Skip to main content
Fig. 1 | BMC Medical Research Methodology

Fig. 1

From: Time dependent hazard ratio estimation using instrumental variables without conditioning on an omitted covariate

Fig. 1

Simulation Results for estimating three-piece time-dependent hazard ratio when the instrument is strong: The nine panes represent each combination of early, mid and late follow-up, with (i) strong positive confounding, (ii) no confounding and (iii) strong negative confounding. Each pane overlays scatterplots of the median estimated bias versus the true hazard ratio, our IV based estimator in black, the standard Cox MPLE in red, the estimator of Wang in green and the estimator of Martinez-Camblor et al. in blue. Each point represents the median of over 1000 estimates obtained from that many randomly generated datasets

Back to article page

BMC Medical Research Methodology

ISSN: 1471-2288

Contact us