The model-building approach in best multiple binary logit model with higher order interaction variables
Binary Logit analysis is commonly used by most researchers but there is no proper procedure or approach of model-building in Binary Logit analysis. Therefore, this research was aimed at developing a procedure to find the best model. An illustration of the model-building approach in Multiple Binary L...
| 第一著者: | |
|---|---|
| フォーマット: | 学位論文 |
| 言語: | 英語 英語 |
| 出版事項: |
2009
|
| 主題: | |
| オンライン･アクセス: | https://eprints.ums.edu.my/id/eprint/43170/1/24%20PAGES.pdf https://eprints.ums.edu.my/id/eprint/43170/2/FULLTEXT.pdf |
| 要約: | Binary Logit analysis is commonly used by most researchers but there is no proper procedure or approach of model-building in Binary Logit analysis. Therefore, this research was aimed at developing a procedure to find the best model. An illustration of the model-building approach in Multiple Binary Logit analysis had been introduced. The dependent variable of Multiple Binary Logit model is a qualitative nature (taking a value of 0 or 1). Besides introducing multiple independent variables into the model, all possible combinations generated interaction variables are included. In order to obtain a set of selected models (with significant variables), a progressive elimination (one by one, least significant first) of the insignificant variables is employed. It was also proposed to use the Modified Eight Selection Criteria (M8SC) by replacing SSE (sum square of error) with Deviance statistic (G²) to finally single out the best model. The proposed procedure was applied to four different case studies. Detailed procedure was exposed, illustrated and explained in the thesis. One important finding that must be mentioned here would be the discovery of how to estimate the missing values. So far, missing values had never been given the recognition and considered unimportant. Therefore, it had been always ignored or eliminated. However, an alternative approach had been found to estimate these missing values. The estimated missing values were then incorporated into the proposed procedure and a new corresponding selected model which is more accurate and reliable was ultimately obtained. An extended work involving these estimating missing values had been included in this work to ensure the best model obtained gives a better estimate and forecast. This work was indeed a refreshing and rewarding experience. The model-building approach through the Multiple Binary Logit model was established. For all case studies, it was found that up to second-order interaction variables were significant in the best model obtained. |
|---|