Table 1.

Summary of some of the actions which may introduce bias and reduce the internal validity of an experiment, the types of biases they introduce, and solutions to avoid these biases.

Threat to internal validity	Bias introduced	Solution	Practical advice
Biased allocation of animals to experimental groups	Selection bias	Randomisation	✓ Do randomly allocate animals to experimental groups using a random number generator (several online tools are available).
			✓ If there are key baseline characteristics which need to be evenly distributed across the groups, conduct randomisation in blocks according to those key baseline characteristics.
			✘ Do not select animals haphazardly or serially, for example, by picking them one by one from a cage and assign to groups or allocating all animals of the same sex or from the same litter to a group.
			✘ Do not move animals between groups even if they did not receive the assigned treatment. The random sequence determines the group allocation. Animals should still be considered a member of their allocated group, and their outcome data (if available) analysed as such. A secondary ‘on-treatment’ analysis is allowable but carries less weight.
		Blinding	✓ Do randomly allocate animals to experimental groups without knowing which group will be receiving which treatment (e.g. label groups 1, 2, and 3).
			✓ Do ask have group allocation conducted by a third party (e.g. a colleague not otherwise involved in the experiment); they should store unblinded allocation securely, in a place not available to other staff involved in the experiment.
			✘ Do not use informative group labels (e.g. do not label the groups ‘exp’ and ‘ctrl’). The allocated group labels should be non-informative (i.e. coded alphanumeric identifiers).
Systematic differences in how animals in different experimental groups are handled or cared for apart from the intervention being tested	Performance bias	Randomisation (performance)	✓ Do perform any interaction with the animals in a randomised order across groups. If animals were properly randomised at allocation and are only identifiable by their number (see later), the order of the animal number can be used.
			✘ Do not perform interactions with the animals in sequence according to their group allocation, for example, sacrificing all animals in group 1 in the morning and group 2 in the afternoon.
		Blinding (performance)	✓ For optimum blinding, label animals only with an alphanumeric identifier (e.g. rats 1A–26X) and avoid being aware which animals belong to the same group. Being aware which animals belong to the same group can cause blinding to fail if, for example, patterns in behaviour become apparent.
			✓ Less optimally, use partial blinding to mask group labels, for example, assign animals to groups ‘A’ and ‘B’ rather than ‘treatment’ and ‘control’.
			✓ Do, if handling or treating the animals requires a member of staff to be unblinded, exclude that person from participating in any other phase of the experiment.
			✓ Do make any decisions on animal welfare concerns blinded to group allocation.
			✘ Do not assign animals to openly labelled groups, for example, ‘treatment’ and ‘control’.
Systematic distortion of the results of a study that occurs when the outcome assessor has knowledge of group allocation	Detection bias	Blinding (outcome assessment)	✓ Do use blinding at the time of outcome assessment, with the help of research assistants or technicians if needed:
			► (Gold standard): label animals only with their alphanumeric identifier and avoid being aware of which animals belong to the same group.
			► (Second best, partial blinding): mask group labels.
			✘ Do not assume that any outcomes can be assessed – whether behavioural or otherwise – in an unbiased unblinded manner.
			✘ Do not remove blinding until analysis is completed. For studies with alphanumeric labels only, first unblind the grouping (i.e. to group A, group B), then perform the analysis, then unblind the identity of groups A and B).
Unequal handling of animal drop-outs between treatment groups	Attrition bias	Pre-define and report exclusions and their criteria	✓ Do pre-define the criteria for any exclusions.
			✓ Do make sure that exclusion criteria are not directly related to the condition or treatment being studied.
			✓ Do ensure that the individual responsible for the decision to exclude an animal is external to the experiment and has no knowledge of group allocation (e.g. the house veterinarian).
			✓ Do clearly report the total number of exclusions and reasons for exclusion for each experimental group.
			✘ Do not decide on criteria for exclusion during or after the experiment.