Mallick, Taslim S. (2009) Conditional weighted generalized quasilikelihood inferences in incomplete longitudinal models for binary and count data. Doctoral (PhD) thesis, Memorial University of Newfoundland.
- Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.
There exists an inverse probability weight (INPW) based unconditional estimating equation approach (a correction to accommodate the missingness nature of the data) for computing unbiased regression estimates in an incomplete longitudinal set-up mainly for binary data. It is however known that this INPW based unconditional estimating equation approach still may produce regression estimates with large bias. It is demonstrated in this thesis that it would be much better to use an INPW based conditional estimating equation approach to obtain unbiased and hence consistent estimates for the regression effects. This approach however requires the longitudinal correlation structure to be known. Under the assumption that the binary or count data follow an autoregressive order-1 [AR(1)] type model, the thesis develops a conditional weighted generalized quasilikelihood (CWGQL) approach that accommodates both missingness and the longitudinal correlation issues properly This appears to be a major improvement over the existing INPW based generalized estimating equation (GEE) approach which either fails to use the longitudinal correlations or uses 'working' correlations approach. Extensive simulation studies are undertaken to examine the relative performance of the proposed CWGQL approach with the existing INPW based GEE approach. Finally the incomplete longitudinal models are generalized to study the survey based incomplete longitudinal data. A stratified finite population is considered to examine the performance of a stratified random sampling (StRS) based CWGQL approach in estimating the regression parameters involved in the finite population for both binary and count data models.
|Item Type:||Thesis (Doctoral (PhD))|
|Additional Information:||Includes bibliographical references (leaves 97-99).|
|Department(s):||Science, Faculty of > Mathematics and Statistics|
|Library of Congress Subject Heading:||Estimation theory; Generalized estimating equations; Regression analysis|
Actions (login required)