«IZA DP No. 3067 How General Is Human Capital? A Task-Based Approach Christina Gathmann Uta Schönberg September 2007 Forschungsinstitut zur Zukunft ...»
2.1 The Concept of Task-Speci…c Human Capital This section de…nes how occupations are related to each other and introduces our concept of taskspeci…c human capital. We assume that output in an occupation is produced by combining multiple tasks, for example negotiating, teaching or managing personnel. These tasks are general in the sense that they are productive in di¤erent occupations. Occupations di¤er in which tasks they require and
that occupation to the extent that occupations place di¤erent values on combinations of skills.
More speci…cally, consider the case of two tasks, denoted by j = A; M. We think of them as manual and analytical tasks. Workers are endowed with a productivity in each task, which we denote j by Tit ; j = A; M: Occupations combine the two tasks in di¤erent ways. For example, one occupation might rely heavily on analytical tasks, a second more on manual tasks, and a third combines the two
the tasks to sum to one, we focus on the relative importance of each task, not on the task intensity of an occupation. We impose this restriction for illustrative purposes, as it simpli…es the notation. None of our empirical results below however require this restriction.
In this framework, we can de…ne the relation between occupations in a straightforward way. Two occupations o and o0 are similar if they employ analytical and manual tasks in similar proportions,
We now describe each component in turn.
Human Capital Accumulation With time in the labor market, individuals become more productive in each task through learning-by-doing. In particular, we assume that Xiot contains three types of human capital: general human capital (Expit ), purely occupation-speci…c human capital (OTit ), and
the three types of human capital. Note that the returns to the three types of human capital vary by occupation–i.e. occupations di¤er in the value placed on general, task- and occupation-speci…c human capital respectively.
General human capital is valuable in all occupations, while occupation-speci…c human capital is fully lost once a worker leaves the occupation. Task-speci…c human capital in contrast is transferable to occupations with similar skill requirements but less so to those that use very di¤erent tasks.5 More speci…cally, we assume that the transferability of skills between the source and destination occupation
acquired skills. Consequently, task-speci…c human capital is neither fully general nor purely speci…c, but partially transferable across occupations.6 These assumptions allow us to collapse the accumulation of skills in multiple tasks into a onedimensional observable measure of task-speci…c human capital, T Tit : In particular, these assumptions
workers’ current and o0 the sequence of past occupations. Dio0 s in turn is an indicator equal to one if individual i worked in occupation o0 in period s and zero otherwise. Hence, task human capital is calculated from occupation tenure in all previous occupations inversely weighted by the distance between the current and previous occupations.
occupational choice, such as Neal (1999) and Pavan (2005), which assume, in line with the assumption Our de…nition of task-speci…c human capital di¤ers from that by Gibbons and Waldman (2006). In their setup, task human capital is speci…c to the job within a …rm and might therefore not be transferable across jobs within the same …rm.
A more general model of occupational choice and human capital accumulation would allow workers to invest separately in task-speci…c skills A and M. For instance, learning a task could depend on the usage of a task in an occupation. If a worker chooses an occupation that mainly specializes in task A; he would mainly accumulate skills in task A: This ties the skill investment decision to the choice of an occupation. See Murphy (1986) or Rosen (1983) for models along these lines.
However, this more general model would not lead to an empirical speci…cation we can estimate with our data.
that speci…c skills fully depreciate upon an occupation switch, that the occupational match is uncorrelated across occupations. In our speci…cation, in contrast, the correlation between the match quality in two occupations depends on the distance between the occupations–which is in line with our concept of task-speci…c skills.
2.2 Wage Determination and Occupational Mobility
Since the concept of task-speci…c human capital is novel, we next clarify the interpretation of the return to task-speci…c human capital, 2o : Consider a worker who has worked for his occupation o for one year. Suppose he is exogenously displaced from his occupation and then randomly assigned to a new
Of course, workers are not randomly allocated into occupations. We assume that workers search over occupations to maximize earnings. In one extreme, as for instance in Miller (1984), Neal (1999) and Pavan (2005), the search process is purely undirected, i.e. the probability of receiving an o¤er from
in each task. In the opposite extreme, workers know the location of their best match already at labor market entry, and …nd employment in this occupation. In this case (and in the absence of productivity shocks), workers would never switch occupations. The reality is probably somewhere in between. In order to ensure that there is some occupational mobility in our set-up, we rule out the second extreme case. However, we do not require assumptions about how exactly the search process looks like: search may be either purely undirected or (partially) directed. For Germany, Fitzenberger and Kunze (2006) and Fitzenberger and Spitz-Öner (2005) argue that search mobility is the most important source of occupational switches.
across occupations, workers are willing to switch occupations only if the gain in match quality compensates for the loss in occupation- and task-speci…c human capital. If, in contrast, the returns to human capital accumulation in the prospective occupation exceed those in the current occupation, workers may voluntarily switch occupations even if they lose speci…c human capital and are worse matched in the new occupation. This is because the new occupation promises higher wage growth in the future than the old one.
Our framework produces a number of novel empirical implications. It implies that, everything else equal, workers are more likely to move to occupations in which they can perform similar tasks as in their previous occupation. The reason is that task-speci…c human capital is more valuable in similar than in
from the current occupation) occur early, rather than late, in the labor market career. This is so for two reasons. First, the accumulation of task-speci…c human capital makes distant occupational switches increasingly costly. Second, with time in the labor market, workers gradually locate better and better occupational matches. It therefore becomes less and less likely that they accept o¤ers from very distant occupations–even if they do not accumulate task-speci…c human capital in the labor market. Since the transferability of task-speci…c human capital as well as the correlation of the match quality across occupations depends on the occupational distance, we also expect that wages at the source occupation are a better predictor for wages at the target occupation if the two occupations require similar tasks. A further advantage of our framework is that allows us to quantify the importance of task-speci…c human capital for individual wage growth, relative to general and occupation-speci…c human capital.
2.3 Comparison with Alternative Approaches Our setup is closely related to the Roy model of occupational sorting (Roy, 1951; Heckman and Sedlacek, 1985). Just like in the Roy model, individuals in our framework sort themselves into occupations
restriction allows us to de…ne how similar occupations are in their skill requirements in a straightforward way. In addition, our framework also incorporates search over occupations into the Roy model.
The framework outlined here is also related to search and matching models of the labor market (Jovanovic, 1979a; 1979b). As in search or matching models, our setup includes a match component that (partially) determines mobility decisions. Whereas search and matching models assume that speci…c skills are fully lost upon an occupation switch and that match qualities uncorrelated across occupations, our set-up also allows for a partial transferability of speci…c skills and a correlation of match qualities across occupations, depending on the occupational distance. This does not only provide new insights into the direction of occupational mobility, but also allows us to analyze the importance of task-speci…c human capital for individual wage growth relative to other forms of human capital.
In a recent paper, Lazear (2003) also sets up a model in which …rms use general skills in di¤erent combinations with …rm-speci…c weights attached to them. In this model, workers are exogenously assigned to a …rm (in our application: occupation) and then choose how much to invest in each skill. Our
- in our opinion more intuitive - approach assumes instead that workers are endowed with a productivity in each task, and then choose the occupation. Furthermore, unlike Lazear (2003), our empirical analyses focus on the transferability of skills across occupations and its implications for occupational mobility and individual wage growth.
3 Data Sources and Descriptive Evidence To study the transferability of skills empirically, we combine two di¤erent data sources from Germany.
Further details on the de…nition of variables and sample construction can be found in Appendix A.
3.1 Data on Tasks Performed in Occupations Our …rst data set contains detailed information on tasks performed in occupations, which we use to characterize how similar occupations are in their skill requirements. The data come from the repeated cross-section German Quali…cation and Career Survey, which is conducted jointly by the Federal Institute for Vocational Education and Training (BIBB) and the Institute for Employment (IAB) to track skill requirements of occupations. The survey, previously used for example by DiNardo and Pischke (1997) and Borghans et al. (2006), is available for four di¤erent years: 1979, 1985, 1991/92 and 1998/99.
Each wave contains information from 30,000 employees between the ages of 16 and 65. In what follows, we restrict our analysis to men since men and women di¤er signi…cantly in their work attachments and occupational choices.
In the survey, individuals are asked whether they perform any of nineteen di¤erent tasks in their job. Tasks vary from repairing and cleaning to buying and selling, teaching, and planning. For each respondent, we know whether he performs a certain task in his job and whether this is his main activity.
Table 1 lists the fraction of workers performing each of the nineteen di¤erent tasks. Following Autor et al. (2003) and Spitz-Öner (2006), we combine the 19 tasks into three aggregate groups: analytical tasks, manual tasks and interactive tasks. On average, 55 percent report performing analytic tasks, 72
percent manual tasks, and 49 percent interactive tasks. The picture for the main task used is similar:
32 percent report analytical tasks, 57 percent manual tasks and 28 percent interactive tasks as their main activity on the job.
The last two columns in Table 1 show the distribution of tasks performed on the job for two popular occupations: teacher and baker. According to our task data, a teacher primarily performs interactive tasks (95.3 percent) with teaching and training others being by far the most important one (91.4 percent). Two other important tasks are correcting texts or data (39.6 percent) and organize, coordinate, manage personnel (39.4 percent). A baker in contrast is a primarily manual occupation (96.4 percent) with manufacturing, producing, installing as the most important task (87.9 percent) followed by teaching and training others (34.3 percent) as well as organizing, coordinating and managing personnel (29.9 percent).
To see how task usage varies across the 64 occupations contained in our data, Table A1 lists the fraction of workers performing manual, analytical, and interactive tasks for all 64 occupations. The table shows that there is a lot of variation in task usage across occupations. For example, while the average use of analytical tasks is 56.3 percent, the mean varies from 16.7 percent as an unskilled construction worker to 92.4 percent for an accountant. We checked whether tasks performed in the same occupation vary across industries and found little support for this conjecture. This results suggest that industries matter little for measuring human capital once we control for the skill set of an occupation, and justi…es our focus on occupations.
3.2 Measuring the Distance between Occupations According to our framework, two occupations have similar skill requirements if they put similar weights on tasks, i.e. individuals perform the same set of tasks. With two tasks, the maximum distance
cannot observe these weights directly, our task data provide us with a closely related measure of the skill content of each occupation.
In particular, the task data described in the previous section tell us the set of skills employed in each occupation. We can then characterize the skill content of each occupation by a 19-dimensional vector qo = (qo1 ; ::::; qoJ ) where qoj denotes the fraction of workers in an occupation performing task j.
We can think of this vector as describing a position in the task space. In equilibrium, an occupation
qoj. To measure the distance between occupations in the task space, we use the angular separation or
uncentered correlation of the vectors qo and qo0 :