DescriptionThe dissertation consists of three self-contained but related essays on causal inference and its applications in e-commerce operations. The first two chapters adopt econometric models, especially causal inference related models, to provide empirical evidence on the value of environmental responsibility and high-quality logistics on e-commerce platforms. According to the needs of empirical study, the last chapter develops an improved causal inference algorithm, which provides more robust estimation results and is conducive to empirical studies. In particular, Chapter 2 empirically studies the impacts of an environmental regulation on secondary markets, which is considered as an important part of sustainable operations. In the paper, we adopt the difference-in-differences with the synthetic control method to study a natural experiment: the implementation of Shanghai's compulsory waste-sorting regulation that took effect on July 1, 2019. We study two key market outcome variables: number of listings and purchase volume. The study finds that due to a strong sense of environmental responsibility, the implementation of a waste-sorting regulation resulted in an 8.42% decrease in the number of resale listings posted by the young generations──that are sensitive to the environmental outcomes──in secondary markets. The second essay, in Chapter 3, studies how and to what extent customers value the high-quality delivery service on a leading online retail platform. Specifically, we employ the difference-in-differences identification to study a natural experiment and analyze 129,448 representative SKUs' market performance on a leading retail platform. The study seeks to quantify the economic value of high-quality logistics services. We find that logistics matters to online platforms: For example, the removal of a high-quality delivery option could reduce sales of Alibaba's retail platform by 14.56%. Last, the third essay, in Chapter 4, develops an improved causal inference algorithm called Stable Synthetic Control (Stable-SC), which combines the synthetic control method with anomaly detection algorithms. The algorithm seeks to mitigate the impact of confounding factors on estimation results, especially in the treatment (or post-treatment) periods. In the study, we further apply the improved algorithm to a few simulated and real-world datasets, and all experimental results demonstrate that our Stable-SC method provides more robust estimation results even under the challenging data environment with serious confounding factors.