In statistics, many statistical tests calculate the correlation between variables and when two variables are found correlated, it is tempting to assume that this indicates that one variable causes the other. That "correlation proves cause-and-effect," is considered a questionable cause of logical error when two events occur together to form a causal relationship. This fallacy is also known as cum hoc ergo propter hoc , Latin for "with this, therefore," and "wrong cause." A similar error, that the event that follows the other is a consequence of the first event, is post hoc ergo propter hoc (Latin for "after this, therefore.") Error.
For example, in the case of many studies, many epidemiologic studies have shown that women who take hormone replacement therapy (HRT) also have a lower than average incidence of coronary artery disease (CHD), leading physicians to propose that HRT is protective against CHD. But randomized controlled trials show that HRT causes a small but statistically significant increase in the risk of CHD. Re-analysis of data from epidemiological studies suggests that women who do HRT are more likely to be from higher socioeconomic groups (ABC1), with better-than-average diet and exercise. The use of HRT and a decrease in the incidence of coronary heart disease is a coincidental effect of a common cause (ie benefits associated with higher socioeconomic status), rather than the direct cause and effect, as might have been expected.
As with any logical error, identifying that the reasoning behind the argument is wrong does not imply that the resulting conclusion is wrong. In the above example, if trials have found that hormone replacement therapy does have a negative incidence of possible coronary heart disease, the assumption of causality will be true, although the logic behind that assumption will still be flawed. Indeed, some go further, using correlation as a basis for testing hypotheses to try to establish a correct causal relationship; examples are the Granger causality test and convergent cross-referencing.
Video Correlation does not imply causation
Usage
In logic, the technical usage of the word "imply" means "is enough the state for". This is the meaning meant by statisticians when they say cause-and-effect is uncertain. Indeed, p implies q has a technical meaning of the conditional material: if p then q is denoted as pÃ, -> q . That is "if the state of p is true, then q follows." In this sense, it is always true to say "Correlation does not imply a cause/effect."
However, in regular use, the word "implies" loosely means suggest rather than requires . The idea that correlation and causation are connected is true; where there is cause-and-effect, there is a possible correlation. Indeed, correlations are used when summarizing causation; an important point is that the conclusion is made after the correlation is confirmed as real and all causal relationships are systematically explored using a substantial set of data.
Maps Correlation does not imply causation
General pattern
For any two correlated events, A and B, the possible different relationships include:
- A causes B (direct cause);
- B causes A (reverse causation);
- A and B are consequences of a common cause, but not mutually causal;
- A and B both cause C, which (explicitly or implicitly) is conditioned. If A and B cause C, why do A and B have to be correlated ?;
- A causes B and B to cause A (two-way or cyclic cause);
- A causes C causing B (indirect cause);
- There is no relationship between A and B; the correlation was accidental.
Thus no conclusions are made about the existence or the direction of the cause-effect relationship only from the fact that A and B are correlated. Determining whether there is a true cause-effect relationship requires further investigation, even when the relationship between A and B is statistically significant, a large effect size is observed, or most variance is described.
An example of the illogical reasoning of the correlation
B causes A (inverse cause or inverse causality)
The reverse cause or reverse causality or the wrong direction is a questionable cause error in which cause and effect is reversed. The cause is said to effect and vice versa.
- Example 1
- The faster the windmill is observed rotating, the more wind is observed.
In this example, the correlation (simultaneity) between windmill activity and wind speed does not mean that wind is caused by a windmill. This is quite the opposite, as suggested by the fact that wind does not need a windmill, while the windmill requires wind to rotate. Winds can be observed in places where there are no windmills or windmills that do not spin - and there is good reason to believe that the winds existed before the invention of the windmill.
- Example 2
- When the country's debt rises above 90% of GDP, growth slows.
- Therefore, high debt causes slow growth.
This argument by Carmen Reinhart and Kenneth Rogoff is denied by Paul Krugman on the basis that they get backward causality: in reality, slow growth causes debt to increase.
- Example 3
- Driving a wheelchair is dangerous, because most people who drive it have an accident.
- Example 4
In other cases it may be just unclear which is the cause and that is the effect. As an example:
- Children who watch as many TVs are the loudest. Obviously, TV makes kids fiercer .
This can easily be otherwise; ie, violent children like watching more TVs than the more rough ones.
- Example 5
The correlation between drug use and psychiatric disorders may be one way: drugs may cause disorders, or people may use drugs to treat themselves for pre-existing conditions. Gateway's drug theory may argue that marijuana use leads to tougher drug use, but the use of hard drugs can lead to cannabis use (see also inversional confusion ). Indeed, in the social sciences where controlled experiments often can not be used to look at causal directions, these errors can trigger long-standing scientific arguments. One example can be found in the educational economy, between the model of filtering and signaling and human capital: it may have innate abilities allowing one to complete education, or that completing education builds one's capabilities.
- Example 6
A historical example of this is that Europeans in the Middle Ages believed that ticks are beneficial to your health, since there are rarely sick fleas in sick people. The reason is that people get sick because the tick goes away. But the real reason is that ticks are very sensitive to body temperature. A small increase in body temperature, such as fever, will make the lice look for another host. Medical thermometers have not been found, so this increase in temperature is rarely noticed. The visible symptoms come later, giving the impression that the ticks go before the person gets sick.
In other cases, two phenomena can each be the cause of some of the others; considering poverty and lack of education, or poor delays and self-esteem. However, the person making the argument based on these two phenomena must be careful to avoid causal errors and circular consequences. Poverty is the cause of the lack of education, but it is not the cause of sole , and vice versa.
The third C factor (causal-common variable) causes A and B
The third culprit (also known as ignoring common causes or a questionable cause ) is a logical error in which the false relationship is confusing because of cause and effect. This confirms that X causes Y when, in fact, X and Y are both caused by Z. This is a variation on the post hoc ergo propter hoc error and the group member of the cause of the error being questioned.
All of these examples relate to the lurking variable, which is the third hidden variable that affects both causes of correlation. Difficulty often also arises where a third factor, although fundamentally different from A and B, is so closely related to A and/or B that it becomes confused with them or it is very difficult to separate them scientifically (see Example 4).
- Example 1
- Sleeping in someone's shoes is highly correlated with waking up with headaches.
- Therefore, sleeping with someone's shoes causes a headache.
The above example does fallacy correlation-implying-cause-effect, because premature concludes that sleeping with one's shoes causes a headache. A more plausible explanation is that both are caused by a third factor, in this case going to sleep drunk, thereby generating a correlation. So the conclusion is wrong.
- Example 2
- Young children who sleep with light are much more likely to develop myopia later on.
- Therefore, sleeping with lights turns on myopia.
This is a scientific example resulting from a study at the University of Pennsylvania Medical Center. Published in the May 13, 1999 edition of Nature, this research received much coverage at the time in popular media. However, subsequent research at Ohio State University did not find that babies who slept with the light above led to the development of myopia. It found a strong association between parental myopia and the development of childhood myopia, also noting that farsighted parents were more likely to leave lights in their children's bedroom. In this case, the cause of both conditions is the parent's myopia, and the above-mentioned conclusion is false.
- Example 3
- As ice cream sales increased, the death rate went up sharply.
- Therefore, the consumption of ice cream causes drowning.
This example fails to recognize the importance of time and temperature for the sale of ice cream. Ice cream is sold during the summer months at a much greater rate than during winter, and during these summer months people are more likely to engage in activities involving water, such as swimming. Increased drowning deaths are only caused by more exposure to water-based activities, not ice cream. The wrong conclusion is stated.
- Example 4
- A hypothetical study shows the relationship between an anxiety test score and a shame score, with a statistical value of r (correlation strength) of.59.
- Therefore, it may be only concluded that shyness, in some parts, causally affects test anxiety.
However, as found in many psychological studies, another variable, "self-awareness score", was found to have a sharper correlation (.73) with shame. This suggests the possibility of a "third variable" problem, however, when the three closely related steps are found, further indicates that each may have a two-way tendency (see "bidirectional variables", above), into a group of mutually correlated values -masing affect each other. to some extent. Therefore, the above simple conclusions may be wrong.
- Example 5
- Since the 1950s, levels of atmospheric CO 2 and the level of obesity have risen sharply.
- Therefore, atmospheric CO 2 causes obesity.
The rich population tends to eat more food and produce more CO 2 .
- Example 6
- HDL ("good") cholesterol is negatively correlated with the incidence of heart attacks.
- Therefore, taking medication to raise HDL reduces the chances of having a heart attack.
Further research calls this conclusion into question. Conversely, perhaps other underlying factors, such as genes, diet and exercise, affect HDL levels and the likelihood of having a heart attack; there is a possibility that drugs may affect directly measurable factors, HDL levels, without affecting the likelihood of a heart attack.
Bidirectional causation: A causes B, and B causes A
Causality is not always one way; in the predator-prey relationship, the number of predators influences the prey number, but the number of prey, ie the food supply, also affects the number of predators.
The relationship between A and B is a coincidence
Both variables are not related at all, but are correlated by chance. The more things are checked, the more likely that two unrelated variables will appear to be related. As an example:
- The result of the last home game by the Washington Redskins before the presidential election predicts the outcome of any presidential election from 1936 to 2000, despite the fact that the outcome of the football game has nothing to do with the election results. This streak eventually broke down in 2004 (or 2012 using an alternative formulation of the original rule).
- Such a collection of coincidences found that for example, there was a 99.79% correlation for the 1999-2009 period between US spending in science, space, and technology; and the number of suicides due to suffocation, suffocation, and hanging.
- The law of Mierscheid, which connects the Social Democratic Party with the German part of the popular vote to the size of crude steel production in West Germany.
- Bald bald Bald leaders: A bald (or very bald) Russian leader has managed to be a bald ("hairy") man, and vice versa, for nearly 200 years.
Determining the causes
In the academic world
The nature of causality is systematically investigated in several disciplines, including philosophy and physics.
In the academic world, there are a large number of theories about causality; The Oxford Handbook of Causation (Beebee, Hitchcock & Menzies 2009) includes 770 pages. Among the more influential theories in philosophy are the Four causes of Aristotle and sometimes Al-Ghazali. David Hume argues that beliefs about causality are based on experience, and the same experience is based on the assumption that the future modeled the past, which in turn can only be based on experience - leading to a circular logic. In conclusion, he asserted that causality is not based on the real reason: only correlations can really be felt. Immanuel Kant, according to Beebee, Hitchcock & amp; Menzies (2009), states that "the causal principle according to which every event has a cause, or follows according to causal law, can not be established through induction as purely empirical claims, because it would lack the strict universality, or necessity".
Outside the field of philosophy, causal theory can be identified in classical mechanics, statistical mechanics, quantum mechanics, spacetime theory, biology, social science, and law. To define correlations as causes in physics, it is commonly understood that their causes and effects must be connected through local mechanisms (eg impact concepts) or nonlocal mechanisms (see field concepts), in accordance with known natural laws.
From the thermodynamic point of view, the universal nature of the cause compared to the effects has been identified through the Second Law of thermodynamics, confirming the ancient, medieval and Cartesian view that "the cause is greater than the effect" for certain thermodynamic cases. free energy. This, in turn, is challenged by popular interpretations of the concept of nonlinear systems and butterfly effects, in which small events cause great effects because, respectively, it is unpredictable and unlikely to trigger a large amount of potential energy.
Causality is interpreted from a counterfactual state
Intuitively, causes seem to require not only correlation, but counterfactual dependence. Suppose a student tests poorly and guesses that the cause is not learning. To prove this, people think about counterfactuals - the same student writes the same test in the same situation but after studying the night before. If one can repeat history, and change only one small thing (making a student study for the exam), then the causes can be observed (by comparing version 1 to version 2). Because one can not repeat history and replay events after making small controlled changes, causation can only be inferred, never known for certain. This is called the Fundamental Issue of Causal Inference - it is impossible to directly observe causal effects.
The main purpose of scientific experiments and statistical methods is to estimate the best counterfactual state in the world. For example, one can run experiments on identical twins that are known to consistently get the same score on their tests. One twin was sent to study for six hours while the others were sent to the amusement park. If their exam scores suddenly deviate heavily, this would be strong evidence that learning (or going to an amusement park) has a causal effect on test scores. In this case, the correlation between learning and test scores will almost certainly cause causation.
Well-designed experimental research replaces individual equality as in the previous example by group equality. The aim is to build two similar groups except for the treatment received by the group. This is achieved by selecting subjects from one population and randomly assigning them to two or more groups. The possibility of groups that behave similarly to each other (on average) increases with the number of subjects in each group. If the group is essentially equivalent except for the treatment they receive, and differences in outcomes for the groups are observed, then this is evidence that treatment is responsible for the outcome, or in other words treatment causing the observed effect. However, the observed effects can also be caused "by chance", for example as a result of random disorders in the population. Statistical tests exist for measuring the probability of incorrectly concluding that observed differences exist when in fact they are not (eg see P-value).
Causality is predicted by trap extrapolation
When experimental studies are not possible and only pre-existing data are available, as is usually the case for example in economics, regression analysis can be used. Factors other than attractive potential causes variables are controlled by incorporating them as supporters other than governors representing interesting variables. A false conclusion about the cause-effect of reverse causation (or a false estimate of the magnitude of causation due to the existence of a two-way cause) can be avoided by using an exogenous explorer (regressor), such as a physical description such as the amount of rainfall ( as determinants, say, futures prices), lagged variables whose value is determined before the value of the dependent variable is determined, the instrumental variable for the explanator (chosen based on known exogeneity), etc. See # Statistics # and Statistics. False correlations due to the mutual influence of the three, common causal variables, are more difficult to avoid: the model must be determined in such a way that there is a theoretical reason to believe that no fundamental underlying cause has been omitted from the model.
Use of correlation as scientific proof
Much of the scientific evidence is based on variable correlations - they are observed to occur together. Scientists are careful to show that correlation does not necessarily mean causation. The assumption that A causes B just because A correlates with B is often not accepted as a valid argument form.
However, sometimes people make the opposite mistake - refusing the correlation completely. This will ignore many important scientific evidence. Because it may be difficult or ethically impossible to run a controlled double-blind study, correlational evidence from several different angles may be useful for predictions even if it fails to provide proof for the cause. For example, social workers may be interested in knowing how child abuse is related to academic performance. While it is unethical to conduct experiments in which children are randomly assigned to receive or not accept harassment, the researcher can look at existing groups using non-experimental correlational designs. If in fact there is a negative correlation between abuse and academic performance, researchers have the potential to use this knowledge from statistical correlations to make predictions about children outside of abused research, although research fails to provide causal evidence that abuse degrades academic performance. The combination of the limited methodology available with the error of the dismissal correlation has sometimes been used against the scientific findings. For example, the tobacco industry has historically relied on the dismissal of correlational evidence to deny links between tobacco and lung cancer, as did biologist and statistician Ronald Fisher.
Correlation is a valuable type of scientific evidence in areas such as medicine, psychology, and sociology. But the first correlation must be confirmed as real, and then any possible causal relationships should be systematically explored. In the end the correlation alone can not be used as evidence for a causal relationship between treatment and benefits, risk factors and disease, or social or economic factors and outcomes. This is one of the most abused types of evidence, for it is easy and even tempting to come to premature conclusions based on the initial appearance of a correlation.
See also
Referensi
Bibliografi
- Beebee, Helen; Hitchcock, Christopher; Menzies, Peter (2009). The Oxford Handbook of Causation . Oxford University Press. ISBN: 978-0-19-162946-4.
- Tufte, Edward R. (2006). "Gaya Kognitif PowerPoint: Meletakkan Korup Di Dalam" (edisi ke-2). Cheshire, Connecticut: Grafik Tekan. ISBNÂ 0-9613921-5-0. Â
Tautan eksternal
- "Seni dan Ilmu sebab dan akibat": slide show dan ceramah tutorial oleh Judea Pearl
- Inferensi kausal dalam statistik: Sebuah ikhtisar, oleh Judea Pearl (September 2009)
- Korelasi Palsu, penelusuran situs, dan menunjukkan korelasi semacam itu.
- Apa Yang Harus Anda Ketahui Tentang Korelasi Statistik
Source of the article : Wikipedia
0 Comments