| 基于独立权重估计二元处理平均因果效应算法 |
 点此下载全文 |
| 引用本文：张雨轩，朱永忠，张洪滔，夏世源.基于独立权重估计二元处理平均因果效应算法[J].计算技术与自动化,2026,(2):10-16 |
| 摘要点击次数: 58 |
| 全文下载次数: 55 |
|
|
| 中文摘要:二元处理中，加权是估计平均因果效应常用方法。倾向得分在传统方法中被广泛使用，但倾向得分模型准确性会影响平均因果效应估计。为此，提出了一种独立权重确定方法。通过样本矩量化加权后各组协变量独立性，将权重确定转化为目标规划问题。求得独立权重可平衡组间差异，且不依赖特定倾向得分模型。并给出使用独立权重估计平均因果效应的敏感性分析方法。数值模拟表明，独立权重估计平均处理效应偏差范围集中，结果稳健。在IHDP数据集上估计结果精度达到99.16%，偏差为0.0338。对比传统方法，该方法偏差最小，精度最高，对初始权重参数选择不敏感，结果稳健。 |
| 中文关键词:因果推断 独立权重 二元处理 平均因果效应 |
| |
| Algorithm for Estimating Average Causal Effect of Binary Treatment Based on Independent Weight Estimation |
|
|
| Abstract:In binary treatment, weighted methods are commonly used to estimate the average causal effect. Propensity scores, which make covariates independent of the grouping variable, are widely used in traditional methods. But the accuracy of the propensity score model affects the estimation of the average causal effect. To address this issue, an independent weight determination method is proposed. By quantifying the independence of covariates in each group after weighting through sample moments, the determination of weights is transformed into a goal programming problem. Obtaining independent weights can balance the differences between groups and does not rely on a specific propensity score model. A sensitivity analysis method for estimating the average causal effect using independent weights is also provided. Numerical simulations show that the deviation range of the average causal effect estimated by independent weights is concentrated, and the results are robust. On the IHDP dataset, the estimation accuracy reaches 99.16%, with a deviation of 0.0338. Compared with traditional methods, this method has the smallest deviation, the highest accuracy, is insensitive to the selection of initial weight parameters, and the results are robust. |
| keywords:causal inference independent weights binary processing average causal effect |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
位访问者,今日访问