Multinomial and Mixed Logit Modeling in the Presence of Heterogeneity: A Two-Period Comparison of Healthcare Provider Choice in Rural China

This study aims at testing the theoretical issue according to which multinomial logit (MNL) would give lower performance than a mixed multinomial logit (MMNL) in the presence of heterogeneity. To do so, we construct two samples of patients surveyed within the same regions in rural China, but of an interval of 18 years, with a difference in preference heterogeneity due to income growth and population aging. With the 1989-1993 sample, both models have predicted price effects; however with the 2004-2006 sample, unlike MMNL, MNL failed to predict price effect. The explanation is that the impact of price on choice became more heterogeneous in the later than the former sample, thus heterogeneity makes a difference between MNL and MMNL. The absence of meaningful divergences of distance effects between the two models can also be explained by the evolution of heterogeneity in distance preferences over the period. The coefficients of price and distances with MMNL are higher than with MNL, indicating stronger price and distance effects in MMNL estimations. Another advantage of MMNL is the possibility to measure the extent of heterogeneity. The findings suggest caution when interpreting estimation results with MNL if heterogeneity is deemed important.