Distinguishing correlation from causation in statistical research

¶ … Correlation and Causation

If research demonstrates a linear correlation between smoking and higher pulse rates it may be tempting to drawn the conclusion that there is a direct relationship indicating that smoking causes higher pulse rates. However, this conclusion may be erroneous as without further information all the research is showing is a linear correlation. To understand why it cannot be assumed that the data indicates a causation it is necessary to look at the difference between correlation and causation, which can be demonstrated with other examples to increase understanding.

Statistical tests are undertaken to show patterns in data, the tests will be used to determine if the results gathered may have been gathered by chance, in order to determine a if there is a correlation. A correlation means that there appears to be a relationship between the ways the data for two variables moves. In the case of the variables discussed above, there is an appears to be a relationship between smoking rates and pulse rate; with pulse rates and smoking rates rising together so if they were plotted on a graph, they would demonstrate a linear relationship. However, correlation may indicate a commonality between the variables, which may be positive, or in some cases may be negative; when one variable increases another decreases, does not prove causation (Curwin and Slater, 2007).

Causation may in the context of cause and effect, where one event is related to another, because it is creating a specific impact. In the argument that smoking increases heart rate; pulse rate), there is the assumption that the smoking is the independent variable, and the pulse rate is the dependent variable, depending upon whether or not the individual smoking, and the level of smoking that is taking place.

Where there is a causal relationship between two or more variables, there will be a correlation in the data measuring those variables (Curwin and Slater, 2007). However, a correlation alone is not sufficient to prove there is causation. In the case above, a first consideration may be the assumption of the dependent and independent factors; as these have not been fully determined, and cannot be determined from the limited data that is provided. For example, if one assumes that the independent variable is the pulse rate, and smoking is the dependent variable, maybe argued that those with a higher pulse rate are more likely to smoke. Therefore, one potential area is the misidentification factors, and the way in which a causal relationship occurs (Dancey and Reidy, 2007).

More importantly, correlation does not necessarily mean that there is causation (Curwin and Slater, 2007). For example, it is possible that there may be other influences that are impacting on both of the variables, impacting on both variables, giving the appearance of a direct relationship, when instead there is a common factor influencing on both smoking and pulse rate. Using a different example, during the summer months there is an increased level of sunburn as people sunbathe, there is also increased level of consumption of ice cream, and an increased use of suntan lotion. There is likely to be a positive correlation between all these factors, but it does not mean that eating ice cream using suntan lotion will cause sunburn. Instead, there is an unidentified (extraneous) factor which is having an impact on all of the variables; the increase sunshine is causing more people to use suntan lotion, and causing more people to eat ice cream, it is this increase in exposure to sun that is also causing sunburn. When only three variables were considered of sunburn, eating ice cream, suntan lotion; an erroneous result may be assumed due to the lack of full information being considered.