**5. Estimation Results**

We conducted the mixed logit regressions using the simulated maximum likelihood estimator. Firstly, the correlation test shows that there is correlation between each pair of the attribute variables. For example, the correlation coefficient between allergen presence labelling and nutrient and compositional change labelling is 0.74. Obviously, the correlation test result is in conflict with the IID assumption of the conditional logit model. Therefore, we run the ML model with correlated normally distributed coefficients. The result of the LR test for the nationwide sample is 3858 and significant at *p* < 0.001, which indicates that the null hypothesis is rejected. In other words, the respondents have heterogeneous preferences. Therefore, the conditional logit model has a poorer fit compared to the ML model.

Table 4 provides the estimation results of respondents' preferences for the enhanced mandatory labelling of the GM soybean oil. The results show that price has a negative effect in all equations, indicating an increase in GM soybean oil price decreases the probability of


a consumer choosing the oil. More importantly, most proposed enhanced GM mandatory labelling information is positive.

**Table 4.** Estimation results of the mixed logit model.

The parameters in the bottom panel of the output are the elements of the lower-triangular matrix L. Standard errors in the parentheses. \*\*\* and \*\* denote statistical significance at the 0.01, 0.05 and 0.10 levels, respectively.

> The coefficients of the three variables of labelling attributes are 0.427 (*p* < 0.01), 0.104, and 0.579 (*p* < 0.01) on a national scale (Table 4). The results sugges<sup>t</sup> that Chinese urban consumers are in favor of the enhanced mandatory labelling of the GM soybean oil. The most attractive and influential labelling attribute is traceability codes, followed by allergen presence labelling with a smaller effect. The least important labelling is nutrient and compositional change labelling, and consumers are more likely to select an alternative based on other enhanced labelling included.

> To better understand consumers' preference for mandatory labelling conveying health and safety attributes, we estimate the WTP values using the parameter estimates from the ML model. The WTP values for each attribute are shown in Table 5. The magnitude of WTP and their ranks are consistent with that of the coefficient estimates from the ML model (Table 4). Positive WTP values represent the amount of money that the consumers are willing to pay for the specific labelling attributes. The highest WTP value is found for the traceability codes. Specifically, the consumers are more likely to pay for the traceability codes nationwide, with a paymen<sup>t</sup> of 8.92 RMB, followed by the allergen presence labelling with a value of 6.57 RMB. Regionally, eastern consumers show a positive preference for all three attributes, with the paymen<sup>t</sup> amounts of 11.24 RMB, 8.87 RMB, and 6.48 RMB for traceability codes, allergen presence labelling, and nutrient and compositional change labelling, respectively. Central consumers only show a positive preference for the traceability codes, i.e., 7.41 RMB. However, western consumers show no preference.


**Table 5.** Willingness to pay for a premium for enhanced mandatory labeling information of the GM soybean oil.

Lower bound and upper bound for 95% confidence interval in the parentheses. \*\*\* denote statistical significance at the 0.01, 0.05 and 0.10 levels, respectively.

> Furthermore, the survey results show that about 62.44% of the urban consumers state that they are willing to pay for the enhanced mandatory labelling (Table 3). For those who are willing to pay, the average WTP is RMB 18.22 for traceability codes, followed by RMB 17.50 for allergen presence labelling. The WTP for nutrient and compositional change labelling is the smallest with a paymen<sup>t</sup> amount of RMB 8.17.
