It is difficult for researchers to get a random sample. If they are performing an experiment on humans, they will impose some kind of treatment on the subjects, and they cannot just go out to each randomly-selected person in the United States say "Hi, take these pills for 6 months. Come to my lab for testing when you finish." They must rely on volunteers, and this introduces bias intro the experiment.
Work Step by Step
For non-human subjects, a random sample may also be difficult to obtain. If the researchers are testing the effectiveness of a new bug spray on cockroaches, for example, they would technically have to take a random sample of cockroaches from all over the country and from both homes and from nature. This is, of course, ridiculous. They will probably use laboratory cockroaches, which they hope will be similar enough to those infesting homes that the results or their laboratory tests could be generalized to all cockroaches.