Proposed by Hans Reichenbach, the common cause principal argues against correlation doesn’t imply causation. It states that if there are two variables I and J and they are correlated, then I causes J or J causes I or there is another variable X that precedes and causes both I and J. 

This theory is not true because it doesn't account for how the data is selected and appeals to our human nature to look for casual explanations whenever there is a pattern. This is an example of collider bias.

University of Michigan - Ann Arbor

When a single node is caused by two other nodes we encounter something called "collider-bias" where one condition explains-away another. 

Since A and C both effect the outcome of B, by conditioning on B, A and C become **dependent**.

$A \rightarrow B \leftarrow C$: The Collider

A correlation between variables indicates a statistical relationship, but it does not provide any evidence about causation, regardless of how strong that relationship may be. Just because two variables move together does not mean that a change in one causes a change in the other. In scientific research, the only definitive way to demonstrate a cause-and-effect relationship is by conducting a controlled experiment.

Correlation Does Not Imply Causation

Pearl, J., & Mackenzie, D. (2018). $\mathit{The \ Book \ of \ Why: \ The \ New \ Science \ of \ Cause \ and \ Effect.}$ Basic Books.

The Book of Why

Cole, S. R., Platt, R. W., Schisterman, E. F., Chu, H., Westreich, D., Richardson, D., & Poole, C. (2010). Illustrating bias due to conditioning on a collider. $\mathit{International \ Journal \ of \ Epidemiology, 39}$, 417-420. [doi.org/10.1093/ije/dyp334](https://doi.org/10.1093/ije/dyp334)
[ncbi.nlm.nih.gov/pmc/articles/PMC2846442/](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2846442/) 

Illustrating bias due to conditioning on a collider

In a path that would impact both variables but also contains a collider, M-bias occurs when the variable in the center of the collider is adjusted for. 

The general structure for M-bias is A <- B -> C <- D -> E where adjusting for C would create bias. 

M-bias

This a subcategory of a collider bias (Collider is the selection of the data). When two factors individually are usually not severe enough to cause some effect but the combination of the two is. Usually both factors are not thought to be correlated with each other in real life settings but in a certain condition they are. This seems to occur because the resulting effect is being adjusted for by not considering the situations where both factors do not occur. This association is caused based on the conditional collider of the environment the data is selected. 

Berkson's Paradox

The statistical procedure of controlling for everthing that can be measured is misguided. 

For instance, in a collider $A\rightarrow B \leftarrow C$ where the path between $A$ and $C$ is initially independent, controlling for B would make them dependnet due to the explain-away effect. This means that a back-door is opened.

Also, controlling for descendents of the varibles we are interested in should also be avoided. This is like "partially" controlling for the interested variable itself. 

Controlling for everything is misguided

Common Cause Principal 

A proxy is used as a substitute measure when a variable that is not directly measurable is needed for analysis.  For a variable to be a proxy for another variable, the proxy variable needs to be highly correlated with the variable of interest.  The proxy can be used for any variables such as a mediator, confounder, collider.  

Proxy

In the Book of Why, some of the classic examples of collider bias are discussed:

1. Celebrity's talent and beauty
2. Birth-weight paradox (Simpson's paradox)
3. Admission-rate bias (Berkson's paradox)

List of Collider Bias examples

Event A directly causes event B, without any hidden contributing factors driving them both.

- A is the cause, B is the effect. 
- A correlation coefficient of 1 usually accompanies a causal relationship because it means that A occuring positively influences/ensures that B occurs. 

Causal Relationship

Events A and B have no impact on the occurence of the other. They do not cause the other.

Neutral Relationship

Instead of event A causing event B, event B causes event A. Sometimes called the 'cart before the horse bias'. 
Eg : If there is a patient at risk for a heart disease that reduces their intake of fatty/unhealthy foods, but then still suffers a heart attack. Their healthier lifestyle cannot be held as a cause for their heart attack, because the reason they shifted to a healthier lifestyle was because they were at high risk for a heart attack.

Reverse Causal relationship

When two heterogeneous data sets are aggregated into one, false correlations between variables can arise due to the differences in populations. 

For example, the correlation between skull length and breadth is statistically significant when measured on all humans, but when separated between male and female, there was no significant correlation.

Spurious Correlation: Aggregated Data

Inheritance of fur color in guinea pigs does not follow Mendelian laws. Sewell Wright postulated that this was because of 'developmental factors' (*d* ). He tried to find *d* by solving mathematical equations where *d* and hereditary factors *h* were the unknowns.

Wright showed that knowing *d* and *h* helped predict correlations, and drew a path diagram that allowed him to identify the causes after knowing correlations.

Reference : Pearl, J., & Mackenzie, D. (2019). The Book of Why. Penguin

S Wright's Guinea Pigs and the "First link between Causality and Probability"

Regression towards the mean is a phenomenon that occurs when resampling a single random variable. 

The phenomenon arises when the first sample of the random variable appears extreme, while the second sample appears much closer to the mean. This result occurs when the first sample is extremely above and below the mean.

Regression to the Mean

Probability is a concept that is largely tied more toward association than it is causation as it in itself does not address the second and third rungs of the ladder of causation (Intervention and Counterfactual Reasoning, respectively). Therefore, to say that "X raises the probability of Y" is faulty as suggesting that it "raises" the probability implies causation, which is not an aspect of probabilistic expressions. Therefore, the proper way to discuss probability according to Judea Pearl would be to state that "If we see X, then the probability of Y increases." This delineates the association without jumping to causation and leaves the causal mechanism open to either interpretation or further research. 

Irreducibility of Causation to Probabilities 

A classic example of a spurious correlation is the observed relationship between ice cream sales and crime rates. While data shows both increase concurrently, one does not cause the other. The association is more likely explained by a third, confounding variable: temperature. During warm weather, more people are outside, leading to increased social interactions and potential for crime. Simultaneously, warm weather drives up the demand for cool treats like ice cream. Therefore, the temperature independently influences both variables, creating a statistical correlation without a direct causal link.

Example of Spurious Correlation: Ice Cream Sales and Crime Rates

Claims of causation based on correlational data are frequently and mistakenly made, particularly in advertisements and news stories. These sources may present a statistical relationship in a way that implies a cause-and-effect link, often to promote a product or a particular narrative. This is sometimes done by maximizing and perhaps overstating the positive aspects of the findings, without acknowledging other possible explanations for the observed correlation.

Misinterpretation of Correlation as Causation in Media

A researcher conducts a study across 100 cities and finds a strong positive correlation between the number of public libraries in a city and the city's annual crime rate. Based on this finding, which of the following conclusions is the most scientifically sound?

Media reports often erroneously imply causation from correlational data, misleading the public. For instance, a headline claiming that eating candy daily as a child leads to violent behavior later in life assumes a causal link based merely on a statistical association. This conclusion ignores potential third variables that could influence both childhood diet and adult behavior. True causal relationships cannot be established from such observational data; they require controlled experiments.

Example of Misinterpreting Correlation: Candy and Violence

The directionality problem is a fundamental reason why correlation does not imply causation. It occurs when two variables, $$X$$ and $$Y$$, are statistically related, but it is impossible to determine the causal direction of the effect. Because neither variable is manipulated by the researcher, one cannot confidently identify whether $$X$$ causes $$Y$$, or if $$Y$$ causes $$X$$. For example, a statistical relationship showing that people who exercise are happier could mean that exercising causes happiness, or conversely, that happiness gives people the energy and desire to exercise.

Directionality Problem

The third-variable problem is a primary reason why correlational research cannot establish causality. It occurs when a statistical relationship between two variables, $$X$$ and $$Y$$, is not due to a direct causal link between them, but rather because some unmeasured third variable, $$Z$$, independently causes both. This creates a spurious correlation. For instance, a statistical relationship between exercise and happiness might exist simply because a third variable, such as overall physical health, simultaneously causes people to exercise more and feel happier.

Learn Before

Related