家計調査への思い出し記入導入のエージェント・ベース・シミュレーションによる評価

抄録

 The result of car sales estimated by Japanese Family Income and Expenditure Survey fluctuate month by month. It is caused by the low sampling ratio. To avoid the expense increase and to achieve more accurate survey, introduction of retrospective survey is one of the potential candidates. This study tries to evaluate the effect, using Agent-Based Simulation.

 The purpose of this study is two-folded: a. to assess the effect of introduction of retrospective survey without any memory recall errors, b. to assess the effect with memory recall errors.

 The simulation model is set according to the sampling method of the real survey, which is not strictly random sampling. Memory recall errors are set to occur according to the random walk model. Each agent has different level of memory recall errors.

 The result for the first purpose is that without memory recall errors, the standard deviation of the estimated results show significant improvement.

 The other result for the second purpose is that with memory recall errors, the average of the estimated results differ from the true value caused by the equalization of seasonal sales pattern.