Association Versus Causation
At its best, one study can show that an association exists between an exposure and an illness. The statistical test used in health studies can be used to demonstrate an association between an exposure and disease. Association means that there is statistical evidence showing a relationship between exposure and illness. This association may indeed be due to the exposure causing the illness, or it may be due to other factors. Even if a study shows an association between exposure and illness, one study alone cannot prove that the exposure caused the disease. To establish a cause-and-effect relationship between exposure and illness, scientists consider many studies over a period of time and more evidence in addition to studies. This is often called the "weight of the evidence" approach.

Often, when determining causation in environmental epidemiology, scientists are guided by the considerations outlined by the scientist Sir Austin Bradford Hill in 1965. Some of these considerations are required to establish a cause-and-effect relationship; other considerations are not required, but if available would strengthen the case for a cause-and-effect relationship. Although these considerations are well regarded, they may evolve as new knowledge emerges.

What Scientists Call
These Considerations		 Questions Posed By
These Considerations
strength of causation is the disease rate among the exposed group much higher than the disease rate among the unexposed group?
consistency has the same association been witnessed by different researchers in different locations in studies conducted at different times?
specificity is one type of exposure linked to one type of disease? Is the disease is not caused by any other factors?
temporality is the exposure happening before the disease?
biological gradient
or dose-response
relationship		 for different levels of exposure, are there varying degrees of risk for the disease?
What Scientists Call
These Considerations		 Questions Posed By
These Considerations
biological plausibility are there research findings that show evidence of a biological mechanism connecting the exposure to the disease?
coherence are the findings consistent with generally accepted knowledge about natural history and biology of the disease?
experiment have there been experiments to show that reducing or eliminating exposure causes a decrease in disease rate?
analogy are associations between similar types of exposures and diseases already well established?
How Statistics Provide Evidence of an Association
Suppose scientists conduct a study looking at the possible association between an exposure and a health outcome, and then publish the results. How do we know if the results reflect a real association between the exposure and disease, or one that plausibly could have occurred by chance?

After a study has been completed, researchers run a statistical test on the results. A statistical test is used to determine whether an exposure is associated with a disease. A statistical test gives you a p-value for these results. The p-value provides information about how likely chance could have produced such an apparent connection between exposure and disease.

P-value is the probability that the study would mistakenly find an association between the exposure and the health outcome when, in reality, the two are not associated. Scientists generally want to avoid this mistake and decide even before a study is conducted that they want their results to achieve a small p-value (usually 5% or less). If their study results generate a small p-value, then scientists will conclude that an exposure and health outcome are associated.

What are the Limitations of Statistical Tools?
A statistical test can give you the wrong result! Even if there is no real association between exposure and disease, a statistical test may indicate that there is an association. The increase in disease rate in the exposed group compared to the unexposed group may be due to chance. The reverse is also true: Even if there is a real association between exposure and disease, a statistical test may indicate that there is no association. The increase in disease rate among the exposed group compared to the unexposed group may not be large enough to generate a small p-value.

In addition, a statistical test is only as good as the quality of the data collected. Factors that can affect data quality include ability to find target population, exposure misclassification, disease misclassification, and confounding. Furthermore, a small sample size in a study could result in the statistical test showing that there is no association between exposure and disease when there really is such an association.

It is important to note that because of the possibility of these errors, epidemiologists do not rely strictly on these tests. More evidence in addition to the results of a statistical test needed to establish whether or not an exposure has caused the disease.