AES Consulting meeting on 10 Jan 2017
Randomly (?) allocated the 140 genotypes to 6 pots each. With 5 plants per pot. Response is 0,0.5,1,1.5,2,2.5,3 (why not multiply by 2?) measuring the how diseased that plant is.
The client is currently fitting
- normality of residuals
- constant variance
Both of these seem easily violated. The residuals are not going to be normally distributed due to the response only taking on 7 values. For genotypes that have low (high) disease, the values will be near zero (3) and there will not be much variability. In contrast, middling disease will result in a values (possibly) spanning the range from 0-3 and therefore have higher variance.
Do a permutation test. Ordinal regression model?
But we aren’t really sure this is at all what you want.