Regression results from mixed logit model
| No interaction | Interacted with Tier 1 | Interacted with Tier 2 | Interacted with Tier 3 | |
|---|---|---|---|---|
| Random coefficients | ||||
| Cigarette tier price—mean | −0.314*** | |||
| (0.049) | ||||
| Cigarette tier price—SD | 0.221*** | |||
| (0.031) | ||||
| Constant—mean | 4.565*** | 5.101*** | 2.558*** | |
| (1.006) | (0.777) | (0. 631) | ||
| Constant—SD | 1.119*** | 1.159*** | 1.524*** | |
| (0.256) | (0.129) | (0.105) | ||
| Fixed coefficients | ||||
| Male | 0.961 | 0.803** | 0.760*** | |
| (0.614) | (0.340) | (0.240) | ||
| Age | −0.087*** | −0.075*** | −0.033*** | |
| (0.009) | (0.008) | (0.007) | ||
| Income: <¥1000 | −3.644*** | −2.692*** | −1.009*** | |
| (0.385) | (0.308) | (0.291) | ||
| Income: ¥1000–2999 | −2.509*** | −1.544*** | −0.512** | |
| (0.310) | (0.267) | (0.262) | ||
| Income: ¥3000–4999 | −1.265*** | −0.648** | −0.386 | |
| (0.324) | (0.286) | (0.281) | ||
| Income:≥¥5000 (omitted) | ||||
| Education: Primary school and below (omitted) | ||||
| Education: Middle school | 0.496 | 0.499** | 0.678*** | |
| (0.340) | (0.243) | (0.212) | ||
| Education: High school | 1.337*** | 1.204*** | 0.899*** | |
| (0.330) | (0.254) | (0.225) | ||
| Education: Beyond high school | 2.505*** | 2.039*** | 1.277*** | |
| (0.380) | (0.310) | (0.260) | ||
| Cigarette consumption at Wave 1 | −0.011 | −0.016** | −0.005 | |
| (0.009) | (0.008) | (0.007) | ||
| Wave fixed effects | Yes | |||
| City fixed effects | Yes | |||
| Number of persons | 3477 | |||
| Number of person-years | 8552 | |||
| Number of observations | 34 208 | |||
Note: Robust SEs, in parentheses, are clustered at the person level and weighted for survey sampling. The omitted category for categorical variables is: income above ¥5000, primary education and below, Wave 1 and Beijing. The model includes tier-specific constants. ‘Interacted with Tier x’ means the coefficient is on the interaction term between the variable listed in the first column with Tier x (x=1, 2, 3, 4), as mixed logit model allows variables that vary over choice alternatives, which is similar to conditional logit.
Significance: *p<0.10, **p<0.05, ***p<0.01.