Why Correlation is not Causation
When two variables correlate - like playing violent video games and having violent thoughts - it seems reasonable to conclude that one variable must be causing the other variable.  However, this is one of the most common flaws in reasoning.  Just because two variables correlate, it does not necessarily mean that one variable is causing another.  Correlation does not imply a cause-and-effect relationship between variables.  To understand why, let's look at a classic example of how variables can correlate with no causal relationship involved.
Think Critically


Last year, there was an alarming number of shark attacks in the seaside town of Surf City, California.  The mayor promised to do something to put a stop to the shark attacks.  To find out what was causing the spike in shark attacks, the mayor turned to the best scientists to study possible causes of the attacks.  After examining hundreds of statistics, the scientists found a stunning correlation:  the number of shart attacks seemed to be strongly linked to ice cream sales.  Whenever ice cream sales went up, so did the number of shark attacks.  The correlation is shown below:



After examining this graph, the mayor decided to ban the sale of ice cream in the town.  "I enjoy ice cream as much as anyone", the mayor said, "but saving lives is more important.  We don't know why ice cream and shark attacks are so strongly correlated, but clearly there is a link.  We need to do everything we can to stop these shark attacks, even if means that we have to ban ice cream."

Do you think banning ice cream sales will reduce the number of shark attacks?
What could explain the strong correlation between ice cream sales and shark attacks?

See if you can solve this puzzle before reading on!
Interpreting Correlations

A positive correlation shows a link between two variables - lets call them X and Y.  However, this does not necessarily mean that X causes Y.  In fact, there are a number of different possibilities.

  • X causes Y - For instance, in the example above, perhaps people who have just eaten ice cream give off a scent that attracts sharks, causing more shark attacks.  So ice cream sales (X) really do cause shark attacks (Y).

  • Y causes X - On the other hand, the relationship may move in the opposite direction.  Perhaps people are very upset after a shark attack, and console themselves by eating ice cream.  So shark attacks (Y) are actually the cause of ice cream sales (X).

  • A third variable, Z, causes both X and Y - Perhaps neither X nor Y cause each other, but there is third variable - let's call it Z - that is responsible for both X and Y.  If you haven't already guessed, this is probably happening in the ice cream example.  The most plausible explanation is that during the hot, summer months, people buy more ice cream, and also tend to go swimming, increasing the number of shark attacks.  Hot weather (Z) causes both ice cream sales (X) and shark attacks (Y). 

The diagram below summarizes the different ways to interpret a correlation:
Think Critically

In the previous lesson, you saw that a study found a positive correlation between playing violent video games and having violent thoughts.

One explanation for this correlation is that playing violent video games (X) cause violent thoughts (Y).  Perhaps all the blood and violence of video games can change how children think about violence in the real world.

Can you think of at least two other explanations for this correlation?

(Hint:  Rather than X being the cause of Y, could Y be the cause of X?  Could there be a third variable, Z, that might cause both X and Y?) 

Scroll to the bottom for answers
A great deal of research in Psychology is correlational.  Unfortunately, many people incorrectly leap to the conclusion that if X and Y are correlated, X must be causing Y.  This can lead to all sorts of bad decisions, like the mayor's decision to ban ice cream sales.  It is very important to be careful in how we interpret correlational findings, and not to jump to any erroneneous conclusions.  Correlation does NOT imply causation.  In fact, finding a correlation should be regarded as a starting point in research, requiring more evidence before a firm explanation is justified.  In fact, if you look hard enough, you can find all sorts of bizarre correlations, as the following examples show:
TOK Link

According to Michael Shermer, "humans are pattern seeking animals, and we are adept at finding patterns, whether they exist or not".  In other words, people are very good at noticing patterns, and creating meaning and explanations for those patterns - even if these explanations may be completely false.  For instance, an athelete who wears red socks one day and then has the best performance of his life may decide to wear red socks for every important match.

Consider the following questions:

1.  Explain why our pattern seeking tendencies may cause people to incorrectly infer that correlation equals causation.

2.  What are the advantages of being a pattern seeking animal?  For instance, how could noticing patterns be useful for our hunter-gatherer ancestors in their struggle to survive?

3.  What are the disadvantages of being a pattern seeking animal?  Think of one example in Psychology and one example in another area of knowledge where being too alert to patterns led to faulty conclusions.

Scroll to the bottom for suggested answers

Correlation vs. Experiments

As you now know, correlational studies have a serious limitation.  Because no variables are manipulated in a correlational study, it is impossible to determine the causal relationship.  However, that does not imply that correlational studies are useless.  In fact, they have several advantages over experiments:

  • Correlational studies are typically easier to carry out than experiments.  Because they simply collect data, they require less time and money.  There is no need to compensate participants for spending hours in a lab, or following up with participants for months or years afterwards.

  • Corrlelational studies can involve a large number of participants.  For instance, it is relatively easy to send out a survey to thousands of people, collecting large quantities of data.

  • Correlational studies avoid many ethical issues, since they do not involve manipulating anyone's behavior. 

On the other hand, experiments typically require a great deal of time and money.  They involve recruiting participants, randomly dividing them into different groups, manipulating an independent variable, and recording a dependent variable.  Thus, participants must volunteer a great deal of time, and must be financially compensated for this.  Some experiments take years to complete, especially when studying effects that take a long time to manifest.  Psychologists must be aware of ethical issues, and refrain from carrying out experiments that cause harm to participants.  Many potentially useful experiments can never be carried out due to ethical concerns. 

However, despite all the difficulties of designing and carring out an experiment, it is the only way to demonstrate a clear, cause-and-effect relationship between variables, and that's why the experiment remains the "gold standard" in Psychological research.




Try it Out

Media reports of scientific research are often guilty of blurring the lines between correlation and causation.  For each of the news articles below,

A) Determine if the research is an experiment or a correlational study

B) If it is an experiment, identify the independent and depedent variable

C) If it is a correlational study, (i) think of two possible explanations for the correlation, (ii) explain how an experiment could be carried out to further investigate the issue


1.   Dogs walked by men are more aggressive  

2.   Sugar rush ... to prison?  Study says lots of candy could lead to violence

3.   Risk of harm from spanking confirmed by analysis of five decades of research


Exam Tip

Whenever you discuss a research study in an exam essay, it is useful to evaluate the research method and explain how the findings of the study should be interpreted.

You should evaluate the following in your discussion:

1,  Is the study an experiment?  If so, you should:

  •  identify the independent and dependent variables, explaining how the                        independent variable was manipulated & how the dependent variable was measured

  •  state that the experiment establishes a causal relationship between the IV and DV

2.  Is the study correlational?  If so, you should:

  • Explain that a correlation between two variables is not enough to establish a               causal relationship

  • Suggest a possible alternative explanation for the findings (for instance, rather 
     than X causing Y, perhaps Y causes X, or a third variable, Z, causes both X and Y

  • Discuss whether an experiment could be carried out to further investigate the 
     findings.  Are there ethical issues that would prevent experimental research?
 


Checklist

  • ​I can explain why correlation does not imply causation

  • I can suggest alternative ways of explaining a correlation (for instance, rather than X causing Y, Y may be causing X, or there may be a hidden third variable)
Quiz Yourself

1.  Which is NOT a reason why a researcher might perform correlational research, rather than carring out an experiment?

(a) Avoiding ethical issues

(b) Having a limited research budget

(c) Reducing the volume of data to be collected

(d) Being able to collect data from a large number of people


2.  Over the last 10 years, the use of smartphones has sharply increased,  and so has the percentage of teens who are obese.  What is NOT a possible explanation for this finding?

(a) Teens who spend more time on their smartphones engage in less physical activity, leading to obesity.

(b) Obese teens find it difficult to perform physical activity, so end up spending more time on their smartphones.

(c) Better technology has led to more smartphone use, as apps are more engaging and useful.  Over the same time, obesity as also increased, but for completely unrelated reasons such as more effective food marketing.

(d) All of the above are plausible explanations



3.  Residents of Italy, Greece and Crete traditionally ate a "Mediterranean diet" high in olive oil, nuts, fish, fruits and vegetables.  Rates of heart disease were much lower in people who had traditionally followed this diet.  Based on this data, which conclusion is justified?

(a) Americans should switch to a Meditteranean diet to improve their health

(b) The Mediterranean diet is healthier than a diet high in processed foods

(c) Salad is healthier with olive oil instead of creamy, high-fat dressing

(d) A number of factors may be responsible for the lower heart disease in the sample


4.  A study has found that married people tend to report higher life satisfaction than single people.  Which of the following explanations involves a third variable being responsible for both higher life satisfaction and marriage?

(a) Better educated people are more likely to earn a high income, which is linked to greater life satisfaction, and also more likely to be married.

(b) Marriage involves emotional support and intimacy, leading to greater life satisfaction.

(c) People who are happier are more likely to get married

(d) People who have lower life satisfaction are more likely to get divorced, skewing the data towards happy people who are still married


Answers

Think Critically

Perhaps children who have more violent thoughts (Y) also like playing violent video games (X).  Or, perhaps, children with violent parents (Z) are more likely to allow their childrent to play violent games (X), and also influence their children to have more violent thoughts (Y).

TOK Link

1.  Being sensitive to patterns can lead to perceiving causal relationships where none actually exist.  For instance, if I wear red socks to my exam and get an A+, I might think that red is my lucky color.  

2.  Noticing patterns could have helped our ancestors survive by staying clear of harm.  If you ate some wild berries and then became very sick, it would be beneficial to see a pattern between eating berries and illness, and avoid those berries in the future.  Even though the illness might be completely unrelated to the berries, it pays to "err on the side of caution".  That is perhaps why evolution has made us very sensitive to finding patterns, even when they do not actually exist.

3.  The human tendency to notice patterns can lead to false conclusions.  For instance, some parents refuse to allow their children to be vaccinated, seeing an imaginary link between vaccinations and autism.  Gamblers often incorrectly beleive that chance events will correct themselves, so a string of Heads on a coin toss will be more likely followed by a Tails.  In the past, many doctors advocated a low-fat diet, even though nowadays some types of fat are considered healthy in moderation.

Try it Out

1.  This is a correlational study.  One explanation is that men are more aggressive, and that dogs learn this aggression from their owners.  However, another explanation could be that men tend to own more aggressive breeds of dogs.  Or, perhaps women are more fearful of walking dogs that behave aggressively, leaving it to men to be the walkers of aggressive dogs.  An experiment could be carried out by randomly assigning dogs to be walked be either men or women, and then recording the level of the dog's aggression on the walk.

2.  This is a correlational study.  One explanation is that children become addicted to sugar, and this sugar high leads to violence.  Alternatively, perhaps children who eat lots of candy do so because they have less self-control and behave more impulsively, which can also lead to violent acts.  Or, perhaps children who eat lots of candy tend to have parents who don't know how to raise them properly.  An experiment could be carried out by randomly assigning children to eat a low-candy or a high-candy diet, and then monitor them for violence.  This would be difficult to carry out for ethical and practical reasons.

3.  This is a correlational study.  One explanation is that spanking children leads to anti-social behavior and mental health problems, because of the trauma of the spanking itself.  Alternatively, perhaps children who frequently misbehave are more likely to be spanked, and frequent misbehavior is an early sign of later anti-social behavior and mental health problems.  Or, perhaps parents who have poorer mental health are more likely to spank their children, and also pass on their mental health problems to their children through their genes.  An experiment would require randomly assigning children to be spanked by their parents, which would be impossible to carry out for ethical reasons.

Quiz Yourself

1 - C, 2 - D, 3 - D, 4 - A