Linked Employer-Employee Data on Firms' Training Costs: Enriching Register based LEE Data with Firm Level Data on Apprenticeship Training

Dietrich, Hans et al. Article Linked Employer-Employee Data on Firms' Training Costs: Enriching Register based LEE Data with Firm Level Data on Apprenticeship Training Schmollers Jahrbuch – Journal of Applied Social Science Studies. Zeitschrift für Wirtschaftsund Sozialwissenschaften Provided in Cooperation with: Duncker & Humblot, Berlin Suggested Citation: Dietrich, Hans et al. (2014) : Linked Employer-Employee Data on Firms' Training Costs: Enriching Register based LEE Data with Firm Level Data on Apprenticeship Training, Schmollers Jahrbuch – Journal of Applied Social Science Studies. Zeitschrift für Wirtschaftsund Sozialwissenschaften, ISSN 1865-5742, Duncker & Humblot, Berlin, Vol. 134, Iss. 2, pp. 237-248, https://doi.org/10.3790/schm.134.2.237 This Version is available at: https://hdl.handle.net/10419/292440 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ European Data Watch This section offers descriptions as well as discussions of data sources that are of interest to social scientists engaged in empirical research or teaching courses that include empirical investigations performed by students. The purpose is to describe the information in the data source, to give examples of questions tackled with the data and to tell how to access the data for research and teaching. We focus on data from German speaking countries that allow international comparative research. While most of the data are at the micro level (individuals, households, or firms), more aggregate data and meta data (for regions, industries, or nations) are included as well. Suggestions for data sources to be described in future columns (or comments on past columns) should be send to: Joachim Wagner, Leuphana University of Lueneburg, Institute of Economics, Campus 4.210, 21332 Lueneburg, Germany, or e-mailed to h[email protected]i. Past “European Data Watch”articles can be downloaded free of charge from the homepage of the German Council for Social and Economic Data (RatSWD) at: http://www.ratswd.de. Linked Employer-Employee Data on Firms’Training Costs: Enriching Register based LEE Data with Firm Level Data on Apprenticeship Training By Hans Dietrich, Holger Alda, Harald Pfeifer, Felix Wenzelmann, Gudrun Schönfeld, Stefan Schiel and Stefan Seth 1. Introduction Despite the general trend towards tertiary education in Germany, more than 60% of each age cohort still enrols in apprenticeship training programs (Bundesministerium für Bildung und Forschung, 2009). Apprenticeship training, thus, remains the dominant educational pathway at the upper secondary level. An important feature of apprenticeship training is the strong link between education and the labour market because a large part of the training takes place Schmollers Jahrbuch 134 (2014), 237 –248 Duncker & Humblot, Berlin Schmollers Jahrbuch 134 (2014) 2 Die freie Verfügbarkeit der Online-Ausgabe dieser Publikation wurde ermöglicht durch das Bundesinstitut für Berufsbildung (BIBB), Stabsstelle »Publikationen und wissenschaftliche Informationsdienste«. OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 within firms. A considerable share of former apprentices are retained by the training firms, and this facilitates the transition of trainees from education to the labour market (for statistics on retention rates see Hartung, 2013). With respect to such labour market transition from an individual perspective, several data sets exist for research purposes. However, data sets representing the firm perspective are limited in their number, and they face severe limitations. The BIBB studies on costs and benefits of apprenticeship training (in the following BIBB CBS) are examples of surveys aimed at filling this gap (see Noll et al., 1983, von Bardeleben et al., 1995, Beicht et al., 2004; Schönfeld et al., 2010). The latest survey of the BIBB CBS was conducted in 2008, reporting training costs and benefits for the reference year 2007. A total of 2,986 firms were interviewed that provided training in 51 occupations. Because the sample was drawn from the Establishment Register (Betriebsdatei) of the Federal Employment Agency (Bundesagentur für Arbeit, BA), the survey data can be merged with register data using an establishment-specific identifier, which, in most cases, corresponds with a firm as a local unit (more in detail see Alda et al. 2013). Technically, the data merge is feasible for all firms observed in the BIBB CBS 2007. However, a precondition for the merging procedure is that firms have formally agreed to the data merge during the survey interview. Two thirds of all of the firms included in CBS 2007 agreed with the data match. The remainder of the paper is structured in the following way. Section 2 first describes the two data sources used for the merge and then addresses potential selectivity problems arising from a firm’s option to decline the data merge. Section 3 discusses the quality of the match between survey and administrative data. Section 4 briefly summarizes the main results of the paper and sketches research perspectives of the matched data. 2. Data Sources of the Linked Employer-Employee Data 2.1 The Cost-Benefit Survey 2007 Based on a comprehensive report of the Sachverständigenkommission Kosten und Finanzierung der Beruflichen Bildung from 1974 (Sachverständigenkommission, 1974) the BIBB Cost and Benefit Surveys were conducted to obtain representative data for the costs and the benefits of apprenticeship training in (West-)Germany. Later on, the surveys were designed to deliver cost and benefit information in a structure which enables differentiation of costs and benefits along structural variables, such as economic sector, occupation, firm size or region. Finally, all cost benefit surveys focused on specific training occupations among all of the approximately 320 regulated training occupations in Germany. For the BIBB CBS survey of 2007, 51 of these occupations were Schmollers Jahrbuch 134 (2014) 2 238 Hans Dietrich et al. OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 chosen based on their representativeness and practical relevance. The 51 occupations represent approximately 70% of all German apprenticeship participants. The survey of the year 2007 includes detailed information about the firm’s costs of training. These include monetary measures of expenditures for apprenticeship wages, costs for (fulland part-time) trainers and costs for training infrastructure (e.g., training facilities and material). For the calculation of the net costs, the studies also measure the benefits of training in the form of productive work performed by apprentices. Finally, the survey inquires about the firm’s cost of recruiting already trained workers from the external labour market. Detailed results of the most recent BIBB CBS 2007 are published in Schönfeld et al. (2010). The BIBB CBS results have been extensively used for theoretical and empirical works in the field of education economics (see, e.g., Acemoglu/ Pischke, 1998 or Mühlemann et al., 2010). For the BIBB CBS 2007, a sample of firms was drawn from the BA-Establishment register. The register contains all firms with at least one employee subject to social security payments. Because all firms fitting this criterion are obliged to register, the representativeness of this pool of addresses can be taken as given. A gross sample of 22,000 firms was drawn, which have provided training in at least one of the 51 occupations of interest in June 2006, the reference period for sample selection. Because the Establishment Register contains occupational information only on a 3-digit level, the survey institute had implemented a screening procedure to ensure that the respective firm was actually providing training in the chosen 4-digit occupation of interest. Approximately 60% of all 22,000 firms provided training in the chosen training occupation. The remaining 40% either did not train in the specific 4-digit occupation or, due to the time lag between sampling procedure and reference period of the interview, did not train apprentices at all. After the screening, approximately 13,000 addresses of the training firms of interest were identified. Field managers used 8,907 of these to conduct the intended number of 3,000 valid interviews. The survey was conducted between April and September 2008, and the average interview duration was 72 min.1 The interview partner was the person responsible for the organisation of training, and/or the firm’s human resource manager. In small firms, the interview was conducted with the owner or managing director of the firm. The field work was administered by infas (Institute for Applied Social Sciences), Bonn. The survey institute infas established several techniques to raise data quality. For example, infas defined confidence intervals for several key variables (e.g., wages and training times) in the CAPI questionnaire. In case the interviewed person stated values outside the confidence interval, the CAPI program looped Schmollers Jahrbuch 134 (2014) 2 Linked Employer-Employee Data on Firms’Training Costs 239 1For more details see infas (2008) and Schönfeld et al. (2010). OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 back to the value and the interviewer had to ensure that the interview partner understood the question correctly. In case of incomlete answers for these crucial variables, an ex-post imputation technique was used to replace missing values. For most of the variables, missing values made up less than 5% of all answers, with somewhat higher shares for wage information on different groups of employees (10% to 15%). For detailed information about the imputation procedures see Alda (2010). The merged data set, which is described in this paper, contains records from those firms that agreed to the merging of the firm specific survey data with the BA register data. Of the 2,986 firms, 2,083 (i.e., almost 70%) provided permission for the merge. Textbox A1 quotes the respective question from BIBB CBS 2007 and Table A1 provides an overview of the topics covered by the survey2. Table A2 displays descriptive information for the main variables for the subsample of merged firms and for the full sample of firms. In order to test whether firms agreeing to the merge differ systematically from those refus the merge, the decision variable (1 for permission, 0 for refusal) was regressed on a set of structural variables, such as firm size, economic sector, region and training occupation (see Table 1). The results show that neither firm size indicators nor the variables of economic activity or region significantly alter or reduce the probability of belonging to the group of firms that did not permit the data matching. Furthermore, none of the dummy variables for the 51 observed occupations are significant at the 5% level. The observed low Pseudo R 2, of barely 0.02 shows that very little of the variance in the firm’s merge permission is explained by the set of structural variables. The probit estimates thus lead to the conclusion that the two groups of firms do not differ systematically with respect to (observed) structural characteristics. Table 1 Probit regression –dependent variable: firms’permission to merge data Model 1 Model 2 Model 3 Model 4 coef se coef se coef se coef se Firm-size: Ref. 50–499 employees 1–9 employees 0.11* 0.063 0.110* 0.064 0.110* 0.064 0.060 0.079 10–49 employees 0.13** 0.062 0.137** 0.062 0.137** 0.062 0.105 0.068 500 and more employees –0.11 0.096 –0.129 0.097 –0.130 0.097 –0.125 0.100 Schmollers Jahrbuch 134 (2014) 2 240 Hans Dietrich et al. 2More detailed information about the set of firm level variables available is given in Alda (2010). OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 Economic sector: Ref. Public administration, education and health Manufacturing, agriculture, mining and quarrying, construction –0.082 0.079 –0.082 0.079 0.007 0.120 Trade –0.102 0.093 –0.102 0.093 –0.026 0.137 Services I –0.037 0.087 –0.037 0.087 0.048 0.115 Services II 0.029 0.101 0.029 0.101 0.022 0.139 Region: Ref. East Germany West Germany –0.005 0.054 0.015 0.055 Occupation: No No No Yes Pseudo R2 0.002 0.003 0.003 0.021 Constant 0.45*** 0.045 0.501***0.077 0.508***0.102 0.315 0.279 Source: BIBB CBS 2007. Level of significance: *** p<0,01, ** p<0,05, * p<0,1. N=2986. 2.2 Establishment History Panel (BHP) and the Integrated Employment Biographies (IEB) The second part of the data is derived from the BA register data. Here, two types of process produced data are available from the 2007 survey records of firms that permitted data matching: a) firm data from the Establishment History Panel (BHP) and b) employee data from the Integrated Employment Biographies (IEB). The Establishment History Panel (BHP) is composed of cross sectional data sets since 1975 for West Germany and 1991 for East Germany. Every cross section contains all of the establishments in Germany that are covered by the Employment History (BeH) on June 30th. These are all establishments with at least one employee liable to social security on the reference date. Since 1999 establishments with no employee liable to social security but with at least one marginal part-time employee are also included. The cross sections can be combined to form a panel. The BHP contains information about the branch of industry and the location of the establishment. Furthermore, the number of employees liable to social security is also included. In addition, marginal part-time employees are also included (since 1999), both in total numbers and broken down by gender, age, occupational status, qualification and nationality. Quartiles of ages and wages are also provided, both for full-time employees alone as well as for all employees (see Spengler, 2008 and http://fdz.iab.de/ en/FDZ_Establishment_Data/Establishment_History_Panel.aspx). The Integrated Employment Biographies (IEB) consists of all individuals in Germany, who are characterized by at least one of the following employment statuses: employment subject to social security (in the data since 1975); marginal part-time employment (in the data since 1999); benefit receipt according Schmollers Jahrbuch 134 (2014) 2 Linked Employer-Employee Data on Firms’Training Costs 241 OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 to the German Social Code III or II (SGB III since 1975, SGB II since 2005); officially registered as job-seeking at the German Federal Employment Agency; (planned) participation in programs of active labour market policies (in the data since 2000). The information delivered from different data sources are consolidated in the IEB. Each employment status is represented on a daily basis (see Dorner et al., 2010). The register data enables the generation of variables that identify the employment background of the apprentice before the start of the apprenticeship, such as prior training qualifications, unemployment durations and schooling level (graduation from high school, Abitur). Register data also provide detailed information regarding the training period itself, such as the apprentice’s training pay, the duration of training and whether the individual received social benefits during the training phase. Finally, the register data contain information about the apprentice’s labour market integration after training. 3. Merging Quality Although the merging of survey data and register data is technically trivial due to the unique firm identifier, the quality of the merge needs to be assessed on the basis of statistical analysis. Several mechanisms may lead to poor quality of the match even if technically the same firm is identified in both data sources: First, different reference periods and/or a reference basis (e.g., mixing up marginal employment with part-time employment) could lead to diverging information about the firm size. Second, interview partners may have difficulties in remembering the correct numbers or values for the staff working in the firm. Third, the firm’s administrative identification number may have changed over time. Although identification numbers are usually adjusted accordingly, a change in the identification number could occur in cases of inor outsourcing or a change of ownership. Fourth, interview partners may only have a rough idea about which parts of the firm are covered by the administrative identification number, which can lead to underor overestimation of the values delivered by the register data. To assess the quality of the firm-level merge, we compare the firm’s number of employees and apprentices reported in each of the two data sources. The number of employees covers all fulland part-time employees. Excluded from both data sources are employees with a marginal employment contract below the tax and social security thresholds (geringfügig Beschäftigte)3. The number Schmollers Jahrbuch 134 (2014) 2 242 Hans Dietrich et al. 3Furthermore, those employees who are not subject to social security payments are not included in the employee variables. These are, for example, freelancers, public servants and short-term trainees. OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 of apprentices includes those apprentices for whom the interviewed firm contributes to social security. Apprentices who are (temporarily) trained by the reporting firm but are not subject to social security payments are not included.4 Figure 1 presents scatter plots of the calculated differences. The plots indicate that the large majority of firms show no or small differences in their reported numbers of employees (left panel) and apprentices (right panel). Source: BIBB CBS and BHP (30. September 2007). Figure 1: Scatter plot of differences in the number of employees and apprentices To be able to differentiate between different merge quality levels, we introduce categories for the quality of fit. We differentiate between a) a perfect fit, b) a good fit (using firm size adjusted confidence bands) and c) a poor fit of the numbers of employees and apprentices reported both in survey and register data. A “perfect fit”is identified when the respective number of employees and apprentices does not diverge at all. A “good fit”is identified in the case of employees if the difference between the number of employees reported by the BIBB CBS and the number of employees reported in the BA data is smaller than half of the respective firm-size category width. In the case of apprentices, a“good fit”refers to a deviation between the two data sources of between 1 and 4 apprentices. A “poor fit”refers to firms outside the thresholds defined for the “good fit”. Finally, we report d) the firms with missing data on employment for which the merge quality cannot be assessed (for a more detailed presentation of the quality assessment see Alda et al., 2014). Table 2 reports the results of the quality assessment. Close to 90% of all merged firms fit perfectly or fit well with respect to their reported numbers of employees and apprentices. For approximately 10% of the merged firms, we find a poor merge quality, while 1% (in the case of employees) and 0.2% (in Schmollers Jahrbuch 134 (2014) 2 Linked Employer-Employee Data on Firms’Training Costs 243 4It can be the case that the interviewed firm takes over (parts of) the training for other firms or other organizational units that are not covered by the firm identifier. OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11 the case of apprentices) are not classified due to problems with missing data (Table 2). Table 2 Quality of fit between surveyand register data based on the number of employees and apprentices Employees Apprentices Share of firms (%) Number of firms Share of firms (%) Number of firms a) Perfect fit 10.4 207 55.9 1117 b) Good fit 78.2 1561 33.2 664 c) Poor fit 10.0 200 10.7 213 d) Missing data 1.5 29 0.2 3 Total 100 1997 100 1997 Source: BIBB CBS and BHP (30. September 2007). Although we find a good fit for both the number of employees and apprentices, a remarkably higher proportion of perfect fits (= identical figures) can be seen in the apprenticeship figures. One reason for this marked difference in the share of perfect fits can be due to the fact that some groups, such as marginal employees, temporary work agency employees, family helpers or proprietors could be reported in the survey but are not included in the number of employees as measured in the administrative data. Furthermore, the number of employees is much more volatile due to labour turnover, when compared to the number of apprentices in the firm. Finally, the person interviewed in the firm was requested to be informed about the costs and the benefits of apprenticeships. Due to this focus of the survey, we can expect more accurate reporting of apprenticeship numbers than of employment numbers. However, for both groups, we find that approximately 10% of the firms diverge strongly with respect to the numbers reported in the survey and in the register based data. The variable identifying the quality of the data fit will be included in the data file accessible at the Research Data Center of the BA in the IAB (FDZ). 4. Summary and Research Perspectives The aim of the paper was to describe the merging of BIBB survey data and IAB process-generated (administrative) data. We discuss potential selectivity problems arising due to the option of firms to decline the data merge. Using a probit model including standard structural control variables, we find no indicaSchmollers Jahrbuch 134 (2014) 2 244 Hans Dietrich et al. OPEN ACCESS | Licensed under CC BY-NC-ND 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.134.2.237 | Generated on 2023-01-16 13:37:11