Statistical Methods for Gender Pay Equity Analysis Every HR Team Can Master

Key points:

• Most companies exceed compliance thresholds: Analysis of customer data shows 82% of companies have gender pay gaps exceeding the 5% threshold that will trigger mandatory EU action in 2026.
• Regression analysis separates discrimination from legitimate differences: Multiple regression analysis controls for factors like experience, location, and performance to isolate gender's true impact on pay.
• Manual statistical analysis creates more problems than it solves: Traditional approaches using Excel or custom coding take weeks to complete, require statistical expertise most HR teams lack, and become outdated before results reach leadership.

Let’s start on a slightly dour note: the gender pay gap still exists, and little is being done to change it.

That’s not something we made up. According to Figures’ analysis of customer data, 82% of companies have gender pay gaps that exceed what EU regulations will allow in 2026.

It’s also not a distant problem you can afford to tackle later. It’s happening right now in your organisation. So, it’s better to get ahead of the curve now, instead of waiting until the new compliance deadlines are banging at your door.

Problem is, where do you start? 

You might use raw compensation data that identifies pay disparities, but it doesn’t tell you why they exist in the first place. Was Peter promoted faster because he’s God's gift to spreadsheets, or because unconscious bias favoured him over equally talented colleagues? 

That’s why you need proper statistical analysis of these numbers. It separates legitimate pay differences – experience levels, performance ratings, role complexity – from potential discrimination that needs immediate attention.

Without proper statistical methods, you’re flying blind. You might waste resources fixing problems that don’t exist while completely missing discrimination that does.

Now, let’s fix that!

Common statistical methods for pay equity analysis

Statistical methods used in pay equity analysis identify and quantify potential pay disparities between demographic groups, particularly gender-based gaps, which are the focus of EU pay transparency regulations.

Your primary goal here is to determine whether pay differences are justifiable and you can do this through various methods. In simple terms, you’ll have to separate legitimate factors – job responsibilities, performance, experience – from potential discrimination patterns.

So, here’s your toolkit for building defensible analysis:

1. Descriptive statistics – the foundation

Before you dive into complex models, you need to understand what you’re working with. Descriptive statistics allow you to get that foothold by showing you the average pay differences and distributions across groups.

You work out those differences by starting with these essential measures:

• Mean and median pay by gender across job categories, levels, and departments.
• Pay ranges and distributions to spot outliers and compression issues.
• Percentile analysis (25th, 50th, 75th percentiles) to understand pay spread.
• Standard deviation to measure variability within groups.

The important thing to remember about descriptive statistics is that while they reveal patterns and structural drivers – like gender imbalances across seniority levels – they can't tell you whether those patterns are justified or discriminatory. 

Think of them as your quick health check that can identify issues before you head in for a deeper investigation. You might be able to identify the raw gaps, but don’t expect them to tell you the whole story.

To show you what we mean, let’s look at a hypothetical.

You’ve crunched the numbers, summarised your findings, and found a 5% average gender pay gap in favour of men.

Now, before you start ringing alarm bells, it’s good to take a step back and ask whether there are any legitimate reasons why this gap exists. Perhaps you have more men working in higher-cost-of-living locations? Or a higher number of men with longer tenures at your business?

Unfortunately, descriptive statistics can only go so far in answering those questions. For a fuller picture, you’ll need a methodology that takes those variables into account. Which brings us nicely onto…

2. Regression analysis – the gold standard

Multiple regression analysis is the statistical method that controls for legitimate pay factors by isolating gender's effect after accounting for level, tenure, and location.

This is where you put your detective cap on. While descriptive statistics show you there might be a problem, regression analysis helps you figure out whether that problem is actually discrimination or just reflects legitimate business factors.

In practice, this means you treat compensation as the outcome you’re trying to explain, then test whether gender affects pay after accounting for everything else that should influence someone’s salary.

The variables that can affect compensation are: 

• Job level and function (senior developers earn more than junior ones).
• Experience and tenure (longer service typically means higher pay).
• Location (London salaries beat Birmingham salaries).
• Performance ratings (top performers should earn more).
• Education and qualifications.

Each of these variables gets its own regression coefficient that shows how much it affects pay. The interpretation depends on how the model is set up:

If we use the log of salary (a common approach in pay equity analysis), coefficients translate into percentage differences. For example, a coefficient of 0.01 for tenure means each additional year at the company is associated with roughly a 1% pay increase.

If we use raw salary values, coefficients translate into absolute amounts (e.g., +€500 per year of tenure).

This is why analysts often prefer log(salary): it makes the results easier to interpret as percentage differences across groups.

Still with us, you brave soul? If yes, let’s continue with p-values.

P-values tell you whether that relationship is statistically significant or just random chance. A low p-value (below 0.05) means the result is statistically significant – the influence you’re seeing is real.

Following those calculations, you’ll have a full picture of what factors influence compensation. Now, taking those numbers, you’ll compare them with a gender coefficient. 

⚠️ A crucial caveat: If you have fewer than 30 employees in any group you're comparing, your results become statistically uncertain. That's why smaller organizations often need alternative methods like median analysis rather than regression.

Let’s take a look at what that might look like:

• A gender coefficient of -0.025 with p-value of 0.02 = Women earn 2.5% less than men, and this difference is statistically significant (not due to chance).
• A gender coefficient of -0.005 with p-value of 0.15 = Women might earn 0.5% less, but this could easily be random variation (often what you'll see with smaller sample sizes).

To reiterate, you need both pieces of information. The coefficient shows the size of the gap, while the p-value tells you whether you can trust that this gap represents a real pattern rather than statistical noise.

When presenting to leadership, you’re saying: “After controlling for legitimate factors like experience and education, women earn X% less than men, and we’re confident this represents a real disparity that needs investigation.”

Courts and regulators expect this methodology for legal defensibility. When you’re presenting findings to leadership or preparing for compliance reviews, regression analysis provides the statistical rigour that stands up to scrutiny.

3. Supporting methods for complete analysis

While regression analysis handles the heavy lifting, sometimes you need additional tools to answer specific questions or work around data limitations.

Cohort analysis

Cohort analysis tracks groups hired or promoted at similar times to spot whether bias happens during hiring, promotion, or both. This method is perfect when you suspect timing-related patterns – like whether your 2022 hiring spree introduced new gaps, or if recent promotions favoured one gender over another.

How it works is that you group employees by when they joined or were promoted, then compare compensation trajectories over time. It’s particularly useful for identifying whether problems stem from initial salary setting (hiring bias) or career progression (promotion bias).

T-tests

T-tests compare average pay between two specific groups. Think marketing versus sales, remote versus office workers, or pre- and post-acquisition employees. This is your go-to when you need a simple, direct comparison that’s easy to explain to leadership.

Unlike regression, which controls for multiple factors simultaneously, t-tests focus on one clear question: “Do these two groups earn different amounts?”

Decomposition analysis

Decomposition analysis breaks your total gap into two clear buckets: the “explainable” portion (legitimate factors) and the “unexplainable” portion (potential discrimination).

The most widely used decomposition technique in pay equity analysis is the Oaxaca-Blinder method. This method shows how the differences in a dependent variable, such as wages, can be separated into explained and unexplained portions by looking at the explanatory variables.

Here's how it works in practice:

Let's say you discover a 10% gender pay gap in your organisation. The Oaxaca-Blinder decomposition might reveal:

• 8% is explainable through differences in experience levels, job roles, and qualifications between groups.
• 2% remains unexplained after accounting for all observable factors – this is the portion that could indicate discrimination.

The differences in the level of observed covariates (the explained component) often account for the majority of the total disparity. However, the unexplained component is often used as a measure for discrimination, but it also subsumes the effects of group differences in unobserved predictors.

This method is brilliant for board presentations because it quantifies exactly how much of your gap comes from structural differences versus potential bias. You might discover that 80% of your gap is explainable through job level differences, leaving only 20% that requires investigation.

⚠️ A word of caution: For this part to be the exact discriminatory part, all the determinants of the outcome should exist in the model, otherwise this part may be over- or under-estimated. That unexplained 20% might include factors you simply haven't measured, like differences in negotiation tactics or unmeasured skills.

These three are the more common supporting methods, but there are also a few more specific ones:

ANOVA (Analysis of Variance)

ANOVA compares pay across multiple groups simultaneously – useful when analysing gender plus race together, or comparing compensation across five different departments at once.

While t-tests handle two-group comparisons, ANOVA manages complex multi-group analysis in a single test, telling you if any groups differ significantly from the others.

Rank sum tests

These tests handle unusual pay distributions like bonuses or commission-heavy roles, where averages can be deeply misleading.

If your sales team has three people earning £200K and twelve earning £45K, the average doesn’t really represent anyone’s actual experience. Rank sum tests focus on the distribution pattern rather than averages, giving you more reliable insights when pay varies dramatically within groups.

Mann-Whitney U test and median-split analysis

These two methods become your reliable option for small teams under 30 people, where regression lacks statistical power, or when dealing with heavily skewed pay distributions. 

The Mann-Whitney U test is particularly useful when your compensation data doesn't follow a normal distribution, which is often the case with salaries and bonuses. Instead of comparing raw salary values, you rank all employees by salary and compare whether one group consistently ranks higher than the other.

Why use this instead of t-tests? If your sales team has three people earning €200K and twelve earning €45K, the average doesn't represent anyone's actual experience. Mann-Whitney focuses on the distribution pattern rather than misleading averages.

Median-split analysis offers another simple approach. Compare who's above and below your median salary – if 80% above the median are men while 80% below are women, that's a clear red flag, no complex statistics required.

🤔 Which one to choose? Most organisations start with descriptive statistics, run regression analysis, and then use these supporting methods to investigate specific concerns or explain findings to different stakeholders.

Two paths to implementation

Now that you understand the statistical methods, you face a choice: build your analysis from scratch or use tools that automate the complexity. 

Path 1: Manual statistical analysis

This is a fully DIY approach that requires a lot of expertise and moving parts. That’s because you’ll be using human statisticians to pore over every number, minimising the intervention of automated software as much as possible.

Manual statistical analysis usually looks like this:

1. Export data to R, Python, or Excel.
2. Write regression code from scratch.
3. Run supplementary analyses separately.
4. Compile results into reports manually.

Keep in mind, this is a heavily simplified version of events – kinda like trying to explain Lord of the Rings as some dudes going on a trip. There are usually teams of employees involved in this type of analysis, including HR professionals, coders, admin professionals, and statisticians working together simultaneously.

Granted, there are some positives to this method:

• Human-driven analysis: Given the complexity of pay equity analysis, having analysts on your side who can make their own interpretations is a significant advantage. 
• Deeper understanding: Unless you hire a third-party analyst, it’s going to be your employees working on the data. This means they’ll have a better understanding of the culture of your company and what variables actually matter to your business. 
• Good option for fledgling businesses: As long as you’re not working with a huge number of employees, manual data analysis can actually be more cost-effective for your business.

Okay, we’ve looked at the advantages; now it’s time to look at the downsides:

• Requires statistical software expertise you might not have on your team.
• Time-intensive and error-prone because people make mistakes.
• Difficult to update regularly as your workforce changes and your business grows.

So what’s the alternative?

Path 2: Automated with Figures

The alternative is letting automated software handle the statistical complexity while you focus on interpreting and acting on the results.

You won’t have to look far. Figures has been built from the ground up to make compensation management less stressful. That includes removing the technical barriers to pay equity analysis.

Here’s what automated analysis looks like with Figures:

• Connect your existing HRIS system (Personio, BambooHR, or 30+ integrations) and let Figures handle the rest. The platform runs regression analysis automatically on clean, normalised data.
• Advanced filtering capabilities let you drill down by role, location, seniority, or custom groups – no more one-size-fits-all analysis that misses departmental nuances.
• Live pay gap tracking updates your results in real-time as people join, leave, or get promoted.

The best thing? You won’t need a group of analysts to crunch the numbers, as Figures does it all for you.

*90s infomercial voice*: But wait, there’s more!

• No statistical expertise required. The platform handles log-linear regression, p-values, and confidence intervals automatically – you just need to understand what the results mean.
• Both adjusted and unadjusted gaps. Compare raw pay differences with context-aware analysis that accounts for experience, tenure, and role complexity.
• Custom stakeholder reports generate tailored insights for different audiences – board-level summaries for executives, detailed breakdowns for HR teams.
• Built-in compliance features handle EU Pay Transparency Directive requirements automatically, with report templates ready for submission to authorities.
• Glass ceiling detection and pay quartile analysis uncover structural trends that basic regression might miss.

💡 Most HR teams discover that understanding statistical methods combined with automated implementation gives them the best outcome – statistical rigour without requiring a team of analysts or months of preparation.

Transform your pay equity analysis with Figures

You now understand the statistical methods that separate legitimate pay differences from potential discrimination. That knowledge puts you ahead of most HR teams, who dive into analysis without understanding what they’re actually measuring.

Understanding these methods helps you ask the right questions and interpret results with confidence. Figures ensures you’re using the right statistical approaches, applying them correctly, and staying compliant — without needing a statistics degree or a team of analysts.

Think about it: would you rather spend weeks wrestling with regression code and data cleaning, or focus on what those results mean for your organisation and how to address any gaps you find?

Figures handles the statistical complexity automatically, updates continuously as your workforce changes, and packages results for different stakeholders. What used to be a months-long statistical project becomes instant, current insights that you can trust and defend.

Book a free demo and see how data-driven pay equity analysis can transform your approach to fair compensation.

{{ cta }}