Saturday, April 18, 2009

On Causality and Correlation in Economics

Causality is perhaps the most fundamental element of empirical evidence available to economists. However, it is also the source of many misconceptions due to its elusive nature.




There are some things we take for granted. The relationship between cause and effect is so deeply rooted in us we sometimes forget they are merely a result of a clever induction made through observation on the repetitive nature of the two.

Causality is a fundamental corner stone to any empirical science. In economics causality is the basis for models and theories observed ranging from basic supply and demand curves to the most sophisticated economic models around.

The social science of Economics is more formal than other social sciences in the sense it includes theoretical models which are not directly dependent on correlations but rather on assumptions of maximization of utility at their basis. Nevertheless, at the end, utility maximization and its behavioral associations are bound to end include causality as a central assumptions of the coherency of nature.

More intricate relationships between various economic indicators such as interest rates, inflation, expectations, government spending, private consumption and many others are based on observations on different correlations between these parameters over time.


The problem of induction


David Hume is a famous Scottish philosopher known for its strict empiricism (A theory in philosophy which asserts all knowledge arises from experience rather than innate ideas and reason). The following may clear up this issue of the sources of knowledge a bit more.

David Hume, being the strict empiricist that he was, pointed out the problem with the logical problem embedded within the inductions we make based on experience. To put it simply, any attempt at proving our inductions to be correct will ultimately be based on induction itself and therefore cannot be constituted a logical reasoning. The argument would look something like:

  • Every time I've dropped an object it fell to the ground

  • Hidden statement: Nature is coherent/consistent or "that which has been is that which will be".

  • Therefore, when dropped, all objects fall to the ground.

To justify my hidden statement I must turn to the same statement assuming that if nature has been consistent so far it will continue to be.

In Hume's eyes there is no logical justification to assume the sun will rise tomorrow simply because it had done so in the past.

Needless to say induction is crucial to science and the problem of induction presented by Hume has received great attention from philosophers ever since. Karl Popper, a famous philosopher of science (among others) suggested a clever way out of this problem by stating science will advance through efforts of disproving existing theories thus strengthening the correct ones and correcting the false ones. For example, the argument that all swans are white is correct until a black swan will be spotted.


Causality and Correlation


David Hume also took a swing at Causality or Causation following his treatment of induction. Hume stays committed to his reasoning all the way to the skeptic conclusion which is unavoidable according to which causality as a concept has no meaning as we cannot conceive the connection between cause and effect in any means available to us (I will explain shortly).

"…It appears that, in single instances of the operation of bodies, we never can, by our utmost scrutiny, discover anything but one event following another, without being able to comprehend any force or power by which the cause operates, or any connection between it and its supposed effect… All events seem entirely loose and separate. One event follows another; but we never can observe any tie between them. They seemed conjoined, but never connected. And as we can have no idea of anything which never appeared to our outward sense or inward sentiment, the necessary conclusion seems to be that we have no idea of connections or force at all, and that these words are absolutely without meaning, when employed either in philosophical reasoning or common life".

Trying to explain cause and effect in terms of experience is bound to lead us to Hume's reasoning. The connection between cause and effect can be described as either proximity in time, proximity in space, an obligatory connection or any combination of these. Still there will always be examples of events which answer to all these criteria and are not cause and effect. Hume gives us the example of the rooster that believes he bring out the sun with his cry every morning.

In order to lead our lives we all assume causality exists and will hopefully continue to govern our structured world. Many false theories however, are the result of assuming causality where only correlation exists. In economics and other social sciences the risk of erring in this way are very significant.

From the explanation above and the rooster example the difference between correlation and causality should be clear by now. The fact two phenomenon display certain similarities in their behavior, either in time, space or a connection does not mean they are, indeed, cause and effect. They may be simply correlated.


Correlation


Correlation is an important statistical figure which indicates the strength and direction of a relationship between two phenomenon or variables. For example, the price of wheat and bread are correlated. Putting the philosophical discussion aside, we believe these prices are correlated because the price of wheat serves as a cause for the price of bread. However, the prices of bread may be correlated with the daily yield on the Iranian government bond. There is no justification to assume causality here.

No less dangerous is assuming, perhaps, that the prices of bread influence the price of wheat. Getting the direction of the correlation wrong is a very easy mistake. Especially when we are researching variables we don't know much about.


Rudi Giuliani's fight against crime


A good example of the problems with assuming causality and using correlation is that of the significant decrease in crime in New York in the 1990's. The crime rate in New York had dropped significantly seemingly due to Mayor Giuliani's strict zero tolerance policy and the "broken window" thesis.

Researches today argue as to the real cause behind the decrease in crime rates in 1990's NY. Apparently some researches today claim Giuliani's success had merely been coincidental and that crime rates all over the country were dropping due to the economic growth in the 1990's and the decrease in the relative share of certain age-groups in the population. Several cities across the US have shown even more dramatic decrease in crime rates whit out Giuliani's fight against crime.

In their fascinating book Freakonomics (which is a must for any economics enthusiast) Steven Levitt and Stephen Dubner are presenting a very unconventional approach to economics and have presented a theory correlating legalization of abortions in America with the famous decrease in crime presented above. According to Levitt and Dubner the legalization of abortions has led to fewer unwanted births and to a decrease of the number of youths in the population exactly in the 1990's which in turn led to less crime and Giuliani's credit for the fight.
Although I agree with Giuliani's "broken window" thesis there is no real way to know what the cause behind the effect was.

This example serves to show how difficult and confusing it can be to prove causality between two variables and to incorporate that into any successful theory.


Conclusion


As always, all theories, especially when it comes to the social sciences need to be taken with a grain, or a whole bag, of salt. Adopting a skeptic attitude towards surprising findings will quickly sharpen your instincts and ability to tell reasonable from fantasy.

Understanding the problems with induction and causality and also the difference between correlation and causality is, perhaps, one of the most important aspects in leading a rational life. It will also help you quickly turn the page on that recent study which shows the amazing relationship between children who ate peanut butter sandwiches and their success in business later in life.


Related posts:

Image by: Dawnzy58

5 comments:

zzboy said...
This comment has been removed by a blog administrator.
Chris said...

Great post! People not understanding the difference between causality and correlation is one of my pet peeves. :)

marvin chester said...

The kernel of Karl Popper's contribution to the philosophy of science is this idea: To be a valid scientific theory its assertions must be 'refutable' or 'falsifiable'. i.e. There must exist experiments which, if verified, would refute the theory. In economics one doesn't do experiments. Neither does one do experiments in astronomy but there exist events, if observed, that would refute theories in astronomy. Some theories in economics are, in fact, not refutable. Marxism is one such as pointed out by Popper in his famous (and charming) essay, 'Science as Falsification'. It's available at:
http://tinyurl.com/dh96v7

sam said...

ok

Cameron Daniels said...

Very good article, absolutely love the grounding in Hume's philosophy, also could have done some Kant, but it is very important. The significance in the layman understanding correlation vs. causality cannot be under-appreciated because of how MANY bad studies there are out there published in reputable newspaper in the like. With the critical eye, readers can see what studies are just B.S. and which ones might have a chance of success.

Example:
I saw a study that said that girls under the age of 14 who watch Comedy Central are much more likely to have a teenage pregnancy. Is this because of Comedy Central or does a girl who watches CC under the age of 14 say something about her?