LMIC Senior Economist Anthony Mantione spoke at the INFONEX Big Data & Analytics for the Public Sector virtual course on improving the timeliness, local nature and granularity of labour market information with web-scraped data.
Improving the Timeliness, Localness and Granularity of Labour Market Information With Web Scraped Data
1. 1
Think about it
Think about the last decision you made
regarding your career. Once you have a
decision in mind, identify some of the
types of information you looked for and
the challenges you faced in so doing.
3. 3
Agenda
1. Brief introduction of the Labour
Market Information Council (LMIC)
2.Most requested types of LMI
3.Identifying and measuring skills
shortages in Canada
a.Traditional approaches and their
limitations
b.New approach leveraging machine
learning
c.Advantages and limitations
4.Current work and resources for more
information
4. Labour Market Information Council
To improve the timeliness, reliability and
accessibility of labour market information to
facilitate decision-making by employers, workers,
job seekers, academics, policy makers, educators,
career practitioners, students, parents and under-
represented populations.
5. 5
Most requested
types of LMI
• Wages
• Skill requirements
• Benefits
• Workplace
environment
• Cost of living
• Current and future
job openings
6. Media outlets express concern over shortages in Canadian Economy
• Historically low national unemployment
rates and shortages major concern prior
to pandemic
• Businesses report difficulty finding
workers1
• Concern over shortage situation paints
depressing picture:
In Canada, labour shortage is “the new normal”: study (CTV News, 2018).
Skilled labour shortages at record high for Canadian small businesses
(Bloomberg, 2018).
Staff shortages create emergency situation for Canadian retailers (Retail
Insider, 2019).
7. Job vacancies at highest level in 2nd quarter of 2019
Statistics Canada. Table 14-10-0325-01 Job vacancies, payroll employees, job vacancy rate, and
average offered hourly wage by provinces and territories, quarterly, unadjusted for seasonality.
DOI: https://doi.org/10.25318/1410032501-eng
8. Are shortages REALLY hindering Canada’s economic prosperity?
Canada has a historically large labour shortage, but job-seekers are still struggling (Huffington Post, 2018)
Is Canada’s skills shortage real, or are businesses to blame? (The Conversation, 2017).
How the myth of a Canadian skill shortage was shattered (The Star, 2014)
McDaniel’s SSHRC study reveals no evidence of labour shortage (University of Lethbridge, 2014).
9. 9
What’s going on?
• 1. Lack of
standardized/univ
ersal approach to
defining skills and
skills-related
issues
• 2. Lack of quality
data to inform
insights
10. 10
Quality attributes
of LMI
• Granularity -- the
number and detail of
categories by which
data can be grouped
• Localness – the level
of geography
• Timeliness – the time
lag from when data
is collected to when
data is available
• Frequency – how
often data are
available (e.g.,
monthly, annually)
11. 11
• 1. Information
not relevant
• 2. Information
outdated
• 3. Information
doesn’t exist
Challenges
using LMI
12. The identification issue: measuring shortages
• In empirical work, shortages have always been
interpreted, or even defined directly, in terms of
difficulties filling vacancies.10
Limitations to using vacancies as a measure for
shortages
• Employer reactions obscure measurement:
• Increased OT
• Later retirement
• Internal recruiting
• Different data reporting agencies yield
different results
13. (Limitations)2: no local, granular data
Primary data source:
• Job Vacancy and Wage Survey
• Introduced in February 2015
• Surveys 100,000 locations every quarter
• Provides job vacancy levels by industry, NOC, and economic
region
Limitations:
• Timeliness: JVWS estimates are published
approximately three months after reference period
• No information below economic region
• Publicly available data very sparse
15. 15 Where do we go from here?
Web scraping
• The process through which information is
gathered from public websites for retrieval
and analysis.
• One application of web scraping is
collecting data from online job boards and
corporate websites and then "cleaning" the
text.
• This process can either be done in-house or
outsourced. Currently, several data analytics
firms (e.g., Vicinity Jobs, Burning
Glass Technologies and TalentNeuron)
collect and analyze job posting data from
multiple Canadian corporate websites and
aggregators (e.g. indeed.ca).
16. Advantages to web scraped data
•Data available in real time
•Data available by detailed location (e.g., city
or town)
•Data provides direct insight on the skills
employers are looking for
17. The process
Collect Clean Structure Extract
Get text data (e.g., job boards, aggregators, corporate sites)
Quality assurance
Pre-processing
Deduplication (50% of data)
Mapping to NOC
Text to vectors (word embedding)
Supervised learning model
Generate insights
19. What insights can this data provide?
Item Type Inciden
ce
1. Communication skills skill 53%
2. Teamwork skill 47%
3. English language Work
requirement
38%
4. Forecasting Work
requirement
34%
5. Data Analysis Work
requirement
22%
6. Decision making Skill 19%
7. EViews Work
requirement
9%
8. Writing Skill 6%
9. MATLAB Work
requirement
3%
20. Advantages (revisited) and Limitations
• Insight into work requirements (including skills)
• No restrictions on level of detail possible
• Large volume of data and localized information
• Inexpensive relative to survey methods
• Data relevant to broad range of stakeholders
• Representativeness and other bias
• Hidden job market
• Job postings not the same as vacancies
• Subject to ML and NLP methods, which vary across data providers
• Unable to determine which work requirements are critical
21. 21
Current work
• Ongoing work with Vicinity
Jobs to improve their
mapping algorithm
• Working with ESDC to
develop a Canadian O*NET
(linking all the elements of
their Skills and Competencies
Taxonomy to occupations)
• Working with STC to develop
new skills survey for
employers
• Updating existing work to
account for structural
changes in the NOC2021
22. 22
Resources for
more information
Insight reports:
• How representative are online job
postings?
• Job skills mapping: Building
concordance between the US
O*NET System and Canada’s NOC
• Through the looking glass:
Assessing skills measures using 21st
century technologies
• What skills do I need? Making the
US O*NET system work for
Canadians
• Bridging the gap between skills and
occupations: A concept note to
identify the skills associated with
NOC
Projects:
• Public opinion research
• Canadian online job
posting dashboard
• WorkWords: Job
Vacancy