I have some data I am trying to analyze but was hoping to get some stats help with it. I'm not sure what the best way to analyze it is.I'm interested in seeing if there is any effect of different dietary treatments on bacterial colonization and shedding in experimentally inoculated chickens with a bacteria. I have 4 diets (A B C & D) in 5 replicates each. After inoculation, bacterial colonization was weekly monitorized by taking faecal samples and bacterial shedding by cloacal swabs during 6 weeks. Samples were submitted to PCR analysis and bacterial load (CFU/g or CFU/cloacal swabs) was extrapolated from Ct values of PCR. My questions are:
- I realize that bacterial load should be considered count variable (values always >0). Is it OK to round the decimals to make it discrete? Values ranged between 0 – 10.000.000.000 CFU/g so I would transform them in to log.
- Which statistical test better suit this experiment to detect any difference in bacterial load between treatments? Data are not normally distributed. Dependant variable: log CFU/g; Independent variables: TREATMENT (A B C D), TIME (1,2,3,4,5,6 weeks of sampling). Another possible independent variable: INOCULATION (Y:those inoculated directly and N:those infected by horizontal transmission).
Until now I was testing the data with non parametric Kruskal-Wallis and only found significant differences regarding time sampling but not with the treatments or inoculation and before making any erroneous conclusions I would like to search better alternatives.I am considering the Non linear mixed model with repeated measures. I am new to mixed models and I am not sure how to include the variables in the model. Any help? Perhaps Fix effect: treatment. Random effect: animal. I read about using linear, quadratic terms for covariate of time. What is it for?Thank you very much; any help would be much appreciated.