Systematic error

Uncertainty associated with imperfect knowledge of calibration, efficiency, backgrounds or other persistent model inputs. Repeating the same measurement does not necessarily average it away. Correlated nuisance parameters can propagate it through a likelihood; estimating such an uncertainty is distinct from asserting that the systematic error has a known sign and magnitude.

If a controlled reversal changes the desired signal's sign while leaving an additive offset unchanged, readings $y_{+}=S+b$ and $y_{-}=-S+b$ separate them by half-difference and half-sum. Offset drift, gain changes, hysteresis or a background that also reverses violate the simple cancellation. Interleaving polarities and independent blank measurements test those assumptions; signal reversal does not cancel an arbitrary systematic error.

If a wanted response changes sign on reversing a controlled setting while an offset remains unchanged, the difference of the reversed readings isolates the response. For $V_{\pm }=\pm RI+V_0$, the estimator $(V_{+}-V_{-})/(2I)$ cancels $V_0$. The method requires offset stability between readings. It does not cancel a common current-scale calibration error, and it does not remove even-in-current Joule heating. A useful reversal test specifies which physical effects change sign, then tests the remaining effects independently.