碩士 / 國立臺北大學 / 統計學系 / 102 / In this study, we use panel survey data form the project “Reasons for adolescent drug abuse” and investigate the effect of missing treatment for multinomial logistic regression. Variables selection and data analysis methods are based on the work in Huang Fang-Mei, Lai Hui-Ying, Wu Chi-Yin (2005) ─ "The impact of family background and personal characteristics of the junior high school educational achievement". The data includes panel survey on students, teachers and parents. Imputation goal is that the mass missing part-- "students’ admitted high school ".
First, using T-test, Chi-square test to screen the significant variables and apply logistic regression to find the significant explanatory variables to predict the missingness. We found that the largest impact variable on the missingness is “ the student's average score”. Second, we use this significant variable to imitate the missing pattern of the original data set and construct 50 groups missing data sets from the baseline. We then perform four missing treatments：Listwise Deletion, Stepwise Logistic Regression Imputation Method, Markov Chain Monte Carlo Single Imputation , Markov Chain Monte Carlo Mutiple Imputation. Finally, perform Multinomial Logistic analysis and compare the result in different missing data treatment with baseline.
We found that stepwise logistic regression imputation fas the closest result with the baseline. Applying this best missing treatment to the original data, the result of the multinomial logistic regression showed quite different from the result without missing treatment. The complete missing treatment process suggests by this research can be a reference for researcher with mass missing data.
| Identifer | oai:union.ndltd.org:TW/102NTPU0337033 |
| Date | January 2014 |
| Creators | LIN, SHU-YANG, 林書揚 |
| Contributors | WANG, HUNG-LUNG, 王鴻龍 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 34 |
Page generated in 0.0022 seconds