土地改良事業における安全性向上効果の推計‐仮想現実的状況下でのセレクションバイアスの修正‐

Abstract

This report proposes the new method to estimate the effectiveness of safety improvements launched by Land Improvement Projects. This new method reveals that estimates from traditional method are often to be biased and inconsistent. Data about the number of accident are collected from the channels with safety improvements or without them. Because each channel cannot be in both states, we cannot observe both states in one channel. This is a kind of missing data problem and often called “counterfactual”. It is this that makes it difficult to accurately estimate the effectiveness of safety improvements. Fortunately, this problem can be solved and accurate estimate is available using the concept of average treatment effect correctly. When the effectiveness of safety improvements is estimated with ATE, we have to consider the possibility of self-selection problem. We need a different way to estimate effectiveness if self-selection occurs. The number of life saved through safety improvements is approximately 1.171 when self-selection is taken account for, compared to 0.909 when self-selection is not. This result shows the possibility that estimated effectiveness for a program is underestimated because of not taking account for self-selection.