Hello!
I am currently situated in Paraguay, where I did research on howler monkeys. I am now in the data analysis fase of my project and pretty stuck.
Background of the project:
I played 5 different sounds (Ship horn, elephant, howling of conspecifics, people talking and a silent control) all once to 9 different monkey groups. The monkey groups were all divided into 3 different habitat types (natural, peri-urban and urban). Within the groups subjects were sampled based on age-sex category (e.g. AM-adult male, JF-juvenile female) and a couple of different reactions were measured, like vocalization, vigilance or movement towards the speaker.
Data collected:
My data was entered in excel with every trial as a row, with columns for the sound, habitat, visibility of the different age-sex categories when doing the trial and a column for every behavior type performed by every age-sex category like picture 1.
I added a column were I divided the behavior performed by a age-sex category, through the amount that age-sex category was visible, creating for example the 'CorTowardsAM' column (cor for corrected).
I then transformed this data to look like picture 2, where ValueMinute is the corrected amount of times a behavior in a particular situation was performed by a particular age-sex category per minute (since the time of sampling was different for a couple of sounds).
Now I do not really know how to analyse this. The main thing I want to know for my research question is if there is a difference in reaction between the habitat types and to a lesser extent between the sound types. Since however, the group compositions vary a lot between groups and also the visibility of subjects varied, I feel like I need to corrert for the differences in groups and age-sex categories.
I am looking now at options of doing multiple different anovas, a linear mixed effects model with habitat and sound as fixed effects and group and subject as random effects. Or maybe I should do an ANCOVA?
Truth is i am a bit lost with my data with so many variables and maybe potential biases.
Picture 3 is a plot from the data to help visualize what I have.
I was hoping if people with more knowledge about how data works could help me a little. Am I on the right track? Is the kind of data I have generally used in a linear mixed effects model? If not, how can I analyse it? If a linear mixed effects model is right, what do I do with the output and do I need to check assumptions beforehand?
Thanks a lot for everyone wanting to look at this long post!
You are dealing with categorical data, so need to use non-parametric tests, such as Chi squared. You do not need to transform your data for these types of tests since the distribution will not be normal anyway, and it's best not to for transparency :)
Like Andrew Paul McKenzie Pegman said, non-parametric tests are the most straightforward way to go. A possible alternative analytical approach would be to use randomization analysis on your field data -- by using your own data to generate a random data distribution (a select and replace sort of function for 1,000, 5,000, 10, 000 iterations... as many as you want...), you can then place your actual data against that randomized distribution to determine if there are any significant patterns in your observed dataset.