Searching for behavioral differences between two groups of horses is my main research goal. All the recordings (as I mentioned in the title) are consisted of categorical variables (nominal in most cases except weather conditions and surrounding sound levels which are ordinal data)*.
The horses are divided in stereotypers and non stereotypers in groups of three (three horses with normal behavior and three that are stereotyping). The horses' behavior is split among 5 major categories:
- Feeding behavior-FD,
- Resting-R,
- Locomotor behavior-L,
- Stereotypical behavior-S,
- finally some other behaviors classified altogether as insignificant because of the low scoring.
By using a $\chi^2$ (chi-square) test, comparing the major behavioral categories of the two groups (except the stereotypical behavior for obvious reasons), I've found that there is a significant behavioral difference ($p=0.008<0.05$, $df=3$) between stereotypers and non stereotypers, especially in feeding and resting behavior, though the standard residual scores of these behaviors ($\pm 1.4$ and $\pm 1.3$ respectively), weren't "significant enough" according to a general rule of thumb that says if the value of the residual lies outside of $\pm 1.96$ then it is significant at $p < .05$.
Question 1: how do I proceed from here? Do I accept that there is a small difference that could let me reject the null hypothesis, but judging from the standard residuals isn't strong enough to make a valid assumption that there is a behavioral difference between the groups? Would it be wise on my part to break the main table in 2x2 contingency tables and check for individual differences in each behavior? Would the results be pointing more clearly in one direction or the other?
Question 2: Last but not least, I'm having a trouble finding the connection between my variables. What I mean by that is, if indeed there is a significant difference in the behavior between stereotypers and non-stereotypers how do I exclude the other parameters (human activity, weather conditions etc) and blame stereotypy for that difference? Do I divide the stereotypers and non-stereotypers and via $\chi^2$ check them separately or do I test them as a whole checking the significance of every behavior concerning the different parameters?
PS*: Here is a portion of my recordings in case you want to have a better idea of what I'm talking about 
PS**: Not much of a stat guy so I'll appreciate very much if you keep it simple. Feel free to ask me anything you want...
