Introduction to Experimental Design

Photo from RMBL Archives
Scientific discovery results from a variety of experiences and processes. Making observations, forming hypotheses, testing hypotheses with experiments, interpreting results, and forming conclusions are a few important components of the scientific process. For example, one might observe the effect of humans on wolf behavior and conduct simple experiments to obtain data to answer questions.
Testable Questions
The very first step in experimental design is to come up with a clear question! Without a clear question (or hypothesis) how will you know which data to collect? To get you thinking about how this process might work, perhaps you wonder if humans affect wolf behavior. One testable hypothesis is that wolves avoid humans – wolves will be observed less frequently in the presence of humans than in the absence of humans.



Photo from RMBL Archives

Experiments are one tool used by scientists to test hypotheses. Experimental design is a critical step in planning experiments so that the data obtained can be analyzed to yield valid and objective conclusions. Good experimental design requires laying out the details of an experiment in advance of collecting data. Well designed experiments maximize the amount of information that can be obtained for a given amount of experimental effort. When designing experiments, we need to think about at least 4 things:

Back to our interest in the effect of humans on wolf behavior… We might design an experiment where we introduce a fake human (e.g. scarecrow) and measure the impact of its presence on wolf behavior. Our unit of replication (data point) is one observation of a specific wolf behavior. Our sample size is the number of wolves observed during the study. The variable is the specific wolf behavior (e.g. time spent hunting or time it takes before the wolf leaves the area). The independent variable is the presence of a fake human versus a control animal (e.g. nonthreatening animal, like a stuffed teddy bear).

Designing a good experimental control can be difficult. In the example above, the problem with a teddy bear control is that it is small, unlike a human. Maybe we could use stuffed deer – they are closer in size to a human. But that too would be a potential problem because wolves could be attracted to stuffed deer.

What else could be used as a control? To better understand experimental design, explore illustrated versions of the published methods sections from two RMBL research projects in Case Study #1 (marmot behavior in the presence of humans) and Case Study #2 (mayfly behavior in the presence of trout chemicals).


Many scientific advancements are made by rejecting hypotheses – nothing is ever proven in science. Statistics enable you to evaluate the results of your experiment. Are your results real or the result of “noise” or random variation in the system? Many different kinds of statistical tests can be performed (e.g. t-tests, ANOVA, etc.) depending on the type of data you have and the question(s) you wish to answer. Statistical analysis can be simple or very complex.

Check out Dr. Dan Blumstein’s Quick-and-Dirty Guide to Statistics for a user-friendly overview of major concepts.

Correlation vs. Causation

Be careful not to confuse correlation and causation in your statistical analyses. Check out this humorous illustration from Vali Chanrasekaran at Bloomberg Businessweek.

Next step – learn about RMBL experiments in animal behavior through Case Study #1 (marmots) or Case Study #2 (mayflies).


Thanks to undergraduate RMBL researcher Rebecca Batzel for her work on this section and to Dr. Blumstein and Dr. Peckarsky for their reviews.