The most common pay gap everyone is familiar with is the unadjusted pay gap. It measures the salary difference between men and women expressed as a percentage of men’s salary. But in Sysarb you can run into two different types of pay gaps:

- Unadjusted pay gap

- Adjusted pay gap

The adjusted pay gap similarly to the unadjusted pay gap shows the salary difference between men and women but accounts for gender-neutral factors that can legitimately influence pay such as grade, experience and performance. The part of the pay gap that can be explained by the selected factors are shown as explained. The remaining part after explanations is the adjusted pay gap.

This remaining gap does not automatically mean something is wrong, but it is the part that usually deserves closer review because it cannot be justified using the pay factors you have selected.

A key principle is that the model does not decide what is fair. Your organisation does. The adjusted pay gap reflects your pay philosophy, not an external judgement.

Sysarb uses regression analysis to understand how salary relates to your selected pay factors across the whole organisation. This same analysis is used consistently to calculate:

An important distinction is that the relationships between pay factors and salary are learned from the entire audit, not from individual teams or roles. This means:

Sysarb calculates the adjusted pay gap using a log-linear regression analysis. This is the standard approach in pay equity analysis and is widely used by both researchers and practitioners.

The adjusted pay gap can be calculated with two different methods:

- The decomposition method

- The direct method

The decomposition method (Blinder–Oaxaca) is the primary method in Sysarb. Its advantage is that it shows not only how much of the pay gap is explained in total, but also how much each selected pay factor contributes to explaining that gap.

In some situations, for example when the data is limited or when pay factors overlap too much, this method can become unstable. In those cases, Sysarb uses the direct method instead. The direct method is more stable, but it can only show how much the selected pay factors explain together, not the contribution of each factor on its own.

Whether a pay factor reduces or increases the adjusted pay gap depends on two things: how that factor is related to salary in your organisation, and how the two groups differ on that factor.

- If a factor is associated with higher pay, and men have higher mean values for that factor, then part of the pay difference can be explained by that factor. In that case, the adjusted pay gap becomes smaller.

- If a factor is associated with higher pay, and women have higher values for that factor, then the factor does not explain the gap. Instead, it can make the adjusted pay gap larger.

The same logic applies if a factor is associated with lower pay. Depending on how the groups differ, the factor may either reduce or increase the adjusted pay gap. This is expected behaviour. It does not mean the model is wrong. It means that, based on your data and the pay factors you selected, that factor does not explain the observed pay difference in the way you might expect.

The adjusted pay gap and predicted salaries are two views of the same model.

At a group level, the model shows how much of the pay gap is explained by differences in pay factors between groups.

At an individual level, the model uses the same factor relationships to calculate a predicted salary for each employee. This predicted salary represents the salary that would be expected for someone with that combination of role characteristics, given your organisation’s current pay structure.

Pay factors fall into two broad types. Understanding the difference helps you choose pay factors more effectively.

Numerical factors have a natural order and size. Moving from one value to the next represents a meaningful step.

Examples:

For numerical factors, the model assumes that each step has a similar effect on salary. Numerical factors are generally stable and efficient, and they rarely cause model failure on their own.

Categorical factors group employees into categories without a natural numeric order.

Examples:

Each category is treated as its own group. This means categorical factors consume more data and increase the risk of small group sizes, especially if you have many categories or combine several categorical factors.

Because of this, categorical factors are the most common cause of model instability when too many are selected.

For something to be counted as pay factor it should explain salary in your organization while being gender neutral. Ultimately, it’s your decision and responsibility to select factors that are compliant with your local legislation or agreements.

What it represents
A job’s complexity and responsibility based on a job evaluation.

Type
​Numerical.

When to use it
Use Grade when job evaluation is a core part of how you structure pay and pay ranges.

HR perspective
Grade is one of the most accepted and defensible explanations for pay differences. It is role-based, transparent, and usually the strongest driver in the model.

What it represents
An alternative way of capturing role complexity, often used instead of formal grading.

Type
​Numerical.

When to use it
Use Position level if it reflects how roles are structured and rewarded in practice.

HR perspective
In most cases, use either Grade or Position level. Using both can blur the explanation and make it harder to communicate how pay is determined.

What it represents
Whether a role includes formal managerial responsibility.

Type
​Categorical.

When to use it
Use it when managerial responsibility is intended to influence pay.

HR perspective
This factor helps avoid comparing managers and non-managers as if they were equivalent, which often improves both fairness and clarity. This factor might share some overlap with grade.

What it represents
Individual performance outcomes used in salary setting.

Type
​Numerical.

When to use it
Use it only if performance is measured consistently and clearly linked to pay.

HR perspective
Performance is common in pay philosophies, but it is also sensitive. If performance ratings vary widely by manager or show systematic differences between groups, including it may hide issues rather than explain them.

What it represents
Age as a proxy for experience.

Type
​Numerical.

When to use it
Only when age/experience is an explicit and accepted part of your pay philosophy.

HR perspective
Age is legally and socially sensitive in many contexts. If the intention is to capture experience, tenure-based factors are often clearer and easier to defend.

What it represents
Time employed in the organisation.

Type
​Numerical.

When to use it
Use it when organisational experience or loyalty is expected to influence pay.

HR perspective
Common in public sector and collectively agreed pay models, but it can also reflect historical patterns. Make sure it represents intent, not inertia.

What it represents
Time spent in the current role.

Type
​Numerical.

When to use it
Use it when competence and contribution are expected to increase with time in role.

HR perspective
Job tenure is often easier to justify than company tenure because it relates directly to role-specific experience. Often shares overlap with company tenure.

What it represents
Broad functional grouping of roles.

Type
​Categorical.

When to use it
Use it when labour market pay differs meaningfully between functions.

HR perspective
Job family helps ensure you compare similar jobs without becoming overly detailed.

What it represents
More specific role groupings within a job family.

Type
​Categorical.

When to use it
Use it when sub-families reflect real and stable market differences and groups are large enough.

HR perspective
Too many small groups can reduce model stability and make results harder to explain.

What it represents
Overall scope and impact level of a role.

Type
​Categorical.

When to use it
Use it when career bands are clearly defined and actively used in pay decisions. For example, when employees with different career bands are paid differently based on their career band.

HR perspective
Career bands are often intuitive for employees and support transparent pay conversations when applied consistently.

What it represents
Where the work is performed.

Type
​Categorical.

When to use it
Use it when location is an intentional driver of pay differences. For example, if you intentionally pay people working in larger cities more than smaller cities.

HR perspective
Locality is important in distributed organisations. If you do not actively differentiate pay by location, including it can weaken the pay story.

What it represents
Type of employment contract.

Type
​Categorical.

When to use it
Use it when a contract is tied to different pay structures or agreements. For example, if collective agreements determine the employee’s pay.

HR perspective
Contract may explain pay differences, but it can also highlight structural risks if one group is overrepresented in certain contracts.

What it represents
Whether employment is full-time or part-time. Employment rate is used determined whether the employee is working full-time or part-time.

Type
​Categorical.

When to use it
Use it only when there are valid, policy-based reasons why full-time and part-time employees are expected to have systematically different salaries.

HR perspective
Because salaries in Sysarb are already normalised to full-time equivalents, this factor should be used cautiously and only when clearly justified by your pay philosophy.

What it represents
Organisation-specific pay drivers, such as certifications, allowances, or critical skills.

Type
​Categorical.

When to use it
Use custom factors when they are clearly defined, consistently applied, and genuinely influence pay.

HR perspective
Custom factors can strengthen transparency when well governed. If a factor is hard to explain, it is usually better treated as an action item than as a statistical adjustment.

It is tempting to select many pay factors to “fully explain” salary differences. In practice, this often makes the model weaker, not stronger.

Regression analysis needs enough employees in each combination of pay factors to learn meaningful relationships. When too many factors are selected, especially categorical ones, the data gets split into many small subgroups. This increases the risk that the model cannot reliably estimate how pay relates to those factors.

This can lead to:

· The model failing to run

· Unreliable results

· Factors being excluded

· Too closely correlating factors

This does not mean your data is wrong. It usually means the model is being asked to explain more detail than the data can support.

As a rule of thumb:

· Select 2–4 core factors that clearly reflect your pay philosophy

· Be especially cautious with multiple categorical factors at the same time

· If the model fails, reduce the number of factors rather than adding more

You can use the export details (for example adjusted R²) to confirm model strength. As guidance, an adjusted R² above 0.5 generally indicates a reliable model.

Most reliable models use:

A strong model is not one that removes the adjusted pay gap entirely, but one that makes the remaining gap meaningful and actionable.
​
​
​