Can you use the chi-squared test if subjects from the groups in the predictor variable can appear multiple times in the response variable categories?
&
Can you use chi-square if you do not accurately know what the ‘expected outcomes’ should be?
I am studying the feeding ecology of sea turtles and am struggling to apply a meaningful statistical test to my data.
I am trying to report where they eat in the water column. I have recorded where individual turtles have taken a bite at a food source in 3 locations: at the sea floor, pelagic (in the middle), and at the surface. Turtles can obviously move and feed in different areas so the data for any individual may include several ‘counts’ in each of the 3 locations. E.g. Turtle 1 may have taken a bite at something 100 times at the sea floor, 2 times in the pelagic zone, and 5 times at the surface. Some turtles did not eat at all.
I have a relatively small sample size of Green turtles (adults n = 9 and juveniles n = 4) I.e. 1 categorical predictor variable with 2 groups. Nb. Also the difference between the adults and juveniles is judged on their size (adults above a certain shell size) so perhaps this variable could be considered as numerical if they were separated by the size?
I want to see if there is a difference in where these turtles ‘mostly’ feed, depending on their ‘group’.
I have looked at a chi-squared or Fisher’s exact test but can this work if the turtles can select more than one of the locations (dependent variable) multiple times?
The other issue with these is that it is not necessarily reasonable to assume that these turtles would be ‘expected’ to feed in any particular place with any real accuracy. I would expect that they would mostly feed on the seafloor (and this is what they did) as they feed from the algae on the bottom. How can I put this as an ‘expected’ outcome percentage for this group without it being fairly arbitrary? Their feeding habits are not well understood, hence the study…
Alternatively, am I looking for a proportions test or perhaps a Generalized linear model? Any advice on this would be greatly appreciated!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire