Pay Equity Analysis: A Practical How-To

Pay Equity Analysis: A Practical How-To

A pay equity analysis answers one deceptively simple question: are people doing comparable work paid comparably, regardless of gender, race, or any other protected characteristic? The headline numbers tell you the question still matters. In Payscale's 2026 Gender Pay Gap Report, women earned $0.82 for every dollar men earned on a raw, uncontrolled basis. Once you control for job, level, and experience, the gap narrows to $0.99 — but that last penny is exactly what a pay equity analysis is built to find. This guide is the hands-on, statistical version of the work: how to build clean comparator groups, choose the right controls, run a regression, and read the output without overreacting or explaining away a real problem. If you want the full program around it, start with the audit; here, we go deep on the analysis itself.

TL;DR — Key takeaways

• A pay equity analysis isolates pay differences after controlling for legitimate factors like job, level, tenure, and location — that residual is what you act on.
• The uncontrolled gap ($0.82 in 2026) measures representation and opportunity; the controlled gap ($0.99) measures pay for comparable work. You need both.
• Multiple regression is the standard method: pay is the outcome, legitimate factors are controls, and the protected-class coefficient is the number that matters.
• Regulators like the EEOC and OFCCP scrutinize statistically significant gaps; roughly two standard deviations is the common flag.
• Clean comparator groups beat clever statistics. Garbage job structure in, garbage analysis out.

Controlled vs. uncontrolled: know which gap you're measuring

Before you touch the data, get clear on the two numbers you can produce, because they answer different questions.

The uncontrolled gap compares average pay between groups with no adjustments. When Payscale reports women earn $0.82 per dollar, that's uncontrolled. It captures everything at once: representation, who holds senior roles, who works in higher-paying functions, and any direct pay differences. It's a powerful signal about opportunity and structure, but it does not isolate discrimination.

The controlled gap compares people doing comparable work after accounting for legitimate, job-related factors. The $0.99 figure is controlled. It strips out the fact that, on average, men hold more of the higher-paid roles, and asks the narrower question: for the same job at the same level with similar experience, is pay equitable?

You need both. The controlled gap tells you whether your pay decisions are fair. The uncontrolled gap tells you whether your structure is fair — whether women and underrepresented groups are stuck below the roles where the money is. Fixing the second usually matters more in the long run, but the first is where legal and remediation risk concentrates.

Step 1: Define comparator groups

This is where most analyses succeed or fail. A regression can only compare people it treats as similar, so you have to define what "similar work" means before you run a single calculation.

The cleanest approach is to group jobs by a consistent, defensible job structure — ideally the output of a point-factor evaluation, where every role is scored against the same compensable factors like skill, effort, responsibility, and working conditions. That gives you internal grades that are comparable across departments, so a senior analyst in finance sits at the same level as a senior analyst in operations if the work is genuinely comparable. If your levels were set by gut feel or inherited titles, your "comparable" groups aren't, and your results won't hold up.

Aim for comparator groups large enough to analyze. A group of three people can't produce a meaningful statistical result. Many teams set a floor of around 30 employees per analysis group; smaller groups get rolled up to a broader level or reviewed manually.

Step 2: Assemble and clean the data

Pull current base pay (and decide separately whether to include bonus, equity, and allowances — analyze them as distinct outcomes, not a blended number). Then gather the legitimate factors you'll control for:

• Job grade or level from your evaluation system
• Job family or function
• Tenure and relevant prior experience
• Location or geographic pay zone
• Performance rating, if it genuinely drives pay
• Full-time/part-time status and FTE

Then the protected characteristics: gender, race/ethnicity, and age where lawful and available. Clean ruthlessly. Standardize location codes, fix duplicate employee records, resolve missing tenure, and reconcile job codes that mean the same thing. A week of data cleanup saves you from a month of defending a flawed result.

Step 3: Run the regression

Multiple regression is the workhorse of pay equity analysis, and the logic is straightforward even if the math isn't.

You build a model where pay is the dependent variable (the thing you're explaining) and your legitimate factors are the control variables. You add the protected-class variable — say, gender — last. The model then estimates how much of the pay difference between men and women remains after the legitimate factors are accounted for. That remaining difference is the coefficient on the gender variable, and it's the number your whole analysis hinges on.

Two outputs matter. The coefficient tells you the size and direction of the adjusted gap — for example, "women are paid 3.1% less than comparable men." The statistical significance (often expressed as a p-value or a standard-deviation measure) tells you how confident you can be that the gap is real and not random noise. Regulators commonly treat a gap of roughly two standard deviations as the threshold for concern, and adjusted gaps that cross that line are where investigations and remediation focus.

If you have enough data, run the regression within job families or levels rather than pooling the whole company into one model. A single company-wide regression can hide a serious problem in one function behind clean numbers everywhere else.

Doing this by hand in a spreadsheet is possible but error-prone at scale. A dedicated pay equity analysis tool — or a comp platform that ties evaluation, leveling, and analysis together — removes the manual stitching that introduces mistakes.

Not sure your job levels are clean enough to analyze? A consistent point-factor structure is the foundation every defensible pay equity analysis sits on. See how PointFactors scores and levels jobs so your comparator groups actually hold together.

Step 4: Interpret results without overreacting

A statistically significant gap is a flag, not a verdict. Before you conclude discrimination, work through the obvious explanations.

Check whether a legitimate factor is missing from your model. If your highest-paid engineers all happen to hold a certification you didn't capture, the model may attribute their pay to gender instead. Check for a handful of outliers — a few recent executive hires can distort a small group. And check the direction: gaps can run either way, and a model will surface both.

But don't talk yourself out of a real result. The classic failure mode is adding control after control until the gap disappears, then declaring victory. If you're controlling for something that is itself a product of bias — like a performance rating system that consistently scores one group lower — you're laundering the problem, not solving it. Controls have to be genuinely job-related and applied consistently.

Step 5: Remediate and re-run

Where you find unexplained, significant gaps, model the cost to close them and prioritize. Most organizations remediate by bringing underpaid employees up — never by cutting anyone's pay — and sequence by the size and significance of each gap.

Then fix the upstream cause so the gap doesn't reopen next year. That usually means tightening how starting pay is set, removing salary history from offers, and anchoring every role to a defensible level. Pay equity isn't a one-time cleanup; re-run the analysis on a regular cadence and at key moments like merit cycles and reorganizations. For the full program — scoping, governance, and how to keep the work defensible — see our pay equity audit guide and the broader pay equity playbook.

A note on regulation and data

You don't run this analysis in a vacuum. Employers with 100 or more employees, and many federal contractors with 50 or more, already report workforce demographic data to the EEOC through the EEO-1 Component 1 collection, under the authority of Title VII. On top of that, the Equal Pay Act and Title VII prohibit compensation discrimination directly, and several states layer on their own pay-data reporting and pay-transparency requirements. Running your own analysis first means you understand your numbers before anyone else asks for them — and you've had the chance to fix what you found.

Frequently asked questions

What's the difference between a pay equity analysis and a pay equity audit? The terms overlap, but a pay equity analysis usually refers to the statistical core — building comparator groups and running the regression. A pay equity audit is the broader program around it: scoping, data governance, remediation, and legal protection. The analysis is a step inside the audit.

Is pay equity the same as internal equity? No. Internal equity is about paying jobs consistently relative to their value inside your organization. Pay equity is the regulatory question of whether protected groups are paid fairly for comparable work. A strong internal structure supports pay equity, but they answer different questions.

Do I need a statistician to run a pay equity analysis? Not always, but you need either statistical expertise or software that handles the regression correctly. For a large or legally sensitive analysis, involve a qualified analyst or counsel — especially when results may carry remediation cost or litigation exposure.

How big a gap should I worry about? There's no universal number, but a common rule of thumb is statistical significance around two standard deviations. Regulators like the EEOC and OFCCP scrutinize significant adjusted gaps, so a result that crosses that threshold warrants investigation and likely remediation.

Why does the gap shrink so much when I add controls? Because controls account for legitimate drivers of pay like level and experience. The uncontrolled gap ($0.82 in 2026) reflects representation; the controlled gap ($0.99) reflects pay for comparable work. A shrinking gap is expected — just make sure your controls are genuinely job-related and not masking bias.

How often should I run a pay equity analysis? At least annually, plus after major events like merit cycles, acquisitions, or reorganizations. Pay drifts as people are hired, promoted, and adjusted, so a one-time analysis goes stale fast.

Should I include bonus and equity, or just base pay? Analyze them separately. Base pay, bonus, and equity are set through different mechanisms and can show different patterns. Blending them into one number hides where the real issue lives.

Pay equity analysis isn't about chasing a perfect score — it's about knowing your numbers, fixing what you find, and being able to show your work. The organizations that do this well treat it as a routine, structured part of running compensation, not a fire drill. Book a PointFactors demo to see how a consistent point-factor foundation makes every pay equity analysis faster, cleaner, and easier to defend.

Justin Hampton is the founder and CEO of PointFactors.