relationship between work experience and firm-occupation distance Table 3 shows the OLS results when occupational intensity is replaced with firm-occupation distance. If the same story holds as for occupational intensity, there should be a positive coefficient. First, note that the remaining coefficients are highly similar to Table 1. However, firm-occupation distance is only positive for low-skilled employees, remains inconclusive or insignificant for medium-skilled employees, and is negative for high-skilled employees. This suggests that when factoring in task similarities with other occupations, the benefits and costs of firm-occupation distance differ by qualification level. This is not the case when repeating the regressions on the female sample, where the coefficient is always negative and significant. However, as discussed, the firm-occupation distance measure may be biased, for instance, due to personal motivations for covering long-distance switches and choosing a new firm. Table 3 The role of the firm-occupation match (OLS) Depvar: Current wage (log)  A  B  C  D  E  Low Qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.110***  0.110***  0.155***  0.108***  0.166***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Firm- occ dist  0.016***  0.020***  0.020***  0.020***  0.020***  (0.003)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        0.000  –0.008***        (0.002)  (0.002)  Firm dist * previous wage          –0.012***          (0.002)  Occ distance    –0.008***  –0.020***  –0.009***  –0.018***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.021***    –0.017***      (0.002)    (0.002)  Current firm size (log)  0.097***  0.097***  0.096***  0.095***  0.094***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.797***  0.795***  0.791***  0.800***  0.793***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.597***  –0.596***  –0.594***  –0.597***  –0.593***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.317***  –0.299***  –0.265***  –0.204***  –0.157**  (0.057)  (0.057)  (0.057)  (0.076)  (0.077)  Observations  32,290  32,290  32,290  31,496  31,496  R-squared  0.323  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.177***  0.169***  0.236***  0.165***  0.244***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Firm- occ dist  –0.019***  –0.003  –0.003  –0.003*  –0.002  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.015***  –0.017***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Current firm size (log)  0.077***  0.076***  0.074***  0.072***  0.070***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.475***  0.463***  0.454***  0.466***  0.455***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.289***  –0.279***  –0.275***  –0.281***  –0.277***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.441***  –0.378***  –0.342***  –0.396***  –0.338***  (0.079)  (0.079)  (0.079)  (0.038)  (0.038)  Observations  100,954  100,954  100,954  98,375  98,375  R-squared  0.404  0.411  0.416  0.414  0.42  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.154***  0.146***  0.152***  0.143***  0.150***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Firm- occ dist  –0.052***  –0.040***  –0.040***  –0.039***  –0.039***  (0.005)  (0.005)  (0.005)  (0.005)  (0.005)  Firm distance        –0.012***  –0.012***        (0.002)  (0.002)  Firm dist * previous wage          –0.001          (0.003)  Occ distance    –0.030***  –0.030***  –0.028***  –0.028***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.003    –0.003      (0.003)    (0.003)  Current firm size (log)  0.082***  0.081***  0.081***  0.078***  0.078***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.311***  0.309***  0.312***  0.311***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.204***  –0.204***  –0.208***  –0.207***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.472  –0.399  –0.402  –0.153  –0.152  (0.302)  (0.296)  (0.296)  (0.135)  (0.135)  Observations  21,668  21,668  21,668  21,378  21,378  R-squared  0.44  0.444  0.444  0.445  0.445  Depvar: Current wage (log)  A  B  C  D  E  Low Qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.110***  0.110***  0.155***  0.108***  0.166***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Firm- occ dist  0.016***  0.020***  0.020***  0.020***  0.020***  (0.003)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        0.000  –0.008***        (0.002)  (0.002)  Firm dist * previous wage          –0.012***          (0.002)  Occ distance    –0.008***  –0.020***  –0.009***  –0.018***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.021***    –0.017***      (0.002)    (0.002)  Current firm size (log)  0.097***  0.097***  0.096***  0.095***  0.094***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.797***  0.795***  0.791***  0.800***  0.793***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.597***  –0.596***  –0.594***  –0.597***  –0.593***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.317***  –0.299***  –0.265***  –0.204***  –0.157**  (0.057)  (0.057)  (0.057)  (0.076)  (0.077)  Observations  32,290  32,290  32,290  31,496  31,496  R-squared  0.323  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.177***  0.169***  0.236***  0.165***  0.244***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Firm- occ dist  –0.019***  –0.003  –0.003  –0.003*  –0.002  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.015***  –0.017***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Current firm size (log)  0.077***  0.076***  0.074***  0.072***  0.070***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.475***  0.463***  0.454***  0.466***  0.455***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.289***  –0.279***  –0.275***  –0.281***  –0.277***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.441***  –0.378***  –0.342***  –0.396***  –0.338***  (0.079)  (0.079)  (0.079)  (0.038)  (0.038)  Observations  100,954  100,954  100,954  98,375  98,375  R-squared  0.404  0.411  0.416  0.414  0.42  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.154***  0.146***  0.152***  0.143***  0.150***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Firm- occ dist  –0.052***  –0.040***  –0.040***  –0.039***  –0.039***  (0.005)  (0.005)  (0.005)  (0.005)  (0.005)  Firm distance        –0.012***  –0.012***        (0.002)  (0.002)  Firm dist * previous wage          –0.001          (0.003)  Occ distance    –0.030***  –0.030***  –0.028***  –0.028***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.003    –0.003      (0.003)    (0.003)  Current firm size (log)  0.082***  0.081***  0.081***  0.078***  0.078***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.311***  0.309***  0.312***  0.311***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.204***  –0.204***  –0.208***  –0.207***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.472  –0.399  –0.402  –0.153  –0.152  (0.302)  (0.296)  (0.296)  (0.135)  (0.135)  Observations  21,668  21,668  21,668  21,378  21,378  R-squared  0.44  0.444  0.444  0.445  0.445  Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of OLS regressions. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(5) are workers with low, (6)–(10) with medium, (11)–(15) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1. The 2SLS results in Table 4 paint a different picture than the OLS regressions, pointing towards a bias of the original estimates. Medium-skilled workers consistently profit from achieving a close match between their own and the firm’s task portfolio, similar to occupational intensity, while low-skilled workers do not seem to be affected. Unfortunately, the instrument does not work for high-skilled workers. The results for female workers are again more consistent with the predictions of the model, showing that low- and medium-skilled workers benefit from small firm-occupation distances while the instrument does not work for high-skilled workers. Table 4 The role of the firm-occupation match (2SLS)   First stage   First stage      A  B  C  D  Low qualification  (1)  (2)  (3)  (4)    Herfindahl Blossfeld  –0.062***    –0.062***    (0.014)    (0.014)    Firm-occ dist    0.158    0.160    (0.151)    (0.151)  Previous wage (log)  0.014***  0.106***  0.014  0.164***  (0.005)  (0.005)  (0.009)  (0.007)  Firm distance  –0.039***  0.005  –0.040***  –0.002  (0.003)  (0.006)  (0.004)  (0.006)  Firm dist * previous wage      –0.001  –0.012***      (0.004)  (0.002)  Occ distance  0.146***  –0.029  0.146***  –0.039*  (0.003)  (0.022)  (0.004)  (0.022)  Occ distance * previous wage      0.001  –0.017***      (0.004)  (0.002)  Current firm size (log)  –0.011*  0.097***  –0.011*  0.096***  (0.007)  (0.005)  (0.007)  (0.005)  Work experience  0.113***  0.784***  0.113***  0.777***  (0.022)  (0.024)  (0.022)  (0.024)  Work experience 2  –0.074***  –0.587***  –0.073***  –0.583***  (0.028)  (0.025)  (0.028)  (0.025)  Constant  0.408***  –0.287**  0.408***  –0.254*  (0.129)  (0.133)  (0.130)  (0.132)  Observations  31,496  31,496  31,496  31,496  R-squared  0.25  0.294  0.25  0.296  F-statistic  18.756***    18.758***      Medium qualification  (5)  (6)  (7)  (8)    Herfindahl Blossfeld  –0.099***    –0.097***    (0.006)    (0.006)    Firm-occ dist    –0.220***    –0.185***    (0.047)    (0.047)  Previous wage (log)  0.004  0.166***  –0.019***  0.240***  (0.004)  (0.003)  (0.005)  (0.005)  Firm distance  –0.017***  –0.019***  –0.015***  –0.019***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      0.017***  –0.010***      (0.002)  (0.002)  Occ distance  0.144***  0.002  0.144***  –0.007  (0.002)  (0.007)  (0.002)  (0.007)  Occ distance * previous wage      –0.001  –0.029***      (0.003)  (0.002)  Current firm size (log)  0.022***  0.076***  0.022***  0.074***  (0.003)  (0.002)  (0.003)  (0.002)  Work experience  0.041***  0.475***  0.045***  0.463***  (0.012)  (0.009)  (0.012)  (0.009)  Work experience 2  –0.02  –0.286***  –0.021*  –0.281***  (0.012)  (0.009)  (0.012)  (0.009)  Constant  –0.599***  0.172***  –0.618***  0.160***  (0.060)  (0.037)  (0.060)  (0.037)  Observations  98,375  98,375  98,375  98,375  R-squared  0.203  0.332  0.204  0.361  F-statistic  302.339***    290.7***    High qualification  (9)  (10)  (11)  (12)    Herfindahl Blossfeld  0.005    0.006    (0.010)    (0.010)    Firm-occ dist    5.159    4.238    (10.308)    (6.996)  Previous wage (log)  0.019***  0.046  –0.007  0.178***  (0.006)  (0.196)  (0.009)  (0.060)  Firm distance  0.004  –0.034  0.002  –0.021  (0.004)  (0.048)  (0.004)  (0.023)  Firm dist * previous wage      0.020***  –0.087      (0.004)  (0.140)  Occ distance  0.116***  –0.629  0.117***  –0.53  (0.005)  (1.192)  (0.005)  (0.821)  Occ distance * previous wage      0.001  –0.007      (0.005)  (0.023)  Current firm size (log)  0.075***  –0.309  0.074***  –0.239  (0.006)  (0.769)  (0.006)  (0.520)  Work experience  0.023  0.191  0.031  0.175  (0.024)  (0.280)  (0.024)  (0.251)  Work experience 2  –0.025  –0.075  –0.031  –0.073  (0.027)  (0.312)  (0.027)  (0.259)  Constant  0.546**  –5.205  0.550**  –4.319  (0.229)  (10.429)  (0.228)  (7.154)  Observations  21,378  21,378  21,378  21,378  R-squared  0.215     0.216     F-statistic  0.259    0.384      First stage   First stage      A  B  C  D  Low qualification  (1)  (2)  (3)  (4)    Herfindahl Blossfeld  –0.062***    –0.062***    (0.014)    (0.014)    Firm-occ dist    0.158    0.160    (0.151)    (0.151)  Previous wage (log)  0.014***  0.106***  0.014  0.164***  (0.005)  (0.005)  (0.009)  (0.007)  Firm distance  –0.039***  0.005  –0.040***  –0.002  (0.003)  (0.006)  (0.004)  (0.006)  Firm dist * previous wage      –0.001  –0.012***      (0.004)  (0.002)  Occ distance  0.146***  –0.029  0.146***  –0.039*  (0.003)  (0.022)  (0.004)  (0.022)  Occ distance * previous wage      0.001  –0.017***      (0.004)  (0.002)  Current firm size (log)  –0.011*  0.097***  –0.011*  0.096***  (0.007)  (0.005)  (0.007)  (0.005)  Work experience  0.113***  0.784***  0.113***  0.777***  (0.022)  (0.024)  (0.022)  (0.024)  Work experience 2  –0.074***  –0.587***  –0.073***  –0.583***  (0.028)  (0.025)  (0.028)  (0.025)  Constant  0.408***  –0.287**  0.408***  –0.254*  (0.129)  (0.133)  (0.130)  (0.132)  Observations  31,496  31,496  31,496  31,496  R-squared  0.25  0.294  0.25  0.296  F-statistic  18.756***    18.758***      Medium qualification  (5)  (6)  (7)  (8)    Herfindahl Blossfeld  –0.099***    –0.097***    (0.006)    (0.006)    Firm-occ dist    –0.220***    –0.185***    (0.047)    (0.047)  Previous wage (log)  0.004  0.166***  –0.019***  0.240***  (0.004)  (0.003)  (0.005)  (0.005)  Firm distance  –0.017***  –0.019***  –0.015***  –0.019***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      0.017***  –0.010***      (0.002)  (0.002)  Occ distance  0.144***  0.002  0.144***  –0.007  (0.002)  (0.007)  (0.002)  (0.007)  Occ distance * previous wage      –0.001  –0.029***      (0.003)  (0.002)  Current firm size (log)  0.022***  0.076***  0.022***  0.074***  (0.003)  (0.002)  (0.003)  (0.002)  Work experience  0.041***  0.475***  0.045***  0.463***  (0.012)  (0.009)  (0.012)  (0.009)  Work experience 2  –0.02  –0.286***  –0.021*  –0.281***  (0.012)  (0.009)  (0.012)  (0.009)  Constant  –0.599***  0.172***  –0.618***  0.160***  (0.060)  (0.037)  (0.060)  (0.037)  Observations  98,375  98,375  98,375  98,375  R-squared  0.203  0.332  0.204  0.361  F-statistic  302.339***    290.7***    High qualification  (9)  (10)  (11)  (12)    Herfindahl Blossfeld  0.005    0.006    (0.010)    (0.010)    Firm-occ dist    5.159    4.238    (10.308)    (6.996)  Previous wage (log)  0.019***  0.046  –0.007  0.178***  (0.006)  (0.196)  (0.009)  (0.060)  Firm distance  0.004  –0.034  0.002  –0.021  (0.004)  (0.048)  (0.004)  (0.023)  Firm dist * previous wage      0.020***  –0.087      (0.004)  (0.140)  Occ distance  0.116***  –0.629  0.117***  –0.53  (0.005)  (1.192)  (0.005)  (0.821)  Occ distance * previous wage      0.001  –0.007      (0.005)  (0.023)  Current firm size (log)  0.075***  –0.309  0.074***  –0.239  (0.006)  (0.769)  (0.006)  (0.520)  Work experience  0.023  0.191  0.031  0.175  (0.024)  (0.280)  (0.024)  (0.251)  Work experience 2  –0.025  –0.075  –0.031  –0.073  (0.027)  (0.312)  (0.027)  (0.259)  Constant  0.546**  –5.205  0.550**  –4.319  (0.229)  (10.429)  (0.228)  (7.154)  Observations  21,378  21,378  21,378  21,378  R-squared  0.215     0.216     F-statistic  0.259    0.384    Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of two-stage least squares regressions. The odd-numbered columns (also, A and C) show the first-stage results. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(4) are workers with low, (5)–(8) with medium, (9)–(12) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1. In sum, workers match themselves to firms with specific task sets but there are important differences across qualification levels regarding the match development over time. Using occupational intensity as a measure generates more robust results, suggesting that this group may play a more prominent role for on-the-job learning than when taking into account all tasks on the firm level. 5. Conclusions Previous work in the field of labour mobility that uses task-based measures to determine job content has helped address several puzzles of labour economists, such as, for instance, skill-biased technological change. Other work with task data has addressed the question of human capital specificity, that is, knowledge that cannot be transferred in the case of job switches. This paper splits occupational and firm knowledge both in two, a specific and a general component, which is done by determining how transferable knowledge is between two firms or two occupations. In addition, it takes into account the role of working in an occupationally intensive firm, that is, a firm with a large amount of specific knowledge. The results reveal the following patterns with regard to how individuals are matched along the career path. First, the majority of switchers travel only small distances between firms. Furthermore, long-distance switches between firms become less likely with increasing work experience, indicating that workers might find better work matches as they move along their career path. Firm and occupational distances—measures for specific knowledge—show a negative relationship with wages, with the exception of low-skilled workers, where firm distance is not always insignificant. Occupational knowledge is of higher importance than firm knowledge. In early career stages, individuals work with a higher share of colleagues in the same occupational group, called occupational intensity, than is the case later on in their employment history. In 2SLS estimations it can be shown that occupational intensity positively affects wages for higher qualification levels and that lower firm-occupation distance reduces costs for medium-skilled workers, supporting the idea of higher wages in firms with a higher demand for an occupation. In addition, the sum of the negative coefficients from increasing both occupational and firm distance by one standard deviation is smaller than the positive coefficient of increasing occupational intensity by one standard deviation. This indicates that long-distance switches can still be rewarding in terms of the awaiting environment at the target firm. In sum, this paper contributes to the literature by showing that the specificity of knowledge on the occupational and firm level is determined by context, and that both firm and occupational knowledge matter for wages after switches. The paper hence confirms previous work by G&S and shows that Lazear’s skill-weights approach holds for firm knowledge. Human capital theory predicts that costs of general on-the-job training should be borne by the worker, while in reality specific training costs seem to be covered partly by workers and partly by firms. If the specificity depends on where workers move next, then this might explain why the empirical studies differ from the theoretical predictions (e.g., Barron and Berger, 1999; Parent, 1999). Further, in a task-based analysis, averaging across occupations (which is the standard procedure) and thereby disregarding firm knowledge implies a loss of information. Supplementary material The data used are confidential, but the replication files are available online at the OUP website, as is an online appendix. The replication files are separated by data sources (IAB for the Sample of Integrated Labour Market Biographies [SIAB] and the Establishment History Panel [BHP], BIBB for the BIBB/BAuA Employment Survey 2006). Funding This work was supported by the German Research Foundation (doctoral scholarship) and the German Academic Exchange Service (fellowship for research stay abroad and conference travel grant). Footnotes 1 In his article, Lazear (2009) provides a real-world example from Silicon Valley that nicely illustrates his basic idea. Note that although one could classify the mentioned skills as occupational, firms will combine them in unique ways that go beyond the occupational arrangement. 2 To facilitate the comparison between that work and the present paper, similar equations have the same numbers. 3 To test whether switches between similar firms are driven by switches between similar occupations, I calculate the distribution of firm switches, hence, excluding occupational switchers. The majority of switches (now up to approximately 90%) take place in the smallest firm distance intervals. Additionally, focusing on long-distance occupational switchers (0.4 or higher) shows that the majority of firm switches are still among similar firms (around approximately 50%), indicating that the switching pattern in terms of firm distance is not driven by switches between similar occupations. The results are available upon request. 4 Note that the standardization of interaction terms changes the null hypothesis and, thereby, complicates the interpretation of the results. Comparison between models is, thus, not possible. It can further lead to coefficients and significance levels that differ from those of an unstandardized model. Nonetheless, the goal of testing the contribution of firm and occupational human capital justifies this approach. Also note that clustered standard errors are not suitable for standardized variables because variables are standardized using the population mean. Standard errors would instead be clustered on the individual level. Thus, the models are estimated with robust standard errors instead. Control regressions with standard errors clustered on the individual level using standardized variables confirm the reported relationships (results available upon request). 5 G&S did not include occupational intensity and firm size, but leaving these variables out does not alter the results. Results are available upon request. 6 The 2SLS for columns A–C of Table 1 are also in line with the results in Table 2. Results are available upon request. Acknowledgements The author is indebted to David Autor, Ljubica Nedelkoska, and Michael Wyrwich, who provided many insightful comments. I thank the members of the Graduate College ‘Economics of Innovative Change’ in Jena and the staff at the Research Data Centre of the Federal Employment Agency for excellent suggestions and support. This research project was carried out while Elisabeth Bublitz was employed at the Friedrich Schiller University Jena and the Hamburg Institute of International Economics. The article’s content is solely the responsibility of the author and does not necessarily represent official views of the author’s affiliations. References Acemoglu D., Autor D. ( 2011) Skills, tasks and technologies: implications for employment and earnings, in Ashenfelter O., Card D., Handbook of Labor Economics, Handbooks in Economics, Volume 4, Part B , Elsevier and University of Chicago Press, Chicago, IL, 1043– 1171. Google Scholar CrossRef Search ADS   Autor D.H., Levy F., Murnane R.J. ( 2003) The skill content of recent technological change: an empirical exploration, Quarterly Journal of Economics , 118, 1279– 1333. Google Scholar CrossRef Search ADS   Backes-Gellner U., Mure J. ( 2005) The skill-weights approach on firm specific human capital: empirical results for Germany, ISU Working Paper Series, Working Paper No. 56, Institute for Strategy and Business Economics, University of Zurich. Baron J.N. ( 1984) Organizational perspectives on stratification, Annual Review of Sociology , 10, 37– 69. Google Scholar CrossRef Search ADS   Barron J.M., Berger M.C. ( 1999) Do workers pay for on-the-job training?, Journal of Human Resources , 34, 235– 52. Google Scholar CrossRef Search ADS   Barron J.M., Berger M.C., Black D.A. ( 1997) How well do we measure training?, Journal of Labor Economics , 15, 507– 28. Google Scholar CrossRef Search ADS   Becker G.S. ( 1962) Investment in human capital: a theoretical analysis, Journal of Political Economy , 70, 9– 49. Google Scholar CrossRef Search ADS   Becker G.S. ( 1964) Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education , University of Chicago Press, Chicago, IL. Bidwell M., Briscoe F. ( 2010) The dynamics of interorganizational careers, Organization Science , 21, 1034– 53. Google Scholar CrossRef Search ADS   Blossfeld H.P. ( 1985) Bildungsexpansion und Berufschancen , Campus, Frankfurt/Main, Mannheim. Borghans L., ter Weel B., Weinberg B.A. ( 2014) People skills and the labor-market outcomes of underrepresented groups, Industrial and Labor Relations Review , 67, 287– 334. Google Scholar CrossRef Search ADS   Dorner M., Heining J., Jacobebbinghaus P., Seth S. ( 2010) Sample of Integrated Labour Market Biographies (SIAB) 1975–2008, FDZ-Methodenreport 09/2010, Research Data Centre of the German Federal Employment Agency at the Institute at the Institute for Employment Research, Nürnberg. Gathmann C., Schönberg U. ( 2010) How general is human capital? A task-based approach, Journal of Labor Economics , 28, 1– 49. Google Scholar CrossRef Search ADS   Geel R., Mure J., Backes-Gellner U. ( 2010) Specificity of occupational training and occupational mobility: an empirical study based on Lazear’s skill-weights approach, Education Economics , 19, 519– 35. Google Scholar CrossRef Search ADS   Green F. ( 2012) Employee involvement, technology, and evolution in job skills: a task-based analysis, Industrial and Labor Relations Review , 65, 36– 67. Google Scholar CrossRef Search ADS   Hall A., Tiemann M. ( 2006) BIBB/BAuA Employment Survey of the Working Population on Qualification and Working Conditions in Germany 2006, suf_1.0; Research Data Center at BIBB (ed.); GESIS Cologne, Germany (data access); Federal Institute of Vocational Education and Training, Bonn, doi:10.4232/1.4820. Hannan M.T. ( 1988) Social change, organizational diversity, and individual careers, in Riley M.W., Social Change and the Life Course , Sage, Newbury Park, CA. Jaffe A.B. ( 1986) Technological opportunity and spillovers of R&D: evidence from firms’ patents, profits, and market value, American Economic Review , 76, 984– 1001. Kambourov G., Manovskii I. ( 2009) Occupational specificity of human capital, International Economic Review , 50, 63– 115. Google Scholar CrossRef Search ADS   Lazear E.P. ( 2009) Firm-specific human capital: a skill-weights approach, Journal of Political Economy , 117, 914– 40. Google Scholar CrossRef Search ADS   Mincer J. ( 1962) On-the-job training: costs, returns, and some implications, Journal of Political Economy , 70, 50– 79. Google Scholar CrossRef Search ADS   Mincer J. ( 1989) Human capital and the labor market: a review of current research, Educational Researcher , 18, 27– 34. Google Scholar CrossRef Search ADS   Neal D. ( 1995) Industry-specific human capital: evidence from displaced workers, Journal of Labor Economics , 13, 653– 77. Google Scholar CrossRef Search ADS   Neal D. ( 1999) The complexity of job mobility among young men, Journal of Labor Economics , 17, 237. Google Scholar CrossRef Search ADS   Nedelkoska L., Neffke F. ( 2011) Skill shortage and skill redundancy: asymmetry in the transferability of skills, DIME Final Conference Paper, DIME Final Conference, 6–8 April, Maastricht. Parent D. ( 1999) Wages and mobility: the impact of employer-provided training, Journal of Labor Economics , 17, 298– 317. Google Scholar CrossRef Search ADS   Parent D. ( 2000) Industry-specific capital and the wage profile: evidence from the National Longitudinal Survey of Youth and the Panel Study of Income Dynamics, Journal of Labor Economics , 18, 306– 23. Google Scholar CrossRef Search ADS   Pavan R. ( 2011) Career choice and wage growth, Journal of Labor Economics , 29, 549– 87. Google Scholar CrossRef Search ADS   Poletaev M., Robinson C. ( 2008) Human capital specificity: evidence from the dictionary of occupational titles and displaced worker surveys, 1984–2000, Journal of Labor Economics , 26, 387– 420. Google Scholar CrossRef Search ADS   Rohrbach-Schmidt D. ( 2009) The BIBB/IAB- and BIBB-BAuA Surveys of the Working Population on Qualification and Working Conditions in Germany, BIBB-FDZ Daten- und Methodenbericht, 1/2009, Bundesinstitut für Berufsbildung (BIBB), Bonn. Rohrbach-Schmidt D., Tiemann M. ( 2013) Changes in workplace tasks in Germany—evaluating skill and task measures, Journal for Labour Market Research , 46, 215– 37. Google Scholar CrossRef Search ADS   Spitz-Oener A. ( 2006) Technical change, job tasks, and rising educational demands: looking outside the wage structure, Journal of Labor Economics , 24, 235– 70. Google Scholar CrossRef Search ADS   Spitz-Oener A. ( 2008) The returns to pencil use revisited, Industrial and Labor Relations Review , 61, 502– 17. © Oxford University Press 2018 All rights reserved This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Oxford Economic Papers Oxford University Press

Matching skills of individuals and firms along the career path

Loading next page...
 
/lp/ou_press/matching-skills-of-individuals-and-firms-along-the-career-path-I4NTfe7hdu
Publisher
Oxford University Press
Copyright
© Oxford University Press 2018 All rights reserved
ISSN
0030-7653
eISSN
1464-3812
D.O.I.
10.1093/oep/gpx056
Publisher site
See Article on Publisher Site

Abstract

Abstract Research since Gary Becker equated specific human capital with firm-specific human capital. This paper divides firm human capital into a specific and a general component to investigate the relationships between firm- and occupation-specific human capital and job switches. Applying the task-based approach, the results show that the degree to which firm knowledge is portable depends on task similarities between the firms. Firm- and occupation-specific knowledge are negatively related to wages in a new job, but achieving a good occupational, instead of firm, match is more important for employees. The occupational intensity, reflecting the overall knowledge composition on the firm level, decreases with experience and can outweigh the costs of covering long task distances. As regards matching, a small distance between the firm and occupational tasks matters for medium-skilled employees. 1. Introduction It is well established that firm-specific knowledge increases with firm tenure and that it is lost when employees switch employers (Becker, 1964). Many studies use firm tenure as a proxy for accumulated firm-specific knowledge. The question remains, however, as to what exactly this knowledge is and whether all firm knowledge is specific and, thus, not transferable. The skill-weights approach (Lazear, 2009) models firm-specific knowledge by letting firms place different weights on general skills. This assumes that a certain amount of all knowledge is transferable across firms. Gathmann and Schönberg (G&S, 2010) test this approach for occupation-specific knowledge by investigating the relationship between occupational knowledge and wages. Following Lazear’s theoretical arguments, G&S show empirically that the amount of specific knowledge and, thus, the number of portable skills, varies between occupations and along the career path. To date, there has been no investigation of the degree to which firm knowledge is portable across establishments. Studies often argue that, in the case of a firm switch, all firm knowledge is lost; thus, the likelihood of firm switches decreases with the accumulation of firm knowledge as measured by increasing firm tenure. However, the importance of firm knowledge as an obstacle to switches would be overestimated if part of the firm knowledge was transferable. Therefore, the aim of this paper is to discover how the firm distance of moves (knowledge differences between firms) varies along the career path and how it relates to wages. In addition, the relative importance of firm and of occupational knowledge for wages is determined, providing an indication for switchers whether it is more important to find a good firm or a good occupational match. Regarding the incentives to switch, the analysis considers on-the-job learning possibilities as proxied by two new variables for the specific knowledge structure of firms: the share of occupational peers on the firm level (occupational intensity) and the match of the firm and occupational task sets (firm-occupation distance). This allows determining to what degree knowledge that is specific to firms can benefit workers who switch jobs. I begin by formally modelling the relationship between specific, non-transferable knowledge and labour mobility. A combination of the approaches of Lazear (2009) and G&S (2010) allows developing one that accounts for firm and occupational knowledge as well as occupational intensity. Both knowledge types can be decomposed into a transferable and a specific component. The model yields several predictions which can then be investigated in the empirical analysis. As stressed by Lazear (2009), firm knowledge is understood as a unique arrangement of skills in that the skills taken alone are not specific but become so through their combination.1 This variation is lost in empirical analyses that only look at the occupational level because, opposite to my analysis, individual-level data is often unavailable. The predictions are empirically tested with the task-based approach that analyses which tasks are performed on the job (cf. Autor et al., 2003; Poletaev and Robinson, 2008; G&S, 2010). The data used for the empirical analyses are from three sources. The Sample of Integrated Labour Market Biographies (SIAB) covers a representative sample of the working population in Germany. Information about firms is drawn from the Establishment History Panel (BHP). Details on occupational skill sets and tasks are provided by the BIBB/BAuA Employment Survey 2006. As a preparatory step, a factor analysis is applied to categorize tasks into groups. Firm knowledge is measured, first, as the task composition of firms which is determined via the occupational task composition of the workforce. Second, the weight of tasks’ combinations shows the importance of single occupations, instead of single tasks (occupational intensity). The first procedure disaggregates occupational information to the task level, while the second procedure allows accounting for the importance of having several tasks embodied in one occupation. Third, the differences between a worker’s and a firm’s task composition are identified to measure an overlap of knowledge (firm-occupation distance). As knowledge naturally differs by qualification levels, this is accounted for by analysing different qualification groups separately. The results are in line with the predictions, confirming that long-distance firm switches occur more often early in the career than later on. Which type of knowledge can be transferred and to what extent depends on the qualification level of workers. In the case of joint firm and occupation switches, firm- and occupation-specific knowledge both matter for wages. The present evidence indicates that firm-specific knowledge matters less than occupation-specific knowledge. Interestingly, it can further be shown that individuals start work in firms that have a relatively higher share of employees in the same occupational group, e.g. a high occupational intensity, and that this share decreases with increasing work experience. Taking this a step further in the OLS estimations reveals that a lower occupational intensity is associated with higher wages. However, workers might strategically select into firms with a certain occupational diversity. To control for this selection problem, occupational intensity is instrumented with occupational diversity on the industry level. Contrary to the OLS results, occupational intensity now shows a positive effect on wages for medium- and high-skilled workers and becomes insignificant for low-skilled employees. Surprisingly, the sizes of the coefficients, which are calculated using standardized variables, suggest that the benefits of occupational intensity can clearly outweigh the costs of both distance measures. This effect supports the notion that a high occupational intensity shows on average a higher demand for an occupational task set which is reflected in higher wages. The overall results generally hold for both men and women. In addition, firm-occupation distance starts at similar levels across qualification levels at the beginning of the employment biography. It then increases for low-skilled, remains stable for medium-skilled, and decreases for high-skilled workers. When estimating 2SLS regressions, medium-skilled workers are affected by higher costs for higher distances while low-skilled seem unaffected. The paper is embedded into the literature on skill-weights and on the task-based approach. Lazear (2009) suggests that all human capital is general; it becomes specific through weights that are firm specific. His approach is novel because, compared to earlier work (cf. Becker, 1962), there is no longer a clear, exogenously given distinction between general and specific human capital and instead they are defined endogenously with observable market parameters. First, empirical tests of Lazear’s model show that the investigated predictions of the model are borne out by the data, making the model a worthwhile basis for the present analysis (Backes-Gellner and Mure, 2005; Geel et al., 2010). The approach of this paper also relates to Neal (1999), who stresses the prevalence of complex job switches, involving a higher degree of changes as measured by simultaneous industrial/occupational (career) and firm moves at younger ages. Also, Neal suggests that careers are chosen first, followed by selection of firms. Although my analysis measures the moves simultaneously and may not make statements on the sequence of job decisions, the relative coefficient sizes confirm that, in terms of the final decision, career changes have a larger impact on wages than firm changes. To date, to estimate which types of switches are more costly, scholars analyse whether specific knowledge is more tied to occupations, firms, or industries. For instance, some scholars are more in favour of industry-specific human capital (Neal, 1995; Parent, 2000) while others prefer the idea of occupation-specific human capital (Poletaev and Robinson, 2008; Kambourov and Manovskii, 2009). Parent (2000) acknowledges that industry-specific human capital might measure something similar to occupation-specific human capital. Pavan (2011) criticizes that the importance of firm-specific human capital is regularly underestimated. With the exception of Poletaev and Robinson (2008), all studies mentioned above use tenure variables for industries, firms, and occupations, which prevents that knowledge can be divided into a sticky and a portable component. The advantage of task data is that they allow moving away from such a generic classification of specific or general skills and measuring instead the degree of similarities (general) or differences (specific) between portfolios. G&S (2010) are first to make use of this approach by combining task-specific human capital and a skill-weights approach to investigate job mobility. Although several of their wage regressions look at simultaneous firm and occupational switches, there is no explicit investigation of the differential impact of firm knowledge. The present paper fills this gap by being the first to conduct an analysis of specific versus general human capital on the firm level using the skill-weights approach, providing a new measure for firm knowledge. In addition, it moves beyond measures of specific knowledge to also capture the impact of a unique skill composition in firms. While specific human capital on the individual level is regarded as costly in the case of a job change, a specific knowledge structure on the firm level could motivate workers to cover long distances. Keeping in mind the transferability of knowledge, the question arises as to what degree careers might be structured to minimize the loss of knowledge and, simultaneously, optimize learning potential and wages. The majority of workers receive additional on-the-job training (Mincer, 1962; for an overview, 1989). Studies have estimated varying incidences of on-the-job training, but there is clear evidence for the importance of formal and also informal training (for an overview, Barron et al., 1997). Naturally, what can be learned depends on the job chosen and the workforce composition at the firm, as shown in organization research (e.g., Baron, 1984; Hannan, 1988), which has long emphasized that organizations shape jobs, e.g. through their size, growth, demography, technology, and unionization. Occupational intensity can be interpreted as a signal sent by the firm that shows either how large the firm’s market power and hence bargaining power is (reflected in lower wages) or how much a firm values an occupational group (as reflected in higher wages). Also, the match between firm and occupation tasks is directly related to occupational intensity but also takes into account other workers in the firm. It reflects the similarities of the own task profile with the complete task composition of the workforce in a firm. In more detail, from a supply-side perspective it can be argued that being around a larger peer group might be costly—because it reduces the uniqueness of own knowledge—but important at the beginning of an occupational career to develop own skills. Learning processes are time-consuming and can hence be costly, as individuals and firms cannot immediately reap the benefits of the investments. As wages decrease, occupational intensity increases (firm-occupation distance decreases). After some years, the importance of workers in the same group will decrease and potentially turn to zero because individuals have accumulated more knowledge, in sum providing fewer incentives to accept costs associated with a high occupational intensity (low firm-occupation distance) and potentially encouraging switches to firms with lower occupational intensities (higher firm-occupation distances). Occupational intensity (firm-occupation distance) should therefore decrease (increase) along the career path and wages should increase as the occupational intensity decreases. Opposite to this but in line with Bidwell and Briscoe (2010), a demand-side perspective suggests that organizations with a higher demand for a bundle of occupational skills also have a higher occupational intensity (lower firm-occupation distance). By providing more complex, challenging task combinations and being willing to also pay higher wages, the firms become very attractive for workers. Assuming that with increasing experience workers become even more attractive to firms would imply that occupational intensity increases (firm-occupation distances decreases) with experience. It further follows that wages increase as occupational intensity increases (firm-occupation distance decreases). Theoretically it appears plausible that both the supply and the demand effect are at work, but which one dominates is an empirical question. The remainder of this paper is organized as follows. Section 2 presents the conceptual framework. In Section 3, the data set and variables are introduced. Section 4 contains the results and a discussion of their implications. Section 5 concludes. 2. Conceptual framework To investigate the relationship between specific human capital and wages, I take as a starting point the conceptual framework by G&S (2010). The final framework allows disentangling the effects that occupational and firm knowledge have on worker productivity. It further helps to interpret the different roles that firm knowledge can play for switchers as regards costs and benefits, which are addressed in terms of firm-specific knowledge and knowledge composition of firms. Both firm and occupational knowledge are divided into a specific and a general component. In the following paragraphs, the focus is on changes to the framework of G&S that were implemented to incorporate firm knowledge.2 Ultimately, the idea is that the previous calculations with task data overestimate the importance of occupational knowledge by disregarding firm knowledge. As regards the terminology, Acemoglu and Autor (2011) suggest distinguishing skills from tasks because ‘a skill is a worker’s endowment of capabilities for performing various tasks’ (Acemoglu and Autor, 2011, p. 1045). They cannot, necessarily, be taken to be equivalent. Now, Lazear refers to skill weights, which in G&S’s approach are labelled task weights. Nonetheless, in light of the similarity of both approaches, it appears reasonable to assume that the idea behind both models is the same and, thus, does not change with labels. Preference is given to tasks because this is also in line with the methodology chosen in the empirical section. It is also acknowledged that the individual has one knowledge base and will not necessarily distinguish between occupational or firm knowledge although they may be of different value. However, in many data sets, information on the individual level is not available and instead researchers rely on information that is added on the occupational level from other sources. Nonetheless, individuals in the same occupation differ as regards the overall accumulated human capital because of different work experiences at different firms, introducing additional variation across individuals within the same occupation. As sketched above, firm knowledge is understood as the result of a unique variation of otherwise general skills, implying that both occupations and firms weigh skills in unique ways that result in a unique skill portfolio. Also, from a firm’s perspective it is of interest what the individual has learned through experience in another firm as regards organizational practices when recruiting personnel. It is hence assumed that firm knowledge can enter in a wage equation separately. In the end, there will be one price but two components contribute separately to it. To illustrate the basic idea, the two knowledge types will be handled separately in the following framework. Suppose that the output in a job is determined by fulfilling a variety of general tasks that become specific by the relative importance attached to them in an occupation and in a firm. Following Lazear and G&S, my approach uses two tasks j, which can be interpreted as analytical and manual tasks (=A, M). The productivity (S) of a worker (i) varies by jobs (g) and by the time spent in the labour market (t). The jobs are a combination of the occupation (o) and the firm (f) for each task (with, for instance, βg=βo+βf). The relative weight β shows the importance of tasks in an occupation o or firm f. G&S suggest that the importance corresponds to the time spent on that task. Worker i’s productivity (measured in log units) in occupation o at firm f and at time t is   ln Sifot=[(βo+βf)tioftA+(1−(βo+βf))tifotM]+αfoXfot. (1) G&S’s original equation ( ln Siot) is adapted to additionally capture the firm dimension ( ln Sifot) in terms of the weights attached to tasks. The task composition on the firm level results from the occupational task structure of the workforce, reflecting an interaction between the occupational and firm level. As mentioned before, this allows capturing the idea that all knowledge is general and that both firm and occupation can attach different weights to this knowledge. In other words, the analysis decomposes the βs into a firm and occupational component which jointly determine the worker’s productivity. It further allows moving a step closer to what the workers know on the individual level because different influences of the career path are taken into account. For instance, a worker might hold an occupation that is specialized in manual tasks but choose to work in a firm that places more weight on analytical tasks. The worker’s skill portfolio then changes while working at the firm when compared to a worker with the same occupation but whose employer puts a higher weight on manual tasks. Again, this variation would normally not be captured in the empirical analyses. It is now possible to calculate the absolute distances between current and previous firms (occupations) by comparing their task weights, for instance, |βf−βf'| ( |βo−βo'|). The more similar firms (occupations) are, the smaller the absolute difference. Although the empirical analyses consider multiple tasks, it is sufficient to consider only two at the moment to illustrate the logic behind the analytical setup. While the weights for task-specific human capital vary on the individual level, worker productivity can also be influenced by the knowledge composition of the firm, as argued above. In other words, task productivity not only arises from the weights placed on individual tasks but also from the weights of the tasks’ combinations. This implies that firms might view certain skills as complementary and prefer these to be used in combination on the individual level. A combination that is very relevant to workers is the one that corresponds to the own occupation and which is captured by Xfot as the share of workers in the same occupational group in a firm or the differences between the own and the firm’s task composition. As outlined above, the relative size of occupational peers or the degree of similar tasks at the current firm can but does not need to be of advantage; the direction of the relationship is determined by the parameter αfo. It takes on positive values ( αfo>0) when the share of firms prepared to pay higher wages for a unique skill composition is larger than the share of firms that pays lower wages, reflecting the trade-off that firms face between attracting suitable workers (higher wages) and making use of their own market power (lower wages). A corollary that can be tested in this context is that as experience increases, occupational intensity/firm-occupation distance can either decrease or increase. Xfot also influences the previously accumulated human capital, as outlined below in eq. (3). Next, the worker’s task productivity in a job tigtj (where go,f}) needs to be determined with   tigtj=tij+γgHigtj(j=A,M), (2) where tij describes the ability of worker i in a certain task j (initial endowment). Higtj includes all previously accumulated human capital of worker i in task j in different jobs. This equation illustrates the influence of previously acquired human capital on both the firm and occupational level, which is combined into one job. Hence, in contrast to G&S, I account for variation on the firm level and, correspondingly, on the occupation level, which is necessary if I want to investigate the difference between firm- and occupation-specific human capital. The equation incorporates the idea that workers gain more knowledge on the job. The degree to which this can be achieved in a certain task t depends on the importance of βg, which is assumed to be captured with the time spent on a task. The more experienced workers are, however, the lesser they can learn. This can be written as   HigtA=βg′Fig′tHigtM=(1−βg′)Fig′t︸ experiencein  priorjobs, (3) where Fig't is the experience of worker i in previous jobs, as indicated by the prime. Since all knowledge is general in this approach, acquired human capital is a function of the combination of task weights on the firm and occupational level. Combining the equations above gives   lnSifot=γfo[(β0+βf)HifotA+(1−(β0+βf))HifotM]︸Tifot+(β0+βf)tiA+(1−(β0+βf))tiM︸mifot+αfoXfot, (4) where γfo measures the returns to task-specific human capital of jobs. Tifot can be observed as a time-variant measure of task-specific human capital; and mifo is the unobservable match to the job that does not vary over time. The equation further includes Xfot, which represents occupational intensity. To investigate labour mobility, wages in different jobs need to be compared. These are determined by multiplying the productivity with the skill or tasks prices of jobs  Pfo that is wifot=Pfo*Sifot. Next, the equation is logarithmized and yields the following expression:   lnwifot=(pfo︸skill pricejob+γfoTifot +mifo︸ln Sifot)+lnαfoXfot︸occupational intensity, (5) where pfo=lnPfo ( pfo=lnPfo). Equation (5) can be used to investigate labour mobility of workers. Like Lazear, G&S suggest a two-period setup where the worker has to decide whether to stay or to switch jobs in the second period. A job switch occurs when   ln wifot︸wages in new job >ln wif'o't︸wage in previousjob . This equation can be rearranged as follows:   pfo-pf'o'+mifo-mif'o'+lnαfoXfot-lnαf'o'Xf'o't+γfoTifot>γf'o'Tif'o't, (6) which shows that what is paid for task-specific human capital in the previous job must be exceeded by the sum of the returns to task-specific human capital in the new job, the difference of skill prices, of the task match, and the improved learning environment. To illustrate the influence of the βs the equation can be rewritten as   pfo-pf'o'+γfo-γf'o'Tif'o't︸wage growth in job+mifo-mif'o'︸task match+lnαfoXfot-lnαf'o'Xf'o't︸changes in task environment>-γfo(βo+βf)-(βo'+βf')HifotA-HifotM︸-γfoTifot-Tif'o'tloss in human capital. (7) The right-hand-side term in eq. (7) shows the loss in task-specific human capital where one can again see the influence of the difference between the βs. The left-hand side is the sum of the difference of the firm task match, the wage growth attributable to an increase in skill prices, and the returns to previously acquired task-specific human capital. In addition, it shows the advantages resulting from a good match in terms of the task environment. When the weights of the knowledge composition are of disadvantage, the term for the task environment could be moved to the right-hand side of the equation, as it adds to the costs associated with switches. 2.1 Empirical predictions The analytical setup yields the following intuitive results, part of which were tested for the case of occupational human capital by G&S but, according to my argument, should simultaneously matter for human capital at the firm level. Although all these predictions should hold equally across qualification levels, the overall knowledge base is expected to differ by the type of knowledge base. It is to be expected that what determines switching behaviour, like the distance of a switch, also differs by qualification level. That is why the qualification categories are investigated separately in the empirical analysis. The first prediction states that less task-specific human capital is lost when the switch takes place between firms that are more similar with regard to their task composition. Therefore, switches occur more often between similar firms. Second, the distance covered in a switch will be the highest early in the career because the likelihood that the left-hand side of eq. (7) is greater than zero becomes less likely with experience. Specifically, during early years of employment, workers are still looking for their best possible match, which might include a certain amount of trial and error. After having spent a longer time in the labour market, workers are less likely to travel long distances because, possibly, they have already found a good match, and as they have accumulated more knowledge, switches become more costly. Third, wages at the source firm are expected to be a better predictor of wages in the target firm if both positions require similar tasks. This follows from the idea that with a higher number of transferable skills, a better match is achieved because distances are shorter. Fourth, both occupation and firm knowledge matter for wages. When investigating joint switches (that is, simultaneous firm and occupation switches), I thus take into account both knowledge types to compare their relative importance. Finally, the task match of the worker and the firm can affect wages negatively because of a time investment in learning by the employee. The theoretical counterargument suggests a positive relationship, showing that a set of occupational skills is more highly valued by firms. The direction of the effect depends on the parameter value of αfo, which will be determined in the empirical analysis. The analysis carried out in this paper is innovative in its methodological measurement of firm knowledge, providing a direct empirical measure for firm-specific knowledge, and it incorporates a unique test of potential (dis-)advantages in terms of learning opportunities arising from the occupational composition of firms. 3. Data 3.1 Data sources Three data sources are accessed for the analysis. The first data set is the weakly anonymous Sample of Integrated Labour Market Biographies (SIAB). Data access was provided via on-site use at the Research Data Centre of the German Federal Employment Agency at the Institute for Employment Research and subsequently remote data access. The SIAB contains a very long observation period (1975–2008) and information on labour market histories of 1.5 million individuals in Germany (Dorner et al., 2010). It is the most comprehensive administrative micro-level data set on employment histories currently available for Germany. In addition, it is possible to link the establishment information of the Establishment History Panel (BHP) to the SIAB. This combination of individual labour market histories (SIAB) and firm employment structure (BHP) makes the data perfectly suited for this analysis. The SIAB provides information on wages and occupations of individuals, and the BHP has information on the occupational categories of all employees in a firm. A detailed description of the final data set is included in the Appendix. In short, I restrict the analysis to men, to employees with an average daily wage of at least 10 euros, and to voluntary switches. The lower the wage is, the higher the likelihood that individuals may hold another job to make a living. For instance, self-employment is not reported to the Research Data Centre of the German Federal Employment Agency and can hence not be controlled for. To identify and later exclude involuntary switches, I start with job switches where simultaneously structural changes occurred in the firm, for instance, a change of ownership or the firm’s exit from the market. This group is augmented with other involuntary switchers, who are identified by receiving unemployment benefits immediately after leaving the firm. Note that in Germany, workers who give notice, in contrast to being given notice, may not receive unemployment benefits for three months. Voluntary switches are more likely to purposefully decide on new firms, while involuntary switchers are more likely to be forced into their new job and hence willing to accept any firm environment. The classification of individuals and firms according to their task sets requires, of course, information on tasks. The BIBB/BAuA Employment Survey 2006 (Hall and Tiemann, 2006; Rohrbach-Schmidt, 2009), which was undertaken in 2005 and 2006 by the Federal Institute for Vocational Education and Training (BIBB) and the Federal Institute for Occupational Safety and Health (BAuA), provides all necessary information. This wave consists of a random sample of 20,000 people who are active in the labour force in Germany. In addition to individual-specific data, the survey includes information on the task requirements of occupations. For further examples using this database, see Spitz-Oener (2006, 2008) and Borghans et al. (2014). The BIBB/BAuA data are merged by occupation (SIAB) or occupational groups (BHP). 3.2 Variables The dependent variable is the logarithm of wage, as proposed in the analytical setup. Wage is measured as gross daily income of employees and reported in euros. Occupational intensity—the share of occupational peers—is calculated by dividing the number of workers in the same occupational group, using Blossfeld categories, by the total number of workers in the firm. The Blossfeld classification, which is the only available unit for occupations on the firm level in the BHP, is based on the three-digit occupation of an individual as specified by the employer in the notification to the social security agencies. Blossfeld first distinguishes between three upper-level groups, namely, production, service, and administration, and second, ranks occupations according to the type of required skills. Accordingly, blue-collar workers who perform simple manual tasks and white-collar workers who provide simple services are regarded as unskilled; blue-collar workers engage in complicated tasks, white-collar workers perform qualified tasks, and semi-professionals are regarded as skilled workers. The third and most highly qualified group includes engineers, technicians, professionals, and managers. The Blossfeld classification thus assigns upper-level groups and then ranks individuals according to their skill requirements. To address a potential bias in the estimation, I further calculate the degree of occupational diversity on an industry level (28 industries) to create a Herfindahl index and use this variable as an instrument for occupational intensity. The idea is that firms located in diverse industries will exhibit lower shares of occupational intensity. There is, however, no empirical evidence that occupational diversity on the industry level has an effect on wages. A detailed description of the instrument variable Herfindahl Blossfeld can be found in Section 4.3. To measure general work experience, I calculate the number of years someone has worked since labour market entry by using information on the exact number of working days, excluding periods of unemployment. It is common practice in wage regressions to include a squared term for work experience because a concave relationship is in line with changes that occur later along the career path. This specification is more restrictive than suggested by the analytical setup but still in line with the general idea. I distinguish three levels of education in the regressions. Low-skilled workers are defined as those who did not pass the Abitur (German university entrance qualification) and have not completed apprenticeship training. This also includes unskilled workers. Medium-skilled workers passed the Abitur and/or have completed nothing above an apprenticeship. High-skilled workers hold a degree from a university or university of applied sciences. Incentives to switch firms can be driven by regional characteristics and, therefore, controls for region types are introduced. Additional controls include, as dummies, years, industry, occupational groups, and the logarithm of establishment size. The summary statistics as well as correlations for the most important variables can be found in Tables A.4 and A.5 in the Appendix. The analyses of employment biographies are carried out separately for men and women because the two groups are known to show significant differences in terms of wages and employment careers. For instance, due to fertility decisions, women are known to behave differently from men in their human capital accumulation over the life cycle. Note that I only report the results for men, leaving the results for women to the discussion at the end of the section. The measures for task distance between occupations, using the BIBB/BAuA data, include both men and women (see Section 4.1). In the regressions, all variables are standardized. 4. Analysis 4.1 A task-based measure for specific human capital of firms The main variable of interest is a measure of the firm- and occupation-specific human capital. G&S group tasks manually into three categories: analytical, manual, and interactive. This categorization makes it possible to combine tasks from different years. As they show, the task content of occupations changes only slightly. In contrast, this paper lets the data structure determine the task groups, which has the advantage of allowing me to take into account more tasks because they do not have to be included in every survey wave. The disadvantage is that this procedure cannot be carried out with every survey because tasks do, and therefore factors would, vary. Thus, I rely on G&S’s result that, over time, task variation in occupations is low, and I instead use a factor analysis. Here, a principal factor analysis shows whether certain tasks need to be clustered on the occupational level in latent variables. The first advantage of this procedure is an easier interpretation of the data due to condensed information and orthogonal factors. In addition, since task level is determined by executing a task regularly or by the degree of expert knowledge required, it takes more than a high value in one task to end up with a high value in a factor. Thus, the factor reflects the task level in a certain domain and the level can change through adjustments of different tasks. Indeed, using exploratory factor analysis on high-dimensional task data is in line with Green (2012), who also discusses the risks associated with classifying tasks by hand. In addition, my analysis does not need to reproduce existing categorizations nor interpret the results in the framework of routine versus non-routine tasks, as done by Autor et al. (2003). Indeed, Rohrbach-Schmidt and Tiemann (2013) show that significant changes in the questionnaire can bias analyses that combine data from all cross-sections of the German data. In sum, although using task data from different survey waves can be of advantage, it is ultimately a trade-off between (1) exogenously determined task categories to which survey questions from different years are assigned; and (2) endogenously determined task categories from one observation period. In the current context, applying factor analysis (option (2)) is considered to be the more appropriate procedure. A selection of 31 survey questions from the BIBB/BAuA Employment Survey 2006 gives information about tasks applied in the employee’s current job. The closest approximation to tasks of individuals in this context is achieved on the occupational level. It is acknowledged that this procedure only allows measuring variation on a more aggregate level, but this is still an improvement to previous tenure measures. The survey question asks respondents to assess the task level that they use in their current position. The calculations of the factor analysis return seven factor variables that explain around 91% of the total variation in 248 occupations (see Table A.2, for an overview, and Appendix A, for details on the data and computations). The factors are then labelled according to their content, which is the combination of certain tasks, placing most emphasis on the variables that load the highest. This is similar to what Poletaev and Robinson (2008) and Nedelkoska and Neffke (2011) do. The factor labels are: intellectual, technological, health, commercial, instruction, production, and protection. To make the occupational classification more transparent, Table A.3 reports the occupations with the highest and lowest values in each factor. The example occupations set out in the table make intuitive sense, thus confirming the plausibility of the principal factor analysis. For instance, the technological factor has a strong focus on the application of technological and manual knowledge, both of which are characteristics of occupations such as aircraft engine mechanic or optometrist. The health factor is most important for various types of medical practitioners and other occupations in the healthcare system. More routine tasks like producing and manufacturing goods, measuring, testing, and operating machines load highest in the production factor, which is where occupations such as machine operators for dairy and paper products are found. The task composition of the workforce is determined with information on the 12 occupational groups by Blossfeld (1985; see Table A.1) and task data. The occupational classification does not allow seeing whether the firm employs workers in the same three-digit occupation as held by the switcher. From an employee perspective, however, it is unlikely that they have detailed information as to all the occupations of prospective co-workers. Thus, the Blossfeld classification appears to be an adequate indicator of one aspect that is driving a (voluntary) job switcher’s decision. Task factors for each Blossfeld group are calculated as follows. First, the average factor value of each task is determined for all occupations that belong to one Blossfeld group ( tb). These task factors are then weighted by multiplying them with the corresponding number of workers in a firm in that Blossfeld group ( nfb). Since the focus is the structure of the workforce, this value is divided with the sum of all weighted task factors to calculate the relative importance of a task factor in a firm.   Task importance in firms= tb*nfb∑b=1ntb*nfb This procedure captures a firm’s unique task composition, which is translated into task weights on the firm level. As soon as larger differences between firms can be detected, it can be expected that workers’ skill portfolios—although they might be holding the same occupations—will differ according to variations on the firm level. The more similar firms are with regard to the task composition, the more firm knowledge can be reapplied by the worker after a switch. Job switchers are, thus, also included because they are part of the firm’s task structure. This procedure returns the relative importance of tasks in a firm to avoid the notion that firm size drives differences. Next, the distance of firms/occupations is determined by using the angular separation or uncentered correlation of two vectors representing two firms/occupations (for details on the computational method, see Jaffe, 1986; G&S, 2010). The equation is   AngSepgg'=1-∑j=1Jqjg*qjg'∑j=1Jqjg2*(∑k=1Jqkg'2)1/2  Distancegg'=1-AngSepgg', where q is the vector of all tasks in a firm/occupation. The measure is slightly adjusted so that a value of 1 (0) means that the firms/occupations are completely different (identical). This distance measure reflects the differences between firms or occupations with regard to their task-specific human capital. The measure for occupations is calculated on the basis of the original factors of occupations from the factor analysis (see Table A.2). For firms, the relative task importance is compared between origin and target firm. For firm-occupation distance, the factors from the firm and the occupation level are laid against each other. 4.2 Transferability of firm and occupational knowledge In what follows, the analysis always distinguishes between qualification levels of employees. This is important because the amount of human capital and, thereby, general and specific knowledge can be expected to differ between groups. First, the share of switches by different firm distance intervals is calculated. The results in Fig. 1 show that the majority of switches involve low firm distances. The largest share of joint occupational and firm switches occurs in the lowest firm distance interval, confirming that switches occur more often between similar firms.3 For the first two intervals, which cover around 90% of the sample, firm distance appears to decrease when qualification level increases. Possibly, workers with higher qualification levels can be more selective in choosing a suitable target firm or low-skilled employees are to a smaller degree affected by firm distance. The figure also reveals, as a control analysis, that the distribution differs for layoffs which cover slightly longer distances than voluntary switchers. As announced, layoffs are thus excluded from the following analysis. Fig. 1 View largeDownload slide Distribution of joint switches across firm distance intervals Fig. 1 View largeDownload slide Distribution of joint switches across firm distance intervals In Fig. 2, I investigate the relationship between firm distance and different years of work experience for all male employees. The graph shows the predicted margins with confidence intervals when firm distance is regressed on work experience. The more experienced workers are, the smaller firm distance becomes. Across qualification groups, the negative trend turns out to be very similar. In general, low-skilled have the largest average firm distance values, followed by medium- and high-skilled workers. The results could be interpreted as evidence that workers accumulate more specific knowledge or achieve a better match along the career path, providing less incentive to cover larger distances and pay associated costs. Fig. 2 View largeDownload slide The relationship between work experience and firm distance Fig. 2 View largeDownload slide The relationship between work experience and firm distance Following, the relationships between task-specific human capital at the occupational as well as firm level and wages are investigated. The analysis focuses on switchers and builds on the analytical framework by estimating the following equation:   lnwifot= γoDot+γfDft+αXoft+βZifot+εifot, where the dependent variable is the logarithm of wage ( lnwifot), D is the distance on the firm or occupational level, Xoft is occupational intensity (replaced with firm-occupation distance in Section 4.3.2), Zifot is a vector of control variables, and ϵ is the error term. The equation is first estimated using ordinary least squares in the baseline, and in the next section using two-stage least squares. All regressions report coefficients based on standardized variables which are needed to compare the relative contribution of occupational and firm knowledge in explaining the variation of the model.4 Past wages are included as a control measure for the reservation wage and, in addition, interactions between past wages and task distance follow the idea that wages at the source firm are expected to be a better predictor of wages in the target firm if both positions require similar tasks. The estimations further include work experience, work experience squared, firm size, as well as dummies for occupational groups, regions, industry, and years. It is acknowledged—for instance by not claiming causal relationships—that the estimation procedure cannot account for endogeneity in the decision to switch jobs, leading to a potential bias in the estimations. Hence, to at least increase homogeneity in the group, the focus remains on voluntary switchers. Nonetheless, the results continue to provide information on the relative importance of firm and occupational knowledge, which is the goal of this exercise and the preparation for the analysis of (dis)advantages of firm knowledge in the following section. Table 1 reports the OLS results by qualification level, following a stepwise inclusion of the variables. The baseline specification in column A shows that previous wage and current firm size contribute positively to the current wage. Work experience has a positive relation with the current wage, but the coefficient decreases over time. Occupational distance shows a negative sign. I continue by replicating the results by G&S, using only occupational distance (columns B–C).5 As mentioned earlier, their wage regressions usually look at simultaneous firm and occupational switches but they do not include a task-based measure for firm knowledge. Across qualification groups, most variables have the expected signs. Occupational distance decreases the current wage. Except for high-skilled workers (column 13), previous wage correlates positively with the current wage but the coefficient decreases with increasing task distance. Columns D and E complement the previous estimations by including the firm distance variables. With one exception for low-skilled workers (column 4), firm distance matters in addition to occupational distance. With the exception of high-skilled employees, the interactions between the distance measures and previous wages are significant (column 15). Whenever both firm and occupational distance measures are significant, the coefficient of occupational distance is roughly twice as large as the one of firm distance, reflecting that a good occupational match is relatively more important. So far, the results confirm that the newly constructed measure for firm knowledge plays a significant role in explaining wages in target firms. In all specifications, occupational intensity contributes negatively to wage, but further evidence is needed to corroborate this finding. From this I can conclude that the skill-weights approach as implemented by G&S, whose results are replicated in the regressions, can and should be extended to firm knowledge. An occupational measure for specific knowledge picks up part of the story but cannot fully capture the costs of losing human capital. On average, the sum of the coefficients suggests that the total costs of specific knowledge are one and a half times as large as when estimated only with occupation-specific variables. Hence, firm-specific human capital costs on average half of the sum of occupation-specific human capital. At this point, instead of proceeding the same way as G&S did, the focus will now shift to a more in-depth investigation of firm knowledge in terms of on-the-job learning. Table 1 Specific knowledge—distance of switches and the correlation of wages (OLS) Depvar: Current wage (log)  A  B  C  D  E  Low qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.111***  0.110***  0.155***  0.109***  0.167***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Occ intensity  –0.009**  –0.009**  –0.008**  –0.009**  –0.008**  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        –0.001  –0.009***        (0.002)  (0.002)  Firm dist * previous wage          –0.013***          (0.002)  Occ distance    –0.006***  –0.017***  –0.006***  –0.015***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.020***    –0.017***      (0.002)    (0.002)  Firm size (log)  0.095***  0.095***  0.094***  0.094***  0.093***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.800***  0.799***  0.794***  0.804***  0.796***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.600***  –0.600***  –0.598***  –0.602***  –0.597***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.014  –0.003  0.032  –0.220***  –0.179***  (0.082)  (0.081)  (0.082)  (0.063)  (0.063)  Observations  32,306  32,306  32,306  31,516  31,516  R-squared  0.324  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.176***  0.167***  0.233***  0.163***  0.242***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Occ intensity  –0.032***  –0.031***  –0.030***  –0.028***  –0.027***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.014***  –0.015***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Firm size (log)  0.068***  0.068***  0.067***  0.065***  0.063***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.476***  0.464***  0.455***  0.467***  0.456***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.292***  –0.282***  –0.278***  –0.285***  –0.281***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.112  –0.064  –0.029  –0.345***  –0.292***  (0.073)  (0.073)  (0.074)  (0.036)  (0.036)  Observations  100,935  100,935  100,935  98,363  98,363  R-squared  0.406  0.413  0.418  0.415  0.421  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.155***  0.145***  0.152***  0.142***  0.151***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Occ intensity  –0.022***  –0.024***  –0.024***  –0.022***  –0.022***  (0.005)  (0.005)  (0.004)  (0.005)  (0.005)  Firm distance        –0.012***  –0.011***        (0.002)  (0.002)  Firm dist * previous wage          –0.003          (0.003)  Occ distance    –0.035***  –0.035***  –0.033***  –0.034***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.004    –0.003      (0.003)    (0.003)  Firm size (log)  0.075***  0.075***  0.075***  0.072***  0.072***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.306***  0.304***  0.308***  0.306***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.201***  –0.200***  –0.205***  –0.203***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.610**  –0.522*  –0.523*  –0.209  –0.209  (0.289)  (0.282)  (0.282)  (0.138)  (0.139)  Observations  21,669  21,669  21,669  21,381  21,381  R-squared  0.437  0.444  0.444  0.445  0.445  Depvar: Current wage (log)  A  B  C  D  E  Low qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.111***  0.110***  0.155***  0.109***  0.167***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Occ intensity  –0.009**  –0.009**  –0.008**  –0.009**  –0.008**  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        –0.001  –0.009***        (0.002)  (0.002)  Firm dist * previous wage          –0.013***          (0.002)  Occ distance    –0.006***  –0.017***  –0.006***  –0.015***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.020***    –0.017***      (0.002)    (0.002)  Firm size (log)  0.095***  0.095***  0.094***  0.094***  0.093***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.800***  0.799***  0.794***  0.804***  0.796***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.600***  –0.600***  –0.598***  –0.602***  –0.597***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.014  –0.003  0.032  –0.220***  –0.179***  (0.082)  (0.081)  (0.082)  (0.063)  (0.063)  Observations  32,306  32,306  32,306  31,516  31,516  R-squared  0.324  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.176***  0.167***  0.233***  0.163***  0.242***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Occ intensity  –0.032***  –0.031***  –0.030***  –0.028***  –0.027***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.014***  –0.015***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Firm size (log)  0.068***  0.068***  0.067***  0.065***  0.063***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.476***  0.464***  0.455***  0.467***  0.456***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.292***  –0.282***  –0.278***  –0.285***  –0.281***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.112  –0.064  –0.029  –0.345***  –0.292***  (0.073)  (0.073)  (0.074)  (0.036)  (0.036)  Observations  100,935  100,935  100,935  98,363  98,363  R-squared  0.406  0.413  0.418  0.415  0.421  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.155***  0.145***  0.152***  0.142***  0.151***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Occ intensity  –0.022***  –0.024***  –0.024***  –0.022***  –0.022***  (0.005)  (0.005)  (0.004)  (0.005)  (0.005)  Firm distance        –0.012***  –0.011***        (0.002)  (0.002)  Firm dist * previous wage          –0.003          (0.003)  Occ distance    –0.035***  –0.035***  –0.033***  –0.034***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.004    –0.003      (0.003)    (0.003)  Firm size (log)  0.075***  0.075***  0.075***  0.072***  0.072***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.306***  0.304***  0.308***  0.306***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.201***  –0.200***  –0.205***  –0.203***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.610**  –0.522*  –0.523*  –0.209  –0.209  (0.289)  (0.282)  (0.282)  (0.138)  (0.139)  Observations  21,669  21,669  21,669  21,381  21,381  R-squared  0.437  0.444  0.444  0.445  0.445  Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of OLS regressions. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(5) are workers with low, (6)–(10) with medium, (11)–(15) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1 4.3 On-the-job learning The descriptive evidence documents the importance of learning in firms. The BIBB/BAuA survey (N = 15,796 with sample restrictions as defined above) shows that 78.2% of the workers need on the job either a longer training or instruction to carry out their current activities and 60% declare to need special courses or trainings. A total of 78.3% often receive support from colleagues and 58.3% from supervisors. In addition, 23.8% have acquired their skills primarily and 37.3% secondarily through experience. In the latter case, this is the answer that was most often chosen among all options. These results confirm that, for the majority of workers, their previously acquired knowledge did not match perfectly the job that they were carrying out. Instead, additional effort was required to learn, for instance, from colleagues while working on the job. Figure 3 shows the relationship between work experience and occupational intensity for all male employees, based on the margins from a regression. The lower the number of observations is—due to either the qualification level or the decreasing number of observations by work experience—the larger the standard errors become. Nonetheless, the first working years, which are of particular interest, show lower variation and provide an interesting indication of the development of occupational intensity. There is a downward trend across all qualification levels, with the exception of the first working years of low-skilled workers, where the margins increase. Fig. 3 View largeDownload slide The relationship between work experience and occupational intensity Fig. 3 View largeDownload slide The relationship between work experience and occupational intensity To address the relationship between wages and learning opportunities in more details, I estimate two-stage least square regressions. Learning opportunities are measured with occupational intensity or the firm-occupation distance. Both measures are closely related but tell different stories. Thus, a high occupational intensity (large share of co-workers in the employee’s occupation) should relate to a low firm-occupation task distance. This follows from the concepts that (1) firms resemble the occupational structure of industries; and (2) similar (different) firm and occupational tasks imply small (large) differences. Due to the negative relationship between occupational intensity and firm-occupation task distance, I would also expect the sign in front of the coefficient to change when compared to the occupational intensity. This new measure allows me to compare two aspects of on-the-job learning: (1) Do workers benefit from co-workers who show the same or a different task portfolio (occupation perspective, measured with occupational intensity)? (2) Do workers benefit from working in a firm that shows a similar or a different task portfolio (firm perspective, measured with firm-occupation distance)? Naturally, the two aspects are closely related because when all co-workers and the employee share the same occupation, the firm-occupation distance is zero. However, firm-occupation distance also captures similarities with the task portfolios of other occupations, not subsuming them into one group. It is thus likely to show smaller changes than occupational intensity when different occupations are added to a firm. Conceptually, it does not make sense to include both measures in a regression because they both measure knowledge concentration with occupational data (either aggregate tasks in occupations or pure task data), and thus they are investigated separately. Without instrumenting these variables, it is unclear whether the observed relationships result from a specific sorting into firms. As discussed before, workers may select firms that pay lower wages because they expect to profit via on-the-job learning. Alternatively, they could expect higher wages when working in more specialized firms. A priori it is unknown whether occupational intensity then serves as a reliable measure for different learning opportunities. For instance, a larger firm-occupation distance may capture other costs related to switches. A higher occupational intensity may reflect a firm’s specialization instead of pure learning opportunities. To ensure that occupational intensity does not proxy another firm characteristic, I use the Herfindahl Blossfeld index as an instrument for learning opportunities (occupational intensity or firm-occupation distance) to measure occupational diversity on an industry level, as opposed to the standard procedure of measuring industrial diversity on a regional level. A Herfindahl index can be understood as a concentration or diversity measure where the minimum value of 0 corresponds with an equal distribution of shares while the maximum value of 1 describes the concentration on one share. The occupational composition of industries should relate to the occupational composition of firms because industrial and firm labour demand should be highly correlated. In the data, the newly created Herfindahl Blossfeld index is significantly positively but moderately related to occupational intensity (r = 0.2943, see Table A.5) and significantly but only weakly related to firm-occupation distance (r = –0.1325). To my knowledge there is no evidence yet that being in an occupationally diverse industry has a direct impact on wages. I therefore start by investigating several potential connections, which one may theoretically expect, and report the results here. The Herfindahl Blossfeld index shows indeed negligible correlations with all the other variables in the analysis, most importantly with wages (r = 0.0436). Technically, an indirect relationship could exist since workers’ wages can be understood as a function of industry productivity which in turn is determined by firm size and the type of workers needed. The calculations show, however, that the correlation between firm size and the Herfindahl Blossfeld index is very low (r = –0.0295, see Table A.5). Also, since the Herfindahl Blossfeld index is calculated as a percentage measure, it should be independent of the number of workers. There is in addition no evidence for an important correlation between workers’ qualification levels and the Herfindahl Blossfeld index, ruling out that the index serves as a proxy for the positive effect of education on wages. In any case, the regressions are carried out separately for qualification level. To avoid that the final results are driven by unobserved, indirect connections or that the Herfindahl Blossfeld index picks up other industrial or firm characteristics, control dummies for the 28 industries of the Herfindahl Blossfeld index, for firm size and for occupational groups (as a measure for type of worker), are included. Therefore, as the Herfindahl Blossfeld varies on the industry and year level and since the regressions include industry controls, the Herfindahl Blossfeld then picks up variation arising from changes in the workforce composition across years. Looking at the development of the index across years and industries shows clear differences between industries, confirming that the variable captures additional industry-specific characteristics. 4.3.1 (Dis-)advantages of occupational intensity in firms Across all qualification levels, the OLS results suggest that occupational intensity and wages correlate negatively. To address potential endogeneity, Table 2 reports the 2SLS results. The odd-numbered columns (also columns A and C) show the first-, the even-numbered columns (also columns B and D) the second-stage regressions. The specifications are the same as the final regressions in columns D and E of Table 1, including all variables of interest.6 The Herfindahl Blossfeld index relates positively to occupational intensity in the first regression in all estimations. The F-statistic on the excluded instrument is always clearly above the threshold of 10. The results for occupational intensity in the second stage clearly differ from the OLS regressions. Occupational intensity becomes insignificant for low-skilled but positively significant for medium- and high-skilled workers. The pattern and signs of the other variables reflect closely the OLS results in Table 1. One important difference is that the coefficients of firm distance increase substantially for medium- and high-skilled employees but are insignificant for low-skilled employees. Nonetheless, occupational distance continues to show larger coefficients than firm distance. Interpreting the results against the construction of the variables shows that the change in the coefficients of occupational intensity is due to an important correlation between firm and industry patterns in workforce composition. In the OLS estimations, occupational intensity also captures labour demand in the form of workforce changes on the industry-year level, leading to biased results. Table 2 The (dis)advantages of occupational intensity (2SLS)   First stage   First stage   Depvar: Current wage (log)  A  B  C  D  Low Qualification  (1)  (2)  (3)  (4)    Occ intensity    –0.063    –0.063    (0.069)    (0.069)  Herfindahl Blossfeld  0.132***    0.132***    (0.014)    (0.014)    Occ distance  –0.019***  –0.007***  –0.011***  –0.016***  (0.003)  (0.002)  (0.004)  (0.002)  Occ distance * previous wage      0.014***  –0.016***      (0.004)  (0.003)  Firm distance  0.055***  0.002  0.055***  –0.006  (0.003)  (0.004)  (0.004)  (0.004)  Firm dist * previous wage      0  –0.013***      (0.004)  (0.002)  Previous wage (log)  –0.002  0.109***  –0.032***  0.165***  (0.006)  (0.004)  (0.010)  (0.007)  Firm size (log)  –0.231***  0.081***  –0.230***  0.080***  (0.007)  (0.016)  (0.007)  (0.016)  Work experience  0.096***  0.809***  0.099***  0.801***  (0.022)  (0.017)  (0.022)  (0.017)  Work experience 2  –0.104***  –0.608***  –0.106***  –0.603***  (0.028)  (0.023)  (0.028)  (0.023)  Constant  –0.672***  –0.163***  –0.695***  –0.129***  (0.096)  (0.038)  (0.096)  (0.038)  Observations  31,516  31,516  31,516  31,516  R-squared  0.208  0.321  0.209  0.324  F-statistic  87.044***    86.772***      Medium qualification  (5)  (6)  (7)  (8)    Occ intensity    0.096***    0.079***    (0.020)    (0.020)  Herfindahl Blossfeld  0.223***    0.223***    (0.007)    (0.007)    Occ distance  –0.001  –0.029***  0.002  –0.034***  (0.002)  (0.001)  (0.002)  (0.001)  Occ distance * previous wage      0.018***  –0.030***      (0.002)  (0.002)  Firm distance  0.051***  –0.020***  0.050***  –0.021***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      –0.007***  –0.012***      (0.002)  (0.001)  Previous wage (log)  –0.028***  0.167***  –0.053***  0.247***  (0.004)  (0.003)  (0.006)  (0.005)  Firm size (log)  –0.249***  0.096***  –0.249***  0.089***  (0.004)  (0.005)  (0.004)  (0.005)  Work experience  –0.088***  0.478***  –0.084***  0.466***  (0.013)  (0.009)  (0.013)  (0.009)  Work experience 2  0.031**  –0.289***  0.029**  –0.284***  (0.013)  (0.009)  (0.013)  (0.009)  Constant  0.374***  0.059**  0.359***  0.065***  (0.067)  (0.023)  (0.068)  (0.023)  Observations  98,363  98,363  98,363  98,363  R-squared  0.194  0.386  0.194  0.399  F-statistic  966.873***    972.649***    High qualification  (9)  (10)  (11)  (12)    Occ intensity    0.127***    0.128***    (0.034)    (0.034)  Herfindahl Blossfeld  0.215***    0.213***    (0.014)    (0.014)    Occ distance  –0.030***  –0.029***  –0.031***  –0.029***  (0.004)  (0.003)  (0.004)  (0.003)  Occ distance * previous wage      0.012***  –0.005      (0.004)  (0.004)  Firm distance  0.061***  –0.021***  0.063***  –0.021***  (0.004)  (0.003)  (0.004)  (0.003)  Firm dist * previous wage      –0.021***  0.001      (0.004)  (0.003)  Previous wage (log)  –0.006  0.143***  –0.001  0.150***  (0.006)  (0.005)  (0.010)  (0.008)  Firm size (log)  –0.149***  0.094***  –0.149***  0.094***  (0.007)  (0.007)  (0.007)  (0.007)  Work experience  –0.076***  0.319***  –0.078***  0.317***  (0.025)  (0.018)  (0.025)  (0.018)  Work experience 2  –0.02  –0.200***  –0.018  –0.199***  (0.027)  (0.019)  (0.027)  (0.019)  Constant  –0.371  –0.002  –0.383  –0.004  (0.292)  (0.111)  (0.290)  (0.111)  Observations  21,381  21,381  21,381  21,381  R-squared  0.316  0.409  0.317  0.409  F-statistic  234.185***     231.983***       First stage   First stage   Depvar: Current wage (log)  A  B  C  D  Low Qualification  (1)  (2)  (3)  (4)    Occ intensity    –0.063    –0.063    (0.069)    (0.069)  Herfindahl Blossfeld  0.132***    0.132***    (0.014)    (0.014)    Occ distance  –0.019***  –0.007***  –0.011***  –0.016***  (0.003)  (0.002)  (0.004)  (0.002)  Occ distance * previous wage      0.014***  –0.016***      (0.004)  (0.003)  Firm distance  0.055***  0.002  0.055***  –0.006  (0.003)  (0.004)  (0.004)  (0.004)  Firm dist * previous wage      0  –0.013***      (0.004)  (0.002)  Previous wage (log)  –0.002  0.109***  –0.032***  0.165***  (0.006)  (0.004)  (0.010)  (0.007)  Firm size (log)  –0.231***  0.081***  –0.230***  0.080***  (0.007)  (0.016)  (0.007)  (0.016)  Work experience  0.096***  0.809***  0.099***  0.801***  (0.022)  (0.017)  (0.022)  (0.017)  Work experience 2  –0.104***  –0.608***  –0.106***  –0.603***  (0.028)  (0.023)  (0.028)  (0.023)  Constant  –0.672***  –0.163***  –0.695***  –0.129***  (0.096)  (0.038)  (0.096)  (0.038)  Observations  31,516  31,516  31,516  31,516  R-squared  0.208  0.321  0.209  0.324  F-statistic  87.044***    86.772***      Medium qualification  (5)  (6)  (7)  (8)    Occ intensity    0.096***    0.079***    (0.020)    (0.020)  Herfindahl Blossfeld  0.223***    0.223***    (0.007)    (0.007)    Occ distance  –0.001  –0.029***  0.002  –0.034***  (0.002)  (0.001)  (0.002)  (0.001)  Occ distance * previous wage      0.018***  –0.030***      (0.002)  (0.002)  Firm distance  0.051***  –0.020***  0.050***  –0.021***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      –0.007***  –0.012***      (0.002)  (0.001)  Previous wage (log)  –0.028***  0.167***  –0.053***  0.247***  (0.004)  (0.003)  (0.006)  (0.005)  Firm size (log)  –0.249***  0.096***  –0.249***  0.089***  (0.004)  (0.005)  (0.004)  (0.005)  Work experience  –0.088***  0.478***  –0.084***  0.466***  (0.013)  (0.009)  (0.013)  (0.009)  Work experience 2  0.031**  –0.289***  0.029**  –0.284***  (0.013)  (0.009)  (0.013)  (0.009)  Constant  0.374***  0.059**  0.359***  0.065***  (0.067)  (0.023)  (0.068)  (0.023)  Observations  98,363  98,363  98,363  98,363  R-squared  0.194  0.386  0.194  0.399  F-statistic  966.873***    972.649***    High qualification  (9)  (10)  (11)  (12)    Occ intensity    0.127***    0.128***    (0.034)    (0.034)  Herfindahl Blossfeld  0.215***    0.213***    (0.014)    (0.014)    Occ distance  –0.030***  –0.029***  –0.031***  –0.029***  (0.004)  (0.003)  (0.004)  (0.003)  Occ distance * previous wage      0.012***  –0.005      (0.004)  (0.004)  Firm distance  0.061***  –0.021***  0.063***  –0.021***  (0.004)  (0.003)  (0.004)  (0.003)  Firm dist * previous wage      –0.021***  0.001      (0.004)  (0.003)  Previous wage (log)  –0.006  0.143***  –0.001  0.150***  (0.006)  (0.005)  (0.010)  (0.008)  Firm size (log)  –0.149***  0.094***  –0.149***  0.094***  (0.007)  (0.007)  (0.007)  (0.007)  Work experience  –0.076***  0.319***  –0.078***  0.317***  (0.025)  (0.018)  (0.025)  (0.018)  Work experience 2  –0.02  –0.200***  –0.018  –0.199***  (0.027)  (0.019)  (0.027)  (0.019)  Constant  –0.371  –0.002  –0.383  –0.004  (0.292)  (0.111)  (0.290)  (0.111)  Observations  21,381  21,381  21,381  21,381  R-squared  0.316  0.409  0.317  0.409  F-statistic  234.185***     231.983***     Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of two-stage least squares regressions. The odd-numbered columns (also, A and C) show the first-stage results. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(4) are workers with low, (5)–(8) with medium, (9)–(12) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1 The reduced-form estimates for this analysis are also in line with the expectations, showing significant, positive coefficients of the Herfindahl Blossfeld index for medium- and high-skilled employees but insignificant coefficients for low-skilled employees in the wage regressions (results are available upon request). Comparing the size of the distance variables and occupational intensity shows that the sum of the negative distance coefficients never amounts to the size of the positive coefficient for occupational intensity. In other words, although an individual might cover costly distances, these costs are outweighed by the benefits of working in an occupationally intensive firm. Going back to eq. (7), one could summarize that since in practice the distance measures and occupational intensity cancel each other out, a joint switch occurs when there is wage growth and/or an improvement in the task match. This is an interesting finding, as there may be various other reasons why individuals switch but nonetheless, on average, the (dis)advantages along knowledge dimension balance out. In sum, instrumenting occupational intensity confirms that the OLS results are biased and that for medium- and high-skilled employees wages in target firms increase with occupational intensity. Thus, in terms of wages, the 2SLS estimates for these groups confirm the demand-side hypothesis by Bidwell and Briscoe, according to which occupationally intensive firms are of advantage to workers. The decreasing share of occupational peers with experience is in line with the supply-side hypothesis. Taking into account that distance and occupational intensity decrease with increasing experience suggests that, when distances are shorter, there is less need to choose occupationally intensive firms as a compensation mechanism. However, this cannot be causally interpreted. Low-skilled workers are not affected by firm knowledge, neither by firm distance nor by occupational intensity—an even stronger result than in the previous models. This again might explain why they cover on average longer firm distances. To verify the robustness of the 2SLS estimations, the same equations were estimated using limited information maximum likelihood (LIML) and generalized method of moments (GMM). The results stay virtually the same. Including the squared term of occupational intensity provides no robust evidence for a non-linear relationship between occupational intensity and wages. All of the previous regressions focus on men. Additional analyses for women show the same patterns. In fact, the positive relationship between occupational intensity and wages can also be confirmed for low-skilled women. All results are available upon request. 4.3.2 Matching task profiles of workers and firms In a final step to better understand on-the-job learning, the focus shifts towards the newly developed measure for the match between occupational and firm tasks. It calculates the distance between the worker’s occupational tasks and the task composition of firms. The data show that for 45.6% (35.3%) of the sample, the most (least) important tasks of workers and firms are identical. As a cross-check, for both tasks the share where the important firm (occupational) task is not the least important occupational (firm) task, implying no match, is clearly above 90%. These results already suggest that employees work at firms with similar task portfolios to their own. Figure 4 depicts the development of the firm-occupation match over the career path, based on results when the firm-occupation distance is regressed on work experience. While, interestingly enough, all qualification levels start at similar levels, the distance increases for low-skilled, remains relatively stable for medium-skilled, and decreases for high-skilled workers. This is interesting, as it suggests that the lower the qualification level, the more likely individuals are to work for firms with a lower overlap of task profiles between firm and occupation when they gain more experience. Fig. 4 View largeDownload slide The relationship between work experience and firm-occupation distance Fig. 4 View largeDownload slide The relationship between work experience and firm-occupation distance Table 3 shows the OLS results when occupational intensity is replaced with firm-occupation distance. If the same story holds as for occupational intensity, there should be a positive coefficient. First, note that the remaining coefficients are highly similar to Table 1. However, firm-occupation distance is only positive for low-skilled employees, remains inconclusive or insignificant for medium-skilled employees, and is negative for high-skilled employees. This suggests that when factoring in task similarities with other occupations, the benefits and costs of firm-occupation distance differ by qualification level. This is not the case when repeating the regressions on the female sample, where the coefficient is always negative and significant. However, as discussed, the firm-occupation distance measure may be biased, for instance, due to personal motivations for covering long-distance switches and choosing a new firm. Table 3 The role of the firm-occupation match (OLS) Depvar: Current wage (log)  A  B  C  D  E  Low Qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.110***  0.110***  0.155***  0.108***  0.166***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Firm- occ dist  0.016***  0.020***  0.020***  0.020***  0.020***  (0.003)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        0.000  –0.008***        (0.002)  (0.002)  Firm dist * previous wage          –0.012***          (0.002)  Occ distance    –0.008***  –0.020***  –0.009***  –0.018***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.021***    –0.017***      (0.002)    (0.002)  Current firm size (log)  0.097***  0.097***  0.096***  0.095***  0.094***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.797***  0.795***  0.791***  0.800***  0.793***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.597***  –0.596***  –0.594***  –0.597***  –0.593***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.317***  –0.299***  –0.265***  –0.204***  –0.157**  (0.057)  (0.057)  (0.057)  (0.076)  (0.077)  Observations  32,290  32,290  32,290  31,496  31,496  R-squared  0.323  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.177***  0.169***  0.236***  0.165***  0.244***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Firm- occ dist  –0.019***  –0.003  –0.003  –0.003*  –0.002  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.015***  –0.017***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Current firm size (log)  0.077***  0.076***  0.074***  0.072***  0.070***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.475***  0.463***  0.454***  0.466***  0.455***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.289***  –0.279***  –0.275***  –0.281***  –0.277***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.441***  –0.378***  –0.342***  –0.396***  –0.338***  (0.079)  (0.079)  (0.079)  (0.038)  (0.038)  Observations  100,954  100,954  100,954  98,375  98,375  R-squared  0.404  0.411  0.416  0.414  0.42  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.154***  0.146***  0.152***  0.143***  0.150***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Firm- occ dist  –0.052***  –0.040***  –0.040***  –0.039***  –0.039***  (0.005)  (0.005)  (0.005)  (0.005)  (0.005)  Firm distance        –0.012***  –0.012***        (0.002)  (0.002)  Firm dist * previous wage          –0.001          (0.003)  Occ distance    –0.030***  –0.030***  –0.028***  –0.028***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.003    –0.003      (0.003)    (0.003)  Current firm size (log)  0.082***  0.081***  0.081***  0.078***  0.078***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.311***  0.309***  0.312***  0.311***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.204***  –0.204***  –0.208***  –0.207***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.472  –0.399  –0.402  –0.153  –0.152  (0.302)  (0.296)  (0.296)  (0.135)  (0.135)  Observations  21,668  21,668  21,668  21,378  21,378  R-squared  0.44  0.444  0.444  0.445  0.445  Depvar: Current wage (log)  A  B  C  D  E  Low Qualification  (1)  (2)  (3)  (4)  (5)    Previous wage (log)  0.110***  0.110***  0.155***  0.108***  0.166***  (0.004)  (0.004)  (0.006)  (0.004)  (0.007)  Firm- occ dist  0.016***  0.020***  0.020***  0.020***  0.020***  (0.003)  (0.004)  (0.004)  (0.004)  (0.004)  Firm distance        0.000  –0.008***        (0.002)  (0.002)  Firm dist * previous wage          –0.012***          (0.002)  Occ distance    –0.008***  –0.020***  –0.009***  –0.018***    (0.002)  (0.002)  (0.002)  (0.002)  Occ distance * previous wage      –0.021***    –0.017***      (0.002)    (0.002)  Current firm size (log)  0.097***  0.097***  0.096***  0.095***  0.094***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.797***  0.795***  0.791***  0.800***  0.793***  (0.016)  (0.016)  (0.016)  (0.016)  (0.016)  Work experience 2  –0.597***  –0.596***  –0.594***  –0.597***  –0.593***  (0.022)  (0.022)  (0.022)  (0.022)  (0.022)  Constant  –0.317***  –0.299***  –0.265***  –0.204***  –0.157**  (0.057)  (0.057)  (0.057)  (0.076)  (0.077)  Observations  32,290  32,290  32,290  31,496  31,496  R-squared  0.323  0.324  0.326  0.326  0.329    Medium qualification  (6)  (7)  (8)  (9)  (10)    Previous wage (log)  0.177***  0.169***  0.236***  0.165***  0.244***  (0.003)  (0.003)  (0.004)  (0.003)  (0.004)  Firm- occ dist  –0.019***  –0.003  –0.003  –0.003*  –0.002  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Firm distance        –0.015***  –0.017***        (0.001)  (0.001)  Firm dist * previous wage          –0.013***          (0.001)  Occ distance    –0.033***  –0.038***  –0.029***  –0.033***    (0.001)  (0.001)  (0.001)  (0.001)  Occ distance * previous wage      –0.033***    –0.029***      (0.002)    (0.002)  Current firm size (log)  0.077***  0.076***  0.074***  0.072***  0.070***  (0.002)  (0.002)  (0.002)  (0.002)  (0.002)  Work experience  0.475***  0.463***  0.454***  0.466***  0.455***  (0.008)  (0.008)  (0.008)  (0.008)  (0.008)  Work experience 2  –0.289***  –0.279***  –0.275***  –0.281***  –0.277***  (0.009)  (0.009)  (0.008)  (0.009)  (0.009)  Constant  –0.441***  –0.378***  –0.342***  –0.396***  –0.338***  (0.079)  (0.079)  (0.079)  (0.038)  (0.038)  Observations  100,954  100,954  100,954  98,375  98,375  R-squared  0.404  0.411  0.416  0.414  0.42  High qualification  (11)  (12)  (13)  (14)  (15)    Previous wage (log)  0.154***  0.146***  0.152***  0.143***  0.150***  (0.005)  (0.005)  (0.007)  (0.005)  (0.008)  Firm- occ dist  –0.052***  –0.040***  –0.040***  –0.039***  –0.039***  (0.005)  (0.005)  (0.005)  (0.005)  (0.005)  Firm distance        –0.012***  –0.012***        (0.002)  (0.002)  Firm dist * previous wage          –0.001          (0.003)  Occ distance    –0.030***  –0.030***  –0.028***  –0.028***    (0.003)  (0.003)  (0.003)  (0.003)  Occ distance * previous wage      –0.003    –0.003      (0.003)    (0.003)  Current firm size (log)  0.082***  0.081***  0.081***  0.078***  0.078***  (0.004)  (0.004)  (0.004)  (0.004)  (0.004)  Work experience  0.333***  0.311***  0.309***  0.312***  0.311***  (0.017)  (0.017)  (0.017)  (0.017)  (0.017)  Work experience 2  –0.223***  –0.204***  –0.204***  –0.208***  –0.207***  (0.019)  (0.019)  (0.019)  (0.019)  (0.019)  Constant  –0.472  –0.399  –0.402  –0.153  –0.152  (0.302)  (0.296)  (0.296)  (0.135)  (0.135)  Observations  21,668  21,668  21,668  21,378  21,378  R-squared  0.44  0.444  0.444  0.445  0.445  Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of OLS regressions. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(5) are workers with low, (6)–(10) with medium, (11)–(15) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1. The 2SLS results in Table 4 paint a different picture than the OLS regressions, pointing towards a bias of the original estimates. Medium-skilled workers consistently profit from achieving a close match between their own and the firm’s task portfolio, similar to occupational intensity, while low-skilled workers do not seem to be affected. Unfortunately, the instrument does not work for high-skilled workers. The results for female workers are again more consistent with the predictions of the model, showing that low- and medium-skilled workers benefit from small firm-occupation distances while the instrument does not work for high-skilled workers. Table 4 The role of the firm-occupation match (2SLS)   First stage   First stage      A  B  C  D  Low qualification  (1)  (2)  (3)  (4)    Herfindahl Blossfeld  –0.062***    –0.062***    (0.014)    (0.014)    Firm-occ dist    0.158    0.160    (0.151)    (0.151)  Previous wage (log)  0.014***  0.106***  0.014  0.164***  (0.005)  (0.005)  (0.009)  (0.007)  Firm distance  –0.039***  0.005  –0.040***  –0.002  (0.003)  (0.006)  (0.004)  (0.006)  Firm dist * previous wage      –0.001  –0.012***      (0.004)  (0.002)  Occ distance  0.146***  –0.029  0.146***  –0.039*  (0.003)  (0.022)  (0.004)  (0.022)  Occ distance * previous wage      0.001  –0.017***      (0.004)  (0.002)  Current firm size (log)  –0.011*  0.097***  –0.011*  0.096***  (0.007)  (0.005)  (0.007)  (0.005)  Work experience  0.113***  0.784***  0.113***  0.777***  (0.022)  (0.024)  (0.022)  (0.024)  Work experience 2  –0.074***  –0.587***  –0.073***  –0.583***  (0.028)  (0.025)  (0.028)  (0.025)  Constant  0.408***  –0.287**  0.408***  –0.254*  (0.129)  (0.133)  (0.130)  (0.132)  Observations  31,496  31,496  31,496  31,496  R-squared  0.25  0.294  0.25  0.296  F-statistic  18.756***    18.758***      Medium qualification  (5)  (6)  (7)  (8)    Herfindahl Blossfeld  –0.099***    –0.097***    (0.006)    (0.006)    Firm-occ dist    –0.220***    –0.185***    (0.047)    (0.047)  Previous wage (log)  0.004  0.166***  –0.019***  0.240***  (0.004)  (0.003)  (0.005)  (0.005)  Firm distance  –0.017***  –0.019***  –0.015***  –0.019***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      0.017***  –0.010***      (0.002)  (0.002)  Occ distance  0.144***  0.002  0.144***  –0.007  (0.002)  (0.007)  (0.002)  (0.007)  Occ distance * previous wage      –0.001  –0.029***      (0.003)  (0.002)  Current firm size (log)  0.022***  0.076***  0.022***  0.074***  (0.003)  (0.002)  (0.003)  (0.002)  Work experience  0.041***  0.475***  0.045***  0.463***  (0.012)  (0.009)  (0.012)  (0.009)  Work experience 2  –0.02  –0.286***  –0.021*  –0.281***  (0.012)  (0.009)  (0.012)  (0.009)  Constant  –0.599***  0.172***  –0.618***  0.160***  (0.060)  (0.037)  (0.060)  (0.037)  Observations  98,375  98,375  98,375  98,375  R-squared  0.203  0.332  0.204  0.361  F-statistic  302.339***    290.7***    High qualification  (9)  (10)  (11)  (12)    Herfindahl Blossfeld  0.005    0.006    (0.010)    (0.010)    Firm-occ dist    5.159    4.238    (10.308)    (6.996)  Previous wage (log)  0.019***  0.046  –0.007  0.178***  (0.006)  (0.196)  (0.009)  (0.060)  Firm distance  0.004  –0.034  0.002  –0.021  (0.004)  (0.048)  (0.004)  (0.023)  Firm dist * previous wage      0.020***  –0.087      (0.004)  (0.140)  Occ distance  0.116***  –0.629  0.117***  –0.53  (0.005)  (1.192)  (0.005)  (0.821)  Occ distance * previous wage      0.001  –0.007      (0.005)  (0.023)  Current firm size (log)  0.075***  –0.309  0.074***  –0.239  (0.006)  (0.769)  (0.006)  (0.520)  Work experience  0.023  0.191  0.031  0.175  (0.024)  (0.280)  (0.024)  (0.251)  Work experience 2  –0.025  –0.075  –0.031  –0.073  (0.027)  (0.312)  (0.027)  (0.259)  Constant  0.546**  –5.205  0.550**  –4.319  (0.229)  (10.429)  (0.228)  (7.154)  Observations  21,378  21,378  21,378  21,378  R-squared  0.215     0.216     F-statistic  0.259    0.384      First stage   First stage      A  B  C  D  Low qualification  (1)  (2)  (3)  (4)    Herfindahl Blossfeld  –0.062***    –0.062***    (0.014)    (0.014)    Firm-occ dist    0.158    0.160    (0.151)    (0.151)  Previous wage (log)  0.014***  0.106***  0.014  0.164***  (0.005)  (0.005)  (0.009)  (0.007)  Firm distance  –0.039***  0.005  –0.040***  –0.002  (0.003)  (0.006)  (0.004)  (0.006)  Firm dist * previous wage      –0.001  –0.012***      (0.004)  (0.002)  Occ distance  0.146***  –0.029  0.146***  –0.039*  (0.003)  (0.022)  (0.004)  (0.022)  Occ distance * previous wage      0.001  –0.017***      (0.004)  (0.002)  Current firm size (log)  –0.011*  0.097***  –0.011*  0.096***  (0.007)  (0.005)  (0.007)  (0.005)  Work experience  0.113***  0.784***  0.113***  0.777***  (0.022)  (0.024)  (0.022)  (0.024)  Work experience 2  –0.074***  –0.587***  –0.073***  –0.583***  (0.028)  (0.025)  (0.028)  (0.025)  Constant  0.408***  –0.287**  0.408***  –0.254*  (0.129)  (0.133)  (0.130)  (0.132)  Observations  31,496  31,496  31,496  31,496  R-squared  0.25  0.294  0.25  0.296  F-statistic  18.756***    18.758***      Medium qualification  (5)  (6)  (7)  (8)    Herfindahl Blossfeld  –0.099***    –0.097***    (0.006)    (0.006)    Firm-occ dist    –0.220***    –0.185***    (0.047)    (0.047)  Previous wage (log)  0.004  0.166***  –0.019***  0.240***  (0.004)  (0.003)  (0.005)  (0.005)  Firm distance  –0.017***  –0.019***  –0.015***  –0.019***  (0.002)  (0.001)  (0.002)  (0.001)  Firm dist * previous wage      0.017***  –0.010***      (0.002)  (0.002)  Occ distance  0.144***  0.002  0.144***  –0.007  (0.002)  (0.007)  (0.002)  (0.007)  Occ distance * previous wage      –0.001  –0.029***      (0.003)  (0.002)  Current firm size (log)  0.022***  0.076***  0.022***  0.074***  (0.003)  (0.002)  (0.003)  (0.002)  Work experience  0.041***  0.475***  0.045***  0.463***  (0.012)  (0.009)  (0.012)  (0.009)  Work experience 2  –0.02  –0.286***  –0.021*  –0.281***  (0.012)  (0.009)  (0.012)  (0.009)  Constant  –0.599***  0.172***  –0.618***  0.160***  (0.060)  (0.037)  (0.060)  (0.037)  Observations  98,375  98,375  98,375  98,375  R-squared  0.203  0.332  0.204  0.361  F-statistic  302.339***    290.7***    High qualification  (9)  (10)  (11)  (12)    Herfindahl Blossfeld  0.005    0.006    (0.010)    (0.010)    Firm-occ dist    5.159    4.238    (10.308)    (6.996)  Previous wage (log)  0.019***  0.046  –0.007  0.178***  (0.006)  (0.196)  (0.009)  (0.060)  Firm distance  0.004  –0.034  0.002  –0.021  (0.004)  (0.048)  (0.004)  (0.023)  Firm dist * previous wage      0.020***  –0.087      (0.004)  (0.140)  Occ distance  0.116***  –0.629  0.117***  –0.53  (0.005)  (1.192)  (0.005)  (0.821)  Occ distance * previous wage      0.001  –0.007      (0.005)  (0.023)  Current firm size (log)  0.075***  –0.309  0.074***  –0.239  (0.006)  (0.769)  (0.006)  (0.520)  Work experience  0.023  0.191  0.031  0.175  (0.024)  (0.280)  (0.024)  (0.251)  Work experience 2  –0.025  –0.075  –0.031  –0.073  (0.027)  (0.312)  (0.027)  (0.259)  Constant  0.546**  –5.205  0.550**  –4.319  (0.229)  (10.429)  (0.228)  (7.154)  Observations  21,378  21,378  21,378  21,378  R-squared  0.215     0.216     F-statistic  0.259    0.384    Source: Author’s calculations. Notes: The dependent variable is the logarithm of the wage in the current job after a joint switch for male employees. The calculations show coefficients for standardized variables of two-stage least squares regressions. The odd-numbered columns (also, A and C) show the first-stage results. Robust standard errors are in parentheses (see footnote 4 for details). All models include controls for occupational field, region, industry, and year. Columns (1)–(4) are workers with low, (5)–(8) with medium, (9)–(12) with high qualification levels. *** p<0.01, ** p<0.05, * p<0.1. In sum, workers match themselves to firms with specific task sets but there are important differences across qualification levels regarding the match development over time. Using occupational intensity as a measure generates more robust results, suggesting that this group may play a more prominent role for on-the-job learning than when taking into account all tasks on the firm level. 5. Conclusions Previous work in the field of labour mobility that uses task-based measures to determine job content has helped address several puzzles of labour economists, such as, for instance, skill-biased technological change. Other work with task data has addressed the question of human capital specificity, that is, knowledge that cannot be transferred in the case of job switches. This paper splits occupational and firm knowledge both in two, a specific and a general component, which is done by determining how transferable knowledge is between two firms or two occupations. In addition, it takes into account the role of working in an occupationally intensive firm, that is, a firm with a large amount of specific knowledge. The results reveal the following patterns with regard to how individuals are matched along the career path. First, the majority of switchers travel only small distances between firms. Furthermore, long-distance switches between firms become less likely with increasing work experience, indicating that workers might find better work matches as they move along their career path. Firm and occupational distances—measures for specific knowledge—show a negative relationship with wages, with the exception of low-skilled workers, where firm distance is not always insignificant. Occupational knowledge is of higher importance than firm knowledge. In early career stages, individuals work with a higher share of colleagues in the same occupational group, called occupational intensity, than is the case later on in their employment history. In 2SLS estimations it can be shown that occupational intensity positively affects wages for higher qualification levels and that lower firm-occupation distance reduces costs for medium-skilled workers, supporting the idea of higher wages in firms with a higher demand for an occupation. In addition, the sum of the negative coefficients from increasing both occupational and firm distance by one standard deviation is smaller than the positive coefficient of increasing occupational intensity by one standard deviation. This indicates that long-distance switches can still be rewarding in terms of the awaiting environment at the target firm. In sum, this paper contributes to the literature by showing that the specificity of knowledge on the occupational and firm level is determined by context, and that both firm and occupational knowledge matter for wages after switches. The paper hence confirms previous work by G&S and shows that Lazear’s skill-weights approach holds for firm knowledge. Human capital theory predicts that costs of general on-the-job training should be borne by the worker, while in reality specific training costs seem to be covered partly by workers and partly by firms. If the specificity depends on where workers move next, then this might explain why the empirical studies differ from the theoretical predictions (e.g., Barron and Berger, 1999; Parent, 1999). Further, in a task-based analysis, averaging across occupations (which is the standard procedure) and thereby disregarding firm knowledge implies a loss of information. Supplementary material The data used are confidential, but the replication files are available online at the OUP website, as is an online appendix. The replication files are separated by data sources (IAB for the Sample of Integrated Labour Market Biographies [SIAB] and the Establishment History Panel [BHP], BIBB for the BIBB/BAuA Employment Survey 2006). Funding This work was supported by the German Research Foundation (doctoral scholarship) and the German Academic Exchange Service (fellowship for research stay abroad and conference travel grant). Footnotes 1 In his article, Lazear (2009) provides a real-world example from Silicon Valley that nicely illustrates his basic idea. Note that although one could classify the mentioned skills as occupational, firms will combine them in unique ways that go beyond the occupational arrangement. 2 To facilitate the comparison between that work and the present paper, similar equations have the same numbers. 3 To test whether switches between similar firms are driven by switches between similar occupations, I calculate the distribution of firm switches, hence, excluding occupational switchers. The majority of switches (now up to approximately 90%) take place in the smallest firm distance intervals. Additionally, focusing on long-distance occupational switchers (0.4 or higher) shows that the majority of firm switches are still among similar firms (around approximately 50%), indicating that the switching pattern in terms of firm distance is not driven by switches between similar occupations. The results are available upon request. 4 Note that the standardization of interaction terms changes the null hypothesis and, thereby, complicates the interpretation of the results. Comparison between models is, thus, not possible. It can further lead to coefficients and significance levels that differ from those of an unstandardized model. Nonetheless, the goal of testing the contribution of firm and occupational human capital justifies this approach. Also note that clustered standard errors are not suitable for standardized variables because variables are standardized using the population mean. Standard errors would instead be clustered on the individual level. Thus, the models are estimated with robust standard errors instead. Control regressions with standard errors clustered on the individual level using standardized variables confirm the reported relationships (results available upon request). 5 G&S did not include occupational intensity and firm size, but leaving these variables out does not alter the results. Results are available upon request. 6 The 2SLS for columns A–C of Table 1 are also in line with the results in Table 2. Results are available upon request. Acknowledgements The author is indebted to David Autor, Ljubica Nedelkoska, and Michael Wyrwich, who provided many insightful comments. I thank the members of the Graduate College ‘Economics of Innovative Change’ in Jena and the staff at the Research Data Centre of the Federal Employment Agency for excellent suggestions and support. This research project was carried out while Elisabeth Bublitz was employed at the Friedrich Schiller University Jena and the Hamburg Institute of International Economics. The article’s content is solely the responsibility of the author and does not necessarily represent official views of the author’s affiliations. References Acemoglu D., Autor D. ( 2011) Skills, tasks and technologies: implications for employment and earnings, in Ashenfelter O., Card D., Handbook of Labor Economics, Handbooks in Economics, Volume 4, Part B , Elsevier and University of Chicago Press, Chicago, IL, 1043– 1171. Google Scholar CrossRef Search ADS   Autor D.H., Levy F., Murnane R.J. ( 2003) The skill content of recent technological change: an empirical exploration, Quarterly Journal of Economics , 118, 1279– 1333. Google Scholar CrossRef Search ADS   Backes-Gellner U., Mure J. ( 2005) The skill-weights approach on firm specific human capital: empirical results for Germany, ISU Working Paper Series, Working Paper No. 56, Institute for Strategy and Business Economics, University of Zurich. Baron J.N. ( 1984) Organizational perspectives on stratification, Annual Review of Sociology , 10, 37– 69. Google Scholar CrossRef Search ADS   Barron J.M., Berger M.C. ( 1999) Do workers pay for on-the-job training?, Journal of Human Resources , 34, 235– 52. Google Scholar CrossRef Search ADS   Barron J.M., Berger M.C., Black D.A. ( 1997) How well do we measure training?, Journal of Labor Economics , 15, 507– 28. Google Scholar CrossRef Search ADS   Becker G.S. ( 1962) Investment in human capital: a theoretical analysis, Journal of Political Economy , 70, 9– 49. Google Scholar CrossRef Search ADS   Becker G.S. ( 1964) Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education , University of Chicago Press, Chicago, IL. Bidwell M., Briscoe F. ( 2010) The dynamics of interorganizational careers, Organization Science , 21, 1034– 53. Google Scholar CrossRef Search ADS   Blossfeld H.P. ( 1985) Bildungsexpansion und Berufschancen , Campus, Frankfurt/Main, Mannheim. Borghans L., ter Weel B., Weinberg B.A. ( 2014) People skills and the labor-market outcomes of underrepresented groups, Industrial and Labor Relations Review , 67, 287– 334. Google Scholar CrossRef Search ADS   Dorner M., Heining J., Jacobebbinghaus P., Seth S. ( 2010) Sample of Integrated Labour Market Biographies (SIAB) 1975–2008, FDZ-Methodenreport 09/2010, Research Data Centre of the German Federal Employment Agency at the Institute at the Institute for Employment Research, Nürnberg. Gathmann C., Schönberg U. ( 2010) How general is human capital? A task-based approach, Journal of Labor Economics , 28, 1– 49. Google Scholar CrossRef Search ADS   Geel R., Mure J., Backes-Gellner U. ( 2010) Specificity of occupational training and occupational mobility: an empirical study based on Lazear’s skill-weights approach, Education Economics , 19, 519– 35. Google Scholar CrossRef Search ADS   Green F. ( 2012) Employee involvement, technology, and evolution in job skills: a task-based analysis, Industrial and Labor Relations Review , 65, 36– 67. Google Scholar CrossRef Search ADS   Hall A., Tiemann M. ( 2006) BIBB/BAuA Employment Survey of the Working Population on Qualification and Working Conditions in Germany 2006, suf_1.0; Research Data Center at BIBB (ed.); GESIS Cologne, Germany (data access); Federal Institute of Vocational Education and Training, Bonn, doi:10.4232/1.4820. Hannan M.T. ( 1988) Social change, organizational diversity, and individual careers, in Riley M.W., Social Change and the Life Course , Sage, Newbury Park, CA. Jaffe A.B. ( 1986) Technological opportunity and spillovers of R&D: evidence from firms’ patents, profits, and market value, American Economic Review , 76, 984– 1001. Kambourov G., Manovskii I. ( 2009) Occupational specificity of human capital, International Economic Review , 50, 63– 115. Google Scholar CrossRef Search ADS   Lazear E.P. ( 2009) Firm-specific human capital: a skill-weights approach, Journal of Political Economy , 117, 914– 40. Google Scholar CrossRef Search ADS   Mincer J. ( 1962) On-the-job training: costs, returns, and some implications, Journal of Political Economy , 70, 50– 79. Google Scholar CrossRef Search ADS   Mincer J. ( 1989) Human capital and the labor market: a review of current research, Educational Researcher , 18, 27– 34. Google Scholar CrossRef Search ADS   Neal D. ( 1995) Industry-specific human capital: evidence from displaced workers, Journal of Labor Economics , 13, 653– 77. Google Scholar CrossRef Search ADS   Neal D. ( 1999) The complexity of job mobility among young men, Journal of Labor Economics , 17, 237. Google Scholar CrossRef Search ADS   Nedelkoska L., Neffke F. ( 2011) Skill shortage and skill redundancy: asymmetry in the transferability of skills, DIME Final Conference Paper, DIME Final Conference, 6–8 April, Maastricht. Parent D. ( 1999) Wages and mobility: the impact of employer-provided training, Journal of Labor Economics , 17, 298– 317. Google Scholar CrossRef Search ADS   Parent D. ( 2000) Industry-specific capital and the wage profile: evidence from the National Longitudinal Survey of Youth and the Panel Study of Income Dynamics, Journal of Labor Economics , 18, 306– 23. Google Scholar CrossRef Search ADS   Pavan R. ( 2011) Career choice and wage growth, Journal of Labor Economics , 29, 549– 87. Google Scholar CrossRef Search ADS   Poletaev M., Robinson C. ( 2008) Human capital specificity: evidence from the dictionary of occupational titles and displaced worker surveys, 1984–2000, Journal of Labor Economics , 26, 387– 420. Google Scholar CrossRef Search ADS   Rohrbach-Schmidt D. ( 2009) The BIBB/IAB- and BIBB-BAuA Surveys of the Working Population on Qualification and Working Conditions in Germany, BIBB-FDZ Daten- und Methodenbericht, 1/2009, Bundesinstitut für Berufsbildung (BIBB), Bonn. Rohrbach-Schmidt D., Tiemann M. ( 2013) Changes in workplace tasks in Germany—evaluating skill and task measures, Journal for Labour Market Research , 46, 215– 37. Google Scholar CrossRef Search ADS   Spitz-Oener A. ( 2006) Technical change, job tasks, and rising educational demands: looking outside the wage structure, Journal of Labor Economics , 24, 235– 70. Google Scholar CrossRef Search ADS   Spitz-Oener A. ( 2008) The returns to pencil use revisited, Industrial and Labor Relations Review , 61, 502– 17. © Oxford University Press 2018 All rights reserved This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Oxford Economic PapersOxford University Press

Published: Apr 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Unlimited reading

Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.

Stay up to date

Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.

Organize your research

It’s easy to organize your research with our built-in tools.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve Freelancer

DeepDyve Pro

Price
FREE
$49/month

$360/year
Save searches from Google Scholar, PubMed
Create lists to organize your research
Export lists, citations
Access to DeepDyve database
Abstract access only
Unlimited access to over
18 million full-text articles
Print
20 pages/month
PDF Discount
20% off