The compounds identified from their mass spectra were compared between experimental groups by multivariate statistical analyses. Principal component analysis was used to define the main peaks to be compared. Discriminant stepwise analysis was used to contrast the chemical composition of different experimental groups. Peaks occurring in fewer than half of individualswere excluded from the statistical analyses. Any peaks shown by principal component analysis to be responsible for a contribution of less than 5% for the first two extracted factors were not used for the discriminant analysis.
Remaining peaks chosen for discriminant analyses had their relative concentration readjusted to 100%. To avoid complications in analysing compositional data, the eakswere transformed according to the formula: Z ¼ ln(Ap/g(Ap)), in which Ap is the peak area, g(Ap) is the geometric mean of the peak in each worker’s group and Z is the transformed peak area