|
|
Analysis of Missing Data: Comparing Performance of Traditional Methods across Mechanisms
Petrúšek, Ivan ; Soukup, Petr (advisor) ; Hendl, Jan (referee)
The objective of this thesis is to evaluate different methods of dealing with missing values in data analysis. The thesis is divided into three major chapters. The first chapter summarizes the theoretical literature on missing data and focuses on missing data mechanisms in particular. The second chapter introduces traditional methods for addressing missing data in sociological research. The third chapter assesses the performance of these methods by analyzing simulated data sets for two variables (income, IQ). For practical analysis (chapter 3), we simulated missing data according to three different mechanisms (MCAR, MAR, NMAR) and varied the proportion of missing values under these mechanisms (10%, 20%, 30%). Then, we applied each of the following four methods of addressing missing values: complete-case analysis, arithmetic mean imputation, regression imputation, and stochastic regression imputation. In order to evaluate the performance of each of these methods we performed correlation and regression analyses for each experimental condition. The results of these simulations are largely in agreement with existing theoretical literature on the subject of missing data. In the case of NMAR, all solution methods provided biased parameter estimates. In the case of MCAR, only complete-case analysis and...
|
|
|
Empirical comparison of imputation methods for missing values in data
Ostrenska, Alona ; Holý, Vladimír (advisor) ; Zouhar, Jan (referee)
Missing values are present in all types of data such as different surveys, socio-scientific information etc. In many applications, it is necessary to replace missing observations to maintain the size of the dataset needed for the statistics. This bachelor thesis at first place introduce the categories of causes of missing data and the problems connected with them. The next step is to acquaint with common methods of imputation of missing values and the explanation of applicating those methods on real data in the context of linear regression. Then the assumptions of linear regression models that are based on data with artificially created missing observations are verified. These observations are removed using the mentioned mechanisms and different proportion of missing, with seven subsequent imputation methods. Regression models constructed based on such imputed datasets are then statically verified. Finally, imputation models are compared using different statistics and visualizations and is suggested possible solution - particular methods in case of a real problem of incomplete data.
|
|
|
Analysis of Missing Data: Comparing Performance of Traditional Methods across Mechanisms
Petrúšek, Ivan ; Soukup, Petr (advisor) ; Hendl, Jan (referee)
The objective of this thesis is to evaluate different methods of dealing with missing values in data analysis. The thesis is divided into three major chapters. The first chapter summarizes the theoretical literature on missing data and focuses on missing data mechanisms in particular. The second chapter introduces traditional methods for addressing missing data in sociological research. The third chapter assesses the performance of these methods by analyzing simulated data sets for two variables (income, IQ). For practical analysis (chapter 3), we simulated missing data according to three different mechanisms (MCAR, MAR, NMAR) and varied the proportion of missing values under these mechanisms (10%, 20%, 30%). Then, we applied each of the following four methods of addressing missing values: complete-case analysis, arithmetic mean imputation, regression imputation, and stochastic regression imputation. In order to evaluate the performance of each of these methods we performed correlation and regression analyses for each experimental condition. The results of these simulations are largely in agreement with existing theoretical literature on the subject of missing data. In the case of NMAR, all solution methods provided biased parameter estimates. In the case of MCAR, only complete-case analysis and...
|
guest ::
