Please consider this experimental result of biological cell counts on differently treated areas:
R2 R3 R4 R5 R6
Replicate1 115421 1582 14989 2997 0
Replicate2 12460 60312 51621 1424 4272
Replicate3 42721 19580 20104 8900 13402
There is a trend in the data, but i have difficulties showing its significance (well, maybe there is none). As you can see, i am working with just three replicates and very high within-sample variance.
I tried linear models with glht() with Tukey Contrasts and correction for high heteroscedasticity:
library(multcomp)
library(sandwich)
library(MASS)
amod <- glm.nb(value ~ variable, data=data4.melt)
amod <- glm(value ~ variable, data=data4.melt, family="poisson")
res <- glht(amod, mcp( variable = "Tukey"), vcov = vcovHC)
for both poisson and negativ binomial distribution assumptions. The result was consistenst between the two models, that only R5 and R2 were different from each other. My PI asked my what the reason was for R6 not behaving like R5. I only could guess its related to one of the replicates being zero and an overall high variance.
Is there a better explanation? And most importantly, did i treat my data right? Is there even a justification for asking such a skewed dataset these questions?
