| Summary: | 碩士 === 國立臺北大學 === 統計學系 === 107 === Most of the general scale construction is analyzed and discussed with complete data. In practice, due to the characteristics of the respondents, part of the sample survey scale may be responded by different respondents. This study takes the satisfaction questionnaire as an example, to explore various possible improvement patterns of questionnaire design and its analysis methods when covering different interviewed groups under multiple issues. The recommendations of the items which can be modified is made according to the empirical data, and provided references for relevant scholars.
Due to the needs of the issue, this study divides questions into 9 types for discussion such as questions answered by all respondents or by the people who have jobs. Each group has the same attribute respondents, makes the Individual analysis relatively simple and accurate.
Through the discriminant test method and the Pearson correlation test method, this study examines the satisfaction questions of survey, to make recommendations for preferential deletion of the items. The reliability analysis is used to measure the consistency, stability and reliability of the questionnaire, by carried out on stratified and whole sample level of questions answered by all respondents. As a result, the Cronbach's α coefficient in whole sample frame is greater than 0.9, and in all stratified levels are greater than 0.6. The cumulative explained variation is 51.7%. Overall, the consistency of the questionnaire is acceptable. This study also recommend the deletion order according to the influence of each item on Cronbach's α.In the aspect of construct validity, using structural equation modeling to analyze each stratified sample, the factor loadings of each item is above 0.5, and the explained variation is above 40%. The factor structure is appropriate. The observation model is poor fit, however the residuals are low makes the model still reasonable fit. In the exploratory factor analysis, the extraction result shows that there are two components needs more consolidations.The consistency of the variables that are re-consolidated into a 6-factor model compared with the original model is acceptable. The extraction results of the each factor are only one component, and each of the evaluation is slightly better than the 5 levels. Indicate that the original survey is a credible survey. If there are too many items, it is recommended to delete more items and the classification of questions could further refined.
|
|---|
