**Data Scaling methods**

Under the main menu of *Models*, click *Scaled By*, and select among the four Scaling methods:

Options of data scaling in GGEbiplot are:

1) **No Scaling. **Replicated data are *not *required.

2)** Scaled by Tester Standard Deviation (SD)**: each value is divided by the *standard deviation* of its corresponding tester (column). This will put all testers roughly the same rang of values. Replicated data are *not *required.

3) **Scaled by Heritabiliy-Adjusted Means, **which replaces the "Divided by Tester Means" in earlier versions. Replicated data are required.

4) **Scaled by LSD (5%), **which replaces the "Divided by Tester Standard Error" in earlier versions. Replicated data are required.

5) **Scaled by Heritability-adjusted SD, **which is a newly added model. Replicated data are required.

**In Depth: Which model is better**

My first choice would be #2, "Scaled by SD", because i) It is the best for visualizing relationships among testers, ii) No replicated is needed, and iii) the adquacy of biplot can be inferred by the uniformity of the length of the vectors. If the vectors of all testers are of similar length, then the biplot is adquate. Otherwise the biplot is not adquate in displaying patterns of the short-vectored testers. However, for this model, the vector length of the environments **cannot **be interpreted as representing the discriminating power of the environments.

When the evaluation of testers is the focus, #5 is most preferred, because the length of the testers in this biplot approximates the heritability of the testers (which is an indication of the discriminating power of the test environments), PROVIDED that the biplot is adequate in displaying the data. Model #4 should be similar to model #5. If replicated data is not available, then #1 can be used in place of #5.