January 2025
Connecting Skills: Using online job postings to unravel the demand for skills in the labour market
This report presents comprehensive insights into the relationships between skills in Canada’s labour market. The content is tailored for researchers, policy analysts, and academics focused on labour and training.
Illustration by Dorothy Leung for LMIC.
Key Findings
In 2023, there were 4,552 unique skills extracted from almost 3.1 million job postings. Notably, the most unique skills appear in a limited number of postings, with only 710 skills appearing in 500 or more.
Social-emotional (or soft) skills account for nine of the 10 most frequently requested skills. Customer service (primarily associated with sales and service roles) is the only occupational skill among the top 10.
Teamwork, communications skills, and customer service are the most frequently requested skill types, appearing in 48%, 39% and 33% of postings, respectively.
Relationships between skills vary across occupations, indicating employers’ different expectations for skills depending on occupation.
Certain skills predict the presence of others in postings. For example, strong writing skills correlate with demand for overall communication skills; being goal-oriented is often requested alongside teamwork skills; and operating a cash register is linked to demands for customer service skills.
Executive summary
Understanding the relationships between skills gives us insights into the needs of the labour market and the complexities of understanding skill demands. This report demonstrates how online job postings serve as a valuable source of labour market information that can be used to better understand the relationships between skills and to highlight trends in skill demands.
Using three replicable methodologies—frequency measurement, association through normalized pointwise mutual information, and predictors based on directional relationships—this report explores skill connections based on mentions of skills in job postings. Understanding the connection(s) between two skills gives us insights into the nuances of skill demand and how the demand for a specific skill can change across circumstances.
To demonstrate our approach, the report includes three case studies focusing on the most in-demand skills:
- Teamwork. As this skill is broadly requested across occupations, offering a point of consistency for exploring how relationships between skills shift with different approaches.
- Communication skills. With varied demand across occupations, these provide an example of how context affects the relationships between skills.
- Customer service skills. Primarily associated with sales and service roles, we explore how specialized skills function within their primary occupational group and in other fields.
These case studies illustrate how advanced methodologies can uncover actionable insights into skill demand and relationships to inform workforce planning, career development and policy-making. By sharing how to use data from online job postings to analyze the relationships between skills, the Labour Market Information Council (LMIC) is setting the stage for new research to address the evolving needs of Canada’s labour market.
Table of contents
Introduction
Understanding the relationships between skills is important for developing responsive workforce strategies and navigating shifts in labour market demand. Online job postings (OJPs) offer a dynamic source of real-time labour market information (LMI) because they capture detailed data on the skills employers seek. Despite this potential, traditional approaches to analyzing OJPs often overlook the nuanced connections between skills.
LMIC undertook this research to address this gap, building on prior work to identify methodologies that better capture the complexities of skill relationships. By focusing on teamwork, communication skills, and customer service, this report illustrates how researchers can use frequency measurement, association (normalized pointwise mutual information), and predictors to uncover actionable insights.
Online job postings in the modern world
An important tool for an efficient labour market
For many Canadians, modern job hunting increasingly relies on OJPs, which provide access to thousands of job postings across sectors. OJPs have enhanced job visibility and access to opportunities, but their sheer volume can make the job search challenging. This often causes job seekers to be overwhelmed by the amount of information, especially when postings lack unique or specific details about skills and requirements.
While OJPs theoretically enable job seekers to identify roles that align with their skills, many postings rely on common terms, particularly regarding skills and requirements. The lack of specificity can make it harder for job seekers to identify positions that are a good fit, leading to wasted efforts and frustration (JDXpert, 2023). For employers, these ambiguities can mean losing out on well-suited candidates: unclear language is reported to be the most confusing part of the application process for young adults. This issue is challenging for young job seekers, but also affects older and neurodivergent workers, who may be discouraged by the vague language often used in postings (Dewar, 2022).
Box 1: What is the difference between a skill and a requirement?
“Skill” and “requirement” are distinct but overlapping terms. Skills encompass the abilities and knowledge needed to perform tasks well, while requirements are included in job postings to express employers’ needs. Requirements may refer to a skill or the components of a skill, such as specific attributes or knowledge. In this report, both terms are used because we worked with data extracted from job postings’ listed requirements to learn more about the overarching skills sought by employers.
Job postings should provide an overview of responsibilities and outline the qualifications the organization is seeking. Even if not explicitly stated, this information provides context about the skills needed to succeed in the role. For example, a requirement may be training or education in a specific field, indicating the type and level of knowledge for the skills required in the position.
For a deeper understanding, see our video, What is a Skill?. It provides clarity on these terms in the context of today’s labour market.
OJPs offer employers easy access to a large pool of job seekers. This means they have a high chance of finding applicants who are a good fit, but must also sift through more applications. This extends the hiring process, which has been reportedly getting longer (it reached an average of 44 days in 2023).
Choosing terminology for job postings is complicated because employers must consider how to communicate to the potential employees and how to optimize postings to reach possible applicants. Search engine optimization incentivizes the use of keywords or common terminology, but over-reliance on keywords can lead to vague job postings, which have been found to put off many strong candidates.
For this report, we work with OJP data to show the potential for learning more about skill demand using novel methods. We include case studies to show how the methods can be applied and interpreted. Through this work, LMIC seeks to support the labour market ecosystem by advancing innovative and replicable methodologies and offering guidance for others looking to take the next steps in this research.
An untapped source of data on labour demand
While the primary purpose of OJPs is to provide information on job openings and attract applicants, they also offer invaluable LMI, providing data that reflects trends in labour demand at a given time. The timeliness of OJPs presents a unique avenue for researchers to monitor labour market trends as they occur; OJPs can offer detailed insights that are not always available through traditional data sources. However, OJPs provide an unstructured data set, and the information collected needs to be validated carefully to ensure that the results of any analysis reflect the labour market accurately.
Many researchers have used OJPs to develop dynamic skill taxonomies (Djumalieva & Sleeman, 2018; Bennet et al., 2022), model job transitions (Dawson et al., 2021), and map skill imbalances (Organisation for Economic Cooperation and Development, 2022). Our approach builds on this previous work by focusing on individual skills and their relationships to other skills, contributing new and reproducible methods that other researchers can use to understand these relationships and how they change in different occupational contexts. This report demonstrates the untapped potential of OJP data as a valuable resource for labour market analysis.
In the following sections, we share insights into which OJP data was useful in our research and how we used it to identify the most in-demand skills. Through three case studies, we also show the importance of context in interpreting skill demand, demonstrating how our methods can be applied to identify nuanced labour market trends.
How LMIC works with online job postings data from Vicinity Jobs
There are multiple providers of OJP data globally and in Canada. Vicinity Jobs (a major provider of OJP data in North America) has supplied LMIC with weekly OJP data going back to January 2018, providing a large-scale, comprehensive dataset that enables the analysis of millions of job postings each year.
From January 2018 to December 2023, Vicinity Jobs catalogued more than 16 million job postings, extracting more than 5,200 unique skills (Table 1). This extensive dataset serves as the foundation for this report’s analysis of skill trends and demands within the labour market.
Table 1: Number of OJPs & unique skills, by year
| Year | Number of job postings | Number of unique skills |
| 2018 | 2,635,139 | 4,542 |
| 2019 | 2,757,830 | 4,309 |
| 2020 | 2,038,614 | 3,994 |
| 2021 | 2,808,399 | 4,097 |
| 2022 | 3,660,890 | 4,576 |
| 2023 | 3,078,987 | 4,552 |
| Total | 16,979,859 | 5,226 |
Source: LMIC calculations using 2023 Vicinity Jobs data
For each job posting, Vicinity Jobs collects and disseminates detailed data, including occupation, location, industry, wages, skill requirements, training and education. To categorize skill requirements effectively, Vicinity Jobs maps information from job postings to their skill taxonomy. Each skill is also allocated to one of four skill groups (Box 2).
Box 2: Vicinity Jobs Skill Taxonomy
A skill taxonomy creates a hierarchy for aggregating skills into higher-level classifications, reducing noise for researchers. Vicinity Jobs utilizes its own internally developed taxonomy with four overarching skill groups (Vicinity Jobs, n.d):
| Skill group | Definition | Examples |
| Social-emotional skills | Skills and attributes that enable individuals to understand and manage their emotions, interact effectively with others, build and maintain relationships, and make responsible decisions
(Often associated with the term “soft skills”) |
Multi-tasking, planning, working with a team, being flexible or speaking languages |
| Occupational skills | Skills related to functions in a specific job or field | Analyzing, doing electrical repairs, designing graphics, providing childcare or serving customers |
| Technology skills | Skills related to technological tools, systems and platforms used to facilitate various processes and tasks within different domains | Knowledge of Microsoft software, quality control tools, customer relations software or Amazon machine learning |
| Tools and equipment | Skills related to physical instruments, machinery and devices used to perform specific tasks, operations or functions within various industries and trades | Ability to work with computers, navigation equipment, forklifts or security devices |
Data limitations
Because OJPs are not designed to be used as data sources, they lack the standardized rigour that would typically be found in survey data. However, Vicinity Jobs mitigates the effects of this by cleaning the data before publication. Despite these efforts, some unavoidable structural limitations remain:
Differences between OJPs and recorded vacancies1
OJPs do not capture all vacancies. Some positions are filled through informal methods like networking or word of mouth rather than being posted online, and a single job posting may represent multiple vacancies, further skewing the gap. These factors introduce variability across job types. For instance, professional and service sector positions are likely to be overrepresented, while trades and manual labour positions are likely to be under-represented, as identified in a 2020 LMIC study.
Omitted requirements
Some occupations have assumed requirements that are not explicitly stated in all postings. For example, a chef job posting may not include “knowledge of cooking techniques” as a requirement because it is an assumed skill for the occupation. These omissions can change the level of co-occurrence between certain terms, reducing their importance in our results.
Data cleaning limitations
Once scraped, OJP data is cleaned and prepared for further use by Vicinity Jobs. Part of this cleaning process involves aligning online job postings with National Occupation Classification (NOC) codes using an algorithm. However, approximately 15% of postings cannot be mapped to a NOC code due to:
-
- no clear equivalent under the NOC system.
- broad or ambiguous job titles and descriptions that prevent precise classification under the NOC standard
Counting skills: Demonstrating how to identify demand for skills based on frequency in online job postings
In 2023, Vicinity Jobs logged approximately 3.1 million job postings with 4,552 unique skills. The demand level for these skills varied considerably across job postings. Many skills are specialized and typically confined to specific occupations or functions. To focus on commonly requested skills for our research, we applied a threshold to identify skills that appeared in at least 500 job postings. The threshold eliminated the most unique skills, leaving only 710 skills—approximately 16% of the total set of unique skills. This suggests that most job posting requirements draw from a smaller subset of unique skills.
While many skills appear in only a small fraction of job postings, some appear much more frequently. The 10 most in-demand skills from the 2023 data were present in 17% to 48% of job postings (Figure 1). These top 10 skills were so widely used that 87% of the 2023 job postings included at least one of them. The popularity of these terms dilutes their impact on job postings, but they are still important clues about the position and the employer (Hickok, 2021).
Figure 1: Percentage of job postings requesting the top 10 skills (2023)
Source: LMIC calculations using 2023 Vicinity Jobs data
Social-emotional skills account for nine of the top 10 most frequently requested skills, with customer service (an occupational skill) being the only exception. Social-emotional skills, also known as soft skills, “refer to character traits and interpersonal skills that characterize a person’s ability to interact effectively with others” (Kenton, 2023). Unlike technical skills, soft skills are hard to assess and often difficult to acquire through formal training.
Job postings by occupation
The demand for labour varies across occupations. In 2023, the number of job postings ranged from 18,948 in natural resources and agriculture to 893,469 in the sales and service occupations.
Exploring how demand for skills varies across occupations yields insight into the role of these skills in the labour market. Figure 2 shows the frequency of requests for the top 10 in-demand skills across occupations in 2023.2
Figure 2: Skill demand by occupation
Source: LMIC calculations using 2023 Vicinity Jobs data
Based on this data, there is notable variation in demand for the top 10 skills across occupations.3 A few skills seem to be highly valued across all major occupational categories. Teamwork is a key example, as the only skill with consistently high demand. References to communication skills are similarly widespread, though not universally required. Customer service, the third most frequently requested skill, is primarily associated with sales and service roles, which also generate the highest volume of postings.
Two of the 10 most frequent skills stand out for their high demand within a single occupation: leadership and customer service. Leadership appears in 55% of postings for legislative and senior management occupations (NOC 1), while customer service appears in 52% of job postings for sales and service occupations (NOC 6). These examples highlight how skill demand can vary widely by occupation, indicating that some skills have unique relevance within specific occupations.
Example skills for exploring trends
Our results so far demonstrate how we can use OJPs to look at skills across occupations, but this data allows us to go beyond describing the demand for skills in occupations. The next section of our report demonstrates how we used OJPs to look at the relationships between skills. To do this, we undertook three case studies to explore our proposed methodology. The three skills we chose are teamwork, communication skills and customer service. We chose these skills because they are the three most in-demand skills, each with a different demand profile across occupations. Table 2 provides additional information on these three skills.
Table 2: Case study details
| Teamwork | Communication skills | Customer service | |
| Skill group | Social-emotional | Social-emotional | Occupational |
| Definition | Teamwork refers to the ability to work with a team to achieve a specific goal and is often tied to a collaborative company culture. | Communication skills refer to the ability to express information to others. | Customer service refers to the ability to provide assistance and/or advice to customers (or clients) as a representative of a company. |
| % of 2023 job postings including a request for this skill | 48% | 39% | 33% |
| Demand by occupation | High demand in all occupations | High demand for some occupations (NOC 0–4) | High demand in one occupation (NOC 6: Sales and service) |
| Examples of alternative language found in job postings |
|
|
|
Methodology for identifying and measuring skill associations
Building on LMIC’s prior research on labour market indicators within OJP data, this report introduces methods to capture skill associations within job postings. In a 2022 LMIC study, we looked at three techniques—raw count ordering, revealed comparative advantage, and term frequency–inverse document frequency—to quantify demand based on individual skill frequencies. Although these techniques were effective for understanding single skills, they did not fully capture relationships between skills.
This report builds on the previous work by using a new approach to quantify the relationships between skills through probabilistic and rule-based measurements of association. This methodology allows for a clearer mapping and understanding of how skills relate to each other in the context of individual occupations.
To explore these connections in more depth, we’ve used three complementary methods in this report—frequency measurement, pointwise mutual information and directional relationships—to show how skills co-occur and interact across occupational contexts.
Box 3: Frequency: A Simple Approach to Measuring Relationships
Our approach involves three methods to examine skill relationships based on job postings. The first approach, frequency measurement, identifies how often two skills appear together, providing insight into co-occurrence patterns without consideration for the independent frequency of each skill. This means frequently requested skills may be over-represented, reducing the useful information we can extract about the relationships between skills. We use frequency measurement for initial insights into the demand for skills, which acts as a baseline to compare the more nuanced results from our proposed methodologies.
Measuring skill connections through pointwise mutual information
The first technique to identify connections between skills is a statistical measure called pointwise mutual information (PMI). It is commonly applied in natural language processing to calculate the association between two words in a text.
Association is a measure of the statistical relationship between two variables. It looks for patterns in the behaviour between variables (or, in our case, skills) (Voxco, 2024).
PMI measures how much more (or less) likely is it that two skills will appear together in a job posting than would be expected by chance. This is done by comparing the probabilities of each skill appearing independently to the likelihood that they appear together. By focusing on the likelihood of skills appearing together versus the likelihood of them appearing independently, researchers can identify meaningful associations that frequency counts alone may miss.
PMI is calculated based on the probability two words co-occur in the text with respect to their individual probabilities of occurring PMI is calculated as:
Where:
p(A,B) = probability of skills A and B both being requested in the same job posting
p(A) = individual probability of skill A being requested in a job posting
p(B) = individual probability of skill B being requested in a job posting
The more frequently two skills appear together as compared to their independent occurrence, the stronger the connection and the higher the PMI value.
A high positive PMI value between two skills indicates a strong association, meaning the presence of one skill significantly increases the likelihood of the other skill appearing in the same posting. This highlights an intrinsic correlation between these skills within the context of job postings.
PMI values have no fixed upper or lower bound. This absence makes them hard to compare and interpret. Traditional PMI is also sensitive to low-frequency words; that is, if one or both words appear infrequently, the PMI may be skewed. To address these concerns, a normalized version of PMI (NPMI) is used. The NPMI is calculated by dividing the PMI value by a normalization factor.4 The NPMI bounds the range of values between –1 (no association, meaning the skills never co-occur) and 1 (complete association, meaning the two terms always co-occur). By bounding the NPMI, we reduce the sensitivity to low-frequency words and improve the interpretability of the results.
Supplementing Pointwise Mutual Information analysis by identifying directional skill relationships
While PMI association provides insight into skill connections by identifying sets of skills that appear together more often than would be expected by chance, it is not a predictive measure. To include the layer of prediction, we study the directional relationship between skills using association rule mining—a technique that uncovers interesting relationships, or associations, between variables in large datasets.
As part of association rule mining, we first identified frequently occurring skill sets in the OJP data. Several algorithms exist for this. We used the frequent pattern growth algorithm. This algorithm identifies recurring skill sets across job postings by finding groupings of skills that frequently co-occur, enabling us to spot common combinations across various job postings.
Using these frequently occurring skill sets, we then extracted meaningful skill association rules and identified patterns of skills occurring together more frequently than would be expected by chance alone. By analyzing these patterns, we determined which skills would likely be sought together in job postings, providing valuable insights into skill dependencies and predictive relationships.
In our analysis, we aimed to understand the directional relationships between skills through a metric called confidence. Confidence is the conditional probability of observing one skill (skill B) in a job when another skill (skill A) is required.
In simpler terms, confidence helps us understand how often one skill is mentioned in a job posting when another skill is required.
Confidence is expressed as the ratio of the co-occurrence of skills A and B to the occurrence of skill A:
Where:
Support(A) = proportion of jobs that require skill A
Support (A U B) = proportion of jobs where skills A and B co-occur
This combined approach provides a detailed and quantitative understanding of how skills are related and offers valuable insights that can help identify key skill sets and trends in the job market. NPMI helps us understand the strength of associations between skills, thereby identifying related skills based on OJPs. Using the confidence metric of the directional relationships, we gain insights into how certain skills predictably co-occur within job contexts.
Case studies on connecting skills: Demonstrating our methodology to quantify skill relationships in online job postings
This report focuses on the three most frequently requested skills in 2023: teamwork, communication skills, and customer service. For each, we explored the connections to other skills and how these connections vary across circumstances and approaches.
Box 4: Understanding how to quantify skill connections
Different methods provide unique insights into the connections between skills.
Frequency simply tells us which skills co-occur most often, showing the number of times two skills appear together as a share of the total postings that mention the focus skill.
Associations, calculated using NPMI, share insight into the strength of the relationship between two skills. A high NPMI value for two skills signals that these are closely linked in job postings and can indicate that employers look for both skills together, or at least that they use both skills to describe the needs of roles.
Predictors define the directional connections between two skills, indicating that, if a given job requires one skill, it’s likely to require the other. The confidence value indicates the likelihood that a job posting requiring skill A will also require skill B. However, because this is a directional relationship, it does not mean that skill B will also predict skill A.
Case study #1: Teamwork
Based on 2023 data, teamwork is a skill in high demand. It appears in 48% of job postings and is mentioned across all occupations. However, the high demand for teamwork masks the importance of teamwork’s connection to other skills, especially when only frequency of co-occurrence is considered.
Teamwork connections: Frequency and association
To explore teamwork’s relationships with other skills, we first looked at the frequency with which skills co-occurred with teamwork. Then, we examined associations using NPMI. For our dataset, Figure 3 shows the skills most frequently co-occurring with teamwork and those most strongly associated with teamwork based on NPMI.
Figure 3: Top skills connected to teamwork by frequency and association
Source: LMIC calculations using 2023 Vicinity Jobs data
Source: LMIC calculations using 2023 Vicinity Jobs data
The first chart shows that teamwork most frequently co-occurs with other high-demand skills (as shown in Figure 3). Although frequency shares a clear picture of popular skill pairings, which is valuable for understanding trends in employer demands and the overall labour market, it doesn’t necessarily mean they have a strong or meaningful relationship. To learn more about these connections, we analyzed the associations between skills based on the NPMI.
For example, customer service often co-occurs with teamwork, yet it ranks 12th in terms of NPMI association. For our data, this suggests that, while teamwork and customer service frequently appear together, they are often paired with a variety of other skills, resulting in a lower NPMI association. The results for a few other skills show similar, but smaller, shifts. These shifts highlight the importance of considering NPMI to capture more nuanced relationships between skills.
NPMI association offers a more refined view of these connections because it shows the strength of the relationship between two skills by comparing the probability of them occurring independently to the probability of them occurring together. Given that teamwork is frequently requested across all job postings, it has a high likelihood of appearing independently. This explains why its highest NPMI association estimate sits at only 0.15.
Predictors for teamwork
NPMI skill associations provide valuable insights into connections between teamwork and other skills, but do not fully capture the specific relationships with other skills due to the broad applicability of teamwork. To address this, we use a predictive measure to understand the directional relationship between teamwork and other skills.
This predictive approach allowed us to determine how other skills, when present, indicate a higher chance of teamwork also being mentioned, providing a clearer, more actionable understanding of skill dynamics in job postings. Table 3 lists the 10 best predictors of teamwork, including the confidence that the presence of the skill in a posting predicts the presence of teamwork in that same posting.
Table 3: Top predictor skills for teamwork based on confidence values
| Confidence | Postings | |
| Goal-oriented | 71% | 66,476 |
| Work under pressure | 70% | 212,757 |
| Self-starter/Self-motivated | 66% | 319,161 |
| Fast-paced setting | 65% | 547,673 |
| Interpersonal Skills | 64% | 541,346 |
| Decision-making | 64% | 468,615 |
| Key performance indicators | 64% | 48,536 |
| Attention to detail | 64% | 550,062 |
| Analytical skills | 63% | 156,238 |
| Writing | 63% | 348,071 |
Source: LMIC calculations using 2023 Vicinity Jobs data
In our study, across all job postings, eight of the 10 best predictors for teamwork are social-emotional skills, most of which are also found among the skills most associated with teamwork. “Key performance indicators” is the only strong predictor in this list that is not also found in the top 20 skills associated with teamwork.
Interestingly, despite having the highest rate of co-occurrence and association with teamwork, communication skills are not a strong predictor for teamwork. These skills co-occur more than any other pair of skills but are still independent enough that there is a weak directional relationship from communication skills to teamwork.
Predicting teamwork across occupations
While teamwork is in high demand across occupations, the top predictors of teamwork vary by occupation. We learned more about what teamwork means in the different contexts by looking at the top predictors of teamwork in 2023 for each occupation, as shown in Table 4.
Table 4: Teamwork top predictors by broad occupational category (NOC 2021)
| Category | Top predictor 1 | Top predictor 2 | Top predictor 3 | |
| NOC 0 | Management | CRM software (75%) | Critical thinking (72%) | Microsoft Access (72%) |
| NOC 1 | Business and finance | Critical thinking (70%) | Decision-making (66%) | Coaching (66%) |
| NOC 2 | Natural and applied sciences | Coaching (74%) | Work under pressure (73%) | Interpersonal skills (73%) |
| NOC 3 | Health | Work under pressure (76%) | Dexterity (74%) | Writing (71%) |
| NOC 4 | Education and government services | Dexterity (85%) | Critical thinking (75%) | Report preparation (72%) |
| NOC 5 | Arts and recreation | Adobe Systems Adobe Creative Suite (77%) | Search engine optimization (73%) | Maya (73%) |
| NOC 6 | Sales and service | Goal-oriented (74%) | Work under pressure (74%) | Interpersonal skills (69%) |
| NOC 7 | Trades and transportation | Interpersonal skills (72%) | Hand-eye coordination (71%) | Work under pressure (70%) |
| NOC 8 | Natural resources and agriculture | Interpersonal skills (82%) | Dexterity (80%) | Team-building (80%) |
| NOC 9 | Manufacturing and utilities | Work under pressure (74%) | Dexterity (69%) | Self-starter/self-motivated (69%) |
Note: The values in brackets show the confidence, or likelihood, that teamwork would be in a job posting based on the presence of the predictor skill.
Source: LMIC calculations using 2023 Vicinity Jobs data
For our dataset, there is little consistency among the top predictors of teamwork across occupations, with strong predictors from social-emotional skills, technologies and occupational skills depending on the occupation. Additionally, confidence levels for top predictors mostly increase when looking at the occupation-specific results compared to the aggregate results. The increased confidence and variation in predictors across occupations highlights the benefit of context when researching trends in the labour market.
Case study #2: Communication skills
This second case study looks at communication skills, which appear in 39% of the job postings. While teamwork is in high demand for all broad-level occupations, communication skills show more variation: the 2023 data show a higher demand in NOCs 0–4 than in NOCs 5–9. The variable demand allowed us to explore how the popularity of the skill in specific occupations affects its relationship to other skills.
Communication connections: Frequency and association
Looking at the skills most frequently co-occurring with communication and those with the strongest NPMI associations yields insights into the types of communication skills in-demand.
Figure 4 shows the skills that occur most frequently with communication (top graph) and those with the strongest NPMI associations (lower graph).
Figure 4: Top skills connected to communication by frequency and association
Source: LMIC calculations using 2023 Vicinity Jobs data
Source: LMIC calculations using 2023 Vicinity Jobs data
As expected, for this dataset, communication skills appear most frequently with other skills in high demand, most commonly appearing with teamwork and customer service. Yet neither of these skills rank in the top 10 skills associated with communication when measured by NPMI.
The strongest associations with communication, as indicated by NPMI, are interpersonal skills and writing. Following these, problem-solving, analytical skills and time management also show strong associations. Writing, a form of communication, appears in only 26% of postings that include communication, ranking eighth in the Figure 4 top graph. However, its high NPMI association with communication—the second-strongest—suggests that postings explicitly request overall communication skills even when including writing as a required skill.
Predictors for communication skills
After identifying the skills most associated with communication skills using NPMI, the next step was to determine which skills were top predictors of communication in these job postings. For each predictor skill, a confidence level (see Table 5) indicates the likelihood that a job posting in our set would include communication if the predictor skill was present.
Table 5 shows the 10 skills that are the strongest predictors of a job posting including communication.
Table 5: Top predictor skills for communication based on confidence values
| Confidence | Postings | |
| Writing | 83% | 348,071 |
| Analytical skills | 78% | 156,238 |
| Presentation skills | 78% | 58,246 |
| Negotiation skills | 78% | 43,951 |
| Conflict management skills | 76% | 34,113 |
| Interpersonal skills | 75% | 541,346 |
| Research skills | 74% | 33,765 |
| Microsoft Access | 73% | 24,372 |
| Multi-tasking | 70% | 214,137 |
| Microsoft Outlook | 70% | 93,044 |
Source: LMIC calculations using 2023 Vicinity Jobs data
Writing stands out as the top predictor for communication skills. The confidence level of 83% means that approximately eight of every 10 postings requesting writing skills also mention communication skills. This strong connection aligns with the high NPMI association between the two. Writing is intrinsically tied to communication, reinforcing the need for writing skills.
The alignment between NPMI association and confidence for top skills shows that the top skills related to communication skills tend to be social-emotional, except for technologies specifically related to the Microsoft Office suite. These results highlight the importance of NPMI and confidence in understanding the nuanced relationships between communication and other competencies.
Predicting communication skills across occupations
We added occupation to the predictors to further see how the demand for communication skills varied within our dataset. Table 6 shows the top predictors of communication skills by broad occupational category.
Table 6: Communication skills top predictors by broad occupational category (NOC 2021)
| Category | Top predictor 1 | Top predictor 2 | Top predictor 3 | |
| NOC 0 | Management | Writing (93%) | Microsoft Access (86%) | Microsoft Windows (83%) |
| NOC 1 | Business and finance | Writing (86%) | Presentation skills (83%) | Interpersonal skills (80%) |
| NOC 2 | Natural and applied sciences | Writing (85%) | Presentation skills (84%) | Negotiation skills (82%) |
| NOC 3 | Health | Analytical skills (88%) | Writing (87%) | Computer skills (85%) |
| NOC 4 | Education and government services | Writing (83%) | Negotiation skills (81%) | Multi-tasking (80%) |
| NOC 5 | Arts and recreation | Interpersonal skills (74%) | Goal-oriented (73%) | Analytical skills (72%) |
| NOC 6 | Sales and service | Writing (82%) | Analytical skills (77%) | Presentation skills (77%) |
| NOC 7 | Trades and transportation | Writing (80%) | Analytical skills (72%) | Microsoft Outlook (65%) |
| NOC 8 | Natural resources and agriculture | Writing (85%) | Analytical skills (81%) | Lean manufacturing (75%) |
| NOC 9 | Manufacturing and utilities | Writing (77%) | Microsoft Outlook (68%) | Analytical skills (68%) |
Source: LMIC calculations using 2023 Vicinity Jobs data
For our data, we found that writing is the best predictor of communication skills for eight out of 10 occupations and the second-best predictor for health occupations. The only major occupational category in which writing is not among the top three predictors for communication skills is NOC 5 (Occupations in art, culture, recreation and sport), where interpersonal skills, goal-oriented nature and analytical skills are the most predictive.
Analytical skills also show a strong relationship to communication skills, ranking as the second-best overall predictor (see Table 5) and appearing in the top predictors for six out of 10 occupational categories (see Table 6).
The consistent association and high confidence values across occupational categories highlight the adaptability and variability of demand for communication in different occupational contexts.
Case study #3: Customer service
Our final case study focuses on customer service, which appears in 33% of the 2023 job postings that we examined. Customer service is primarily associated with jobs in sales and service occupations. We wanted to explore how a skill that’s typically linked with a particular field functions in other areas. Customer service is an ideal skill to examine for this purpose.
Figure 5 shows the skills that co-occur most frequently with customer service (top graph) and those that are most strongly associated with it, according to NPMI (lower graph).
Figure 5: Top skills connected to customer service, by frequency and association
Source: LMIC calculations using 2023 Vicinity Jobs data
Source: LMIC calculations using 2023 Vicinity Jobs data
While customer service frequently co-occurs with social-emotional skills, the NPMI results show a broader range of associations. For this dataset, customer service is actually less associated with social-emotional skills than the skills in our other case studies, with only one social-emotional skill appearing in the top 10 associations. This highlights the finding that a more diverse mix of skills is associated with customer service; in fact, all four skill groups are represented.
Association by occupation for customer service
Customer service is an occupational skill; it’s closely related to sales and service occupations but is also found in postings across other occupations. This makes customer service ideal for studying how trends in sales and service occupations compare to those in all other occupations. We conducted additional research in this final case study to explore that.
The comparison of skill associations across occupations starts by looking at the skills most associated with customer service for sales and service occupations (Figure 6, top graph) and then examining its associations in all other occupations (Figure 6, lower graph).
Figure 6: Top skills associated with customer service, by occupation
Source: LMIC calculations using 2023 Vicinity Jobs data
Source: LMIC calculations using 2023 Vicinity Jobs data
Our results show different trends for sales and service occupations as compared to the set of all other occupations. Only three skills—computer terminal, cash handling, and ability to cope with a fast-paced setting—appear in the top 10 associations for both groups. In sales and service occupations, the top three skills associated with customer service are interpersonal, teamwork and communication skills, which are all social-emotional skills. This highlights the importance of people skills in customer service roles within this occupational group.
In contrast, the skills associated with customer service outside of sales and service occupations reflect functions generally linked to sales and service occupations. This includes skills related to managing transactions (such as using cash registers, handling cash, and working with point-of-sale software and terminals) and managing a storefront (such as setting up visual merchandising and using inventory control systems).
Predictors for customer service
Table 7: Top predictor skills for customer service based on confidence values
| Confidence | Postings | |
| Cash registers | 75% | 49,244 |
| Sales | 70% | 105,875 |
| Work under pressure | 49% | 212,757 |
| Fast-paced setting | 49% | 547,673 |
| Multi-tasking | 48% | 214,137 |
| Work scheduling | 46% | 85,925 |
| Inventory management | 46% | 225,946 |
| Goal-oriented | 45% | 66,476 |
| Occupational health and safety | 44% | 208,929 |
| Flexibility | 43% | 617,348 |
Source: LMIC’s calculations using 2023 Vicinity Jobs data
Overall, for our study, mentions of cash registers or sales are the top predictors of customer service being included in a job posting, with confidence at or above 70%. These two skills stand out because all other predictors fall below 50% confidence. For context, a confidence level of 50% means that customer service only appears in every other posting that includes the predictor skill.
Table 8: Top predictor skills of customer service, by occupation
| Top predictors (sales and service only) | Confidence | Top predictors (set of all other occupations) | Confidence |
| Point-of-sale (POS) software | 79% | Multi-tasking | 41% |
| POS systems | 77% | Work under pressure | 40% |
| Cash registers | 73% | Inventory management | 39% |
| Interpersonal skills | 72% | Office administration | 38% |
| Problem-solving | 71% | Microsoft Outlook | 37% |
| Occupational health and safety | 70% | Fast-paced setting | 36% |
| Sales | 70% | Ordering of supplies and equipment | 35% |
| Microsoft Outlook | 70% | Attention to detail | 35% |
| Microsoft Office | 67% | Handling heavy loads | 34% |
| Microsoft Suite | 67% | Scheduling | 34% |
Source: LMIC calculations using 2023 Vicinity Jobs data
In this analysis, confidence for predictors varies significantly between sales and service occupations and the set of other fields. In sales and service roles, the top predictors’ confidence levels range from 67% to 79%, showcasing the strong demand for customer service skills within this field. However, in the set of other occupations, confidence is much lower (i.e., the highest confidence rating—for multi-tasking—is only 41%). This lower confidence indicates that, outside of sales and service positions, customer service skills may not be as strongly linked to specific predictor skills.
Both NPMI association and predictors give valuable perspectives on the relationships between skills, but they serve different purposes. NPMI identifies the strength of relationships between skills, which can be useful for understanding skill dimensions and identifying transferable skills. Predictors highlight one-way directional relationships, helping to identify job postings in a specific field (sales and service) that would likely show interest in customer service skills.
Each approach has unique applications. Therefore, the choice of which to use depends on the end goal. For those looking at career transitions, NPMI associations may reveal transferable skills. For those focused on job requirements within a specific occupational group, predictors offer insights into which postings represent roles most likely to benefit from customer service skills.
What does this mean? Why does it matter?
Using OJPs as a source of LMI, this report has introduced novel ways to measure and understand skill relationships. Through our case studies, we demonstrated how these methods can uncover the relationships between skills in job postings to enable a deeper understanding of labour market needs.
The frequency count method allows us to identify the skills that most frequently co-occur. However, while straightforward, frequency counting alone can be misleading because of the uneven demand for skills. This imbalance can create the perception that high-demand skills are closely connected, even if they lack a meaningful relationship in practice.
To adjust for the imbalance in skill demand, we can (and did) look to skill associations. This technique examines how skills connect to each other based on their joint use in OJP text. Put simply, it leverages the idea that employers frequently seek candidates who are proficient in sets of related skills. This allowed us to look at the natural relationships between skills within job postings.
Associations also tell us about the dimensions of a skill, based on overlaps with the demands for other skills. This can be useful for translating a skill between contexts, for example when changing careers.
In this work, we also demonstrated the value of looking at skill predictors, which help identify where a skill would be a good fit based on the presence of other skills in a posting. Predictors are useful when looking for postings that relate to a specific skill, even if it is not directly mentioned.
These methods revealed nuanced information. For instance, we found that our studied skills mostly co-occur with social-emotional skills. Due to the high demand for them, social-emotional skills are more likely to co-occur with each other and other skills. Interestingly, we learned that teamwork is the skill that most frequently appears with other social-emotional skills, and that the presence of teamwork actually predicts an increase in the average number of social-emotional skills in a posting.
Association analyses revealed that teamwork has low values of association to other skills. This might be because teamwork is so frequently in demand, which would increase the likelihood that it would simply appear independently, thereby lowering the possible association value. The lower associations limited how much we could learn about the relationship between teamwork and other skills; it doesn’t seem to form strong, specific pairings. However, we did find that most skills associated with teamwork are still social-emotional skills.
This trend continued when looking at predictors for teamwork, where eight of the ten best predictors for teamwork are social-emotional skills.
For communication, we saw strong connections with writing regardless of the research method used. This indicates that skills related to communication form a more consistent group than those seen for teamwork or customer service.
The skills most relevant to communication—writing, analytical skills, interpersonal skills and multi-tasking—consistently appear in both predictors and associations. Further, predictions of communication remain relatively consistent across occupations. This shows a consistent skill profile, which helps with clearly defining what employers mean when requesting communication skills in job postings.
Customer service, the subject of our third case study, also shows frequent co-occurrence with social-emotional skills, but presents more variation in its skill relationships in our other approaches. A unique aspect of customer service is that most of its overall predictors have low confidence measurements, with only two predictor skills having a confidence value greater than 50%.
Given that customer service is highly related to sales and service occupations, we looked at how its results for associations and predictors differed in the specific context of sales and service occupations versus the set of all other occupations. Within sales and service occupations, the top skills associated with customer service are all social-emotional skills, emphasizing the importance of people skills in those occupations. Predictors also had higher confidence results in the sales and service context, while there was lower confidence in the analysis conducted with the set of all other occupations. These lower rates of confidence tell us that there is likely little consistency in how customer service is approached outside of sales and service occupations.
Understanding these nuanced relationships, including the strength of associations and the directional relationships between skills, can foster a better understanding of skill demand trends in the labour market. Further analysis with granularity, such as occupational breakdowns, could reveal information that might otherwise be obscured in the aggregate.
The way forward
LMIC will continue to develop best practices and guidelines for understanding skill demand through online job postings
This report provides a springboard for further research, demonstrating how researchers can use OJPs as a source of LMI to inform decisions and strategic workforce development initiatives. The methodology presented in this report could be used to replicate this exercise across other dimensions from OJPs.
For instance, future studies could probe deeper by looking at trends related to specific characteristics in job postings, perhaps by analyzing the impact of education and training requirements or exploring the differences in skill requirements across various work settings (e.g., on-site, hybrid and remote).
This methodology could also be used to compare trends in different labour markets, such as across regions, or to provide focused analysis on specific skill groups, such as skills related to artificial intelligence (AI). By examining the relationships between AI-related skills and other skills, we could identify key competencies that are frequently required together and uncover emerging skill combinations that are gaining traction.
Another approach could be to explore occupations at a more granular level than this initial work. This could be done by incorporating more detailed occupational classifications (e.g., using the five-digit NOC code) to deepen our understanding of the changing nature of skills closely associated with specific occupations. This, in turn, could facilitate a more comprehensive understanding of occupational pathways.
The methodology presented in this report opens avenues for a range of research possibilities, including a thorough mapping of skills across occupations, cluster analyses to learn about the natural groupings of skills, or longitudinal studies to track the evolution of trends over time.
This research has established the foundation for a comprehensive approach to understanding the nuanced relationships between skills and their impact on labour market trends, providing valuable insights that might otherwise be overlooked.
Acknowledgements
This report was prepared by Laura Adkins-Hackett and Sukriti Trehan for LMIC.
For more information about this report, please contact research@lmic-cimt.ca.
How to cite this report
Adkins-Hackett, L. & Trehan, S. (2024). Connecting skills: Using online job postings to unravel demand for skills in the labour market. Ottawa: Labour Market Information Council (LMIC).
Endnotes
1 Recorded vacancies are provided by Statistics Canada through the Job Vacancy and Wage Survey.
2 For occupation, we use the one-digit broad occupational category from the National Occupation Classification, published in partnership between Employment and Social Development Canada and Statistics Canada. Vicinity Jobs has mapped 83% of job postings to their broad occupational category.
3 Frequency was determined by looking at the job postings for each occupation in isolation, so that the significant variation in the number of postings per occupation would not affect the results.
4 The normalization factor used for our NPMI calculations is referred to as self-information and is computed using the formula - log2 p(A, B)
References
Bennett, F., Escudero, V., Liepmann, H., & Podjanin, A. (2022). Using online vacancy and job applicants’ data to study skills dynamics. International Labour Organization. https://webapps.ilo.org/static/english/intserv/working-papers/wp075/index.html
Dawson, N., Williams, M-A., & Rizouio, M-A. (2021). Skill-driven recommendations for job transition pathways. PLoS ONE 16(8): e0254722. https://doi.org/10.1371/journal.pone.0254722
Dewar, J. (2022, November 16). 3 ways coded language is costing you applicants. LinkedIn. https://www.linkedin.com/business/talent/blog/talent-acquisition/ways-coded-language-is-costing-you-applicants
Djumalieva, J., & Sleeman, C. (2018). An open and data-driven taxonomy of skills extracted from online job adverts. Economic Statistics Centre of Excellence.
Hickok, H. (2021, July 29). The coded words of job adverts that reveal important clues. BBC. https://www.bbc.com/worklife/article/20210727-the-coded-words-of-job-adverts-that-reveal-important-clues
JDXpert. (2023, July 11). The job posting mirage: How misleading descriptions drive away top talent. JDXpert. https://blog.jdxpert.com/the-job-posting-mirage-how-misleading-descriptions-drive-away-top-talent
Kenton, W. (2024, August 26). What are soft skills? Definition, importance, and examples. Investopedia. https://www.investopedia.com/terms/s/soft-skills.asp#toc-what-are-soft-skills
LMIC. (2020, October). How representative are online job postings? LMI Insights, no. 36. https://lmic-cimt.ca/publications-all/lmi-insight-report-no-36/
LMIC. (2022, April). Developing indicators for skill demand. LMI Insights, no.. https://lmic-cimt.ca/developing-indicators-for-skill-demand/
Organisation for Economic Cooperation and Development. (2022). Skills for the digital transition: Assessing recent trends using big data. https://doi.org/10.1787/38c36777-en
Vicinity Jobs. (n.d.). Vicinity Jobs skills taxonomy. Vicinity Jobs. https://vicinityjobs.net/trends-user-guide
Voxco. (2024). Difference between association and causation. Voxco. https://www.voxco.com/blog/association-vs-causation/