Greetings my brain-box disciples! How’s life going? Fabulous.
This week I will be touching on a wholly dull and atypically uncontroversial topic, correlations!
Having studied statistics for one and a half years I feel you are all qualified to tell me what this graph suggests, take a moment if you need it. Done? Being the kind of learned student who brushes their teeth every morning with more difficult conundrums than this, you will probably scoff at my ignorant younger self. I would’ve taken one look at this and confidently declared that the more lemons the USA had imported, the fewer fatal highway crashes there would be as a result. Thanks past me, not the first time you’ve screwed me over! Many graphs like this exist and are quite funny to look at, apart from one linking an increase in the number of pirates and global warming…never mess with pirate stats.
You no doubt have correctly concluded that correlation does not imply causation. Maybe you’re curious as to how this blog is even worth your time? Am I going to tell you anything vaguely worthwhile? I believe this is useful to society at large, as franky, many people are not aware of the fallacies that surround correlation (not to say my blog is specifically useful to society, but the general publication of factually correct information). I refuse to live in a world where the news print a graph and use it to justify to the unlearned something that is statistically unsound. So you can receive this blog in 3 ways:
- A ineffectual attempt to educate the masses
- A gentle exploration of the usefulness of correlation designs
- A shorter blog than normal to trawl through
Correlational research cannot imply causation as that is not what it is trying to find. Correlations are purely used to see if there is any discernible relationship between variables, not whether A caused B. Two issues arise once you try to venture down that avenue:
Direction – Did A cause B or B cause A? Does a broken family cause depression or does depression cause a broken family?
Third variable issue – Are both variables influenced by a third, unknown variable?
Therefore, because correlations do not give any information in regard to the direction of the relationship or even whether it is an influential relationship, it is an invalid assumption that we can learn anything causal from a correlation.
A second issue is related to the assumed linearity of a correlation, critics have suggested that the Pearson Correlation does not completely characterise the relationship between the two variables. Anscombe demonstrated this with his quartet, a set of scatterplots that all have the same mean, standard deviation, correlation, and regression line. You will note that all of the datasets are completely different and that the numbers have not necessarily explained the relationship.
So what are the values of correlational research? If it can’t tell us anything about the cause, is it a largely redundant statistical endeavour? Stanovich (2007) suggests that firstly, most scientific hypotheses are stated in terms of correlation and are therefore crucial to studies; secondly, that correlation research can indicate the correct direction for research into causation, whether the results of a study produce a strong correlation or not. If we can link two variables together, then we can darn sure design further research to explore that relationship. Likewise if two variables do not correlate then we can take our research elsewhere. So while correlation studies don’t find the root variable themselves, they are a key stepping-stone to finding it.
Will try and rack my brain for something super awesome to blog about next time.
Laterz geek squad.
psychopolemic 