Linked Employer–Employee Data from XING and the Mannheim Enterprise Panel

2.1 Employers and Employments

In the first step, employer and user employment data are exported from the data warehouse of XING. This involves about 1.9 M employers, whereby not every observation has to be a valid or active employer. These observations are not directly excluded, as we are also interested in historical data. XING users have deposited about 46 M employment data points. About one-third of the employments are linked to the XING employer database which is an internal database by XING that lists employers. The employment data are partly maintained by the users and include the employer name, employer URL, industry, employment type, career level, and field of activity (discipline). Some of the fields are optional, e.g. the employer URL. Thus, we also export employment data points with unclear quality. XING’s employer data show specific features: First, the employer database is not standardised or linked to a uniform database like the MUP. Second, outdated, duplicate or invalid/fake employers are listed. For example, insolvent or non-active employers. Third, operating sites, subsidiaries, and employer groups can have their own XING profiles. Fourth, some employers are not from Germany or the DACH region (Germany, Austria, Switzerland). To provide examples of the difference between valid German and available employers: The employer-level country data on XING is available for around 1.1 M employers, and, of these, 450 K have entered a country other than Germany. About 650 K of the 1.9 M employers have an invalid employer name, e.g. names consist exclusively of dots.

The data preparation consists of two consecutive steps. First, the employers (1.9 M) and employments (46 M) need to be linked to the Mannheim Enterprise Panel (MUP). This is done with the SearchEngine tool (Doherr 2023) developed by Thorsten Doherr. The linking relies on the text fields ‘employer name’ or ‘employer URL’, if available. Employer profiles and employment data are separately linked with the tool, as their quality is different. We assume that the data from the employer profiles have a higher quality than the employment data. As a result, we receive candidates from the Mannheim Enterprise Panel for the XING employer profiles and employers mentioned in employments on XING. Each candidate has a unique identifier called ‘crefo’ that represents the employer identifier of the MUP. There are 86 million candidates for the employment data points based on the employer name and 12 million based on the URL. Furthermore, about 187 thousand employer profiles are linked to MUP employers. Second, the MUP candidates for the employments are enriched with additional employer-level data like exit dates, if available. Some outliers are removed, e.g. implausible data like employments before 1900 and after 2020 (after the data export), leading to a subset of 44 million employments. Then, we apply the ‘group-crefo’, which combines affiliated employers within the MUP, e.g. subsidiaries. Lastly, binary indicators are created, which, for example, indicate if a ‘crefo’ exists in the Markus database.^[4]

Next, a five-step heuristic is used to select the best candidate from the MUP for each employment data point (see Table 1). As a result, about 21 million employments are matched to the MUP. Thus, the matching rate is about 47 %. However, we deliberately extracted fewer matches than possible to ensure higher quality, as there is a trade-off between the observation count and data quality of the matches. For legal reasons, all employments from users without a public profile must be removed. We export a list of all publicly available user profiles (privacy setting) to ensure that we use only publicly accessible employments as this is a legal requirement by New Work SE. Public employments can be viewed and saved by every visitor of XING. Each employment data point has a user reference, which is used to remove non-public data. Around 31 million of 46 million employment data points are publicly available. This reduces the number of disambiguated employments to 14 million. We do not delete the unmatched employments of users with public profiles but assign an artificial employer identifier based on the employer name.

The employer data consists of three types: Employers only identified in XING, employers only identified in the MUP and employers identified in both databases. In the following step, we are only interested in employers listed in public employments linked to the MUP or non-matched employers (XING). For the subsequent statistical analyses, we only use the matched employers. About 1.5 million unique employers have been successfully linked to the MUP. The number refers to the matched employers listed in XING employments because we do not count employers with a profile but are not mentioned in at least one employment. Furthermore, employers with an artificial identifier are not counted as well. In the MUP, some variables have missing values. The missing rates for the variables of the matched employers are: Founding date (14 %), district id (4 %), legal form ( < 1 %), two-digit NACE code (12 %), exit data ( < 1 %). The employment counts are rather sparse. For example, 23 % of matched employers have not even one employment count from 2000 to 2021. However, many of these employers were not yet or are no longer active. A possible solution is imputing missing numbers or carrying forward the last existing employment count. About 9 million employers are identified in the XING database without finding a link to the MUP, e.g. organisations outside the business enterprise sector or foreign employers. This number is overestimated because unmatched employers like ‘ZEW’ and ‘ZEW Leibniz Centre’ do not receive the same identifier. These observations are not used in the following analyses.

2.2 Flows

Employee flows between employers are extracted from the employment data on XING. For this, we select the employment data of users with a public profile and define a flow as the switch between two successive employments. Temporal breaks between employments, such as unemployment, are ignored. For example, if an individual was unemployed for three years between two employments, then a flow between the employment before and after unemployment exists in the data set. We create an employer identifier (‘missing’) for the initial entry and final exit, e.g. for employment starters or retired individuals. Furthermore, only user profiles that list at least two employments are used. The described process leads to about 21 M flows for the matched and unmatched employments. Flows extracted from matched employments comprise about 9 M observations, where 7 M observations are flows between employers (see Table 2). Each flow entails the year of the employment change and the old and new employment characteristics.

3.1 Employees and Their Employments

Figure 1 shows selected characteristics of the platform users. The analyses intend to improve our understanding of the users active on the platform XING. We restrict the analysis to subsets of the data: employments have to be publicly available, matched to the MUP, and the respective characteristic, e.g. the career stage, needs to be available. The most recent employment per user is utilised for the discipline, career stage, and employment type analyses. For the analysis of the user experience, the oldest employment per user is considered. Our findings suggest: First, most employees have between zero and 19 years of professional experience. Few users have more than 39 years of professional experience as German employees must work for 35 or 45 years to be considered long-term or very long-term insured to receive their full pensions. Primarily, highly educated employees are present on XING, who often start their careers later and, thus, work for fewer years overall. Second, the disciplines Sales, Administration, and IT & Data have the most employees. Surprisingly, few employees are working in Production, R&D, and Finance. There may be a certain bias because the respective employees rarely use the platform (e.g. for Production). Third, most employees have professional experience but no managerial position. Managing directors and directors are found least often. Fourth, the majority of the employees work full-time. Civil servants and unpaid workers are by far the least frequent. Overall, we can conclude that the XING data are not representative of German employees.

Figure 1:

XING users with respect to employment characteristics. Note: (a): per experience class; (b): per discipline; (c): per career stage; (d): per employment type. For the analyses, only matched employments from publicly accessible user profiles are used. Source: TUM and ZEW based on XING data. Own illustration.

Figure 2 shows the number of employments by employer size, region, industry, and founding period. First, small employers ( < 10 employees) are most often mentioned in employments. We expected this pattern as a large proportion of employees in Germany work for small and medium-sized enterprises (SMEs).^[5] The smallest number of employments is found for employers with 250–999 employees. Second, most employments are linked to employers in North Rhine-Westphalia, Bavaria, and Baden-Württemberg. The fewest employments are found in Saarland, Mecklenburg-Western Pomerania, and Bremen. The results are plausible because they have the highest/lowest population counts. Third, most employments are linked to limited liability companies (GmbH) and liberal professions. Furthermore, employments within the Freelance, scientific and technical services (S&T Services) industry and for employers founded before 1990 are the most frequent. Fewer employments exist in the hospitality and transport industries and for employers founded since 2010.

Figure 2:

Employment counts (XING) with respect to median employer size (top left), region (top right), employer industry (bottom left), and employer founding year (bottom right). Only matched and public employments are used. Source: TUM and ZEW based on data from XING and MUP. Own illustration.

3.2 Employers

Figure 3 shows selected MUP characteristics for the matched XING employers. The employers are extracted from the XING employments that were matched to the MUP. The XING employer data set cannot be compared directly with the total stock of German employers in the MUP. E.g. the XING data contain employers that are no longer economically active. Our findings suggest: First, most employers on XING have less than 10 employees, and only a few employers have at least 250 employees. Second, most employers are found in the industry Freelance, Scientific and Technical Services (69–75); the fewest employers are in the industry transport (49–53). The numbers in parentheses are the two-digit NACE codes.^[6] Third, most employers are founded in ‘2000–2009’ or ‘2010 & later’; there is a small decrease in the number of founded employers ‘before 2000’. Fourth, most employers are located in the German regions of North Rhine-Westphalia, Bavaria, and Baden-Württemberg (in the former territory of West Germany). The fewest are located in Mecklenburg-Western Pomerania, Bremen, and Saarland. Many regions of the former German Democratic Republic (GDR) are in the lower half of the ranking, illustrating an east-west divide. The cities of Berlin and Hamburg are in the midfield of the ranking. In summary, the XING employer data seem plausible and, with some limitations, representative of Germany.

Figure 3:

Number of XING employers with respect to MUP characteristics; based on matched employers. Note: (a): per median employer size group; (b): per industry; (c): per founding year group; (d): per German region. Source: TUM and ZEW based on XING and MUP data. Own illustration.

3.3 Employee Flows

Figure 4 shows the employee flows of XING users by their employment characteristics, i.e. career level, discipline, employment type, employer size, industry, and region. First, we analyse the flows between employers with respect to career levels. Most users transfer out of employments with professional experience. Users usually switch employments within an experience class or into a higher experience class, e.g. from young professional to professional. Employees rarely reduce their experience level. For example, if an employee is moving into a new industry. Second, we analyse the flows between employers for employment disciplines. Most users switch employments within their discipline. However, we find flows for all discipline combinations. Third, we analyse the flows between employers by their employment type. Here, we find a deviating pattern: Many of the employment types show a flow into the ‘Full-time’ class.

Figure 4:

Employee flows as chord diagrams. Top left: Career levels. Top right: employment discipline, middle left: employment type, middle right: employer size, bottom left: employer industry, bottom right: Employer region. Flows are based on matched employments from public user profiles. Source: TUM and ZEW based on XING and MUP data. Own illustration.

Further validity checks show that civil servants move only occasionally into unpaid employments, and young professionals rarely move into (managing) director positions. Fourth, many flows occur within the same employer size class, and relatively few flows are found for employment changes to large-sized employers. Fifth, many employees are moving within industries. However, employers in the Freelance, Scientific and Technical Services (69–75) industry receive substantial inflows from all other industries. Sixth, the majority of employees change employers within their region. However, the regions of Bavaria, North Rhine-Westphalia, and Baden-Württemberg have substantial inflows from all other regions.