A control in an experiment helps to eliminate variables irrelevant to the experiment. Imagine an experiment in which a researcher tries to test the effect of a new drug on a group of patients. In the real world there are many variables other than the new drug -- for instance, the patient's age, fitness, race, gender, diet -- that can affect the progress of the disease. In a situation like this, the researcher typically has a control group of patients with the same disease and the same characteristics (such as age, fitness, race) as the experimental group. But the control group does not get the new treatment. This way, if an effect is seen in both groups, it is clear that it is not due to the drug. On the other hand, if an effect is seen only in the experimental group, it is likely the result of the new treatment.
Sometimes experiments deliver unexpected results. In cases like these, controls can be designed to tease out the source of error. Suppose the researcher in the experiment described above finds the new drug has no effect at all. Is this a genuine finding, or does it indicate an error in the way the experiment was conducted? The researcher suspects that the way the drug was given is the reason it had no effect. She can then design a control, such that a drug with known effects is given to a group of patients using the suspect method. If the known effects are not observed, it is clear that the way the drug was given is at fault.
There are many advanced statistical methods such as regression, ANOVA and t-tests that can be used to analyze experimental data and the validity of conclusions drawn from them. ANOVA and t-tests can be used to sort out the differences in effect on two separate treatment groups. Regression can be used to determine the relative effects of different independent variables. These methods can only be used, however, when experiments have been performed with sufficient rigor and in a controlled fashion. The analysis is only as good as the data.
The goal of a scientific experiment is often to identify causal relationships between variables. Determine whether a given independent variable directly influences a dependent variable. For example, does the amount of light it is exposed to determine how many inches a plant grows? It is possible to answer this question only if the experimenter has controlled for other factors, such as temperature, amount of water, type of plant and so on. Experimental controls make it possible to narrow down the number of variables and use statistical methods to discover causal relationships between variables.