A comparison of best-worst scaling marginal and rank methods

This study compares the marginal and the rank methods for analysing best-worst scaling case 2 data using a simulated and an empirical example dataset. Simulation results suggest the rank method improves accuracy of estimates compared to the marginal method as measured by bias and mean square error. The rank method reduced bias on average by 48% across all coefficient estimates as compared to the marginal-CL method. Results from the empirical example align with those of the simulation and added robustness to the findings.