Table 3:

Utility function, class membership, and diagnostics from latent class, latent variable multinomial logistic model (LCLVMNL) model of NIQ smokers’ experimental choices. Robust standard error (robust to heteroscedasticity).

	Class 1: Non-switchers			Class 2: Switchers
Utility Function	Estimate	Rob.std.err	Rob.t-ratio(0)	Estimate	Rob.std.err	Rob.t-ratio(0)
Pod E-cigarette	−5.078	0.275	−18.446	−0.129	0.092	−1.405
Disposable E-cigarette	−3.525	0.173	−20.322	−0.363	0.087	−4.193
Cigarette Menthol Smoker	0.374	0.291	1.285	0.168	0.128	1.315
Cheaper	0.062	0.010	6.545	0.062	0.010	6.545
Expensive	−0.024	0.011	−2.154	−0.024	0.011	−2.154
Helps You Quit	−0.469	0.126	−3.714	−0.032	0.032	−1.007
Healthier Than Cigs	−0.289	0.118	−2.452	0.015	0.029	0.516
No Nicotine	0.654	0.112	5.831	0.170	0.028	6.158
Less Nicotine	0.499	0.162	3.088	0.054	0.055	0.973
More Nicotine	0.132	0.198	0.666	0.070	0.047	1.492
Menthol	−1.047	0.227	−4.620	−0.431	0.065	−6.619
Fruit	−1.723	0.198	−8.711	−0.342	0.050	−6.796
Sweet	−0.579	0.150	−3.854	−0.357	0.050	−7.173
Class membership probability	Estimate	Rob.std.err.	Rob.t-ratio(0)
Constant class 1	1.287	0.113	11.395
Older	0.240	0.126	1.906
Female	0.081	0.117	0.693
Education	−0.647	0.118	−5.484
Income	−0.412	0.172	−2.390
Black	−0.428	0.178	−2.409
Asian	−1.413	0.527	−2.681
Hispanic	−0.027	0.244	−0.109
E-cigarette user	−1.508	0.135	−11.145
tau(Latent variable health behaviours)	0.272	0.082	3.329
Class shares	0.68			0.32
Diagnostics
Individuals	2,000
Observations	24,000
Estimated parameters	61
LL(0)	−26,366.69
LL(fitted)	−19,953.66

Robust-t-ratio(0) – Robust t-ratio (versus 0). LL(0) – log-likelihood when all parameters are set to zero. LL(fitted) – log-likelihood of fitted model. Tau is a latent variable to control for the effect on health behaviours from COVID – see Appendix for methods and results.