Fig. 2: DAG and plots of IVs. | Nature Computational Science

Fig. 2: DAG and plots of IVs.

From: Quantifying causality in data science with quasi-experiments

Fig. 2

a, Graphical representation of how IVs vary X to measure its causal effect on Y. The crossed edges depict assumptions needed for valid IV inference: the exclusion restriction (crossed grey edge) as well as no unmeasured confounding between IV and Y (crossed red edge). b, Histogram of causal effect estimates across 100 simulated datasets for both IV (orange) and typical regression (blue) when the true treatment effect (dashed line) of X on Y is confounded. c, Histogram of causal effect estimates across 100 simulated datasets for both IV (orange) and typical regression (blue) when the true treatment effect (dashed line) of X on Y is confounded and the exclusion restriction is violated.

Back to article page