Occupation Coding of Job Titles: Iterative Development of an Automated Coding Algorithm for the Canadian National...

Area of Science:

• Occupational Health
• Public Health
• Data Science

Background:

• Identifying social determinants, including occupation, is crucial in research.
• Manual coding of occupation data using systems like the Canadian National Occupational Classification (NOC) is time-consuming and prone to errors.
• Automating this coding process is essential for efficient data analysis.

Purpose of the Study:

• To develop and evaluate a rigorous algorithm, ACA-NOC, for automated coding of the Canadian NOC (2016) system.
• To utilize job titles and industry information as input for the algorithm.
• To benchmark and iteratively improve algorithm performance using manually coded datasets.

Main Methods:

• Developed the ACA-NOC algorithm based on the NOC (2016) structure.
• Implemented various search strategies (exact, near, like, any, weak match) to identify matching NOC codes.
• Applied a filtering step based on the NOC's hierarchical structure to refine code selection.

Main Results:

• The ACA-NOC algorithm achieved 58.7% accuracy at the four-digit NOC code level for over 500 manually coded entries.
• Accuracy improved with broader categories: 65.0% (three-digit), 72.3% (two-digit), and 81.6% (one-digit).
• The algorithm demonstrated effectiveness on real-world data.

Conclusions:

• The ACA-NOC algorithm provides a rigorous and automated method for coding the Canadian NOC system.
• It enables researchers to code occupation data efficiently and cost-effectively for further analysis.
• The algorithm shows state-of-the-art performance and is extensible for larger datasets.