How We identify cause-effect relationships given evidence: An exploratory study

Abstract: For well-informed decision-making and precise predictions, the ability to discern causality is imperative. An innate understanding of cause-and-effect relationships in daily life can be deceptive and lead to incorrect interpretations of correlations between variables. In literature estimates of the capacity to identify causal relationships from factual information remain rare. This paper builds upon Kendall and Charles’s 2022, study of the influence of persuasive false narratives on subjects’ inference within a linear cause-effect schema. Drawing on Oprea’s research in 2020, we extend Kendall and Charles (2022) to non-linear patterns and the relationship between accuracy and complexity. In our framework participants face repeated tasks with data generation processes (DGP) of varying structure with 3 variables (Eberhardt, 2017) that they need to deduce by observing the data. Our findings reveal that even in the simplest case with three variables, the concept of conditional independence poses a significant challenge in correctly identifying cause-and-effect relationships, with accuracy levels only marginally exceeding random guessing (10-20 percent improvement). Moreover, different DGPs exhibit varied accuracy that does not align with standard t-complexity (Oprea 2020), where more connections imply greater complexity. Accuracy in our task is rather determined by the type of data source of the variable (observed or intervening) and how much the relationship or lack thereof matches the rest of the relationships in the DGP.
Instructions and demo are available by link

Kendall CW and Charles C (2022) Causal narratives. Technical report, National Bureau of Economic Research
Eberhardt F (2017) Introduction to the foundations of causal discovery. International Journal of Data Science and Analytics 3: 81–91
Oprea R (2020) What makes a rule complex? American economic review 110(12): 3913–3951.