k | σ_{j} ≠ const 1 | σ_{j} = const 2 | |
---|---|---|---|
MG | LOD(H_{2k}/H_{1k}) (df) | 2.954 (8) | 3.403 (8) |
P | 0.095 | 0.015 | |
MA | LOD(H_{2k}/H_{1k}) (df) | 1.983 (2) | 1.817 (2) |
P | 0.010 | 0.015 | |
MG | LOD(H_{21}/H_{22}) (df) | 21.64 (8) | |
P | 0.000 | ||
MA | LOD(H_{21}/H_{22}) (df) | 20.25 (8) | |
P | 0.000 | ||
(df) | 0.971 (6) | 1.586 (6) | |
P | 0.60 | 0.29 |
We used the index k to denote two types of models corresponding to equal (k = 1) vs. nonequal (k = 2) residual variances across environments. Therefore, H_{11} and H_{12} hypotheses here assume the presence of a QTL effect with constant and varying residual variances, respectively. Correspondingly, H_{21} and H_{22} assume the presence of a QTL with varying effect and constant and varying residual variances, respectively. To test whether the two models, MA or MG, differ significantly, provided H_{2} {a_{j} ≠ const} is true, both situations, i.e., with and , were considered using LOD score, (see the last section of the table).