df = Degrees of freedom; A0 = intercept coefficient;
A1 = slope coefficient; Err[Ai] =
standard error of coefficient ‘Ai’; p -value =
probability Ai ≠ 0 (< 0.05); R² = adjusted
correlation coefficient; σy = standard error of the
y-estimate; F-ratio = variance ratio (model/residuals). SeeTable 1 for abbreviations.
errors of the y-estimate (σy ≤ 0.2) for calculating
ln(IPR) directly from T data. Results from ANOVA
demonstrated high variance (F-) ratios (model/residuals), indicating
that the goodness
of fit of the model equations was high. However, examination of the data
in Figure 1 revealed small degrees of curvature in the data for
MeC18:2. Figure S1 is a residual plot in the supporting
information (residuals = IPR – IPA[IP calculated from the corresponding Model A equation])
that revealed a non-random pattern, suggesting that a linear model may
be insufficient for correlating IPR to T. When the data
for MeC18:2 were fitted to a second-order polynomial by regression
analysis, the results showed that R² increased from 0.985 to 0.997 and
F-ratio from 328 to 850 relative to the linear model summarized inTable 2 . All three regression coefficients had p -value ≤
0.02 and σy decreased from 0.1 to 0.06. Nevertheless, it
was desired to maintain consistency in the application of Model
A to correlate IPR-T data for the five FAME studied
herein. Therefore, the linear model equations summarized inTable 2 were employed to calculate shelf-life data at 25 °C for
all FAME, including MeC18:2.
Shown in Figure 2 is a confirmation graph of
IPA values versus experimental
IPR values. The dashed line drawn through the data
represents the IPA = IPR line. All data
points appear in