In Employment in 2030, we are creating an employment forecast for Canadian occupations. It is driven by both the insights of the labour market experts that we gathered in workshops across the country and the skills, abilities, and knowledge requisite for each profession. Our aim is to help governments, educators, and Canadians in general, attain a clearer picture of the skills that will help to future-proof everyone in the face of complex forces, ranging from automation and offshoring to resource scarcity.
With 500 national occupations (NOCs), it would have been an unwieldy task for experts to examine how each one would change by the year 2030. So—how did we select the occupations that would drive this study? We considered several factors, including the skills data available, the workshop design and constraints, and the information needs of the machine learning model that we will apply to expand our forecast.
After scanning the available skill classifications, we selected the skills, abilities, and knowledge features from O*NET (Occupational Information Network) to represent the worker attributes that are important in each occupation. Since this database focuses on US occupation groups, the Brookfield Institute created a crosswalk that matches US codes to their Canadian counterparts. This matching approach uses both the International Standard Classification of Occupations (ISCO) and the job titles contained in each occupation to link the NOC and US codes.
We conducted six workshops across the country with the goal of gathering quality data from expert participants to inform our machine-learning model. Having each workshop in a different city, from St. John’s to Whitehorse, had two major implications: the workshop design and delivery needed to be as consistent as possible, and we had to consider potential regional differences.
Selecting occupations to drive a forecast
We designed a data-driven and reproducible method to ensure that we included occupations important to both the model and the labour market. At each workshop, we asked experts to rate 20 occupations chosen through a two-stage approach. Fifteen were benchmark occupations presented to all participants across the country, and five were regional occupations unique to each workshop.
We chose these occupations because they best represent the combination of skills, abilities, and knowledge present in the Canadian job market. Our benchmarks include financial managers and electrical mechanics, among others (scroll down to see our full list). With our selection, we aimed to inform our model, reduce its uncertainty, and allow for comparison between regional workshops.
Our starting data consisted of 485 Canadian occupations that we matched with 120 skill, ability and knowledge scores from O*NET. Our first step was to reduce these 120 variables to 21 using a single value decomposition (SVD) technique. These combine the original variables in a way that accounts for over 90 percent of the variation in O*NET scores among all the occupations. Why did we want a smaller group of variables? With this reduced data, it was easier to cluster our occupations into 15 groups based on the similarities of their skill, ability and knowledge requirements using k-means clustering. After careful review, we saw that this process created meaningful groups of occupations that share either responsibility levels or broad sector characteristics. For example, one cluster includes most managerial codes while another is comprised of occupations in education and social service provision.
From each group, we selected a benchmark occupation that was most representative of the skills required for the occupations included in that group. In more technical terms, we selected the occupation that was closest to the centre of each cluster. In the case of a tie between occupations, we selected the one with the highest employment. However, there were also cases where additional criteria were necessary to ensure that we gathered high-quality responses from our workshop participants. We chose the next closest occupation as the benchmark occupation when:
- The occupation selected though this process included jobs “not elsewhere classified” (e.g. Other professional engineers not elsewhere classified). We believe that the lack of specificity in both the occupational description and the associated skill profile would not lead to a good participant assessment.
- The historical employment data for the occupation selected through the process was affected by changes in the national classification. If the employment estimates were inflated because of changes in the 2011 NOC structure, the graphs provided to participants may have deterred and misguided participants. We also considered this issue with regional occupations.