Designing the Experiment

Once you have a research question and a clear and testable a priori hypothesis, you can design an experiment to test it.


Ideally, experiments should be conducted in such a way that the experimenter has control over every variable that might have an influence on your results (in reality this is much harder than it sounds!). A variable is any factor that might affect the outcome of the experiment. The experimenter therefore manipulates the independent variable and observes the effects of this manipulation on the dependent (or response) variable.

For example, if the goal is to determine the effects of temperature on plant height after 14 days, then height is the dependent variable because it "depends" on the temperature to which the plant is exposed. All other variables must be controlled or held constant, to the extent possible. Temperature is the independent variable because that is the variable which is manipulated by the experimenter.

The ideal way to perform such an experiment is to arrange a set of tests that are identical in all ways (light, soil moisture, etc.) except for the one specific factor that is being tested (in this case, temperature). Thus, for our plant experiment, a greenhouse with multiple temperature-control chambers would be ideal; each chamber would host a different temperature "treatment group". It's worth noting here that we don't always have the option to do what's ideal, but that doesn't mean we shouldn't work to control as much as possible. We should also consider carefully how the things we didn't or couldn't control might be affecting the results of our experiment.


Crucially, one of the treatment groups must serve as a control group against which all other treatment groups are compared. The importance of the control group cannot be over emphasized. It is essential to know how the system you are investigating works under normal circumstances (i.e., before you started messing with it), before you can be sure the results obtained from the experimentation are actually due to the manipulation of the independent variable(s).

To continue our example above, if you wanted to investigate the effects of temperature (the independent variable) on the height of plants after 14 days (the dependent variable), you would measure the heights of plants grown at their normal, expected temperature (most likely room temperature) as the control group, and then compare the data collected from this group to the heights of plants exposed to higher and / or lower temperatures, depending on what you're hoping to learn. The control group would provide the "normal standard" against which the other treatment groups would be compared.

Sample Size

Another important rule governing experimentation is that each treatment group (which includes the control group) should include a decent number of individual test subjects or "replicates" (and ideally an equal number of individuals in each group). The more replicates you include in your experiment, or in other words the larger your "sample size" (indicated by the letter n) per treatment group, the more confident you would be in your results and the more power your study has. However, in most situations, increasing the number of replicates increases the cost and / or logistical difficulty of the experiment.

The key is to have a sufficient number of replicates per group to ensure your experiment has the power to detect meaningful treatment effects (if they exist). Determining what the minimum sample size per group should be is beyond our scope here, but for our purposes you can assume that three is the bare minimum, and ten or more is desirable.

Variation & Random Assignment

Biological variation is the inherent differences among organisms in a study that arise due to differences in genetic makeup, age, sex, health, etc. This natural variation has the potential to obscure or confuse experimental treatment effects. Thus, it is important to attempt to minimize this variation when designing your experiment (e.g., by using organisms of the same age, sex, etc.). Even when potential sources of variation are accounted for, it is crucial that subjects be randomly assigned to the treatment groups, so that any inherent variation among individuals will be distributed at random among all treatments, including the control group.