The Occupational Panel for Germany

2.1 Data Basis

The Occupational Panel allows to analyse the development of important occupational characteristics and socio-structural features of occupations over time. It is mainly based on two data bases, the IAB Employment History (IAB Beschäftigtenhistorik (BeH) V10.04.00, Nürnberg 2019) and the occupational expert data base, BERUFENET. Furthermore, we use information from the IAB Establishment History Panel (BHP 2019 version 2, total population).

The BeH contains all social security notifications that employers have to submit for their employees who are subject to social security contributions as well as for marginal part-time employees. Periods in which persons worked as civil servants or self-employed are therefore not included. The data includes individual characteristics like age, gender or qualification, employment-related information such as the type of occupation and information on the establishment such as sector.

BERUFENET^[1] is an expert database of the Federal Employment Agency which provides information regarding all occupations in Germany online and free of charge. BERUFENET is used in particular for vocational guidance or job placement and currently comprises approx. 4.000 individual occupations. For instance, it includes information regarding the tasks in the respective occupations, work equipment used, work conditions, required training or legal regulations.

The Establishment History Panel (BHP)^[2] contains panel data at the establishment-level. These data offer yearly information about establishments’ workforces, earnings distributions, sectors and locations. For our purposes, it is important to note that it contains information on the age of the establishment and the size.

2.2 Data Preparation

We use information on employees subject to social security contributions for the years 2012–2018. We focus on the employment group without special characteristics that means we exclude marginal part-time employees, and further groups like apprentices, trainees and working students. Moreover, we exclude individuals aged younger than 16 years and older than 65 years because they are not yet or no longer in employment. For the purpose of the analysis, the information on employees subject to social insurance contributions from the BeH was aggregated at the occupational level for each year.

We use the occupational group (three-digit level) combined with the requirement level (fifth digit) based on the 2010 German Classification of Occupations – ‘Klassifikation der Berufe 2010 (Kldb2010)’ (Paulus and Matthes 2013). The requirement level differentiates (1) unskilled/semiskilled workers, (2) skilled workers, (3) specialists and (4) experts. Unskilled or semi-skilled occupations normally require no vocational qualification, whereas specialist occupations imply at least two years of vocational training. The prerequisites for complex specialist occupations are qualifications such as master craftsman or technician or an equivalent technician school or college graduation but also graduation from a professional academy or a university bachelor’s degree. Highly complex occupations require a completed university degree of at least four years. For the aggregation, we use full-time equivalents for weighting: full-time employees get a weight of 1 and part-time employees a weight of 0.5. We also do some data preparations such as dropping small occupations with less than 100 employees in 2012 due to anonymisation reasons. As no information is available on military services, we do not include them.

2.3 Variables

Table 1 provides an overview of all variables included in the occupational panel. We consider a rich set of variables for each occupation aggregated at the three-digit level and fifth digit (kldb10_3plus5) and for each year of 2012–2018. Furthermore, we consider full-time equivalents (see Section 2.2) and a headcount for every employee.

As individual characteristics, we first regard age. We distinguish three categories: the share of employees aged younger than 30 years, the share of employees aged 30 to 50 years and the share of employees aged 50 years and older. Next, the panel also includes the share by gender (men/women) and the share by nationality (German/foreign). Qualification is measured by three categories: share of employees with low or missing education, share of employees with medium education and share of employees with high education. As the variable education provides many missing values in the observation period, we use the imputation procedure (IP 3) by Fitzenberger et al. (2005).

Second, employment-related information comprises the days in employment and the tenure in years. Furthermore, we consider the share of employees by temporary employment (permanent/fixed-term contract), by temporary agency work and by working time (full-time). The occupational panel also includes a large range of wage variables (mean and median) for full-time employed individuals for total and by gender – imputed and not imputed. The employee history contains information on the daily wage of employees in euros. Values above the income threshold in the statutory pension insurance are censored. This affects about ten percent of full-time employees. To avoid biased results when aggregating the occupational wages, we follow the imputation procedure described in Dustmann et al. (2009). More specifically, we estimate censored regressions for each year (we use age, tenure, skill, establishment size, sector, region, type of region, foreigner, fixed-term contract, temporary agency employment as covariates), separated by gender, and impute the censored wages. Besides the imputed values, we also include the original wage information for comparison reasons. If the wage information is based on less than 20 observations, we do not show these values, but set them to missing (this is the case for the wages of female workers in three occupations and for the wages of male workers in two occupations).

Furthermore, we consider the share of employees by sector. We use the 1-digit level of WZ 08 resulting in 21 different sectors. We also aggregate the 21 sectors into three industry types: primary (sector 1), secondary (sector 2–6) and tertiary sector (sector 7–21).

Third, we use establishment-related variables. We consider the share of employees by establishment size and age by distinguishing three categories respectively: establishment size 1–49, establishment size 50–449 and establishment size >500 as well as establishment age 0–10 years, establishment age 11–20 years and establishment age >20 years.

Fourth, we also consider regional variables: the share of employees by regional type (core cities/urbanized districts, rural districts with features of concentration, rural districts-sparsely populated), by federal states as well as by a categorisation of the federal states into Northern, Western, Eastern and Southern federal states.

Finally, as occupational variables we consider the task type, the substitution potential, as well as the Digital-Tools Index. For the calculation of the task types and the substitution potential, the so-called requirement matrix for the year 2013 and 2016 from the BERUFENET is used, which assigns approximately 8000 tasks to approximately 4000 occupations. The requirement matrix assigns each single occupation the tasks to perform in this respective occupation. In line with Autor et al. (2003), Dengler et al. (2014) assign each task to five task types: analytical non-routine tasks, interactive non-routine tasks, cognitive routine tasks, manual routine tasks and manual non-routine tasks. The decision of whether a task is to be regarded as substitutable corresponds to the distinction between routine task and non-routine task in the task-based approach (Dengler et al. 2014). The term ‘routine’ means that an activity can be broken down into machine-programmable sub-elements and can be replaced by machines. To evaluate how strongly a certain occupation can potentially be substituted by computers or computer-controlled machines, we consider the share of routine tasks in the occupations, i.e., the share of tasks that can potentially be substituted by computers or computer-controlled machines. We interpret the share of routine tasks in the occupations as a measure for the substitutability of these occupations – labelled as substitution potentials (Dengler and Matthes 2018a). The assessment is only about the current technical feasibility. Legal and ethical obstacles, cost considerations or preferences cannot be considered, but may prevent automation. Dengler and Matthes (2018a) calculated the task types and substitution potentials for 2013. To account for changing task profiles within occupations over time, new digital technologies and new occupations, they update the task types and substitution potentials each three years (Dengler and Matthes 2018a, 2021). In the occupational panel, we can consider the task types and substitution potentials for 2013 and 2016.

Deeply connected with task-based indicators, but less considered are tool-based indicators, i.e. information about the work equipment that is necessary to apply tasks. In terms of the level of digitalisation, the common work tools of an occupation are a promising mean to measure the digitalisation of the specific occupation. Therefore, we also include the Digital-Tools Index (DTOX) to the occupation panel. This index provides the percentage of digital tools relative to the total number of tools within every single occupation. So far, the DTOX is only available for 2017 due to data limitations. Nevertheless, the index has been already successfully applied in studies of Antoni et al. (2019) and Genz et al. (2019).

The data source from which we identify digital tools is also the BERUFENET. The key section of the BERUFENET to identify digital tools is the section on work equipment/tools (German: ‘Arbeitsgegenstaende’). We use a unique BERUFENET data extract of the Federal Employment Agency from 2017. This extract facilitates analyses of tools for 2963 occupations. The definition of tools is very broad in BERUFENET. It covers an overall number of tools of about 14,333 tools. After narrowing down this broad set for further analysis we use 5919 tools. Given the large number of potential tools to analyse for identification of digital tools, a semi-automatic text mining approach identifies these digital tools. The procedure is based on a text mining approach developed and introduced by Janser (2019). The list of tools of the BERUFENET, provided by the Federal Employment Agency, is the source for this computational content analysis. The text mining is based on the following working definitions, retrieved by a comprehensive literature review and own conceptual works:

1. IT-aided tools are tools that are electronically based or supported, such as computers, printers, text processing software, electronic machines that are not explicitly dedicated to an industry 4.0 feature.

2. IT-integrated tools are tools that are electronically based or supported AND that are explicitly dedicated to an industry 4.0 feature, such as 3D printers, machine learning algorithms or mobile robot clusters.

Both definitions are represented in the index, as there are four DTOX variations:

1. dtox_1_2017 DTOX 1 Total (IT-aided AND -integrated Tools)

2. dtox_2_2017 DTOX 2 (only) IT-aided tools (2.0 and 3.0 tech)

3. dtox_3_2017 DTOX 3 (only) IT-integrated tools (4.0 tech)

4. dtox_4_2017 DTOX 4 Non-digital tools

2.4 Data Access

Access to the data is possible via the IAB homepage: https://iab.de/en/daten/iab-occupational-panel. The data access is free of charge.