L'impact des méthodes de traitement des valeurs manquantes sur les qualités psychométriques d'échelles de mesure de type Likert

Authors: Rousseau, Michel
Advisor: Bertrand, Richard
Abstract: The presence of missing answers for some items of a scale of measurement is a phenomenon which any researcher is suitable to meet during his work. Although bias that an inadequate treatment of this non-response can cause are known since nearly 30 years (Rubin, 1976), knowledge of the effectiveness of the various missing values treatment is still very restricted. The present study aims at making knowledge and practices concerning the treatment of the missing values evolve in the context of Likert type scale. The fundamental problem that missing values pose is that it is impossible not to take it into account at the time of the application of a method of statistical analysis, the majority of these methods having been developed to treat matrices of complete data. The models of measurement used in the analysis of Likert type scale data do not escape from this reality. Two models of measurement are studied more in-depth in this project, the classical test model and the Samejima graded model. The main objective of the research undertaken is to evaluate the effectiveness of five missing values treatment, including the multiple imputation method. Moreover, it was aimed to evaluate the impact of the number of subjects, the number of items and the proportion of the missing values on the effectiveness of the methods. The results of this research seem to suggest that the effectiveness of multiple imputation is higher than the other methods, although depending on the model of measurement considered, other simpler methods seem also effective. In conclusion, it is important to note that because no method of treatment can eliminate completely the bias caused by the presence of missing values, it would be preferable to prevent rather than to cure.
Document Type: Thèse de doctorat
Issue Date: 2006
Open Access Date: 12 April 2018
Permalink: http://hdl.handle.net/20.500.11794/18669
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
SizeFormat 
23426.pdf1.47 MBAdobe PDFView/Open
All documents in CorpusUL are protected by Copyright Act of Canada.