py-why/dowhy

Add asymptotic confidence intervals for average treatment effect for linear regression with effect modifiers

Open

#336 创建于 2021年11月19日

在 GitHub 查看
 (5 评论) (0 反应) (0 负责人)Python (6,453 star) (883 fork)batch import
good first issue

描述

Refer to #335. DoWhy defaults to bootstrap confidence interval. It will be good to implement a computationally efficient confidence interval method.

With effect modifiers, the average treatment effect is a linear combination of parameters. For example, for y=a + b1t + b2t.x1 where x1 is the effect modifier, the ATE of t on y is b1+b2.mean(x1). Both b1 and b2 coefficients are random variables and so is mean(x1), so need to compute the standard error of the combined quantity.

This is a good start, for anyone who would like to contribute: https://scholar.princeton.edu/sites/default/files/jmummolo/files/interaction_models_jm.pdf

贡献者指南

技术栈
python
领域
data
议题类型
feature
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
3
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-2 days
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
stale
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
Causal inference fundamentalsLinear algebraStatistics basics
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
55
研究方向
Study the method described in the linked PDF for computing standard errors of linear combinations of coefficients. Look at the existing confidence interval implementation in DoWhy (likely in the `estimators` module) to understand the current bootstrap approach. Implement a new method that computes asymptotic confidence intervals for ATE with effect modifiers, ensuring it integrates with the existing API.