Experimentation and variables

In many experiments, the aim is to provide evidence for causality. If x causes y, we expect, repeatably, to find that a change in x results in a change in x. Hence, the ideal experiment of this kind involves measurement of x, the dependent (measured) variable, at one or more values of x, the independent variable, and subsequent demonstration of some relationship between them. Experiments therefore involve comparisons of the results of treatments - changes in the independent variable as applied to an experimental subject. The change is engineered by the experimenter under controlled conditions. Experimental subjects given the same treatment are known as replicates.

Interpretation of experiments is seldom clear-cut because uncontrolled variables always change when treatments are given.

Confounding variables
These increase or decrease systematically as the independent variable increases or decreases. Their effects are known as systematic variation. This form of variation can be disentangled from that caused directly by treatments by incorporating appropriate controls in the experiment. A control is really just another treatment where a potentially confounding variable is adjusted so that its effects, if any, can be taken into account. The results from a control may therefore allow an alternative hypothesis to be rejected. There are often many potential controls for any experiment.

The consequence of systematic variation is that you can never be certain that the treatment, and the treatment alone, has caused an observed result. By careful design, you can, however, 'minimize the uncertainty' involved in your conclusion. Methods available include:

• Ensuring, through experimental design, that the independent variable is the only major factor that changes in any treatment.
• Incorporating appropriate controls to show that potential confounding variables have little or no effect.
• Selecting experimental subjects randomly to cancel out systematic variation arising from biased selection.
• Matching or pairing individuals among treatments so that differences in response due to their initial status are eliminated.
• Arranging subjects and treatments randomly so that responses to systematic differences in conditions do not influence the results.
• Ensuring that experimental conditions are uniform so that responses to systematic differences in conditions are minimized.

Nuisance variables
These are uncontrolled variables which cause differences in the value of y independently of the value of x, resulting in random variation. Nuisance variables are not common in chemistry except where molecules from natural sources are used. To reduce and assess the consequences of nuisance variables:

• incorporate replicates to allow random variation to be quantified;
• choose experimental subjects that are as similar as possible;
• control random fluctuations in environmental conditions.