Preliminary Remarks on the Notion of Causation
The concept of causation lacks a precise legal definition. According to jurisprudence, it falls within the sovereign discretion of judges to assess the existence of a causal link based on the chronology of events, the simultaneity of the fault and damage, the economic connection between behavior and harm, etc.
In other words, it is not possible to establish an irrefutable causal link.
Therefore, for the plaintiff, the task is to convince the judge of the existence of a causal link, while for the defendant, it is to persuade the judge of its absence.
In our view, the most effective method to refute a causal link is to restrict it to related but lesser concepts: coincidence, concomitance, or correlation.
The Rule of the Three Cs
Depending on the efforts made by the plaintiff to defend the existence of a causal link between alleged faults and quantified damages, causation can be criticized from three angles:
Coincidence, defined as the simultaneous occurrence of the wrongful event and the harmful event, where the logical/economic link between the alleged fault and the quantified harm is not evident, and there is no recurrence between the alleged fault and quantified harm.
If the plaintiff provides sufficient accounting elements, it is then possible to propose a regression analysis between the alleged fault and its precedents (such as acts of denigration) on one hand, and the damage (such as a decrease in margin) on the other. In the absence of a correlation between the two elements, one can conclude a coincidence.
Concomitance, where the plaintiff has demonstrated a logical and proportionate economic link between the alleged fault and the quantified harm (damage). In this case, the idea is to invoke the concomitance of several possible causes and the damage.
For example, if acts of denigration were directed against a restaurant owner in 2020, it would be relatively easy to invoke the concomitance of the COVID-19 pandemic and denigration as causes of the decrease in revenue. The argumentation can be complemented by calculating correlation coefficients (Spearman or Pearson), showing either no correlation between the alleged fault and quantified harm or a correlation weaker than with another variable (COVID-19 in our example).
Correlation, where the plaintiff has additionally established a correlation between the alleged fault and the quantified harm (damage), or even demonstrated a causal relationship through a mathematical model.
Limits of Correlation Coefficients
Before discussing the limits of causal models, it is important to note that correlation does not imply causation. Positive correlation between two events can result from chance, inverse causation (event Y causes event X), systemic relationships, or hidden causal variables (the stork effect).
An example of the stork effect involves the observation that birth rates in areas with nesting storks were higher than in the rest of the country. However, storks do not bring babies; the explanation lies in the preference of storks to nest in rural villages with higher birth rates.
Limits of Causal Models
If the plaintiff defends the hypothesis of causation based on more complex models (aiming to control hidden causal relationships), it is essential to criticize the construction of these models.
To do so, the fundamental concept to consider is the Kullback-Leibler divergence, which measures the deficiency (or its deviation from reality) of a model based on biases in its construction sample (such as omitted or unobserved variables).