I have data from a plant disease screen in a greenhouse. 145 plants in a greenhouse were inoculated with bacteria over the course of three days, and the amount of bacteria in the plant assessed at a later date, to determine to what extent each plant is resistant to the bacteria. the data appears quite clearly to be weibull distributed.

fit.weibull = fitdist(AOD_raw_data$`Average of all reps`[2:145], "weibull")

enter image description here

There are two important independent variables in this data, the day on which the plants were inoculated, and the "provenance" (the place where the plants were sourced from). Plants were sourced from three different provenances, and the number of plants from each one does differ.

I would like to use a GLM to analyse my data, to account for the confounding variation of the day on which they were inoculated, and tell me to what extent provenance has a significant effect on disease. I believe i have successfully done this using the GAMLSS package.

my_model2 <- gamlss(AOD_raw_data$`Average of all reps`[2:145] ~ AOD_raw_data$Provenance[2:145] + AOD_raw_data$Batch[2:145], data = AOD_raw_data, family = WEI())

the Q-Q plot and hoslem goodness of fit test appears to indicate that my GLM as it is is good:

> hoslem.test(AOD_raw_data$`Average of all reps`[2:145], fitted(my_model2))

Hosmer and Lemeshow goodness of fit (GOF) test
`data:  AOD_raw_data$Average of all reps[2:145], fitted(my_model2)
  X-squared = -42.139, df = 8, p-value = 1`

enter image description here

However, the quantile against index/fitted values plots give me pause, i've never used plots like this before and do not really know how to interpret them, so i can't really tell if i'm looking at unacceptable heteroscedasticity. I'd be very gratefull if someone could give me a second opinion.

Kind Regards, Tom



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.