I'm currently in the final stages of my PhD and have some queries regarding data analysis on behavioural data.

Details of my set-up are as follows:

Open Field

n=12

1 primary outcome and 3 secondary

3-4 experimental groups - thigmotaxis in models of experimental pain

My approach to date has been to first establish normality, and then use the relevant ANOVA followed by a post-hoc test (dependent on ANOVA used).

I normally just plot columns, with mean and StErr... but am moving away from this after thinking more about the data - some are continuous and others discrete (i.e. number of rears/frequency).

I am not sure this is the best way of presenting the data.I have experimented with binning into quartiles and plotting that, which nicely illustrates the behavioural shift but is no use for further quantification.

So now I am moving towards a combination of boxplots/dotplots as this best illustrates the spread of behaviour and how it is affected by different treatments.

There seem to be differences in variation between the groups and I wonder whether this would be more relevent. I've read Gordon Drummonds excellent series on statistical reporting in the British Journal of Pharmacology, in which he mentions permuations as a way of looking at variation, and in fact he suggests that standard T-tests and ANOVAs are not always the most relevant tests.

Do you use permutations to look at behavioural variation?

How do you represent your behavioural data? Is a standard bar chart with SD/StErr is enough?

Essentially, I'd like to move towards a style of reporting more analogous to clinical data reporting - I already work within the ARRIVE guidelines - and feel that it's the outliers and other expected data points are most interesting in terms of explaining variation in clinical presentation.