Animal health studies are designed to compare two or more treatments. Attempts are made to obtain equal number of subjects within each of several sites to balance the design for optimal power. Unfortunately, a balanced design cannot always be obtained resulting in unbalanced center size. This imbalance has a detrimental effect on the power of any tests performed. Three methods are considered that may mitigate this problem: balanced clinic sizes, centralized randomization, and weighted analysis.

In order to balance clinic sizes the overall enrollment period may need to be longer. This allows smaller clinics to obtain the number of subjects that match, or come close to matching, the numbers in the large clinics.

Centralized randomization helps prevent imbalance because, by monitoring in real time, the accrual of subjects within each clinic permits corrective measures to be taken if some sites are slow recruiters. It may also assist in the prevention of statistical biases.

In the majority of Animal health efficacy studies unweighted analyses are used. Unweighted analyses assume that the standard deviation is consistent over all sites. This may not always be true, especially in unbalanced designs where a weighted analysis may be a more efficient method. If the standard deviation of the error term is not constant across all sites, the use of weights that are inversely proportional to the variance at each site would provide the most precise parameter estimates. One disadvantage is that a weighted analysis assumes that the weights are known, however, this is rarely the case in the real world, so estimated weights must be used instead. When these weights are estimated from small numbers of observations, the results can adversely affect the outcome.