The estimated coecients and standard errors for four separate demand specications are shown
in Table 6; the means and standard deviations of the variables used in estimation are reported in
Appendix Table A8. Model 1 uses a simple specication where only chronic TrOOP, premium,
brand xed-eects and an enhanced plan xed eect, with random coecients as specied above,
are included in the utility equation. We add variables incrementally in the following columns; Model
4 is our full specication. In all models the switch parameter estimates indicate that consumers
are more likely to switch plans if they receive premium or coverage shocks or have an acute shock
to their health. Women, nonwhite, lower-income and older enrollees have lower threshold values
to trigger awareness, and hence are more likely to switch. These results are consistent with the
probit regression estimates in Appendix Table A6 and also with intuition. Overspending mistakes
are more costly for older enrollees who spend a higher fraction of their income on drugs and for
lower-income enrollees for whom the excess spending is more burdensome. For this reason they
tend to require smaller prompts in order to re-optimize their choice.