Chapter 15: Reasoning About Causation

Chapter 15: Reasoning About Causation

In many situations, causes are correlated with their effects. 

• An event C is said to be positively correlated with E when the presence of C increases the probability that E will also occur. 
• C is said to be negatively correlated with E when C decreases the probability of E. 
• If C has no effect on the probability of E, then C is not correlated with E, or C is independent of E. So for example, the appearance of lightning is positively correlated with thunder, negatively correlated with a clear sky, and presumably not at all correlated with the day of the week. 

Correlation is about how often two things are associated with each other, so there is no thunder without lightning. This is 100% or a perfect correlation. 

• Smoking is positively correlated with lung cancer, but obviously, not all smokers will get cancer. 
• A low correlation between two types of events does not rule out causation in particular instances. A hunter might fail to shoot his prey most of the time, but when he succeeds his shot will be the cause of the animal's death. 

Even if C does not cause E, there can be many reasons why C is positively correlated with E. Here are the main possibilities: 

• The correlation between C and E is purely an accident.
• E causes C and not the other way round.
• C does not cause E but they are the effects of a common cause.
• The main cause of E is some side effect of C rather than C itself directly.
• Why Correlation is not Causation
1. Accidental Correlation
2. Suppose I have been in only one car accident my whole life, and that was the only time I ever wore red trousers. There is a perfect correlation between the colour of my trousers and my being involved in a car accident, but this is just a coincidence. Correlation data are more useful when they involve a large range of cases. 
3. But still, we need to be careful. It has been suggested that the sea level in Venice and the cost of bread in Britain have both been generally on the rise in the past two centuries (Sober, 1988). But it is rather implausible to think that the correlation is due to some underlying causal connection between the two cases. The correlation is presumably an accident due to the fact that both have been steadily increasing for a long time for very different reasons. 

2. The Casual Direction is Reversed

Sometimes C is correlated with E not because C causes E, but because E causes C. Drug users are more likely to suffer from psychiatric problems. This might be because drug use is the cause, but perhaps preexisting psychiatric problems cause people to turn to drugs. Correlation by itself does not tell us which of these two stories (if any) is correct. 

• Scientific research might tell us only that drug use is positively correlated with psychiatric problems, but reporters who should know better might instead write, "drugs make you depressed and crazy." The word make suggests causation but this might not actually be supported by the data. 
• When two causal factors or variables reinforce each other, we have a causal loop. For example, there is a correlation between health and GDP growth in many countries. This is because, on one hand, healthy citizens work longer and better, contributing to economic growth. On the other hand, higher GDP brings about better living conditions and medical care, improving health as a result. So health and GDP growth are mutually reinforcing.
• A vicious circle happens when a causal loop makes a bad situation even worse. Take stage fright for example. Becoming nervous and stressed when you are performing can make you perform less well, and this might in turn make you even more

3. Hidden Common Causes

Sometimes C and E are correlated not because one causes the other but because there is a hidden condition X that causes both C and E. 

• For example, children who wear bigger shoes tend to have better reading skills. Do shoes somehow promote brain growth? Presumably not. The more mundane explanation is that older children read better, and they have bigger feet. So growth is the hidden common cause that leads to both bigger shoes and better reading skills. 
• Or suppose the drinking of bottled water instead of tap water is correlated with healthier children. Is this because bottled water is cleaner, and ordinary tap water contains harmful impurities? Not necessarily. Perhaps this is just because wealthier parents can afford higher-quality care and food for their children, and being more cautious, they choose to buy bottled water even if tap water is just as good. 

4. Causation Due to Side Effects

In some cases where C correlates with E because C occurs together with some other condition or side effect that actually causes E. The causal contribution from C to E might be nonexistent or of lesser importance. 

• The placebo effect (placebo effect is the psychological or physiological improvement seen in a patient after receiving a fake treatment or therapy that has no actual therapeutic effect.) is a good example of causation due to side effects. It refers to the real or felt improvement in a patient's condition that is due to beliefs about the treatment rather than the medical efficacy of the treatment itself. It is suggested that when patients believe that they have taken medicine, this is enough to make many of them feel better or suffer from less pain, even when the treatment being given has no medical benefit. In fact, it has been reported that a larger pill has a more pronounced placebo effect, and coloured pills are better than white ones and injections are even better!

In scientific research, it is important to investigate side effects to ensure that experimental results are reliable. For example, studying captive animals might not give a true picture of the behaviour of wild animals because putting animals in a confined environment might change their habits, which is a form of side effect. 

The following are other cases of side effect causation relating to human beings.

• An example in social science and industrial psychology is the Hawthorne effect. This refers to the fact that people tend to change their behaviour when they know they are being studied. In particular, they might work harder or perform better in an experimental setting.  
• People react to new things differently, and this produces a novelty effect. Once the novelty wears off, their behaviour might return to normal. For example, some schools claim that students behave better and learn better when they switch their pale school uniforms to colourful shirts.
• The pygmalion effect originated from a study where teachers were told that some of their students were above average even though they were randomly selected with the same average abilities. But the subjective expectation of the teachers somehow led to better performance by these students later on. 

2. Good Evidence for Causation

To establish causation, it is important to eliminate alternative hypotheses. But what kind of positive evidence can we obtain to support causation? 

1. Look for Covariation and Manipulability

• If changes in one event correspond to changes in another event, then this makes it more probable that one causes the other. When we suspect smoking causes lung cancer, the fact that cigarette smokers have a higher cancer rate than nonsmokers is only one piece of evidence. It becomes even more convincing when it is discovered that the death rate from lung cancer increases linearly with the number of cigarettes smoked per day. 
• Covariation is even stronger evidence when it can be directly manipulated and not just passively observed—we vary some aspects of the cause and see how it affects the effect. For example, hitting the key of a piano causes a sound to be made. We can be sure of the causal connection because we can change the timing and the loudness of the sound by controlling when and how we hit the piano key. This makes it extremely unlikely that the correlation is accidental or due to some other explanation. In reality, manipulating correlation can sometimes be difficult or even unethical to do. To study how smoking leads to lung cancer, it would be immoral to request some subjects to smoke more cigarettes and see if they are more likely to get cancer! 

2. Look for a Reliable Model of the Causal Mechanism 

A causal mechanism is a series of objects, processes, or events that explain how a cause leads to its effects. Using the piano as an example, hitting the key causes a felt-covered hammer to strike a steel string. This causes the string to vibrate, and the vibration in turn causes air molecules to move, which is the sound we hear. This causal process explains how the keys can create music, and a break- down in any step of the causal process might result in no sound being produced. 

• A causal mechanism explains why there is causation and helps us make predictions about what would happen when the system changes. For example, the story about the causal mechanism in a piano explains why a louder sound is heard when we hit the key harder because this means the string would vibrate with a larger amplitude, making a louder sound. 

3. Causation is Complicated

The world is a complicated place and events can interact with each other, often making it difficult if not impossible to find the one true cause. Here are some useful terms for making more fine-grained distinctions between causes: 

• Causal relevance: Suppose a student failed a course. She might have been lazy or had personal problems. Or perhaps she was ill on the day of the exam. All these factors could have contributed to her failure. They were all causally relevant, each being a cause of her failure but none being the cause. The most important one is the primary or central cause. 
• Causally necessary and sufficient conditions: X is causally necessary for Y when Y would not happen without X, and X is causally sufficient for Y when X by itself is enough for Y. Water is causally necessary but not sufficient for our survival, and moving electric charges are sufficient but not necessary for the presence of a magnetic field. But X can be causally relevant to Y even if X is neither causally necessary nor sufficient for Y. 
• Triggers: A triggering cause (or trigger) is a cause that starts off a chain of events leading to an effect. Whereas a structural cause (or standing condition) is a background condition that is causally relevant to the effect but which on its own is not sufficient for it. For example, an electric spark in a kitchen with a gas leak can result in an explosion. Here, the spark is the trigger, and the flammable gas is the standing condition. 
• Proximity: A proximate cause happened at a time near the occurrence of the effect, whereas a distal cause happened much earlier. 
• Randomness and causal determination: A random event is one that is not causally determined by what happened earlier. To say that an event is determined is to say that it must occur given what has happened earlier and the physical laws of our universe. 

END OF THE PART