Understand Consistency Labels
Consistency Labels tell you how reliably a benchmark represents the market. Every benchmark in Market Data includes a Consistency Label based on the variability of the underlying compensation data, not just the sample size.
Two benchmarks can have the same median value and the same number of data points but very different distributions. If data points cluster tightly around the median, the benchmark is a reliable signal. If data points are spread widely above and below the median, the benchmark is less predictable. Consistency Labels capture this difference so you can factor it into your compensation decisions.
Consistency Labels replaced the earlier Confidence Labels system, expanding coverage to all compensation types and adding more granularity.
Consistency Label levels
Each benchmark receives one of six Consistency Labels based on its confidence interval (the margin of error around the benchmark value). Labels apply to all compensation types: base salary, total cash, variable pay, and equity.
| Label | Cash confidence interval | Equity confidence interval |
|---|---|---|
| Excellent | 0-5% | 0-5% |
| Very High | 5-10% | 5-10% |
| High | 10-20% | 10-20% |
| Medium | 20-25%+ | 20-50% |
| Low | 25%+ | 50%+ |
| Insufficient Data | Benchmark does not meet accuracy standards | Benchmark does not meet accuracy standards |
Equity benchmarks use a wider confidence interval range than cash benchmarks. Across the market, there is significantly more variation in how companies compensate with equity than with cash. It typically takes around 10 times the number of data points in an equity benchmark to reach the same consistency level as a base salary benchmark. The wider range accounts for this difference so that equity benchmarks are not disproportionately labeled as Low or Insufficient Data.
When a benchmark is labeled Insufficient Data, Pave does not display percentile values for that combination of filters.
Where to find Consistency Labels
In the table view: Each benchmark displays its Consistency Label. Hover over the label to see the confidence interval, sample size, company count, and whether the benchmark is raw or calculated.
In exports: Market Data Pro reports include columns for the Consistency Label, confidence interval, and sample size for each benchmark.
How to use Consistency Labels
Consistency Labels help you decide how much weight to give a benchmark when making compensation decisions.
- A benchmark labeled Excellent or Very High can be used at face value with a high degree of confidence
- A benchmark labeled Medium or Low can still inform your decisions, but the underlying data has more variability. You may want to apply additional judgment based on your compensation philosophy, existing pay ranges, and company stage
- Consistency Labels are not a replacement for company-specific range spreads on compensation bands. They describe market variability, not your internal pay strategy
FAQ
Why does a benchmark have Low consistency when the sample size is large?
A large sample size does not guarantee high consistency. When you view data with broad filters (for example, US-All locations and All Companies), the sample includes employees across very different markets. Compensation for the same role can vary significantly between San Francisco and Chicago, or between a seed-stage startup and a large public company. This variation widens the confidence interval even with many data points.
To improve consistency, apply location and company stage filters. As you narrow the profile of employees in the benchmark, sample size decreases but consistency often increases because the remaining data points are more comparable.
How do Consistency Labels apply to calculated benchmarks?
Consistency Labels apply to both raw and calculated benchmarks. The label reflects the variability of the underlying data regardless of whether the benchmark was computed from raw data or computed by Pave's models. A calculated benchmark labeled Low means there is high variation in how the market compensates that role, not that the calculation is unreliable.
For more on how calculated benchmarks work, see Understand calculated benchmarks.