I am designing a small scale trial to test dose response to a vaccine in an animal model. The animals in question are fish, and the vaccine is administered by injection. Lets say that I have already conducted a power analysis and estimated that I need 120 fish per treatment group to observe the effect that I'm interested in.
Our fish holding facility allows for up to 12 tanks, each housing up to 60 fish. Each tank operates completely independently, with its own temperature control, pumps, water supply etc. In a trial with six treatment groups (i.e. five vaccine doses and a placebo control), this allows for a couple of different experimental designs.
In the first design, individual tanks, each containing 60 fish, are assigned to one of the six treatment groups. This gives duplicate tanks for each treatment group, with a total of 120 fish per treatment. My understanding is that in this case, the experimental units are the tanks, so the entire trial has effectively been conducted twice, with a sample size of 60 fish per treatment.
In the second design individual fish are injected with one of the six vaccine treatments, and tagged to identify which treatment they received. The tagged and vaccinated fish are then distributed around the room such that each of the 12 tanks houses 10 fish from each treatment group. The total number of fish per treatment is still 120, and the total number of fish per tank is still 60, but now the experimental unit is the individual fish, rather than the tanks. This means that the trial has been conducted once (i.e no replication of the overall design), but with a sample size of 120 fish per treatment.
I have a strong preference to use the second design, as experience in the past has shown unexpected events such as an equipment failure or disease outbreak in an individual tank can severely compromise the experiment if each tank contains only one treatment group. Even in the absence of serious problems, small variations in temperature, lighting, water flow etc can exist between tanks, which can potentially influence the outcome from one tank to the next. The second design seems to overcome these issues by spreading each treatment group out over 12 tanks.
So, my questions are - (1) Am I correct in my understanding of the definition of the experimental units and replicates in each trial design? (2) Is it acceptable to spread a treatment group across multiple tanks, while still considering each animal as a single experimental unit?
I guess my answer is coming too late, but still thought to share my view: For your second choice (question 2), the plus is all other factors are common except the treatment since all the treatment groups will be housed in one tank. However, my concern is what if one group of fish receiving treatment X has a negative effect (e.g., social dominance) on another group of fish receiving treatment Y. For your question 1, if fish are not tagged, tank is your experimental unit. Duplicates will not be sufficient. You could run power analysis to determine what combination is best - having more number of fish per tank would reduce tank variability, however, you would still need certain number of replicates. If you had tagged fish, you can subgroup/nest them in tanks or block the individual responses by tanks during analysis. I hope this gives you some idea.
for fish you should replicate and take average wight for each replicate as one animal
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size