Experimental Design, Variables, and Controls

Lesson 2.1 · 13 min

Independent, dependent, and confounding variables

The independent variable is what you deliberately change or compare. The dependent variable is what you measure to see if the independent variable had an effect. Every other factor that could plausibly influence the outcome is a candidate confounding variable, and your job before starting is to list as many of them as you can.

Confounding variables are dangerous precisely because they are invisible when ignored. Temperature, timing of measurement, handling technique, batch of material, and even the order in which samples are processed can all move an outcome and masquerade as an effect of the thing you actually meant to study.

A design that changes more than one variable at a time, without a plan for separating their effects, cannot tell you which change produced which result. This is the single most common structural failure in informal research.

Confounding variable:

A factor other than the one under study that could independently influence the outcome, making cause hard to isolate.

Lesson 2.2 · 13 min

Controls and meaningful comparison

A control condition tells you what the outcome looks like in the absence of the variable you are studying, or under a known reference condition. Without a control, an observed change has no baseline to be judged against, and you cannot rule out that it would have happened anyway.

Different questions call for different controls: a negative control to show what happens with nothing applied, a vehicle control to isolate a carrier's effect from the substance itself, and sometimes a positive control to confirm your measurement system can detect a real effect at all.

A design with no control is not a lesser version of an experiment, it is not an experiment in the comparative sense. It can still produce an observation worth writing down, but it cannot support a claim of cause and effect.

Lesson 2.3 · 12 min

Choosing measurement methods

The measurement method should match the outcome, not the researcher's convenience. A subjective, unblinded self-report is a weak measure for an outcome that could be influenced by expectation. A quantitative instrument reading, recorded on a fixed schedule, is a stronger measure where one is available.

Every measurement method has a resolution limit and a source of error. Knowing what your instrument or method can and cannot distinguish keeps you from reporting more precision, or more certainty, than the method supports.

Decide on the measurement method and the schedule for taking measurements before starting, and hold to that schedule. Measuring more often when things look promising and less often when they do not is a subtle way that bias enters supposedly objective records.