Statistics for Experimental Biologists

Home

Topic index

Key books

External links

Book

Is the information below useful? Chapter 2 discusses how to design experiments to deal with litter effects and many other factors that can influence the outcome.




Are litter effects something you need to consider?


It is not unusual for one litter of rats or mice to be assigned to a control group, while another litter (born on the same or a different day) is assigned to a treated group. There are two problems with this. First, because treatments were applied to litters, this makes the litters—and not the individual animals—the experimental unit. This is because randomisation and treatment assignment occurred at the level of the litters and not at the level of the individual animals. Second, and more importantly, there may be differences between litters, with animals from one litter having higher values of the outcome variable than animals from another litter; in other words, there might be a litter effect. This also means that two animals from the same litter will tend to be more similar than two animals from different litters, and thus animals within a litter are not completely independent, as most standard analyses would assume or require. Therefore assigning a whole litter to a treatment, while administratively easy, means that treatments are partly or completely confounded with litter. If there is only one litter in each treatment, then litter and treatment are completely confounded, and there is no way to tell whether any observed difference is due to the treatment or due to natural variation between litters.

This point is discussed extensively by Lazic and Essioux (2013), where the presence of litter effects was demonstrated on a number of variables. So, are litter effects important for the main outcome variables that you routinely measure? Have you even checked? Litter effects may always be present, and so it will always be necessary to account for them. It is also possible that litter effects are not important in general, but on one particular day something happened (e.g. one pregnant dam had an infection, or was stressed, etc.), which made the offspring different compared to other litters. Even if litter effects are not important in general, the experiment should be designed to protect you from these one-off events.

Options for dealing with litter effects

The simplest option is to use only one animal from each litter. The other animals can be used for a different experiment or by other members of the lab group or department. The advantage of this is that standard statistical tests can be used, and litter effects can be ignored because they are now subsumed in the animal-to-animal variation. A disadvantage is that this might introduce a large amount of unexplained variation in the data, which will reduce the power of the analysis. A second option is to use all (or most) animals, but ensure that animals from each litter are represented in each experimental condition, which is shown in the table below (here it is assumed that there were at least six animals per litter).


Experimental design

This layout is referred to as a randomised complete block design (RCBD). The litters are the "blocks", which is another variable that is introduced into the experiment to account for a known source of variation, and usually there is no experimental interest in it. "Randomised" refers to the fact that within each litter, the six animals were randomly assigned to one of the three conditions, and "complete" indicates that every treatment appears in every block (in this case the design is nicely balanced).

The key message of this article is that it is important to avoid confounding any potential litter effects with treatment effects because there is no way to disentangle the effect of one from the effect of the other. An appropriately designed experiment will protect you from this, and it is likely that in many cases, an incorrect analysis will still give a reasonably correct answer (but the appropriate analysis is always preferable).



References


Lazic SE, Essioux L (2013). Improving basic and translational science by accounting for litter-to-litter variation in animal models. BMC Neurosci 14:37. [PDF]