The data values are often dependent when there are physical relationships between the experimental units, such as the manner in which pots of plants are placed on a greenhouse bench, the physical proximity of test animals in a laboratory,
having multiple animals feed from the same container, or the location of experimental plots in a field. To minimize the possibility of experimental biases, dependency in the data, and to obtain valid reference distributions, it is necessary to randomly assign the treatments to the experimental units. However, the random assignment of treatments to experimental units does not completely eliminate the problem of correlated data values.