Articles
HARVESTING PERIOD AND VARIATIONS IN TUSCAN OLIVE OIL COMPOSITION: A MULTIVARIATE APPROACH.
Article number
356_50
Pages
233 – 238
Language
Abstract
During a 3 years (1988/89–1990/91) trial, 236 samples of tuscan olive oil, representing the typical regional production, and obtained from olive fruits harvested in 3 different periods (before 15 November, after 1 December, and the 2 weeks in between) were collected.
For each sample, free acidity and UV-spectrometry were analized; fatty acids, minor polar compounds and tocopherols contents were also determined so that every observation was defined by 32 analysis variables.
A simple logistic regression model for each variable was calculated within each year.
Variables not showing consistent parameters in every year’s model (same sign in 2 years’ parameters estimates, significant at 88% level, without a likewise significant estimate of opposite sign, in the remaining year) were discarded and multiple logistic regression models (from 12 to 3 selected variables included) were built.
The best simple regression models were related with total and 7 different minor polar compounds, free acidity, palmitic acid, total tocopherols.
A multiple logistic regression model was then selected; it included: palmitic acid, total minor polar compounds, Tyrosol, and tocopherols; the resulting Concordant Pairs were 81% (’88/89), 92% (’89/90), 89% (’90/91), and 83% (3-years model). Cross-validated classification models, based on multivariate linear discriminant analysis, were calculated on the same data set and variables: among the observations from 1st and 3rd harvesting period, 59% (’88/89), 77% (’89/90), 66% (’90/91) were correctly classified, and only 12%, 0% and 9% were respectively swapped; these percents settled to 66% and 8% for the model calculated considernig the 3 years together.
For each sample, free acidity and UV-spectrometry were analized; fatty acids, minor polar compounds and tocopherols contents were also determined so that every observation was defined by 32 analysis variables.
A simple logistic regression model for each variable was calculated within each year.
Variables not showing consistent parameters in every year’s model (same sign in 2 years’ parameters estimates, significant at 88% level, without a likewise significant estimate of opposite sign, in the remaining year) were discarded and multiple logistic regression models (from 12 to 3 selected variables included) were built.
The best simple regression models were related with total and 7 different minor polar compounds, free acidity, palmitic acid, total tocopherols.
A multiple logistic regression model was then selected; it included: palmitic acid, total minor polar compounds, Tyrosol, and tocopherols; the resulting Concordant Pairs were 81% (’88/89), 92% (’89/90), 89% (’90/91), and 83% (3-years model). Cross-validated classification models, based on multivariate linear discriminant analysis, were calculated on the same data set and variables: among the observations from 1st and 3rd harvesting period, 59% (’88/89), 77% (’89/90), 66% (’90/91) were correctly classified, and only 12%, 0% and 9% were respectively swapped; these percents settled to 66% and 8% for the model calculated considernig the 3 years together.
Publication
Authors
S. Alessandri, A. Cimato, A. Mattei, G. Modi
Keywords
fatty acids, phenols, pherols, multivariate models
Online Articles (93)