Let me start this answer with a scenario: I want to study what type of substrate the American cockroach prefers to stand on when given a choice of 6 different surfaces. This question might be simply an academic exercise, or it may be part of a research program on understanding the habits of American cockroaches in order to control them better. For instance, a pest extermination company may wish to know what surfaces these cockroaches prefer in order to build the best lures or traps possible, or what sorts of material these animals are likely to congregate on.

If I take a typical rectangular classroom, with a door in one corner and windows on one side, as the research space in a science building, that will serve as the space for the experiment. Then I put ten containers in the room; I space them apart equally and put a patchwork of each of the 6 substrates randomly ordered in the bottom of each container. I can then put ten cameras over the containers and monitor the movements (the choices) of 10 animals in the 10 containers.

So I have done everything correctly so far! The animals are 1 to a container so they do not interfere with each other. The containers are equally far apart, the substrates (say: glass, astroturf, low carpet, styrofoam as examples) are randomized in the containers.

But what about those extraneous variables - let me name a few:

- lighting: where are the lights in the room? Where is the shade? Will the animals seek the darkest areas regardless of the substrate type?

- windows: same as lighting

- air vents: what is the airflow in the room? do the animals prefer warmer or cooler temperatures, are they sensitive to moving air versus still air?

- vibration: is the building near a road, is one side of the room more shakey than the other due to vehicles driving by? What about people walking down the hallway next to the room? Is there an HVAC control unit or fan in the ceiling?

- magnetic fields: one building I worked in had Chemistry on the floor above and they ran MRI machines as part of their research. So the entire building experienced strong, intermittent magnetic fields

- slope: are the building, floor, and container floor level? Animals may be sensitive to slope more than humans.

- repeated trials: if I run more than one animal in a container I should clean off the substrate due to chemical cues left by the previous animal, and clean with a material that will remove the types of substances these animals leave behind

- substrate color: I am testing substrate preference and this is presumably texture, but if the materials I have used are different colors or even light vs dark are the animals responding to that more than substrate texture?

As you can see, there are many possible variables that could impact the choices the animals make. And there are more: possible inbreeding of the cockroach colony, age of the cockroaches, water and food needs of the cockroaches and so one.

So what is a person to do? My view is a strong experimental design. And in this case, that means as many replicates as possible and as many randomization steps as practical in the experimental design. Operationally this means:

1. Doing a pilot study to streamline materials and methods. Pilot data should also be used to generate the statistical power of the study. From this value, an equation can be used to generate the sample size needed to obtain a power of 0.7 or 0.8 which is considered high.

2. Obtaining the number of animals needed for the study

3. Randomizing how the animals are taken from their holding area

4. Randomizing where animals are placed in the experiment containers

5. Randomizing the direction the animals are facing when placed in the containers

6. Randomizing what order animals are placed in containers as how a person moves through the room may have an impact

Finally, but still before the study begins, determine the rules that will be followed during the experiment for all possible outcomes. For instance, what if an animal is standing 1/2 on one substrate and 1/2 on another? What if an animal does not move during the study period? What if an animal escapes the container? What if an animal dies during the study period? What will be the data notation in each of these instances? These rules should be written down and not changed before the study begins.

If these steps are followed, the impact of variables that were not considered or measured during the study should be canceled out. The value and power of randomization is that randomization spreads the impact of variables that were not measured across possible outcomes (or experimental groups in studies where there are multiple treatments). For instance, if the animals seek out dark places, the randomization of substrate types in the containers, the random placement of animals in the containers, and the randomization of the direction the animals are facing when placed in the containers should all help distribute the impact of that effect across substate types. And indeed the American cockroach is negatively phototactic (avoids light), so this will be an issue in this hypothetical experiment.

I think the best way to mitigate extraneous variables is to have robust and multiple randomization steps built into the experimetnal design. I hope this helps!