Selecting optimum model weights by comparing sum relative errors, this is
relerr(), of weighted calibration() models as
suggested by Almeida et al. (2002).
Usage
weight_select(x, weights, ...)
# S3 method for class 'calibration'
weight_select(x, weights = NULL, ...)
relerr(x)
# S3 method for class 'calibration'
relerr(x)Arguments
- x
an object of class '
calibration'.- weights
a list of weights to be added to the default weights to be checked. These are
1/concentration^0.5,1/concentration^1,1/concentration^2,1/signal^0.5,1/signal^1, and1/signal^2.- ...
further arguments passed to
calibration().
Value
weight_select() produces a matrix with differently weighted
'calibration' models ordered by sum relative errors.
relerr() compares the nominal concentrations with those predicted by
the calibration model.
Details
If calibration data is not homoscedastic, a weighted least squares linear
calibration model may be applied to counteract the influence of high
concentrations on the regression model. This, in turn, typically improves the
accuracy at the lower end of the calibration curve (Almeida et al., 2002).
weight_select uses sum relative errors (relerr) to
find the best weight as suggested by Almeida et al. (2002). Predefined
weights include 1/concentration^0.5, 1/concentration^1,
1/concentration^2, 1/signal^0.5, 1/signal^1, and
1/signal^2 (see calibration() for details).
References
Almeida, A.M.D., Castel-Branco, M.M., & Falcao, A.C. (2002). Linear regression for calibration lines revisited: weighting schemes for bioanalytical methods. Journal of Chromatography B, 774(2), 215-222. doi:10.1016/S1570-0232(02)00244-1 .
See also
Other calibration:
calibration(),
din32645,
icp,
matrix_effect(),
neitzel2003,
phenolics
Examples
data(din32645)
din <- calibration(Area ~ Conc, data = din32645)
weight_select(din)
#> Sum relative error Adj. R-squared
#> 1/Conc^2.0 0.6109420 0.9841021
#> 1/Conc^1.0 0.6928994 0.9863181
#> 1/Conc^0.5 0.7463664 0.9854612
#> 1/Area^2.0 0.7545672 0.9839309
#> 1/Area^1.0 0.7795665 0.9842662
#> 1/Area^0.5 0.7896535 0.9838375
#> din 0.7977949 0.9829773
#> NULL 0.7977949 0.9829773
relerr(din)
#> [1] 0.19879314 0.07756144 -0.15397708 -0.06895849 0.06692175 0.04503979
#> [7] -0.04717936 -0.03639084 0.07526913 -0.02770383