This is accomplished by introducing a probability density function for attributes that are assumed to have heterogeneous value among respondents, h(β), enabling the estimation of the magnitude of this heterogeneity. We separated each of the attribute levels into binary-coded indicator variables.