Correlational Research
Correlational research examines the degree to which two variables have a relationship to one another. In examining correlational research we attend to two factors: strength and direction.
The correlation coefficient tells us the strength of the relationship between two variables. The strength of the relationship is independent of whether the coefficient is positive or negative. The closer the coefficient is to 0, the weaker the relationship between the two variables, with a coefficient of 0 indicating no relationship. The closer the coefficient is to 1, the stronger the relationship between the two variables. A coefficient of 1 would indicate that two variables are perfectly correlated. For example, a correlation coefficient of -.92 is stronger than one of +.41.
If two variables are positively correlated, we would expect to see an increase in one variable as another variable increases. For instance, height and weight tend to be positively correlated. Taller people tend to weigh more, and vice-versa. If two variables are negatively correlated, we expect to see a decrease in one variable as the other increases. For example, this study found that time spent watching television and life expectancy are negatively correlated. The more time spent watching television, the lower we expect life expectancy to be.
While correlational research can illuminate relationships between two variables, correlation does not necessarily indicate a causal relationship. To determine causality we need to use an experimental design. In part, this is because confounding variables might give us the appearance of a relationship where none exists. For instance, if the number of televisions per household is positively correlated to the life expectancy when looking at data by country, it might be tempting to conclude that more televisions = longer life expectancy. This would be an illusory correlation. In this case, the relationship has a confounding variable, standard of living, which would allow for access to both consumer electronics and quality health care.
The correlation coefficient tells us the strength of the relationship between two variables. The strength of the relationship is independent of whether the coefficient is positive or negative. The closer the coefficient is to 0, the weaker the relationship between the two variables, with a coefficient of 0 indicating no relationship. The closer the coefficient is to 1, the stronger the relationship between the two variables. A coefficient of 1 would indicate that two variables are perfectly correlated. For example, a correlation coefficient of -.92 is stronger than one of +.41.
If two variables are positively correlated, we would expect to see an increase in one variable as another variable increases. For instance, height and weight tend to be positively correlated. Taller people tend to weigh more, and vice-versa. If two variables are negatively correlated, we expect to see a decrease in one variable as the other increases. For example, this study found that time spent watching television and life expectancy are negatively correlated. The more time spent watching television, the lower we expect life expectancy to be.
While correlational research can illuminate relationships between two variables, correlation does not necessarily indicate a causal relationship. To determine causality we need to use an experimental design. In part, this is because confounding variables might give us the appearance of a relationship where none exists. For instance, if the number of televisions per household is positively correlated to the life expectancy when looking at data by country, it might be tempting to conclude that more televisions = longer life expectancy. This would be an illusory correlation. In this case, the relationship has a confounding variable, standard of living, which would allow for access to both consumer electronics and quality health care.
Participant Selection |
Does Football Cause Traumatic Brain Injury? |
|
Random Selection
One of the keys to a correlational study is to ensure that data from the study is representative of the population being studied. For our purposes, a population consists of the totality of the group from which we can draw samples. For instance, if we were to conduct a correlational study of survey data from Mountain View High School students, the population for our study would be students at Mountain View High School. Because it is often impractical to be able to collect data from everyone in a population, studies often seek to use representative samples. Random selection means that anyone in the population has an equal chance of being selected to be part of a sample. Random selection is important because we want to minimize other potential factors that might skew our data and give us an incomplete or inaccurate view of a relationship. This doesn't mean that we can't use and learn from other sampling methods, it simply means that studies using those methods may not be as convincing as those using random selection. |
Below are links to two treatments of the same study. One is the study abstract from an article in the Journal of the American Medical Association (JAMA), one of the premier medical journals in the United States. The other is from a piece about the study published in the New York Times. One of the study's key findings is that of 111 brains of former NFL players that were examined, 110 of those brains had evidence of a degenerative disease linked to repeated blows to the head. Taken by itself, this would seem to be pretty strong evidence that playing in the NFL leads to degenerative brain disease. While that may be the case, as both of the pieces below point out, we can not make that conclusion from the data that we have.
First, correlation does not imply causation. That is, just because two things happened together (individuals played in the NFL and had traumatic brain injury) does not by itself tell us that one led to the other. Of course, this data isn't useless either. The fact that the correlation is so strong means that other studies might seek to confirm or add further evidence to the link. Second, the sample used was a convenience sample. In this instance, the players whose brains were examined were those whose families had donated their brains after death to be examined. Players who exhibited symptoms consistent with brain injury might be more likely to have had their brains donated. Again, this doesn't nullify the study - but it does give us yet another factor to consider if we try to draw conclusions from this study. So what good is this study if we can't draw hard conclusions from it? The collection of data such as we saw in this study is useful to direct future research. Researchers can use this data to form and refine hypotheses that can help them focus on what to look for and where. Clinicopathological Evaluation of Chronic Traumatic Encephalopathy in Players of American Football - JAMA 110 NFL Brains - New York Times |
