Task data

In order to assess the structure of offshoring costs and the related demand shifts in the offshoring process, it has to be clear how tasks are assumed to be bundled into occupations. In the theory outlined in the previous section, there is a number of intermediate inputs, each produced by one occupation through the performance of a specific bundle of tasks. Empirically, it is a challenge to find an adequate approximation in available data. In this paper, the unit of analysis for task-related offshoring costs is the level of a 2-digit occupation, of which 74 in the German ”KldB 88” classification will be used.⁷ Note again that this is not meant to strictly define one occupation as one task but rather as a bundle of them. Moreover, the tasks to be performed within an occupation are usually fixed by a specific work contract which

7These 74 occupations exclude agricultural and military occupations.

is assumed to be binding at least in the short-run.⁸ Thus, from a firm perspective such an occupation seems like a natural unit when restructuring employment. This renders the offshoring costs of an occupation a crucial determinant in the offshoring decision of firms - just as described in the model above. In this paper, these costs are seen as being determined by the distribution of differently costly tasks within individual occupations. That is, the more routine or non-interactive tasks are performed within an occupation, the less costly it is to offshore. These lower costs can, for instance, be related to routine tasks being easier to supervise from a distance or the fact that they are less likely to produce complicated problems which need direct and costly intervention. Highly interactive tasks, beyond those requiring physical presence, are also more costly to do at a distance. While internet based communication has constantly reduced the costs of international communication, there are still many situations in which communication across time zones about complex problems remains costly.

Given the above considerations, the empirical equivalent to the range of jobs from the model in the previous section is a vector of 74 occupations ordered according to their offshoring costs - the latter being based on their share ofR-tasks in total tasks.

The task shares within occupations are represented by an average of task intensities at the individual level. For Germany, the best data on this topic comes from the

”BIBB/IAB-Employment Survey 1998/99”.⁹ This database has previously been used by Spitz-Oener (2006) and Becker et al. (2013), among many others, and has proven to be the source of choice for task related information. The database holds

survey-8In the long run, the task composition of occupations is likely changing. Examples are found in Spitz-Oener (2006) where evidence for long run trends over several decades is presented. An update of the analysis of long run task changes can be found in Antonczyk et al. (2009). These long-run changes relate to the work content as such, however. They are not directly related to offshoring costs.

9The data used is the BIBB/IAB study: ”Acquisition and Application of Occupational Qualifica-tions 1998/99”, provided to the author by GESIS Cologne, Germany. No. ZA3379. Datafile version

results describing the tasks individual workers perform. It is thus a direct account of observed work contents. This sets the data used here apart from classifications based on external expert assessments of an occupation’s typical work content such as the O*NET data base developed for the United States. For the analysis in this paper, 13 different activities are grouped into eitherN-tasks (non-routine and interactive) or R-tasks (routine and non-interactive).¹⁰ Individual-level task intensities are calculated and subsequently aggregated to the 2-digit occupational level as simple mean values.

The individual level task intensities are derived as:

λⁱ(k) = number N-tasks performed by individual i in occupation k

number of all tasks performed by individuali in k (4.9) These individual specific task intensities are averaged within each occupation as λ(k) =P

λⁱ(k)/L(k).¹¹

This calculation of tasks intensities is a variant of a method proposed by An-tonczyk et al. (2009). The measure yields an approximation of how an individual splits her job into different tasks. Thus, the measures of task intensities within all 74 occupations considered add up to one.¹² Naturally, (1−λ(k)) describes the share of R-tasks in total tasks. It is similar to, yet distinct from approaches using cate-gory specific intensities such as Spitz-Oener (2006) or Becker et al. (2013), where the number of type-τ tasks (with τ ∈routine, non−routine) an individual performs is related to all possible type-τ tasks.

10The grouping is a further aggregation of the five categories in Spitz-Oener (2006), with non-routine analytical, non-non-routine interactive, and non-non-routine manual being summarized in the N -group with higher costs of offshoring. Routine manual and routine cognitive tasks form the lower costR-group. The individual tasks grouped into these categories are very similar to the ones used in Spitz-Oener (2006). The individual tasks are also listed in the appendix.

11The task intensity λ(k) is directly and positively related to the theoretical expression φ_k. Yet, since this empirical measure holds no capital, it is denoted differently.

12Further details on the data used and the calculation of the task intensities can be found in the appendix.