The Missing Data's Effect On Micro Econometrics

Micro econometrics is an interdisciplinary subject between economics and statistics; it is the economic theory and statistical method which research on micro-data mainly about economic information of individual's families and companies. In recent years, domestic micro econometrics focuses on construction, evaluation and testing of models. However, data, the foundation of modeling, is hardly involved in this subject. Theory, mathematics, and data are the three most important factors to the success of economic modeling. Whether a economic model is valid directly relies on sampling, especially relies on the quality of sample data. Missing data is a key factor to the quality of sampling. In the social investigations of individuals, organizations, and governments, missing data is a common problem. While there are lots of researches on micro econometrics, the missing data does not get enough attention.Therefore, the:thesis begins with the data deficiency, discusses the reason and mechanism of data deficiency, and discourses the impact of the data deficiency on micro econometric. Concerning surveys, whether there is difference of characteristic between response and nonresponse will influence sampling differently. If the data deficiency is totally random, then only the precision will be reduced while deviation will not be caused. However, if the deficiency in caused on purpose or the deficiency focuses on certain group, there will be obvious unrandom difference between the response and nonresponse, so, they will not share the same quantitative distribution characteristic. As the result, the sample with complete data can not represent the characteristic of entire group. At this time, the deficiency of data will not only increase the variance of estimator, but also the error of estimate.This paper is about how to compensate the data deficiency's influence on micro econometrics models. This method is to fill the missing data by observing the distribution of data. The result of estimation is accurate and efficient. In the case of famers'revenue and consumption, this paper compares the result of regression with expectation maximization to that without expectation maximization. Either in linear regression model or binary selective model, the result of regression is much better than that of listwise deletion and mean imputation。To a certain content, the method can eliminate the deviation hidden by data and represent the difference of all levels in the results. Thus, this method will produce more accurate conclusions.