The most important factor of the statistical research of social welfare is the accuracy. To ensure accuracy, errors occurred by a variety of causes should be lessened and whole processes of investigation should be thoroughly controlled. Research data, which lacks accuracy, generates biases and has no effective value even though the result data fulfills conditions of relevance and comparability. In order to lessen errors and ensure correctness in the course of investigation, it is needed to figure out the causes of errors and review indicators that can quantifiably explain the elements affecting accuracy by each stages of research investigation. In this study, I intend to look into the range of error and non-response error, the typical patterns of non-sampling errors frequently occurred in processes of planning, implementation, analysis interpretation of the investigation to find out the ways to correct errors.