Aunir Glossary - Bias

Bias, or mean prediction error, is the average difference between reference chemistry results and NIR predicted results for a set of samples.

Bias can also be seen as a consistent overestimation or underestimation of predicted values by NIR. If bias is found to be large then a bias correction might be justified. This is normally determined by a student t test*.

Differences in NIR instruments, reference chemistry methods and sample sets are amongst the most common reasons why we might observe a bias in our NIR predictions. It is important to remember to remove outlier results and to use a representative test sample set before bias calculations.

* For a sufficiently large sample size (n>15) if t  is more than 2 then bias is considered significant.

 

Reference:

Fearn, Tom. "Bias." NIR News 9.6 (1998): 4-0.