To test a hypothesis in the right way, scientists must precisely design experiments that will bring them the necessary information. Experimental design is therefore a key part of scientific work, especially if the scientist is studying a complicated issue where it is necessary to approach it carefully. Even the experiment that seems to test a given hypothesis perfectly might still be prone to a certain type of unexpected error or produces misleading results due to an incorrect design. To effectively avoid such faulty experimental design, scientists generally follow these rules when planning experiments:
The experiment should be unambiguous and controlled. As a rule of thumb, if we are investigating the effect of certain factor (e.g., the ability of an antibiotic to kill specific type of bacteria), it is important to compare the experimental setup in which this factor is present with the experiment in which it is not present. We call this second type of experiment, which differs from the main test in only one factor, a control experiment. Since both experiments differ only in this respect, we can attribute the different results to the effect of this particular factor (in case of antibiotics, we would cultivate two groups of bacterial cells in the same environment and at the same temperature, but we would add an antibiotic to only one group). If our experiment wasn’t unambiguous, we would not be able to evaluate its outcome with certainty (if we treated bacterial cells with two antibiotics at the same time, we would not know whether the first antibiotic, the second antibiotic, or the combination of both killed them). If we didn't have a control in the experiment, we wouldn't be able to tell which factor caused the observed result (if we added an antibiotic to the cells but didn't have control cells, we wouldn't know whether the bacteria were not previously infected with a virus that killed them regardless of the presence of antibiotic). Since a properly designed control experiment can be sometimes hard to set up, in complex experimental designs we can encounter a whole series of control experiments that cover all possible uncertainties.
Test subjects should be randomly distributed. The random distribution of tested people, laboratory animals or cells that we use in experiments is important because if we do not select our test groups carefully, we may end up with unwanted significant differences between them (e.g., in one group there is larger proportion of older participants). In such case, the observed results could possibly be due to differences in age, rather than the effect of the tested substance. The more complex the studied phenomenon is, the more properties (often unpredictable) of the test subjects might influence the results of the experiment.
The experiment should be reproducible. Another guiding principle of experimental design is that if we have truly identified a factor that is responsible for a particular experimental outcome, then that factor will be still effective if our experiment is replicated by other scientists in other laboratories. If it is not, that would indicate that not only the studied factor (e.g., antibiotic) but also another unknown circumstance, present only in our laboratory, is responsible for the observed results. For this reason, it is important to successfully repeat every significant experiment at least in one's own laboratory (ideally it should be confirmed by an independent team of scientists elsewhere).