Using the LSMEANS option of Proc GLM, the multivariate geometric means were determined to be significantly different at unadjusted p < 0.05 across a limited number of a priori contrasts.
For exposure biomarkers whose medians exhibited differences with respect to these characteristics [see Supplemental Materials, Table 1 (http://www.ehponline.org/docs/2006/9488/suppl.pdf)], we then examined whether the multivariate geometric means were significantly different across characteristic levels using the LSMEANS option of the GLM procedure.
In men, age-adjusted geometric means of HbA1c for the quartile categories of GGT (≤ 23, 24 35, 36 58 and ≥ 59 IU/L) were 5.00, 5.08, 5.16 and 5.20 respectively (trend P < 10 15), and multivariate-adjusted geometric means were 4.99, 5.07, 5.16 and 5.23 respectively (trend P < 10 18).
In women, age-adjusted geometric means of HbA1c according to the quartile categories (≤ 15, 16 20, 21 29 and ≥ 30 IU/L) were 5.00, 5.06, 5.13 and 5.17 respectively (trend P < 10 32), and multivariate-adjusted geometric means were 5.00, 5.07, 5.12 and 5.16 respectively (trend P < 10 26).
In a second step, continuous target variables were analyzed with multivariate linear models and adjusted means were given as least square means (LS-means).
Multivariate linear regression (least-squares means) was used to identify independent factors associated with HRQOL in the statistical models; analyses included variables noted from the literature as potentially associated with quality of life (e.g., age, sex, education, and income).
Least-squares means were used for correlation analysis of traits using the Multivariate Platform.
Specifically, we simulated traits that followed a multivariate normal distribution with mean being equal to one-fifth, one-third, one-half, or one-half of the QTL effects estimated from these traits at markers 10 (At1g31580-11), 27 (At2g14260-2), 42 (At3g10720-6), or 55 (andg56360-7) and variance covariance being equal to the residual variance covariance estimated from those traits at the marker.
The group centroids (the multivariate means) for the two groups were 1.594 for level 1 and −0.874 for level 3. Table 6 shows the resulting standardized canonical discriminant function coefficients for each of the categories on the discriminant function (these coefficients are similar to beta weights in a regression analysis).
multivariate emission probability densities, are for uni-, meansvarianceseans, variancovarianceiance matrices referring to the uni- resp.
