Correlation
And yet again I have to find from somewhere an ‘interesting’ blog topic to write about so this weeks chosen gauntlet is correlation. Enjoy!
Correlation best describes the form of data that is produced from observational research. Observational research is where no variables are altered (except in controlled observations, but these are pretty much field experiments, but I digress) and individuals are just watched doing as they are doing to get real data, not something that has been ‘grown’ in a lab. After observational data you are left with correlational data that simply tells you that A could predict B and vice versa. There is no causality in correlational data as no manipulation of variables has taken place so you cannot tell whether it was variable A that predicts B or whether variable C is actually a better predictor. This leads us to our first disadvantage of correlation.
Without causality, correlation can only really tell us that there may be a relationship. Also the relationship shown on a scatter graph may mean nothing as the data could be completely down to chance and we would not know. An example of this (although silly) is as follows: A correlational data set shows that large quantitites of chocolate eaten (A) directly influences obesity (B). However the data set does not show that the people who ate large quantities of chocolate studied in this procedure were also incredibly heavy drinkers or ate other high fat foods. However the correlation data set shows a near perfect relationship between chocolate and obesity without looking at all the other factors. But I am being very nitpicky! There are some positives to correlational data.
The major positive is that correlational data steers future research. Observational and correlational research is a cheaper way to explore whether or not an area of research is worth studying. In the same breath I would like to add that even though correlational data cannot show causality it CAN show when there is no causality. If a correlational data set comes back show little to no relationship, then you know as a researcher that there is not much point in continuing with this research. This can save time and money for researchers.
To conclude although correlation cannot show causality or true relationships between variables, it can give us a strong indicator as to whether there will be a relationship. Correlation is a cheap alternative to running a fully blown experiment only then to find out that there was no relationship after all. Same time next week (and another week)!
These websites are some great (but simple) resources for correlation:
http://www.intropsych.com/ch01_psychology_and_science/correlation_and_prediction.html
http://www.experiment-resources.com/correlation-and-causation.html
8 responses to “Correlation”
Trackbacks / Pingbacks
- - February 20, 2012
- - February 22, 2012
- - February 22, 2012
- - March 10, 2012
psud6d 
I liked the point you make when you say that correlations steer future research because I myself think this is a very valid point, because surely if there is a relationship between two things then surely it would spur people on to find a common factor or a cause. Even though you say that a correlation can show no cause and save the researcher time and money, this may be an experimental error and surely the researcher will want to spend time making sure that they got correct results, which may cost them even more money and time. I really enjoyed reading your blog!!
I enjoyed your argument this week 🙂 as you say correlation helps to point us in the right direction even if they do not show causation. For example a study conducted by McNeal and Cimbolic (1986)* found a positive correlational relationship between depression and low serotonin levels, which actually turned out to be a very important contributor to a more effective treatment of depression and depressive symptoms. Researchers were then able to introduce antidepressants (like Selective Serotonin Reuptake Inhibitors (SSRI’s)) that help to relieve the dreadful depressive symptoms by increasing serotonin in the brain. Without correlational studies like this we may not have made the link between the two!
However, on the other hand research has highlighted the more negative side to correlation. Gentile and Anderson (2003)** found that there was a positive correlation between playing video games of a violent nature and levels of aggression and violence in children. So we can see that there is a relationship between the two variables, but… How do we know that the violent video games are responsible for the aggression we see in the children who play them? Well, we don’t as we can not say that one thing definitely causes the other. The bi-directional model shows us that violent video games and aggression levels are linked we can not be sure which way the relationship goes. Are the video games responsible for the aggression or do aggressive children tend to choose to play violent video games?
Coreelation is one of those things, it’s good and bad.. We don’t want to accept everything it says without question but at the same time we should completely disregard the contributions it has made towards psychology 🙂
*McNeal, E. T. and Cimbolic, P. (1986) Antidepressants and biochemical theories of depression
** Gentile, D.A. and Anderson, C.A. (2003). Violent video games: the newest media violence hazard
Hey great blog It strange how many blogs I’ve written and read about correlation and yet it has never occurred to me that they can show no causation. In an ideal world we wouldn’t do correlation studies because of their lack of ability to show causation but then this isn’t an ideal world and correlation are very cost effective and are extremely easy to run. The biggest problem as far as I am concerned with correlation is the lack of qualitative it produces. I believe that qualitative data produces far better quality data than quantitative data which is much more frequently used in research today.
Having seen all the ways correlation can’t show causation, I found it interesting to find out that a researcher, Kenny (1979), has set out a strict set of circumstances in which correlation can in fact indicate causation.
They are: time precedence, relationship and nonspuriousness. In time precedence, the cause of the behaviour etc. must precede the effect (i.e. treatment). Relationship usually requires the use of statistical tests, as ‘relationship’ refers to the possibility the results could have occurred by chance, which is hard to judge without robust measures. Finally, in nonspuriousness, there must not be an external factor present which can account for the link between the two variables.
Kenny, D. (1979). Correlation and Causality. New York: John Wiley.