I have a series of animal behavioral data observations for 5 animals and 5 measured behaviors (duration of behavior and frequency of behavior) sampled 3x/weekly for 12 weeks. At week 7, a new ingredient was added to the animal diet, and behavioral observations continued as before.
What I would like to see is a change in behavior, pre vs. post. As each animal is very unique, grouping them did not provide statistical significance (basic t-test, average of all animals before vs. average of all animals after for each behavior). So I need to analyze each animal individually, and see if measured behaviors changed for that animal after the diet intervention.
I am struggling with a couple of concepts:
Over time, the animals became accustomed to the observer, so you would expect changes in behavior. Second, when the diet is changed, it would take time for it to make it through to the intestines-you would not expect an immediate behavioral change, if there was one to be found.
I'm looking for advice as to how to organize and analyze my data to potentially see differential effects of the diet.
You should use a mixed model to analyse these data. Mixed models let you specify a random effect, in your case the different subjects. The model will "understand" that each subject represents a cohesive block of measures, and should help you with the differences between them.
To address the problem of change in behaviour as a function of how they become accustomed to the experimenter, you could use "sampling day" as a variable. You should have 36 measures for around 90 days, so you could code your sampling day as a continuous variable and observe the change along the days. this would also let you see a difference between days after the change in diet.
your final models should look something like this (R syntax):
duration of behaviour ~ sampling day (1-90) * diet type (A or B) + (1|subject)
and
frequency of behaviour ~ sampling day (1-90) * diet type (A or B) + (1|subject)
note that with frequencies you should use a poisson error structure
as a final note, I do not know how these model will work with 5 subjects, because they should be generally used for a bigger sample size, but it is worth a try!
If you have any questions, please feel free to ask!
- Badges
- Science topic