Hello,

So Root Mean Squared Error Calibration (RMSEC) is derived from an average of how well (or how close) the data points are to a calibration line using all data points (samples). Root Mean Squared Error Cross Validation (RMSECV) is an internal method to 'test' your model (or calibration model) to make sure it isn't badly skewed by a data point or if the model is over fitted, or if there are any outliers. RMSECV is determined by the validation method chosen. The most popular being a 'leave one out' method. Here a data point is removed when building a 'new' calibration model with a new RMSEC, then the data point is put back but a different data point is removed and new calibration model is made with a new RMSEC . The two RMSECscan be averaged giving us the beginings of RMSECV. This prcess is repeated until every point has a chance at being left out. The RMSECs are collected and averaged giving an overall RMSECV. This should also show you if any there are any outlier or data points that are skewing your calibration line. RMSECV should always be slightly higher than RMSEC.

Both methods use the RMSE calculation in where you get the square root of sample deviations from the mean (squared)over the number of samples.

So we square the difference of the prepared samples to that the model predicted for you samples (how far a data points deviates from the line), divide it by the the number of samples in the set, and get the square root of it.

Hello,

So Root Mean Squared Error Calibration (RMSEC) is derived from an average of how well (or how close) the data points are to a calibration line using all data points (samples). Root Mean Squared Error Cross Validation (RMSECV) is an internal method to 'test' your model (or calibration model) to make sure it isn't badly skewed by a data point or if the model is over fitted, or if there are any outliers. RMSECV is determined by the validation method chosen. The most popular being a 'leave one out' method. Here a data point is removed when building a 'new' calibration model with a new RMSEC, then the data point is put back but a different data point is removed and new calibration model is made with a new RMSEC . The two RMSECscan be averaged giving us the beginings of RMSECV. This prcess is repeated until every point has a chance at being left out. The RMSECs are collected and averaged giving an overall RMSECV. This should also show you if any there are any outlier or data points that are skewing your calibration line. RMSECV should always be slightly higher than RMSEC.

Both methods use the RMSE calculation in where you get the square root of sample deviations from the mean (squared)over the number of samples.

So we square the difference of the prepared samples to that the model predicted for you samples (how far a data points deviates from the line), divide it by the the number of samples in the set, and get the square root of it.

Hi，

Pól MacFhionnghaile has clearlly shown the differences between RMSEC and RMSECV. But, I think you can also find some references to master it deeply.

The following one is for your reference.

Faber N M, Song X H, Hopke P K. Sample-specific standard error of prediction for partial least squares regression[J]. TrAC Trends in Analytical Chemistry, 2003, 22(5): 330-334.

Best,

RMSEC is the calibration error, which measures goodness of fit between your data and the calibration model. RMSECV gives a good estimate on how the model built on the data set at hand performs for unknown cases. Lastly you have the MSEP or RMSEP: prediction error, i.e. measured on real cases and compared to reference values obtained for these.