Table 1.

Performance of Hicks et al.’s (2018) PF approach and linear regression when estimating the cumulative effect of a time-varying exposure (simulations based on 10,000 replications)

Type of Dynamic Selection		PF Approach		Linear Regression
True State Dependence ($T_1\to T_2$)	Exposure-Induced Confounding ($T_1\to X_2$)	Bias	RMSE	Bias	RMSE
No	No	−0.0002	0.061	−0.0001	0.061
Yes	No	−0.0004	0.056	−0.0004	0.055
No	Yes	0.0561	0.084	0.0561	0.083
Yes	Yes	0.0559	0.078	0.0558	0.078

Notes: RMSE = the root mean squared error. The cumulative effect is defined here as the average marginal effect of a unit increase in the exposure at each time point. The Monte Carlo standard errors on the biases are each about 0.0006.