UVM Theses and Dissertations
Format:
Print
Author:
Putname, Matthew E.
Dept./Program:
Community Development and Applied Economics
Year:
2013
Degree:
MS
Abstract:
Obesity is a growing problem in the United States and the world. Numerous studies have looked at the drivers of obesity but are usually limited by disciplinary foci or the unavailability of a complete data set from which to draw conclusions. This thesis seeks to go beyond these previous limitations by statistically matching two large national data sets, the National Health and Nutrition Examination Survey (NHANES) and the American Time Use Survey (ATUS) that contain information on both time use, caloric intake, and food behavior, physical activity and sedentary behavior.
The first article discusses the use of propensity score statistical matching (PSSM) to fuse the data sets. PSSM uses a discrete choice model to predict probability of group membership. The advantages of propensity score matching are that it minimizes bias and creates a single balancing score on which to match individuals, nullifying the dimentionality problem.
PSSM requires that variables in both data sets be measured on a similar scale and requires a significant amount of preprocessing. Two key steps prior to PSSM in this instance were to identify under-reporters of calories in the NHANES and account for missing data within both the NHANES and ATUS.
The Goldberg cutoff is a statistical technique to identify under-reporters (outliers) based on Basal Metabolic Rate (BMR), calorie intake and activity level. Missing data is a problem for most surveys and has often been ignored as techniques to address missing data have created as many problems as they have solved. However, the now relevant technique of multiple imputation (MI) is a balanced way of dealing with missing data that minimizes bias, error and los of information.
Article two is a structural equation model (SEM) predicting the association of diet, activity level, food behavior, and sleep patterns on probability of being overweight. The SEM add-on to SPSS, Amos is used for the estimation.
Propensity score matching does seem to provide a useful match. Results from the SEM are in line with the literature showing that age, diet, and activity level are important drivers behind obesity. Not in line with the current consensus the effects of food behavior and sleep were mixed but largely insignificant.
The first article discusses the use of propensity score statistical matching (PSSM) to fuse the data sets. PSSM uses a discrete choice model to predict probability of group membership. The advantages of propensity score matching are that it minimizes bias and creates a single balancing score on which to match individuals, nullifying the dimentionality problem.
PSSM requires that variables in both data sets be measured on a similar scale and requires a significant amount of preprocessing. Two key steps prior to PSSM in this instance were to identify under-reporters of calories in the NHANES and account for missing data within both the NHANES and ATUS.
The Goldberg cutoff is a statistical technique to identify under-reporters (outliers) based on Basal Metabolic Rate (BMR), calorie intake and activity level. Missing data is a problem for most surveys and has often been ignored as techniques to address missing data have created as many problems as they have solved. However, the now relevant technique of multiple imputation (MI) is a balanced way of dealing with missing data that minimizes bias, error and los of information.
Article two is a structural equation model (SEM) predicting the association of diet, activity level, food behavior, and sleep patterns on probability of being overweight. The SEM add-on to SPSS, Amos is used for the estimation.
Propensity score matching does seem to provide a useful match. Results from the SEM are in line with the literature showing that age, diet, and activity level are important drivers behind obesity. Not in line with the current consensus the effects of food behavior and sleep were mixed but largely insignificant.