Hello.
I want to test the preference of fish species X to stimuli in an aquarium. One side of the aquarium had stimulus A and the other side had stimulus B. I recorded the position of the fish (near A or B) every 10 sec for 1 min. So there were 6 records for each time of the experiment. I conducted the experiment with 10 fish and each individual was tested for 5 times (one time per day). So, I got total 300 records. I want to test whether fish species X has preference to the given stimuli or just moved by random by using Chi-square test.
Can I pool all data together (all fish and all replications) for analysis? If not, how to handle it?
Thank you.
What is your outcome variable, distance from A (or B), a dichotomized variable for which side closer too (which doesn't seem a good measure), or something else?
Without a bit more description what your question makes me think about is the drift rate from diffusion models, but with only 6 measurements per unit probably the best start is more basic longitudinal models. If x_t is location (say distance from A) at t, it sounds like a plausible non-random hypothesis would be x_t - x_(t-1) is the same sign as x_(t-1) - x_(t-2), but it might be more valuable having a quantitative hypothesis to evaluate. Depends on the subject area (which I don't know).
ps. On pooling (good choice of words for the example) all data, I expect no, but depends what you mean. If you mean treat each observation as independent this will leave out what I assume is important variation for the individual fish. If you mean use a single aggregate value for each fish, and then analyze the data, you'd lose some information, but can answer some queries.
Do not pretend to have 300 data points. You have ten fish over five days of the study. Having only six values to average over a single minute of a day is not very comprehensive, but I can see where you might say it is a singular average position for the fish.
Your ten fish, I hope, were randomly chosen. If that is the case, then you have ten independent longitudinal samples. Each of these samples would be a mean +/- standard error of position based on the average position for each day of the study. I think Chi squared would likely be violated by this small sample size, though I have not run through any numbers.
Hope this helps
Ryan Werdel , he's got 300 data points, they just aren't independent.
Daniel Wright In the general sense, certainly he has 300 data points. The reason I said otherwise is to discourage the tendency to assume these data can be analyzed by any means. Especially for a test like Chi squared, doing so would instigate error due to pseudoreplication.
So you say he should not "pretend" to have something that he "certainly" has. It is good he wasn't pretending!
