Correlation designs

If a study is criticised because it doesn’t show cause and effect it’s probably a correlational study.

An example: For example we could look for a correlation between IQ and performance at GCSE or A-level.  Common sense would perhaps tell us that students that have higher IQs are more likely to perform well at GCSE.

Correlations do not have an IV and DV.  Nothing is manipulated as with an experiment.  A correlational analysis involves the comparison of two co-variables.  As one increases what happens to the other?  Is there some form of association or relationship?

Types of correlation

Positive: The most common; as one variable increases so does the other, e.g. IQ and GCSE score in the example above.

Negative: As one variable increases the other decreases, e.g. it might be fair to assume that the higher your stress levels the lower your life expectancy.  Again we are unable to show cause and effect.  As mentioned frequently in ‘Stress,’ illnesses could be due to secondary habits such as smoking, poor diet etc.

  • Correlations allow us to study links between variables that could not be studied in any other way.  We could not inflict so much stress on a person that we endanger their life.  However, we can use a correlational analysis to show a possible link between the two occurring naturally.
  • If a correlation is found a possible cause and effect relationship can be checked using experimental methods.  No correlation would tend to suggest no such relationship.
  • Economical and fast: large amounts of data can be compared quickly and cheaply, e.g. by using a questionnaire to collect data.
  • Correlations (like experiments) should be easy enough to replicate so findings can be checked for reliability.

Scattergraphs

Correlations are best illustrated using scattergrams. The two co-variables are plotted (one across the x axis and the other up the y – it doesn’t matter which variable goes on which axis).  Any link can immediately be seen. A perfect positive correlation has a coefficient of + 1.0, no correlation has a coefficient of 0. In the real world neither of these extremes usually exists.  Coefficients lie somewhere between 0 and +1.0 for positive correlations and between 0 and -1.0 for negative.  The nearer 1.0 the higher the correlation.  The exam board expect you to be able to estimate a correlation coefficient (see examples below).

Correlation vs Causation

Sales of ice cream are closely correlated to drownings in swimming pools and to the number of shark attacks! According to QI, in the UK, in the 20th century hair length was closely correlated to performance on the Stock Market, as were lengths of women’s skirts! Clearly there is no obvious causal factors in any of these, rather third variables causing both. Even reputable broadcasters report what at first glance may appear to be bona fide relationships such as the correlation between clumsiness in childhood and later, adult obesity.


Brain2

Go Further