Confounding Variables In Statistical Testing

The idea of science is to use observational analysis to identify the cause of a particular change. For example, scientist may want to test the effects of gravity on the movement of an apple. In order to be able to measure the effects of gravity, it is important that only gravity is able to influence the movement of the apple. If measurements were taken at the same time as the wind is blowing for example, not all of the effect could be attributed to gravity. The wind is therefore a confounding variable, because it has an influence on the outcome, but it is not of primary interest to the researchers. If the effect of the wind is known, and its speed measurable, adjustment for the wind can be made and gravity measured. However, often in science confounding variables are not known, and so results are influenced by unknown confounding variables. Cohort studies very commonly have both known and unknown confounding variables influencing the outcome.

Confounding variables are often a problem in epidemiological research. Epidemiological research is interested in studying populations over time, and as such associations between that population and a particular variable can be found. For example, it might be found that over time, those who consume more flavonoids in their diet have a lower risk of cardiovascular disease. This is interesting but care must be taken in interpreting the results because cause and effect cannot be ascribed. While the flavonoids could be the cause of the protection from cardiovascular disease, another identified variable could easily be the causative factor. In fact, eating high intakes of flavonoids is suggestive of a high fruit and vegetable intake and so the flavonoids could just be a marker for a healthy lifestyle. Controlling for known confounding variables, such as fibre, would lessen the risk of drawing an incorrect conclusion, but does not eliminate it.

RdB confounding variables