When performing data analysis using a structural equation model, the main concern is the estimation and testing of path coefficients. Testing methods for path coefficients include the t-value method, Satorra-Bentler's scaling testing statistic method, and Bollen-Stine's bootstrap method. All of these methods are asymptotic testing methods that assume a normal distribution of data, and may provide different testing results depending on the type of given data and model. In this study, the testing power (1-type 2 errors) was calculated under various circumstances using simulations for the testing methods that are mainly used, and the performance of these testing methods was evaluated through the results. The ML method was found to have robust properties in deviation from normal distribution in terms of estimation and power. Also, in most situations, the power of the Satorra-Bentler scaling testing was greater than that of the ML method. And in some situations, when the sample size is small, the power of the bootstrap method based on WLS was slightly smaller than that of other methods. Although the results of this study cannot be generalized in all situations, it is thought that they can be referenced in various fields of research using structural equation models.