Handling pay equity in small groups

High-level recommendations

Working with small groups introduces statistical limitations that affect the reliability of pay equity analysis. Keep the following principles in mind before defining your categories of workers or running any model.

Headcount per group: Aim for at least 20–30 employees of each gender across the full dataset to establish meaningful patterns. Within each category of workers, target at least 5 employees per gender.
Gender balance: Equal numbers of men and women produce the most reliable results, but the analysis can work when the smaller gender group represents at least 15–25% of the larger. This is just enough to work with, but results may still be unreliable.
Structural distribution: If most women concentrate in one job family or grade, the model cannot separate gender effects from job structure, making it difficult to draw valid conclusions.

Defining categories of workers

The core principle in defining worker categories is to determine what constitutes work of equal or substantially similar value.

According to the EU Pay Transparency Directive (Directive (EU) 2023/970), the criteria used to define a category of worker should include four factors: skills, effort, responsibility, and working conditions. Local legislation may impose specific category definitions, and worker representatives may need to approve them. Where categories have not yet been defined, we recommend a broad approach based on grades or role levels, ideally grounded in a job evaluation methodology.

Job family requires particular care. It should reflect the four factors above in line with the Directive. When based on job architecture and for organizations without job grades, job family can be an option, but it may carry risks if not well calibrated. For example, male-dominated or female-dominated job families can prove problematic.

Avoid categories with fewer than 5–10 employees. You can combine levels and groups as long as they represent similar skills, effort, responsibility, and working conditions – and as long as that combination can be justified and documented.

For further guidance on defining categories of workers, see Getting ready for the EUPTD with PayAnalytics.

Simulating different category definitions

Before committing to a category structure, test it in a spreadsheet to identify problem areas early:

Build a simple table in Excel: job family × gender × headcount.
Highlight any rows with fewer than 5–10 people, or rows containing only men or only women.

The example below shows how this analysis can be expanded into a full risk evaluation dashboard across different category definitions:

Risk evaluation dashboard with worker categories included

Combine groups smartly. If you have grades that represent substantially similar work, such as Grade 4A and Grade 4B, group them together. Roles with different pay logic, like engineers and salespeople, should stay separate, as combining them will produce unfair comparisons.
You can also check several different combinations to find the structure that balances statistical validity with fairness.

When groups are still too small

Manual average pay review

When a group contains only a handful of employees of one gender, the model has limited information to calculate the pay gap. In those cases, we recommend traditional comparisons instead of regression modeling. Compare the average salary of women and men within the same group. For instance, five women in Marketing earn an average of €52,000, while eight men earn an average of €55,000. That difference can become the starting point for investigation.

PayAnalytics can prove helpful in such investigations, as it offers the Group compensation distribution view, showing average pay and the full distribution by gender within each group:

Group compensation distribution by gender

Company example

Think of a company with 200 employees, 27 of whom are women. Only two women work in Grade 6; the rest sit across Grades 2–5. The two women in Grade 6 earn €58,000 and €60,000, while most men in Grade 6 earn between €61,000 and €65,000.

To investigate, start by examining other variables that may not appear in the model. Tenure, for instance. Are the two women newer in their roles? If they have 2 years of experience compared to 10 years for their male peers, does that explain the difference?

Next, review what data is missing. If performance scores were not included in the dataset but are relevant in this organization, add them and re-run the analysis. If nothing explains the gap, you may have a pay equity issue.

The Employee plot in PayAnalytics helps visualize how individual salaries relate to Time in Role by gender:

Employee plot by Time in Role and gender

When the system does not provide pay recommendations

PayAnalytics sometimes stops giving raise suggestions when groups are too small and the gender mix is highly uneven. The applicable threshold appears in the system settings. See Interpreting pay gaps for further explanation.

In those cases, a manual review process applies:

Compare individually: Match each woman with men in the same grade and assess whether the pay difference has a clear explanation.
Document the reasoning: Record findings for each employee, for example: "Employee A earns less due to shorter tenure; Employee B is an unexplained gap."
Act on real gaps: Where no justification exists for Employee B's lower salary, correct it.
Track over time: A single year's snapshot data can be misleading. The same pay pattern observed year after year, even in small groups, provides stronger evidence of an issue.

Practical implementation tips

Working with 20–60 employees per dataset

The viability of regression modeling depends entirely on the granularity of the job architecture. A highly detailed model – splitting a small workforce into 15 job families across 10 grades and locations – produces micro-cohorts of 1–2 people, making it statistically impossible to identify meaningful patterns or isolate the true drivers of pay.

Conversely, by consolidating into a more broadly defined structure – such as 4 job families – you can often uncover clear patterns even in a group of 60 employees. Ultimately, no fixed "headcount rule" applies; the goal is to find the right balance between a model detailed enough to be fair and broad enough to be statistically valid. Where that balance proves unachievable, we recommend a manual average pay review rather than regression modeling.

Including small countries and legal entities

To ensure compliance with varying local thresholds and reporting requirements, consider the following data strategies:

Strategic data partitioning: Segment your data by country, either by maintaining distinct datasets or by using subgroup analysis with country as the primary subgroup.
Threshold-based segmentation:
- Reporting group: Create a primary dataset for countries that meet the mandatory reporting threshold – typically >100 employees, though some national transpositions suggest 50.
- Monitoring group: For countries below the threshold, group them separately. While the 5% pay gap per category reporting obligation does not apply to these countries, you should manually review outliers to prepare for potential right-to-information requests from employees, as no threshold is relevant here.
  
  The Pay Equity Analysis in the platform provides the key metrics and outlier detail needed to identify and document individual cases ahead of such inquiries:
  
  Pay gaps, compensation model, and outliers in the Pay Equity Analysis

Using a single model for all of Europe

If you are considering a single model for all of Europe, remember that it depends on your global pay philosophy, and we advise caution. The technical reality is that if you build a single pan-European model and use country as a coefficient – treating it as an objective factor – that variable applies globally across the entire population. This assumes that the relationship between variables is uniform across borders, which in turn can lead to less reliable predictions.

Before consolidating into a single model, we recommend that you take into account job value alignment and local market dynamics. The key evaluative questions to ask yourself are: Does a Grade 3 role in Country A truly represent the same level of responsibility and market value as a Grade 3 in Country B? Are the drivers of pay, such as tenure versus performance, weighted differently by local custom or collective bargaining agreements? Your answers should determine whether a consolidated or country-level approach better reflects the reality of pay in your organization.