G7 Think carefully about when to randomize, especially when you are not blind to condition

Human behaviour is surprisingly sensitive to influence. You may find to your surprise that participants tell you in your post-experimental interviews that, in effect, their behaviour has been influenced by some extremely subtle feature of the design or of your behaviour as experimenter. (For more on post-experimental interviews, see Section 13.9.4 of the book.) For this reason, we are often concerned that even subtle features of our behaviour might differ between conditions when we interact with participants and that this might influence their behaviour. Whenever possible, therefore, it is useful for the person interacting with the participant to be blind to conditions, i.e., to not know what condition the participant is in. In some types of research it is essential that both the researcher and the participant are blind to condition. Such experiments are called double blind experiments.

Of course, practically, especially as a student, this may not be possible. The experimenter may often be the researcher too. Under these circumstances, you should leave it to the last possible minute before discovering which condition your participant is in. For example, run the experiment without looking to see what condition the participant will be assigned to until the point at which the manipulation has to be made. At least that way you can guarantee that your behaviour is not different between conditions up to the point of randomization (although, of course, it may differ in subtle and not so subtle ways afterwards - an important potential confound). (To find out more about confounding variables, why they are important and how to control for them, see Section 9.1.3 of the book.)

If you are using different versions of questionnaires or a computer to run the experiment it might be possible to remain blind to condition throughout. If you are working as a group, think about ways of running the experiment that involve the different group members so as to allow one or more of you to be blind to condition when running participants. (However, make sure that you do not create a confound by having different experimenters running the different conditions!) To understand more about this problem and its importance, read up on experimenter bias and demand characteristics. (You can find recommended reading at the rear of the book.)