According to reject preference regularity, we then estimated the models to elicit the marginal utility of each attribute separately for all three regions. However, accounted for differences in variances, log likelihood functions (represented by equation (5)) of each region were multiplied by the relative scale parameters identified by combined data model. The scale parameters multiplied to the log likelihood function were 1.441, 1.229, and 1.000 for North region, Northeast region, and Central region, respectively. These values represent smaller error variances for the North and Northeast regions relative to the Central region responses. Following these steps, coefficient vectors of each region model were controlled for differences in scale, so they can be directly compared. Table 3 shows the estimated results of combined models and regional models.