  Confidence and prediction bands, error bars and extrapolation.     Extrapolation.  Confidence and prediction bands and extrapolation. 
|
The NIST reference dataset contains 27 datasets for validating nonlinear regression analysis software. The datasets were distributed by level of difficulty: 8 lower, 11 average and 8 higher.
Here, will be presented the LAB Fit results for the higher level. For each dataset a function is indicated and the initial values are suggested (two groups).
The two groups were tested by LAB Fit. For most of the datasets, the uncertainty could be given with 1 or 2 significant digits. However, the LAB Fit results were
presented with 3 digits for the uncertainty. The value 1 was assumed for the "POWER" parameter. If there was not success, another value was specified. The default value for the tolerance (1.0E-06) was assumed. Although LAB Fit calculates the covariance matrix (for error propagation purposes, confidence and prediction bands), the results won't be presented because the SRD / NIST doesn't certify it.
BENNETT5 - Dataset, initial and certified values y = a * ( b + x ) ** ( -1 / c ) a = -2524 ± 297 b = 46.74 ± 1.25 c = 0.9322 ± 0.0203 Obs: the results agree with the validated values by the SRD of the NIST (initial values suggested: primary [POWER = 300] and secondary [POWER = 300]).
|