B₀ = intercept coefficient; B₁ = slope coefficient; Err[B_i] = standard error of coefficient ‘B_i’; p -value = probability B_i ≠ 0 (< 0.05). See Tables 1 and 2 for abbreviations.

The linear correlations obtained from application of Model B to IP_R-T⁻¹ demonstrated in the high R² coefficients, low σ_y (≤ 0.2) and high F-ratios (≥ 368), all indicators that the model equations had relatively high goodness of fit for each FAME. Similar to the results from applying Model A , the Model B results for MeC18:2 in Figure 4exhibited some curvature in the data. The residual plot for MeC18:2 (seeFigure S2 in the supporting information) revealed a non-random pattern and fitting the data to a second-order polynomial (data not shown) resulted in slight improvements for R² (0.997), F-ratio (848) and σ_y (0.06). However, the three regression coefficients had p -value ≥ 0.09, indicating they were not significant to the correlation of the data. Given these results, the linear model described by Model B was determined to be sufficient for correlating IP_R-T⁻¹ data for MeC18:2. Therefore, the linear correlation results in Table 4 were employed to calculate shelf-life results at 25 °C for all FAME studied in the present work.

A conformation graph showing IP_B data calculated from application of the Model B equations versus experimental IP_R data is presented Figure 5 . Analogous to the results in Figure 2 , all data points are in close proximity to the dashed line representing IP_B = IP_R. This suggested