Recently, choice experiments have been employed in the field of environmental evaluation. The characteristics of conjoint analysis including choice experiments are as follows: first, conjoint analysis can evaluate the value of imaginary goods since this technique uses stated preference data. Secondly, conjoint analysis can calculate the value of each attribute as the marginal rate of substitution between attributes and price.
Conditional logit model is applied to identify utility functions from data collected by choice experiences, discrete choice travel cost model and so on.However, in generally, conditional logit model has two assumptions: linearity of part worth and preferential independence. Preferential independence implies that total utility is the linear summation of part worth. These assumptions lead to the constant marginal utility and the zero estimation of interaction between attributes.
The purpose of this paper is to relax these assumptions and to estimate the value in consideration of the nonlinearity of part worth and the interaction between attributes. For this purpose, we have proposed a conditional logit model with a neural network structure (NNLogit model). The unique qualities and advantages of NNLogit model are to calculate the utilities of profiles directly without formulating utilities as linear functions and to use these data as training data of neural network.
As the verification of NNLogit model, first, the performance of this model has been assessed by using Monte Carlo simulation. From the results of assessment, it is clear that NNLogit model has high performance and availability since this model can incorporate nonlinearity into part worth and evaluate the interaction. Secondly, NNLogit model has been applied to estimate the recreational demands of a forest. In this empirical study, NNLogit model can estimate the various values depending on various statuses of recreational sites, though we obtain only constant values by applying conditional logit model. These estimation results from NNLogit model are useful in marketing research.
抄録全体を表示