What is the skill adjacencies model?
This skill adjacencies model assesses the strength of the relationship between two skills based on how often they are found together in job postings and the uniqueness of their relationship. We apply this model to a number of skills and generate a network graph that represents these relationships.
The skills adjacency model also includes information on the demand for primary, secondary, and tertiary skills. Learn more about our demand data.
What can I use this data for?
You can use the skills adjacencies model to identify opportunities to train talent with related skills to learn a desired skill.
Acquire: Where do you get your skills data?
To identify skills, we monitor job postings from thousands of sources, including job boards, corporate sites, partner feeds, news sites, staffing websites, and applicant tracking systems. Every day, we process an average of 1.3 million job postings in 22 different languages. Learn more about our skills data.
Organize: How do you organize the skills data?
Analyze: How do you calculate the skill adjacency model?
We determine the primary skill by identifying the skill found in the greatest number of job postings related to your Talent Profile (but irrespective of your selected time period, locations, and employers). The default primary skill can be replaced with any skill that falls within the 100 most common skills for your Talent Profile.
Secondary skills are those with the highest relatedness scores for the primary skill, while tertiary skills are those with the highest relatedness scores for the secondary skills. We calculate the “relatedness score” of two skills using a scaled variant of chi-squared, a statistical test for independence. You can learn more about our chi squared test in the More section.
The relatedness score is calculated using job postings found over the past four years for the functions and occupations in your Talent Profile. These job postings are not narrowed down by other attributes like your selected time period, location, employer, or other Talent Profile filters. Why? If your Talent Profile doesn’t include a function or occupation, then we evaluate all job postings from the past four years.
However, it’s important to note that all of your search criteria, including time period, location, employer, and other Talent Profile filters are taken into account when calculating demand for the primary, secondary, and tertiary skills.
Deliver: How do you communicate data in the skill adjacency model?
The primary skill is represented in orange, the secondary skills in dark blue, and the tertiary skills in bright blue.
The relatedness score for a pair of skills (primary and secondary, or secondary and tertiary) can range from 0 to 1. A score of 0 means that two skills have no discernable relationship, while a score of 1 means that every time one skill is found in a job posting, the other is as well, and vice versa. The relatedness score determines which skills appear in the model
We represent demand for a skill as a count and percentage job postings returned for your search. Demand is influenced by your searched Talent Profile, employers, and locations and determines the size of the bubbles in the model.
By default, the model includes both hard and soft skills, though the results can be filtered to show only soft skills. For each skill, we show demand for your searched locations and your searched employers (“the market”). However, the model can be adjusted to show demand at your organization only.
More about skill adjacencies:
What is the difference between hard and soft skills?
All skills in our system are categorized either as hard or soft skills. Hard skills are typically tools, technologies, and subject matter knowledge, while soft skills are typically occupational traits, like “hardworking”, and work requirements, like “ability to travel.” By default, we display hard and soft skills together, though you can adjust the results to show only hard skills.
Why am I seeing a skill in the results that doesn't align with my search?
The skills you see are those found in job postings that match your search criteria. You may want to make sure the filters you’ve applied (like function and occupation) are relevant to your role.
There may also be instances where our system mistakenly identifies a word phrase from a job description as a skill. Often, this content comes from the sections of the job posting that discuss the company and its values, the job benefits, or advancement potential. We are working on solutions to identify and isolate these parts of job descriptions and remove these false positives from our database. In the meantime, please feel free to report erroneous skills by emailing TNSupport@gartner.com.
What kind of chi-squared test do you use to determine relatedness?
Typically, chi-squared tests are used to determine whether the relationship between two variables can or cannot be attributed to pure chance. In our model, we apply the chi-squared test in an inverse fashion to identify pairs of skills that often appear in job postings together, indicated by high chi squared value between them.
Chi squared values do not naturally fall between 0 and 1, and very highly linked skills may have values in the tens of millions, so we normalize the data so that all values fall on a comparable 0 to 1 scale.
A score of 1 means that postings containing skill A always include skill B, and vice versa. For example, we would expect “cooking” and “baking” to have a relatively high relatedness score. A score of 0 means that skills A and B have no significant relationship, and that any appearance of these two skills in the same job posting is statistically random. We would expect a skill like “Spanish language” to have a near-zero score to hard skills such as “graphic design” because language skills are needed across many professions, not just design roles.
Why aren’t location, employer, or certain Talent Profile attributes taken into account when calculating relatedness or determining the primary skill?
Function and occupation (within a Talent Profile) are the only attributes that determine which job postings the relatedness calculation takes into account. We believe that none of our other search criteria (location, employer, experience level, etc.) should have a meaningful impact on the relationship between two skills.
To use an analogy: baking fruit pies and baking meat pies are two related skills, and bakers who only know how to make one version could fairly easily learn how to make the other. If we were to take location into account when calculating relatedness, the two types of pies would be more closely related in the United Kingdom, where meat pies are more common, and less closely related in the United States, where fruit pies predominate.
This regional variation is misleading and detracts from the purpose of the model, which is meant to measure how easily someone with one skill can learn another.