Is this Causation or Correlation?

Oct. 17, 2020, 7:01 a.m.

Blog Image

Separating correlation from causation can be tricky. However, this distinction is vitally important for interpretation of any statistics. It is all too easy - and counterproductive - to make changes to a process when correlation has been misinterpreted as causation. Using the three established criteria of association, temporality, and nonspuriousness can help make this distinction easier.

A recent article A case series of pediatric croup with COVID-19 in the American Journal of Emergency Medicine highlighted three pediatric patients with Croup and COVID-19. Was this correlation or causation? Did these children have croup caused by COVID-19, or, did they simply have both Croup and COVID-19? Did the authors have sufficient evidence to prove causation? In this blog post I will review how these authors used the three principles of Association, Temporality, and Nonspuriousness to make an epidemiological hypothesis that Croup can be a manifestation of COVID-19.

1. Association

The Cochrane Collaboration definition of causal effect: "An association between two characteristics that can be demonstrated to be due to cause and effect, i.e. a change in one causes the change in the other," shows the importance of association as the first step in determining causation. It is basic human nature to make associations, and most of us are superb at it. Humans use pattern recognition as an important tool for navigating our complex lives.

Randomized trials are the ultimate way to demonstrate association. By controlling one variable, and measuring another, it is elementary to demonstrate the association. However, when randomized trials are not possible, observational evidence can show association.

In the Croup Study, the authors have done an excellent job of showing association. The three cases were explained in detail giving ample evidence that the diagnosis of croup was correct. In addition, the study gives sufficient details to believe that the microbiological diagnosis of COVID-19 was correct.

However, simply having an association is not causation. A very interesting article Storks Deliver Babies (p= 0.008) gives an excellent example of how easy it can be to find association.

Think About Association
Think About Association

2. Temporality

The principle of temporality states that the cause must come before the effect. This seems logical, of course, but can be easily missed in the design of epidemiological studies.

Again, this is a situation that is best addressed by randomized experiments, where we can specifically expose the experimental subject to a treatment first, and then look for the effect later.

The COVID/Croup study is cross-sectional observational study which makes assessing temporality difficult. The three patients appear to have presented to the hospital having both the disease(croup) and the suspected infection(COVID-19) simultaneously. The authors do make a valiant effort to address temporality by constructing careful narratives of their cases including reconstruction retrospectively potential exposures.

Nonetheless, proving association and temporality is not enough to prove causation.

Time Order
Assess Temporality

3. Nonspuriousness

Assessing nonspuriousness involves ensuring that the relationship between two variables is not actually due to some other hidden third variable. For instance, imagine a survey of reading level and number of dental cavities in schoolchildren. This is likely to show that reading level increases with number of cavities. Why? Likely this is related to a third spurious variable - the child's age. As age increases, the number of cavities and reading level increase together.

This is another area where randomized experiments reign supreme. By randomizing participants we are giving the best possible chance of making sure that the spurious factors are equal in both the treated and untreated groups.

Obviously, in the COVID-19 / Croup paper, we cannot randomize children to get or not get COVID-19. However, the authors did an excellent job of trying to eliminate spurious factors. The authors give excellent case detail to help the reader sort out what other factors in the clinical history could lead to spurious findings. For instance, the authors present the viral test results of the children, which in all three cases did not show any evidence of coinfection.

Although the authors make a compelling argument by demonstrating no other apparent cause for croup in these children, a study of three cases is simply to small to prove nonspuriousness. For instance, many cases of croup - probably around 20% - do not have a microbiological diagnosis - making it a distinct possibility that the finding of COVID-19 in these three croup patients may have been spurious.


Although having a known mechanism for the phenomenon in question is not strictly necessary from a statistical point of view, it can certainly make the argument more convincing. Although many important scientific findings were observed first, and explained later, in most cases modern science relies heavily on developing a theoretical mechanism supported by current literature before claiming causality. This is the critical "could this possibly be true?" part of any manuscript. Being able to explain a mechanism is one of the most important parts of the introduction of a manuscript. In randomized experiments the choice of treatment is almost always guided by a theoretical mechanism which is developed during the study planning.

In this paper, it seems likely that COVID-19 could be a causative agent for croup. Croup is after all a viral disease.

And So.. Causation or Correlation?

In the end does this paper support that croup is caused by COVID-19? I believe the authors have done an excellent job of addressing association, temporality, and nonspuriousness in this short paper of only three cases. For certain, this paper should serve as a springboard to look into the role of COVID-19 in croup. However, with a small paper of only three cases there are a number of ways in which we may be confusion correlation with causation.

Do you want a visual reminder to help you assess correlation versus causation? Get our Causation versus Correlation infographic.

By Jeffrey Franc


If you liked this post, please share it on Twitter or Facebook.


You must be registered as a user to add a comment.

Please login or create a FREE account here

Are You Ready to Increase Your Research Quality and Impact Factor?

Sign up for our mailing list and you will get monthly email updates and special offers.

Email Address

Protected by reCAPTCHA