Can correlations show causality?

It is a well known fact that correlations indicate a predictive relationship between two variables. For example, you can show a correlation about how the sales of ice cream positively correlates with the increase in weather temperature. Many of you will be familiar with the conventional saying that correlations do not imply causality. However, is this statement solely correct?

The long-term stigma that correlations cannot infer a causal relationship between two variables is largely supported. However, that is not to say that correlations cannot indicate the potential of a causal relationship. Take for instance the above example about ice cream sales. We cannot directly suggest from a correlation that hot weather causes people to buy more ice cream. However, one could strongly predict from the correlation that hot weather could cause an increase in ice cream sales.

However, the problem with this is that causal relationships may be underlying, indirect or unknown. What’s more, high correlations have a tendency to overlap with identity relations, where no actual causation exists. I will concede however that some variable relationships can be causally transparent, for example the relationship between male teenagers’ age and the amount of facial hair they have. One would assume that the older the boys get, the more facial hair they will have. On the other hand, other relationships are not so clear cut. Take for example if you were looking at the relationship between gender and depression. It is much harder to draw a logically prediction from these two variables. To predict a causal relationship between these variables is ambitious at best.

As a result, I do not believe that it can be conclusively stated that correlations show causality, this is because establishing a correlation between two variables is not enough to suggest that a causal process exists. However, correlations can, on occasions, indicate potential causal relationships between two variables but this effect is inconsistent across all correlations.

Advertisements

6 responses to this post.

  1. Hi there!

    I enjoyed reading your blog. I thought what you could have added in your blog, is the problem the public can face when researchers publish their results and try and trick the public. For example, sometimes the researcher finds a strong correlation and reports in their results that there is a significant effect say to r= 0.8. However this doesn’t show that there is a significant effect, or the direction in which one variable, may affect the other. A positive correlation simply tells us that as one variable increased the other variable increased.

    Here is a current example: The sun published an article stating that parents who smoke cause their teenagers to be delinquent. However the survey into which the researchers looked at crime rates and parental smoking only gave a correlation. It could be that delinquent behavior in the teenagers, caused the parents to be stressed which in turn caused them to smoke. Here is a link to the article: http://www.criticalthinking.org.uk/smokingparents/
    Further more, because all other variables such as economic status, family size, age and number of parents was not controlled, you cannot conclude that delinquent behavior was solely down to parental smoking. It could be that it is down to a number of variables such as those from single parent families or those with teenage parents are more likely to have delinquent teenagers.

    I know you touch upon this in your conclusion but you could argue that if a research study was to find a strong correlation, this could lead researchers then to conduct a study in which they try and find the cause and effect. For example, you find a strong correlation between increased exercise and losing weight. A researcher could then conduct a study where one group (the experimental group) is given the same diet and sleep (control other variables which may contribute to weight loss) and 7 hours of exercise a week. The control group is treated the same as the experimental, so they have the same amount of sleep and same diet, however they do no exercise a week. After say a total of 8 weeks, the weights of each participant could be collected in and if there was a significant difference between the experimental group and control, you could then conclude that by exercising 7 hours a week, it causes weight loss. You can know see how an original correlation led to finding a cause and effect relationship, however the original correlation does not show causation as other variables were not controlled.

    Reply

  2. correlation predicts two variables true.but does not show casuality,but it indicate causal relationship between two variables like people buying icecream in relation to hot weather which comes with dehydration and thirst to balance body electrolytes lost but the effect of correllation on causality is inconsistent across all correllated variables.

    Reply

  3. I found your take on this subject interesting. I have read so many blogs that simply talk about how correlation can never show causality and that’s it. I like that you point out that though correlation doesn’t SHOW causation, it can indicate possible causation. If it was always so clear cut that correlation doesn’t show causation (other than the fact we are constantly told that) then the question wouldn’t be raised.

    Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: