D0 = Intercept coefficient; D1 = slope
coefficient; Err[Di] = standard error of coefficient
‘Di’; p -value = probability Di ≠
0 (< 0.05). See Tables 1-3 for abbreviations.
coefficients (D0) that were very low (between −0.3 and
+0.3) and ANOVA indicated they were not significant (p -value
> 0.05) allowing D0 = 0 to be assumed. The
slope coefficients (D1) were significant (p -value
<< 0.05) with values for four of the FAME being
close to unity (0.996-1.05) indicating good confirmation of these model
equations. Similar to the results for Model A , theModel B confirmation equation for MeC18:2 yielded the lowest
D1 coefficient (0.93), suggesting that its model
equation under predicted IPB, with respect to
experimental IPR values. This was confirmed for four of
the six data pairs evaluated for MeC18:2, where IPR> IPB at the same four temperatures (50,
60, 90 and 100 °C [323.15, 333.15, 363.15 and 373.15 K]) as observed
above for its Model A equation.
The MAD and RMSD results calculated from (IPR,
IPB) data pairs are presented in Table 5 .
Similar to the results for Model A , the confirmation results
for Model B exhibited generally low values, MAD = 0.081-0.61
and RMSD = 0.12-0.86. MeC18:2 had the highest MAD and RMSD values,
suggesting its predicted IPB values were influenced by
the lower D1 coefficient in its confirmation equation.
The results for the residuals (IPR −
IPB) calculated from application of the Model Btype equations to the data for the five FAME are presented inFigure 3 . Overlayed with these results are residuals
(IPR − IPA) obtained from theModel A equations (discussed earlier). In general, the
residuals for Model B were ≤ 1.47 for all 26 data pairs with 21
data pairs having residuals ≤ 0.44. The highest residual (1.47) was for
MeC18:1 at T = 100 °C (373.15 K). Summarizing the Model Bresults, the application of this type of model equation to
IPR-T−1 data did an excellent job
predicting IPB values within the range of measurement
temperatures corresponding to the five FAME studied in the present work.
Direct comparison of results from applying Model A andModel B type equations to IPR data for the five
FAME showed that the calculation of IPB values from theModel B equations yielded more accurate results. This is best
demonstrated by the deviation results obtained from confirmation
analyses performed on both models. First, comparing the data plotted inFigures 2 and 5 shows that IPA data
points tended to be more separated from the dashed line than the
IPB results, especially when IPR> 15 h. This observation is supported by the residuals data
overlayed for Models A and B in Figure 3 .
These results demonstrate that residuals were generally larger in the
IPA data than the IPB data. Finally,
comparing MAD and RMSD data in Tables 3 and 5 shows
that values were consistently lower for Model B results for
four of the five FAME studied herein. The exception was observed for
MeC18:1, where the Model A results yielded MAD = 0.42 and RMSD
= 0.62, values that were slightly lower than those for the Model
B results (0.61 and 0.86).