- standard sample: A control or standard sample is a way of coping simultaneously with abundant uncontrollable or even unknown variables that might otherwise bias the measurements or mask the target relationship. Repeated measurements of this standard indicate how much data variability is generated by other variables. Sometimes one must accept that the observations are relative rather than absolute, because of the biasing effects of uncontrolled variables. Often, however, a suite of standards can allow calibration of the observations. They also can enable comparison to measurements by other investigators.
- control group: Dividing the experiment into two groups can demonstrate the effect of a variable of interest, even when many uncontrollable or unknown variables are present. Use of a control group is standard in social sciences such as psychology, but it may also be useful wherever one must cope with several uncontrolled variables.
The two groups should be as similar as possible, except that the problem variable is missing from the ‘control group’ and present in the ‘experiment group’. Such an experiment is called a controlled experiment. Note that this term does not mean that the experiment is under control (almost all experiments are controlled in that sense), but that it employs experiment and control groups.
The two-group experiment described here is the simplest controlled experiment; often it is not the most efficient experiment. Multivariate experiments using a factorial design permit one to explore the possible effects of several variables and their interactions in one experiment, rather than in sequential experiments. Design of such experiments is described in most statistics books.
An essential ingredient of most controlled experiments is randomization. Random assignment of individual samples to the two groups avoids bias and permits statistical determination of confidence levels for the effect of the variable of interest. For example, drug tests routinely use a controlled experiment with randomization and double blinds: subjects are randomly assigned to receive either the drug or a placebo, and neither the subject nor the drug dispenser knows which type is received.
Understanding the effects of acid rain on lakes has been hampered by the complexity of lake systems and the very high variability among lakes. Thus even when a lake with rising acid levels undergoes ecologic change, it is not possible to establish causality between the two.
Recent experiments in Canada have been able to detect ecologic changes caused by only minor acidification. They demonstrated, for example, that acidification causes a decrease in species diversity without changing total biomass -- an observation consistent with the more general ecologic pattern that environmental extremes affect diversity but not necessarily numbers. The experiments used the following drastic technique: choose a remote region of Canada where lakes are plentiful, select a pair of environmentally similar lakes, make one of each pair a control and deliberately acidify the second lake, then monitor the changes in both [Luoma, 1992].
- randomization: If an experimental design randomly selects samples or randomly matches treatment to samples, then potential biasing effects of uncontrolled variables are converted into random unbiased error that can be averaged out. For example, time often is an influential variable, because instruments may drift or subtle changes may creep into the experimental setup. By randomizing the sequence of sample measurements, the investigator can prevent undetected temporal changes from biasing the result. Randomization is the most powerful tool for dealing with uncontrolled variables;
