Hi!

I'm a PhD student and struggling with the analysis of my data. Here's a quick introduction to it:

Aim of the study is to do a physiological validation of faecal cortisol metabolites as a measure of adrenocortical activity with foxes. The most widely used experiment of physiological validation is ACTH challenge test where the HPA axis is stimulated in a pharmacologically standardized way. In the test the animals are treated with ACTH injection, and in the cases of valid methods there will be a significant increase in FCM levels after species-specific delay time. The baseline levels will be determined from the samples collected before ACTH adiministration from same animals so each animal can be used as its own control

So I did the ACTH challenge test for foxes and the data looks like expexted but there is still some issues with data analysis I haven't found solution and I hope I can find answer from here.

I am using SPSS 25 and linear mixed model.There is five factors in the model:

A. metabolite concentration (ng/g)

B. treatment (1=control before injection 2=treatment after injection)

C.Time period (hours from the injection the sample is collected, faeces sample collection were done for every three hours for the next 26h)

D.gender of the animal (1=male 2=female)

E. Animal ID

Our hypothesis is to confirm that the metabolite is affected by treatment and Time period (treatment) while there should be a clear peak in metabolite concentrations after the delay time. And there is.The faeces samples are collected in two 24 hour periods, at same time of the day so that the concentrations of each time periods are comparable between control and treatment. The data consists of 20 animals and their faeces samples collected, there is 9 time periods for sample collections per both treatments so in perfect data there should be 20x9+20x9=360 faeces samples altogether. However, you cannot make the animals to defecate regularly in every three hours for two days so our data has a lot of missing samples and our fragmented data is the main reason to use LMM.

We have made this kind of model in LMM:

Subject: animal ID

Repeated: treatment, gender, time period

Dependent: ng/g

Fixed factors: treatment (main effect), aikaperiodi(treatment) (main effect), sukupuoli (main effect)

Animal ID (random effect)

I would like to confirm couple of things which are still unclear for me:

1. Due to huge individual variation especially after the ACTH injection the data is not normally distributed nor variances are equal. Should one of these assumptions be met or is LMM 'robust' enough? If assumptions must be met what kind of data transformation I should carry out?

2.What is the 'repeated covariance type' I should select for this data?

3.How I should set the random factor or is there a need to set one? I have understood that only random in my model is the animal ID but does LMM take is already as random when it is set as subject in the beginning? If I need to set it as random factor separately, there is two different boxes, 'model' and 'combinations'; what does these mean? and what mean if I decide to include the intercept? and what should I choose as covariance type?

4. I have received the warning "Iteration was terminated but convergence has not been achieved. The MIXED procedure continues despite this warning. Subsequent results produced are based on the last iteration. Validity of the model fit is uncertain." what does this actually mean?

5. Also there is something weird I cannot completely describe, sometimes when I keep the model exactly same but just change the order of the fixed effect it may result to much lower information criteria number but when I open the data again and do exactly the same syntax it does not give me the low IC numbers anymore. So there is something I just dont get.

I really would appreciate if someone could help me with this issues I am dealing with.