Identifying average treatment effects (ATE) from quasi-experimental panel data has become one of the most important yet challenging endeavors for social scientists. The difficulty lies in accurately estimating the counterfactual outcomes for the potentially treated units in the absence of treatment. Perhaps the most popular method to estimate average treatment effects is the Difference-in-Differences (DID) method. The key assumption of the DID method is that outcomes of the treated units would have followed a path parallel to the control units in the absence of treatment and violation of this ``parallel lines" assumption will result in biased estimates. This dissertation consists of three essays, which either build on existing methods (essay 1 and 3) or propose a new method (essay 2) that can be used even when the ``parallel lines" assumption of DID does not hold. In essay 1, we derive the asymptotic distribution of the HCW method, which is computationally simple as it only involves least squares regressions. However, in cases where treatment and control units are positively correlated, the HCW method may have less predictive efficiency than other methods such as the synthetic control and modified synthetic control method, which impose the restriction that weights are non-negative. The popular synthetic control method additionally imposes the restriction that the weights sum to one, which can be a helpful regularization condition when there are many control units. In essay 3, we provide the inference theory for both the synthetic control and modified synthetic control method through projection theory and propose a computational algorithm using subsampling to compute the confidence intervals. In order to apply the HCW method, synthetic control method and modified synthetic control method, the number of control units needs to be smaller than the pre-treatment sample size. In essay 2, we propose the augmented DID method, which can be used where there are many treatment and control units, but is less flexible than the three aforementioned methods. In short, this dissertation provides several methods and their inference procedures to identify average treatment effects. Which method should be used when depends on the structure of the data.