New methods for job and occupation classification

This dissertation addresses the measurement of occupation in surveys. Many surveys ask respondents about their occupation with open-ended questions. The verbal answers are typically coded after the interview into official classifications (e.g., the 2008 International Standard Classification of Occupations or the 2010 German Classification of Occupations). This process is known to be time-consuming and prone to errors. To counter both issues, the first paper of the dissertation develops and tests a software prototype, which searches for candidate job titles at the time of the interview. A small set of relevant jobs are suggested based on the respondents’ initial verbal input, allowing respondents to select the most appropriate job on their own. A second paper compares various statistical learning algorithms to optimize the suggestions. A novel algorithm was developed employing Bayesian principles, improving the suggestions further. In a third paper, 1226 work activity descriptions were created based on close inspection of the official occupational classifications. These work activity descriptions can be used as answer options in an improved version of the prototype.