Failure to identify correctly the experimental unit is a common mistake which can result in incorrect conclusions. The experimental unit is the physical entity which can be assigned, at random, to a treatment. Commonly it is an individual animal. The experimental unit is also the unit of statistical analysis. However, any two experimental units must be capable of receiving different treatments. Thus, if mice in a cage are given a treatment in the diet, the cage of animals rather than the individual animal is the experimental unit as mice in the cage can not have different treatments, and they may be more similar than mice in different cages. This means that the p-values in the statistical analysis may be incorrect if it is assumed that the mouse is the experimental unit. In this case the statistical analysis should normally be done using the mean of all the animals in the cage.
Experimental units may be:
The individual animal:
The breeding female and litter:
The cage of animals
A part of an animal:
An animal for a period of time:
More than one type of experimental unit in an experiment.
A fancy tri-coloured guinea-pig.
This is the most common case. Individual animals are assigned to the treatments, but it must be possible for any two animals to receive different treatments. If the treatment is given to several animals in the same cage in the diet or water then individuals within a cage can not receive different treatments, so the individual animal is not the experimental unit.
In teratogenesis experiments the pregnant female is treated with the test substance (or vehicle), but measurements are made on individual pups after birth. Because individual pups within a litter can not have received different treatments the experimental unit is the whole litter and the unit of statistical analysis is the litter mean or proportion affected, possibly weighted by litter size. If there are, say, 10 treated and 10 control females, then the statistical tests will be based on a group size of 10, not on the numbers of individual pups
If the treatment is incorporated into the diet or water then any two animals within the same cage can not be assigned to different diets so the experimental unit will be the cage of animals. Cage effects (differences between cages on the same treatment) can occur as a result of fighting, sub-clinical infection or differences in environment due to cage position. The statistical analysis will then be based on the mean of all animals within a cage, and “N” will be the number of cages, not the number of animals. However, if the treatment can be given to individual animals within a cage, say by injection or gavage, then the individual animal may be the experimental unit.
If the experimental treatment is the topical application of a substance to the shaved skin of an animal, then it may be possible to divide an area of skin into a number of patches which can receive different treatments. In this case it is the patch on the back of an animal that is the experimental unit. Similarly, some paired organs may be considered as the experimental unit if they can be assigned to different treatments. If individual cells, say in the brain, can be given different stimuli and recordings of response are made, then individual cells are the experimental unit because different cells can receive different treatments. In this case it may be possible to get a lot of experimental units out of a single individual. However, usually it is wise to use three or four animals (this amounts to a randomised block or repeated measures experimental design) in case individuals behave differently.
In a crossover experiment where an animal may be assigned to a treatment for a period of time, then rested and assigned to another treatment, the experimental unit is the animal for a period of time.
Occasionally “split plot” or “nested” experiments are designed where there are two classes of experimental unit. For example, a researcher may want to compare two diets each with or without a vitamin supplement on growth rate in mice. Suppose 20 cages each containing two mice are assigned to the project, with 10 cages being assigned at random to each diet. Suppose also that the vitamin supplement can be given individually by injection or gavage to one of the two mice, with the other one receiving the vehicle as a control. In this case the cage is the experimental unit for comparing the two diets using the average growth rate of the two mice in the cage, but the mouse is the experimental unit for comparing the effect of the vitamin supplement. With this design it will also be possible to find out whether the response to the vitamin depends on the diet. Split plot designs can be economical and powerful, but the statistical analysis can become complicated. Expert advice should be obtained before starting such an experiment.