Experiment Design -- Double Blind

Return to the Main Math 160 Topics page Revised November, 2013

A double Blind experimental design is one where neither subjects receiving the experimental treatments nor the experimenters (those who are giving the experimental treatments) know which experimental treatment is being applied to the subject. The goal of this design is to eliminate any possible effect that might be caused by the subjects or the experimenters knowing which experimental treatment they are getting.

This design is not possible in all cases. However, it is possible and it is preferred in many clinical trials. For example, consider the situation where we were to run an experiment to investigate the usefulness of a certain medication. We would like to give the medication to a group of subjects so that we can see the effect of the medication. To measure that effect we would need another, similar, group of subjects who do not get the medication. Of course, we want to treat all subjects in an identical fashion. Therefore, we come up with two types of pills, one that contains the medicine we are testing and the other that, although it looks just like the former pill, does not contain any active medical substances. This latter pill is called a placebo. Subjects in the study are randomly assigned to receive the real pills or the placebo. At the end of the treatment period we can compare the medical outcomes of the two groups. Such a design is a "blind" study because the subjects do not know if they are being given the medicine or the placebo.

As described so far, this process is not a "double blind" study because the people handing out the pills to the subjects may still know which pill they are distributing. The concern is that if the people running the experiment know which type of pill they are administering, then they will is some subtle way convey expectations related to the type of pill being given to the experimental subjects. Clearly, someone has to know which pills are being given to which subjects. In a "double blind" study the experimenter knows this but the experimenter is not the person interacting with the subject in giving the subject the pills or even in collecting and evaluating the result of the experiment. In the case described above, the experimenter could prepare the pills, putting the assigned pills into containers that differ from one another only by a unique number on the container. The experimenter would randomly assign the true medication and the placebo to these numbered containers. Then the experimenter could randomly assign the containers to the various subjects. The experimenter would then turn over the actual distribution of the containers to someone who did not have access to the information that indicates if a particular container holds the medication or if it holds the placebo. Furthermore, after the specified time for the treatment, the evaluation of the subjects, that is the measure of the effectiveness of the medication, would be collected by someone who also does not have information about whether or not the subject got the real medication. It is only after all the results have been collected that the experimenter can associate the results with the kind of treatment administered.

A double blind experimental design protects the experiment from the effects of having the administration of the experiment influence the results. This is a good thing. The additional cost of this is to involve others to manage the experiment. As noted above, it is not always possible to run a double blind experiment. If we were investigating the possible advantages of using advanced calculators in a basic statistics class it would be impossible to have the instructors "administer" the lessons, some using the calculators and others not using them, without having the instructors know that they are or are not using the calculator.

Return to the Main Math 160 Topics page

©Roger M. Palay
Saline, MI 48176
November, 2013