pretrained_models {occupationMeasurement}

Pretrained ML models to be used with the package.

Description

Pretrained ML models to be used with the package.

Usage

pretrained_models

Format

A nested list with pretrained machine learning models:

similarity_based_reasoning

list. Contains pretrained models to be used with algo_similarity_based_reasoning().

similarity_based_reasoning$wordwise

list. Contains the pretrained model to be used for providing suggestions using full wordwise matching.

similarity_based_reasoning$substring

list. Contains the pretrained model to be used for providing suggestions using substring matching.

This training data always predicts a 5-digit code from the 2010 German Classification of Occupations, with some exceptions: -0004 stands for 'Not precise enough/uncodable', -0006 stands for 'Multiple Jobs', -0012 stands for 'Blue-collar workers', -0019 stands for 'Volunteer/Social Service', and -0030 stands for 'Student assistant'.

Source

Data from the following surveys were pooled:

Antoni, M., Drasch, K., Kleinert, C., Matthes, B., Ruland, M. and Trahms, A. (2010): Arbeiten und Lernen im Wandel * Teil 1: Überblick über die Studie, FDZ-Methodenreport 05/2010, Forschungsdatenzentrum der Bundesagentur für Arbeit im Institut für Arbeitsmarkt- und Berufsforschung, Nuremberg.

Rohrbach-Schmidt, D., Hall, A. (2013): BIBB/BAuA Employment Survey 2012, BIBB-FDZ Data and Methodological Reports Nr. 1/2013. Version 4.1, Federal Institute for Vocational Education and Training (Research Data Centre), Bonn.

Lange, C., Finger, J., Allen, J., Born, S., Hoebel, J., Kuhnert, R., Müters, S., Thelen, J., Schmich, P., Varga, M., von der Lippe, E., Wetzstein, M., Ziese, T. (2017): Implementation of the European Health Interview Survey (EHIS) into the German Health Update (GEDA), Archives of Public Health, 75, 1–14.

Hoffmann, R., Lange, M., Butschalowsky, H., Houben, R., Schmich, P., Allen, J., Kuhnert, R., Schaffrath Rosario, A., Gößwald, A. (2018): KiGGS Wave 2 Cross-Sectional Study—Participant Acquisition, Response Rates and Representativeness, Journal of Health Monitoring, 3, 78–91. (only wave 2)

Trappmann, M., Beste, J., Bethmann, A., Müller, G. (2013): The PASS Panel Survey after Six Waves, Journal for Labour Market Research, 46, 275–281. (only wave 10)

Job titles were taken from the following publication:

Bundesagentur für Arbeit (2019). Gesamtberufsliste der Bundesagentur für Arbeit. Stand: 03.01.2019. https://download-portal.arbeitsagentur.de/files/.

Basically, leaving some data anonymization steps aside, we count for each job title from the Gesamtberufsliste (and some additional titles/texts) how many responses from all surveys are similar to this job title, separately for each coded category. Similarity is calculated in two ways, implying that we obtain two different counts: SubstringSimilarity refers to situations where the job title from the Gesamtberufsliste is a substring of the verbal answer; WordwiseSimilarity refers to situations where a word from the verbal answer is identical to a job title from the Gesamtberufsliste, except that one character from this word is allowed to change (Levenshtein distance). These counts are available as two separate files in the data-raw/training-data/ directory of this package. The algorithm to create these counts is available inside an R-package at https://github.com/malsch/occupationCoding, along with further documentation.

train_similarity_based_reasoning() is then used to train the ML models. See data-raw/pretrained_models.R for the raw counts and further details.

See Also

algo_similarity_based_reasoning(), train_similarity_based_reasoning(), https://github.com/malsch/occupationCoding