1) Data analysis can lead to misleading conclusions if analysts do not account for causation and simply observe correlations in data. Correlation does not necessarily imply causation. 2) When analyzing treatments or interventions, it is important to control for confounding variables but not mediating variables to accurately assess causal relationships. 3) Variables that are actually colliders (common effects of other variables) can introduce spurious correlations if those other common causes are not accounted for. 4) Techniques from causal inference and probabilistic graphical models, like do-calculus, can help data scientists properly reason about and interpret causal effects and the results of interventions based on observational data.