Attribute non-attendance in the choice experiment with machine Learning: WTP for organic apples in Germany

Abstract Choice experiments (CE) have been widely applied as an important method for estimating consumer preferences and predicting consumer behaviors. However, it is well known that the non-attended attributes in CE could lead to some bias and estimation inefficiency. Therefore, it is crucial to accurately filter out the attribute non-attendance (ANA) and to ensure that there is no effect from the change in correlation structure while increasing the fitting degree of the estimation model. In addition to the general approaches for identifying ANA, we applied a machine learning algorithm, namely, the Least Absolute Shrinkage and Selection Operator (LASSO), to tackle this problem. The approach is empirically applied to analyze consumers’ willingness to pay (WTP) for organic apples in Germany where promoting organic agriculture is an important policy. The results show that when the penalty factor ( λ ) takes different values, the attribute of Organic can been screened out when countries of origin and size are considered. It implies that ANA might be linked to niche market or hypothetical market. Furthermore, German consumers prefer apples produced in France and Italy over those produced in China, paying a premium of €1.53 and €1.40, respectively. Meanwhile, apple size is negatively correlated with consumers’ purchase preferences. Lasso performs efficiently and conveniently in identifying ANA in CE.