Variance Indicator
Statistical evaluation measured against regression model outputs isolates the unmodeled dispersion of sensor calibration readings. Calculating the residual variance reveals how effectively a chosen mathematical function describes physical sensor behavior. This metric serves as a foundational input when calculating fit quality and estimating prediction intervals across sensor operational ranges.
Scatter Analysis
Computation involves summing the squared differences between observed sensor outputs and calculated model predictions, divided by the available degrees of freedom. A high value indicates that the chosen mathematical model fails to capture underlying physical sensor non-linearities, or that the measurement system suffers from excessive random noise. Examining residual plots across the calibration range reveals whether residual errors are randomly distributed or follow systematic patterns indicative of missing polynomial terms or hysteresis effects.
Model Adequacy
Selecting the appropriate polynomial order requires monitoring changes in unexplained variance as terms are added. If adding a higher-order term yields a negligible reduction in residual variance, the lower-order model is retained to avoid overfitting calibration data.
Uncertainty Budgeting
Calibration certificates incorporate residual scatter into the overall expanded uncertainty budget for digitized sensor instruments. Qualification testing enforces upper thresholds on residual variance to ensure that interpolation error does not compromise specified sensor accuracy.