Transfer Learning for Spatial Autoregressive Models with Application to U.S. Presidential Election Prediction
针对美国总统选举预测中空间数据样本量小的问题，提出一种空间自回归模型的迁移学习框架tranSAR，利用相似源数据提升估计精度，并在摇摆州预测中优于传统方法。
It is important to incorporate spatial geographic information into U.S. presidential election analysis, especially for swing states. The state-level analysis also faces significant challenges of limited spatial data availability. To address the challenges of spatial dependence and small sample sizes in predicting U.S. presidential election results using spatially dependent data, we propose a novel transfer learning framework within the SAR model, called as tranSAR. Classical SAR model estimation often loses accuracy with small target data samples. Our framework enhances estimation and prediction by leveraging information from similar source data. We introduce a two-stage algorithm, consisting of a transferring stage and a debiasing stage, to estimate parameters and establish theoretical convergence rates for the estimators. Additionally, if the informative source data are unknown, we propose a transferable source detection algorithm using spatial residual bootstrap to maintain spatial dependence and derive its detection consistency. Simulation studies show our algorithm substantially improves the classical two-stage least squares estimator. We demonstrate our method’s effectiveness in predicting outcomes in U.S. presidential swing states, where it outperforms traditional methods. In addition, our tranSAR model predicts that the Republican Party would win the 2024 U.S. presidential election.