We are going to do a behaviour trial with fish. We will video record each fish individually behaviour after different stimuli. I was wondering how many individuals should we used for the correct replication levels for each treatment? Thank you very much
Hi Marta,
This is a bit of a tricky question because there really is no "magic number". It mostly come down the questions are you are trying to answer and the level of certainty that you would like to achieve.
If you have the resources (time, money, fish, etc.) then more replicates are always going to be a bonus - to a point. But there comes a point when you essentially reach replicate saturation, where new data points don't really tell you any new information. So, I guess the easiest answer is to provide a minimum number of replicates. Statistically the absolute minimum would be 3 replicates, but that is still less than ideal.
Depending on how many different treatments you have, I would recommend at least 5 individuals per treatment (even this is pretty small). But, if you have plenty of fish at your disposal, I would certainly recommend that you increase this sample size as your study allows.
Thank you so much for you answer. My question is more about time, because we have 7 treatments and probably we can only perform one fish per day, so it will take many days to running the entire trial. We will video record fish and taking some swimming velocity parameters. Usually we don't work with individuals but with tanks with many fish inside, and we always use 3 replicates as minimum, but since here the replicates are the individuals and not a tank, that is why my question raise. thank you!
You might suggest to know if your sample is representative to make an accumulated curve of fish species based on the sampling days or transect numbers.
Hi Marta,
That's not such a tricky question.
Basically, Bradley said it all.
If you want a suggestion, I would say that around 10 is good but if you can reach 20 individuals it should be comfortably enough.
Hi Marta,
As others have said, it depends somewhat on what your objectives are for the experiments. However, in our experience you can get a lot of variation between individuals when you're working with fish and so to be able to make some reasonable conclusions I would agree with José that you should be aiming for at least 10 and ideally more per treatment. This will also allow for the fact that some fish just won't do what you want them to!
Hi Marta,
In a project about the effects of cadmium P. lineatus behaviors we used one between 6 and 7 individuals per treatment. Overall, the results were consistent, but as Paul Franklin commented on, in some evaluated parameters, there were large variations among individuals in the same group. Good luck!
Dear Marta,
I agree that the number of replicates you need depends on your experimental setup and objectives, especially on
1) how much variation between individuals you expect for a particular trait, and
2) how much you expect your treatments to affect that trait.
When you develop your experimental design, you could/should consider using a power analysis. This will allow you to determine which sample size is required to detect a certain effect (or rather the degree of it) with a given degree of confidence.
To run such an analysis, you need info on the above-mentioned two criteria. In particular, what is the expected standard deviation for your data and based on that, what effect size is relevant for your question? You can either scout literature to get an idea on those values, or use your own experience / preliminary data to make a judgement.
For example, given a standard deviation of approximately 70 and an effect size of 0.86, the minimum sample size to detect that effect would be n = 22 when you assume a significance level of 0.05 and a power of 80% (i.e. the probability of finding that effect). This is in case of a very basic experimental design (t-test).
You can run such a power analysis in free software like R. Using the pwr-package, the command to do the above calculation would be: pwr.t.test(n=NULL, d=0.86, sig.level = 0.05, power = 0.8)
The exact calculation will differ depending on your experimental design (t-test, ANOVA,...) so you would have to take that into account as well, but hopefully this already gives you an idea.
All the best,
Eli
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Writing