I'm analyzing data from experiments, where in a net pen we exposed fish to different noise frequencies. However I'm uncertain as to what statistical tests I should use to test for possible differences before/during/after the exposure.
We have data in one second intervals of the area of the school of fish and the velocity of the centroid of the school (plus X/Y coordinates), I'm interested in testing for differences within and between different exposures. Data for each exposure starts 60 seconds before the onset of sound, which lasts for about two minutes, and ends with a 60 second tail after the noise has ended.
Should one use time series analysis, like ARIMA, ARMA or ARIMAX to test the data, or something different altogether? At the moment I'm using SPSS to test the data.
Dear Timo Tikka,
I believe using time series methods are not suitable for this study. You are looking for the effect of different noise exposure on fish, and time series methods won't give you any answer to reach that goal.
You may consider at least one of specific seconds during the 2 minutes of exposure. You may choose only one specific time before, and one after the noise exposure.
Now, by using SPSS you may apply comparison of two means test. It will show you for which noise the difference is significant.
Good luck.
your unit is in seconds i guess , therefore the appropriate test for mean , or quantitative date is t-test or f test depending if it is less the 30 subjects or less.
Dear Timo Tikka,
in my opinion before time series analysis it is necessary to build two regressions. in the first regression the dependent variable is the velocity of the centroid of the school of fish and the independent variable is the noise frequency; in the second regression the dependent variable is the area of the school of fish and the independent variable is the noise frequency. We can assume that the dependent variables will also depend on two variables: the power of the noise source and distance from the noise source.
Your experimental unit is the tank. Tanks are replicates. The volume of the school, and the location have a great deal of noise. So you want to know if (on average) the school responds to the noise. To get an average, you need to combine the measures for each period: before, during, and after noise. I think initially this is a paired t-test because you have before-during pairs. The after treatment results may be more problematic if there is a period of time wherein the fish slowly return to normal. You could try during-after and before-after pairs, and see what you get.
As described, these data are probably not suitable for any time series analysis. The second-to-second variability within and between schools of fish I would expect to be too high relative to the number of replicates to make such an analysis meaningful. Of course that assessment should be qualified by an admission that I do not work with fish.
I agree with the above responses; assuming you don't have any specific predictions about when during the 2-minute exposure your differences would happen, it makes the most sense to just take the mean of that whole time window. If you prefer to make multiple point-wise comparisons across the whole time window, you could instead use something like a temporal cluster-based permutation test, which has been developed for EEG data (Maris & Oostenveld, 2007, Journal of Neuroscience Methods), which actually have a kind of similar structure as your data (multiple measurements of some outcome, taken over a certain time period at a high sampling rate, including both some time before a critical stimulus event and some time after a critical stimulus event).
Stephen,
What an interesting idea, and especially cool if there is sufficient data to support using the method. The article can be found at:
https://www.researchgate.net/publication/6316066_Nonparametric_statistical_testing_of_EEG-_and_MEG-data
One great thing about RG is that you sometimes find articles from other disciplines that you would never ever find otherwise.
Article Nonparametric statistical testing of EEG- and MEG-data
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS