This dissertation concerns the post-selection bias issue in statistical inference on treatment effects when a large number of covariates are present in a linear or partially linear model. While the estimation bias in an under-fitted model is well understood, we address a lesser known bias that arises from an over-fitted model. We show that the over-fitting bias can be reduced or eliminated through data splitting, and more importantly, smoothing over random data splits or bootstrap-induced splits can be pursued to mitigate the efficiency loss. We also discuss some of the existing methods for debiased inference and provide insights into their intrinsic bias-variance trade-off, which leads to an improvement in bias controls. Based on these insights, we thoroughly study the connections between our current framework and average treatment effects estimation under the Neyman-Rubin causal model. A careful analysis shows that the post-selection bias issue can exist in a wider range of treatment effect estimation procedures. Under appropriate conditions, we show that our proposed estimators for the treatment effects are asymptotically normal and their variances can be well estimated. We discuss the pros and cons of various methods both theoretically and empirically, and show that the proposed methods are valuable options in post-selection inference.