Read Description Carefully
alikh86
correlation__cause.pptxCorrelation & Cause
Do we need a concept of cause & effect? How is cause different from correlation?
What is a correlation?
See the link to “Correlation and Margin of Error” provided under the menu item, “Week 7”.
“5. Sometimes two things or events are clearly associated or linked. Where you find X, you will often or always find Y. A mutual relationship such as this, in which two things are frequently or invariably found together is called a correlation.”
Correlations are patterns that occur between variables, but are all such patterns meaningful? Do they necessarily indicate a cause-and-effect relationship?
“6. Sometimes, but not always, a correlation is a sign of a cause-and-effect relationship.”
“9. Correlation can always be a result of mere coincidence, and most of the time it is.” (That “most of the time it is” is controversial.
But more controversial is whether or not we need to move from correlation to cause-and-effect at all. The next slides will further investigate these questions.
See also the electronic reserves article, “A Beginner’s Guide to Scientific Method”, p. 34, Quick Review 3.1, “Causation & Correlation” on different types of correlations. You are not required to know this.
Why isn’t a correlation necessarily a cause?
See the bottom handwritten part of the link “Correlation & Margin of error”, as well as the recommended link, “Correlation, Causation, & Confusion”.
CHANCE: A correlation can be due to chance, as already mentioned on the previous slide: Examples: the divorce rate varies with the import of apples. This does not imply that the divorce rate has any effect on the import of apples or that the import of apples has any effect on the divorce rate. Similarly, the murder rate goes up in Central Park in NYC when the # of ice cream cones that are eaten goes up. This does not mean that murders cause the eating of ice cream cones or that eating ice cream cones cause murders. These are chance occurrences or coincidences.
One way to understand chance: If you throw a box of rice in the air, some of the rice that lands on the floor will scatter and some of it will cluster. The clusters are random – you did not design it that way, it simply happens
The concept of random clustering often forms the basis of legal defenses by large corporations that are sued by plaintiffs who are complaining that they have been harmed by the corporations’ activities, such as dumping hazardous waste. The corporations will claim that, for example, the cluster of cancer cases in the area of the hazardous waste is simply a random cluster. However, there is the possibility that the presence of the hazardous waste caused, or was a causal factor, in the cancer cases. However, what it means to say “cause” is tricky.
Why isn’t a correlation necessarily a cause? (Continued)
2) REVERSE CAUSATION: Even if there is a cause and effect relationship between 2 factors in a correlation, it is not always clear when factor A causes factor B or when factor B causes factor A.
We often assume that we can figure this out from the temporal order of events. If A occurred first, then it caused B.
However, temporal order alone does not determine a cause and effect relationship. For example, if a black cat passes the classroom door, and 10 seconds later the movie projector crashes to the floor, the black cat did not cause the movie projector to fall, despite the sequence of events. Falsely attributing cause in this instance is known as the fallacy (error in reasoning) of post hoc ergo propter.
Moreover, the temporal sequence is not always clear when 2 factors vary together in a correlation. Thus the problem of which factor caused which factor remains. Example: There is a correlation that senior citizens who have pets are healthier. Does that mean that we should make sure that senior citizens have pets, because having pets causes or produces better health? Or does that mean that senior citizens who are already healthy are better able to take care of their pets, which would imply better health causes senior citizens to have pets?
This is important because we do not know how to intervene to change outcomes, if we do not know what the causal factors are. See later slides.
Why isn’t a correlation necessarily a cause? (continued)
COMMON CAUSE: (or Confounding Cause): There is the possibility that there is not a direct relationship between factor A and factor B, in either direction, but that they share a third underlying common cause, which independently affects each of these factors.
Take the previous example of the correlation that senior citizens who have pets are healthier. What if there is a common underlying cause of wealth that makes it possible for senior citizens to be healthier, and for senior citizens to be able to afford to have pets? Wealth would be factor C that does not appear in the original correlation.
Why isn’t a correlation necessarily a cause? (continued)
SELECTION BIAS: We arrive at causes often by generalizing from samples of a targeted population. For example, we observe a sample of smokers (and, as we shall see, we compare this sample of smokers to a sample of non-smokers) to see whether or not they come down with lung cancer over a period of 20 years or so. However, that means that it is important as to how we select our sample: Are the smokers in our sample also coffee drinkers? If so, coffee might be the cause of the lung cancer. Are the smokers in our sample only people who are older and already more prone to get ill? Only males? Only people who live in a city that is full of smog, which could be a factor in causing lung cancer? Would this bias the results of our study/observation? Not only do we need to select a sample that reflects the characteristics of the target population, but we must be able to control for other possible factors that could be alternative causes.
Insofar as causes are attributed through inductive generalizations, they are never absolutely certain, about which more will stated on the next slide.
Insofar as inductive generalizations depend upon samples, it becomes important as to how we select that sample, both in size and representativeness of the target population. (See later slides.)
5. THERE CAN BE MULTIPLE CAUSAL FACTORS.
When are correlations in the absence of a causal relationship useful?
See p. 29 of “Correlation, Causation & Confusion”
Symptoms of illness reveal patterns that help us diagnose illness, by which we can identify an underlying problem (that is, the illness).
Similarly “certain economic indicators may presage a recession” that also identify an underlying problem.
Both of the above are “markers”: although they identify underlying problems, “changing the marker itself may have no effect on the condition. For example, fever often precedes full-blown chickenpox, but while medication to reduce the fever may make the patient feel better, they have no impact on the infection.” [p. 29]
“Insurance companies are interested in correlations between risk factors and adverse outcomes, regardless of causation. For example, if a certain model of car is at higher risk of accident, then an insurance company will charge more to insure a car of that type”, regardless of why that model is a higher risk.
Correlations are useful in making predictions (also a type of induction).
So what is so important about “cause”?
Identifying causes helps us decide how to intervene to change outcomes, because we can sometimes explain how some effect or outcome is produced by reference to its “causes”. Key words here are:
- PRODUCE: causes produce effects.
- EXPLANATION: causes are a means to explain how effects are produced.
- INTERVENTION: how we try to intervene to produce different outcomes: For example, by trying to explain how global warming is produced from greenhouse gases, we can try to intervene by reducing greenhouse gases or by capturing carbon or…..
If we have incorrectly identified the causes, or if we mistake a correlation for a causal relationship, our interventions will not be effective.
Chris Anderson, former- editor-in-chief of Wired magazine, thinks correlation is enough, & we do not need causation.
See p. 32 in “Correlation, Causation & Confusion”;
“Counterfactuals” “require considering something other than what in fact happened,” in other words, something that we do not actually observe.
“To assess that A caused B we need to consider a counterfactual: What would have happened if A had been different? To evaluate whether your neighbor’s dieting caused his weight loss, we need to consider what would have happened had he not dieted, and so on.” [p. 32] Perhaps it was his jogging that caused his weight loss. Can we replay the event by omitting the dieting? Then omitting the jogging?
However, “we can only observe what actually happened, not what might have happened” had things gone differently or if we could somehow replay history. We might try to “replay history” by feeding into computer models or video games all the factors that we think may be relevant to the outcome (of losing weight, of the war, etc.), and then manipulate the factors – omit some, change others, etc. However, this only works if we have considered all relevant variables or factors in the first place.
We do not observe the cause and effect relationship itself; we only observe the correlation which sometimes serves as a clue to a cause and effect relationship. Thus, we infer causes (and do not deduce them). We try to do this through random controlled experiments, which will be addressed in another power point.
Can we actually observe causes producing effects?
So what is a cause anyway?
Link to “Correlation & Margin of Error” under Week 6: “When speaking about causality in an entire population, we usually mean that X results in a higher rate of Y in the population, not that every individual who uses X will get Y.”
For example, we can compare two groups or populations, those who smoke cigarettes and those who do not. Over 20 – 40 years, the rate of lung cancer among the smokers is higher than the rate of lung cancer among the non-smokers. This does not mean that everyone who smokes will get lung cancer and it does not mean that everybody who does not smoke will avoid lung cancer. In this way, the statement that “smoking cigarettes causes lung cancer” does not necessarily apply to each individual smoker. After all, this is difficult to test, because some smokers might die of other causes, such as car accidents or heart disease, before they would have gotten lung cancer. (The problem of counterfactuals again.)
This is a contemporary way of looking at cause and effect, which combines cause-and-effect reasoning with statistics. It is not the only way.
More ways of defining cause – Koch’s postulates
See the electronic reserves article, “The Elements of Reasoning”, p. 103, the chart “Koch’s Postulates”: (the pages in this article are in the incorrect order)
“Robert Koch developed a set of rules…to determine when a particular organism is the cause of [that is, produces] a particular disease. The organism must
Be present in every case of the disease.
Be isolated from a case of the disease and grown in a pure culture.
Produce a case of the disease when inoculated into a susceptible animal. [deliberately inject the material into a person or animal in order to see if they get the disease]
Be recovered from the diseased animal.
Here is a real attempt to show how a specific factor A produces the effect, B, although we might not sometimes think it is unethical to do this. However, this does not address other possible causal factors that might require intervention for preventing certain diseases. For instance, cholera occurs in unsanitary water conditions. Should we also be considering sanitation?
More ways of defining cause – necessary & sufficient conditions
See the electronic reserves article, “The Elements of Reasoning”, pp. 102 – 103 (the pages are in the incorrect order).
“A factor is a necessary condition for an event if the event does not occur in the absence of the factor. Everyone who gets polio has been infected with the polio virus, but not everyone infected with the polio virus gets polio. Thus, the polio virus is a necessary, but not sufficient condition for the disease.”
“We usually do not speak of just any necessary condition as a cause. The presence of oxygen in the air was a necessary condition for World War II, yet we would consider it absurd to mention oxygen as the cause of the war. We regard such conditions as trivial with respect to explanation because they constitute relatively constant background conditions. Unusual factors or those varying from case to case…are the necessary conditions we [often – see later slides] identify as causes.”
CAUSE AS BOTH NECESSARY AND SUFFICIENT CONDITION: “Very few (if any) actual relationships in ordinary experience can satisfy” this way of construing cause, especially in the social and biological sciences. Laws of physics , such as Newton’s laws, are sometimes treated separately.
P. 103 gives an example of how diamonds are produced from both carbon and great pressure as relying on both necessary & sufficient conditions, but even this example doesn’t always hold up, given new techniques for producing diamonds.
More ways of defining cause – Contributing factors as “causes”
See the electronic reserves article, “The Elements of Reasoning”, pp. 104 – 105.
“Each factor making up a set of factors sufficient (or necessary) for the occurrence of an event can be called a contributing factor. Thus, the methane has that accumulated in the room was a contributing factor of the explosion that occurred. So, too, was the oxygen present in the room, as well as the spark that set off the methane.”
“The factor that we mention as the cause of an event is rarely one we consider sufficient or even necessary for such events.” We select one factor, depending upon “our aims and interests. Legal, practical, medical, or moral concerns may influence our selection.” Below our 3 out of many ways to make this selection:
1. TRIGGERING FACTOR: (Proximate – or nearest – cause): “The one that occurs last and completes the causal chain…producing the event.” Example: “Halton caused the explosion when he turned on the light” [in the room full of methane], because it was the last event to occur before the explosion.
2. UNUSUAL FACTOR: Example: “The explosion was caused by an accumulation of gas”, because that was the unusual factor, while “oxygen is usually in the air, and people ordinarily turn on lights.”
3. CONTROLLABLE FACTOR: the factor by which we can intervene in the event: Example: “Cholesterol causes heart disease” says “that eating foods high in saturated fats increases the chances of developing heart disease. Hence, anyone wanting to avoid heart disease should avoid eating such foods.” (In other words, cholesterol is one risk factor for heart disease that we can control through proper interventions.) However, there are other risk factors for heart disease, such as heredity, and ordinarily we do not call it the cause, because we cannot directly control heredity.
Types of Explanations (rather than arguments)
See electronic reserves article, “A Beginner’s Guide to Scientific Method”, p. 39, Quick Review 3.2: “Ways of Explaining”.
CAUSES: To explain one thing or event by reference to another, antecedent thing or event. EXAMPLES: “Debris from last night’s windstorm caused the power outage.” “Excessive alcohol consumption can damage the liver.” But these are incomplete explanations, because we could ask “how did debris cause the power outage?” “how does alcohol damage the liver?” That is why we refine our explanations with the following:
CAUSAL MECHANISMS: To explain by citing intervening causal factors, factors that explain the effects of a more distant cause. EXAMPLES: “Debris from the storm severed several power lines thus causing last night’s power outage.”
LAWS: We leave this out of our present discussion, although it can play an important role in laws of physics and medical diagnoses.
UNDERLYING PROCESSES: To explain something by reference to the working of its component parts. EXAMPLE: “The chest pain and breathing difficulty symptomatic of pneumonia results from an infection of the lung tissue. The tiny air sacs of which the lungs are composed…fill with inflammatory fluid caused by the infection. As a result, the flow of oxygen through the alveolar (air sac) walls is greatly impaired.”
FUNCTION: To explain something by reference to the role it fulfills in some larger enterprise: EXAMPLES: “Many species of birds build their nests in high places – trees, cliffs, etc. - to protect their young from predators.” “The lungs serve as means of both introducing oxygen into and removing carbon dioxide from the blood stream.”
Are causes aspects of reality? Aristotle v. Hume
Are causes aspects of reality? Or are they ways we try to simply predict what will happen?
ARISTOTLE (BCE): ancient Greek philosopher and scientist, who had four conceptions of “cause”, by which he tried to explain how a thing came to be. Although he applied this to natural things (that is, substances), his examples were often drawn from things we produce: For instance, what is the cause of this statue?
Material cause: the bronze from which it was made;
Formal cause: the shape which the bronze took, that is, perhaps the shape of a man on a horse;
Final cause: the purpose for which the statue was made: Perhaps, to commemorate a war hero.
Efficient cause: the motion or agent that brought it about, in this case, the sculpture in his activity of sculpting.
The scientific revolution in the 1500’s and beyond tried to eliminate final causes, as, during the middle ages, they were used in this way: “The purpose of the rain is to make the grass grow; the purpose of the grass growing is to feed the cattle; the purpose of feeding the cattle is to feed humans.” However, Aristotle sometimes applied “final causes” in the way we still sometimes explain things through “functions”. (See previous slide). The scientific revolution tried to reduce cause to basically “efficient causes.”
HUME: (1700’s): a British philosopher, who was important to laying out the problems of counterfactuals and that we infer causes, rather than directly observe them. (See earlier slides.) Hume did not think that causation actually exists “in the world”, but was a “habit of our minds, a connection we draw between two events we have observed in succession many times.” [p. 23, “Correlation, Causation & Confusion”] Therefore, he attached no necessity to causes.
