This experiment tested the null hypothesis that 2 methods for measuring the length of an animal are not significantly different from each other.

Subjects (people) were assigned 1 of the methods with which to measure 5 animal replicas (not live animals) i.e. each subject produced 5 measurements (a repeated measure). Each measurement was compared to the ‘true length’ of the animal replica, to calculate % difference (i.e. how accurate the method is).

• Response variable: % difference

• Random factors: subjectID and animalID

• Fixed factor: method

Response data were arcsine square-root transformed (to improve the residuals vs fitted values plot). Some responses were zero, meaning they were exactly the same as the ‘true length’. We used the nlme package in R to analyse the data, as follows:

model.1 <- lme(percentdifference~method*animalID, random=list(~1 | subject, ~1 | animalID), weights=varFixed(~(percentdifference+1)),data=data)

anova(model.1)

Is this the correct approach? Is there something else that should be considered with this kind of data?

Thank you very much.