Randomization

Abstract

There are three aspects of randomization in statistics that are considered here. The first aspect is randomization as part of a sampling design to estimate one or more parameters for a statistical population such as all the farms in a certain area of a country, based on information obtained about the parameters from only a part of the population. The second aspect is using randomization as part of an experimental design to ensure that the allocation of treatment levels to the experimental units is not biased in any way. For example, the test of a new drug for relieving the symptoms of a disease might involve this drug being randomly allocated to half of a group of patients, while the other half of the patients receive a standard drug that is used for the disease. Finally, the third aspect is using randomization to test some statistical hypothesis. For example, to see if there is a significant difference between two drugs for the treatment of a disease in terms of some suitable outcome measure, the observed mean difference between means for this outcome measure might be compared to the distribution of mean differences that is obtained by randomly reallocating the observed values of the measure to the drugs. The null hypothesis being tested would then be that each of the observed values of the measure was equally likely to have occurred with each of the two drugs. Copyright © 2010 John Wiley & Sons, Inc.