I have a data set containing root, shoot and seedling growth measurements from 5 treatment groups (at different concentrations, 3 replicates) including control. There were no separate control groups for each treatment group. I have the following doubts (The code for analysis is given below, the data set is attached):
1) SHOULD I PERFORM ONE-WAY ANOVA AND POST HOC ON THE SAME TRANSFORMED DATA SET (IF FITTED WELL)
2) CAN I BELIEVE WHAT I HAVE DONE IN TRIAL 2 IS CORRECT?
3) CAN WE JUSTIFY RUNNING SUCH ANOVA WITH DATA SET CONTAINING ONLY ONE CONTROL FOR ALL THE TREATMENT GROUPS WITH MULTIPLE LEVELS (WITH DIFFERENT CONCENTRATIONS FOR EACH TREATMENT)?
4) CAN I CONSIDER VARIABLES ("ROOT", "SHOOT", "SEEDLING") A THIRD FACTOR (AS THEY REPRESENT DIFFERENT RESPONSES RECEIVING SAME TREATMENTS)?
TRIAL 1)
Two-way anova between concentration and treatment
md2$concm1
shapiro.test(resid(m1)) ## assumption violated
However I continued ANOVA on log-transformed data as follows:
aov(log(length+1)~conc*treat,data=md2)->m1
shapiro.test(resid(m1))## assumption violated.
TRIAL 2)
Then I split the data and repeated one-way ANOVA with the subset of data with control and each treatment separately
as follows:
rbind(md2[md2$treat=="control"&md2$variable=="root",],md2[md2$treat=="t1"&md2$variable=="root",])->t1
rbind(md2[md2$treat=="control"&md2$variable=="root",],md2[md2$treat=="t2"&md2$variable=="root",])->t2
rbind(md2[md2$treat=="control"&md2$variable=="root",],md2[md2$treat=="t3"&md2$variable=="root",])->t3
rbind(md2[md2$treat=="control"&md2$variable=="root",],md2[md2$treat=="t4"&md2$variable=="root",])->t4
md2$concm1
shapiro.test(resid(m1)) ## assumption not violated