We currently offer two randomization methods in Studies:
Block - this is suitable for allocating animals into groups based on a single metric
Clustered - this is suitable for randomizing based on multiple metrics / attributes simultaneously.
Block
Animals are ordered from smallest to largest based on the selected metric.
Animals are separated into blocks. The size of these blocks is determined by the number of treatment groups in the study. For example, if there are 5 groups, the 5 animals with the smallest measurement will be in block 1, the next 5 animals will be in block 2 and so forth.
Starting at block 1, we sequentially allocate an animal to each group
Animal 1 (smallest measurement) - Group 1
Animal 2 - Group 2
Animal 3 - Group 3
Animal 4 - Group 4
Animal 5 - Group 5
When a new block is reached, the assignment order is reversed
Animal 6 - Group 5
Animal 7 - Group 4
Animal 8 - Group 3
Animal 9 - Group 2
Animal 10 - Group 1
This process is repeated until all animals have been allocated to a group.
If the number of animals being assigned to each group is different, we allocate until the smaller groups are filled, then distribute the remaining animals to the groups with space remaining. This naturally results in animals with the largest measurement being assigned to the larger groups.
Clustered
Animals are arranged into cohorts:
If no attribute is selected, there will always be one cohort.
If an attribute is selected, animals will be organized into cohorts based on the number of options available. For example, if "sex" is selected as the attribute, the animals will be separated into 2 cohorts: male and female.
Within each cohort, k-means is used to separate animals into clusters based on the selected measurements. We currently allow up to three measurements to be selected at the same time.
For each group, we divide the capacity of the group by the number of cohorts (eg 30 animals per group and 2 cohorts gives an "initial cohort capacity" of 15)
Select the first cluster and allocate the first animal from that cluster to a group at random. We then cycle through the groups until no animals remain in the cluster.
We cycle through the remaining clusters and repeat the steps above until all animals have been assigned.
Note, if there is uneven cohort sizing (eg. more males than females), the initial cohort capacity may not be hit for all groups.
We move to the next cohort and repeat this process.
In the last cohort, some animals may be remaining. This can happen for a number of reasons including uneven group sizing, uneven cohort sizing etc. These animals will be allocated at random to groups with space remaining until the capacity of each group is hit.