Effective Degrees of Freedom

Effective degrees of freedom (ν_eff) quantify how reliable the combined uncertainty estimate is, considering the individual degrees of freedom of each contributing component. The concept is necessary because Type A components (based on limited samples) have finite degrees of freedom, while Type B components (based on complete specifications or large datasets) may have effectively infinite degrees of freedom. The combined degrees of freedom determine whether the standard coverage factor of k=2 provides true 95% coverage.

The Welch-Satterthwaite formula calculates effective degrees of freedom from the individual standard uncertainties and their degrees of freedom. Type A components with n observations have (n-1) degrees of freedom. Type B components are often assigned large (infinite) degrees of freedom when based on reliable information, or finite degrees of freedom when the estimate is less certain. The resulting ν_eff is used with the t-distribution to find the appropriate coverage factor for the desired confidence level.

For calibration uncertainty budgets, effective degrees of freedom matter most when Type A components with few observations are dominant contributors. If ν_eff is large (typically >30), k=2 provides approximately 95% coverage and no special treatment is needed. If ν_eff is small (e.g., less than 10), a larger coverage factor from the t-distribution is needed to achieve 95% coverage. This situation arises when calibration repeatability is evaluated from only a few measurements and dominates the uncertainty budget.