one-on-one support
Waiting to Get Your Dissertation Accepted?
Last Updated on: 3rd February 2024, 01:33 am
In many scientific investigations, experiments, or research projects (including dissertations and theses), we’re interested in establishing some form of causation. We might express that objective in the form of “does factor A cause effect B?” This, generally, would be the gold standard, but it is also somewhat and sometimes challenging to prove.
In other situations, the questions may be less stringent, as in the following: “Is A related to B?” Or, “Does A predict B?” Or, “Are A and B correlated?”
In this article, I’ll talk about causation—what it means, how it differs from correlation, what it takes to prove it, and what if we cannot convincingly prove it?
Definitions
Let’s define the terms.
First, we’re talking about two conditions, A and B. These might be events, effects, or occurrences.
If A is correlated to B (and vice versa), then there exists a mutual or reciprocal relationship, an orderly connection, between A and B. For example, after dawn, the sun rises in the sky and the volume of traffic (number of vehicles per mile) increases. The conditions are correlated.
On the other hand, if A causes B, that means that A brings about B. Or, that A produces an effect that we call B. Or, that A is the reason for B.
In this situation, A might be an observable condition or a controllable factor. B is some response or effect, that we hypothesize is caused (at least in part) by A.
The idea that A causes B is a subtle change from our definition of correlation, when A and B occur (it seems) simultaneously.
For example, we apply a flame to a vessel of water which causes the water (and the vessel) to increase in temperature. Or, increasing ambient air temperature might cause the temperature of a body of water to increase. These events are not just occurring simultaneously.
In these examples, we can more clearly grasp the distinction between causation and correlation. An external source or condition causes water to heat up. In contrast, the sun does not cause the number of cars on the freeway to increase, but these two activities occur together—they are correlated. There here may be some intervening factor—more on that later.
What’s the Logical Process to Prove Causation?
Perhaps we’re aware of or we observe a phenomenon. Something is happening! We want to know, why? Let’s run through a logical sequence of questions and criteria.
We have an idea about two events, conditions, or factors. So, we postulate a relationship between them. Then observe the phenomenon.
At first, we think A and B are related. We may rely on theoretical or practical reasons that suggest a relationship between them. Or, we may observe that A and B occur simultaneously. This is empirical association.
In the beginning, we’re not sure if A and B are statistically correlated. Or, if one causes the other. And, if one causes the other, we may not even be sure which causes what.
We just know that they occur at the same time.
We may be interested in determining if A causes B (or, in the beginning, if either causes the other). If so, then the first question—or, criterion—is, are they correlated? We compute the correlation coefficient between A and B, based on whatever data we have collected.
To be a bit more scientific, we conduct a designed experiment, formulating the problem based on sound information and theory. We generate controlled data from rigorous scientific sampling, and use parametric hypothesis testing to strengthen the evidence of the relationship.
Over 50% of doctoral candidates don’t finish their dissertations.
If A and B are not correlated (using a statistical test), then we have evidence that there is nothing going on other than a random, spurious, unexplained simultaneity.
Our analysis may alter our understanding of the phenomenon. Or, it may generate strong statistical evidence to corroborate our theory, that a true relationship exists between A and B.
In this case, we’re interested in causation. So, the next step is to determine if A precedes B in time (temporal sequencing or temporal priority).
Temporal sequencing is the second criterion for causation. It stands to reason, if A does not precede B in time, it cannot be the cause of B.
Up to this point, we have a situation in which A and B seem to occur simultaneously. And, they are statistically related, with strong evidence. And, further, A precedes B in time.
That is not enough for a convincing, credible conclusion that A causes B. What is needed at this point is theory or, more emphatically, subject matter expertise.
Subject matter expertise based on theory, technical knowledge, and experience, provides an explanation of what the observations, experiments, and statistical analysis are telling us. The expertise might be hard science (physics, engineering). Or, social science (psychological, sociological, historical, or economic evidence).
Subject matter expertise explains more than the fact that B consistently occurs following A. It explains the relationship, conditions, and process in terms of cause and effect—how A causes B. It is the final piece of evidence that enables us to make a credible claim that A causes B.
Real-World Problems
There are those who argue for an additional criterion for determining if A causes B. The criterion is that A is the sole cause of B.
It is a simplistic criterion. Surely, if in reality A is the sole cause of B, it simplifies the task of gathering evidence of the relationship. But, it is unrealistic.
Most real-world phenomena and problems are complex. In complex scientific phenomena, the issues of causation and correlation are not always obvious and rarely simple.
We may think that A causes B based on theory or experience, but it is not necessarily clear nor always credibly proven. There may be many factors (C, D, etc.) that cause (wholly or in part) the effect, B. So, making a claim about A as a cause is not so clear, and may be incomplete.
There are two situations that complicate the declaration that A causes B. The first is that there may be an intermediate condition, or an intervening factor at work.
Consider the case in which the sun rises (A) and traffic increases (B). These conditions are easily correlated. And A precedes B in time. But we have sufficient experience to know that A is not a direct cause of B. There is at least one intervening factor or condition. The sun rises. People wake up and prepare for work. They get in their cars and travel to work, generally in the morning. Traffic increases.
So, people working in the daylight, going to work in the morning, in measurable numbers (C) is the cause of B. And, A can be considered a cause of C.
Another situation which typifies the real-world is complexity.
There are many reasons (conditions) why the temperature of the ocean changes: ambient air temperature, currents, tides, sun angle, distance of the sun, and so on (factors A, D, E, F, . . .). They all are direct causes of ocean temperature. They are influencing ocean temperature simultaneously, not sequentially. And, they interact between and among themselves.
Human behavioral and sociological sciences are even more complex and difficult to analyze. There are seemingly limitless known factors (causes?) and unknown influences and interactions related to behavior.
In the complex, real world, adding to the multiple causes of effects, there may be many intervening factors involved. And, of course, interactions among factors.
There may be many direct influences on or causes of B. Or there may be many other factors that create the conditions for B to occur. Or intervening factors that explain B. In any case, A may be a cause, but it is seldom the sole cause. In most cases, it is far too simplistic to expect or claim A to be the sole cause of B.
How to Determine Causation in a Complex World
In spite of the complexity, we still strive to determine causation in some research problems. But, given the criteria and the complexity, how is this done? The simple answer is to follow a rigorous and objective scientific process as I have laid out thus far. Specifically . . .
- Identify, clearly, the question to be answered.
- Formulate a postulated arrangement of factors, effects, and conditions to be measured and analyzed.
- Consider existing theory as it relates to the factors, effects, and conditions.
- Start with simultaneity. Do A and B occur simultaneously?
- Then, consider correlation. Are A and B correlated, statistically?
- If neither simultaneity nor correlation is present, stop; and perhaps re-craft your theory.
- Next, check for temporal sequencing. Does A precede B in time?
- Consider other factors and conditions besides A, which might influence B.
- Plan and execute rigorous, scientific, designed experiments or samples to collect data. Control for factors and conditions known or suspected to influence, predict, correlate with, or cause B. Use double-blind experiments when human behavior is involved.
- Use power analysis to plan appropriate sample sizes and guard against Type I and II statistical errors.
- Measure factors and effects as precisely as possible
- Perform rigorous multi-variate analysis and predictive, mathematical model-building to determine the significant predictors of B (which may be causes).
- Expect multiple explanations (predictors, causes) of B. Expect, interpret, and understand interactions between predictors.
- Employ subjective matter expertise to explain analytical results in practical, scientific terms.
The mathematical problem, when there may be multiple causes of an effect, is this:
Final Thoughts
We’re often interested in the relationship between factors, conditions, and predictors on the one hand, and effects, outcomes, and responses on the other hand. We want to know what the predictors are of real-world phenomena. What the influencers are.
In some cases, we also want to know if the predictors are causes of the effects.
There are well-established criteria for credibly claiming causation. And sound, objective analytical processes and steps.
Use quantitative analysis, wisely. But, combine it with subject matter expertise.
Causation is not an insurmountable objective. But, it entails more effort and analysis to prove it.
If we have correlation, without causation—all is not lost. At least, if we use a rigorous process, we understand the phenomenon better. And, we can always continue the exploration.
References
Anderson, B. (n.d.). Establishing causation. https://info.umkc.edu/drbanderson/establishing- causation/#:~:text=To%20establish%20 causation%20you%20need,the%20X%20%2D%3E%20Y%20relationship
Dictionary.com. (2022). https://www.dictionary.com
Jaffe, A. (2010), Correlation, causation, and association: What does it all mean? Psychology Today. https://www.psychologytoday.com/us/blog/all-about-addiction/201003/correlation-causation-and-association-what-does-it-all-mean
Kelleher, A. (2016). If correlation doesn’t imply causation, then what does? https://medium.com/causal-data-science/if-correlation-doesnt-imply-causation-then-what-does-c74f20d26438
