Et si nous arrêtions d'utiliser systématiquement l’analyse en composantes principales (PCA)? Efficacité des méthodes d’extraction de facteurs dans l’analyse factorielle exploratoire (EFA).

Research output: Contribution to conferenceOtherpeer-review

Abstract

[automatically translated] Fabrigar et al. (1999) and, more Worthington and Whittaker (2006) described the limitations of using BCP as factors extraction method in the EFA procedures. The authors emphasize that the abuse of this practice is probably explained by the fact that this method is presented as the first choice in most data analysis software. The objective of this work is to show that there is still an excess of use in the practice of the PCA and on the other, to give basic guidelines for choosing the correct method for E FA. This work identifies the key research published between 2011 and 2013 in journals with impact factor in which the EFA is conducted with the PCA. It highlights as this practice is still used incorrectly in the field of psychological research. Of a total of 1203 articles on the adaptation and validation of psychological scales, 162 (13.5%) used only PCA and 265 (22%) was used improperly PCA, associated to other analytical methods. The 776 articles (64.5%) used other factors extraction methods (or PCA to validate structures to a single factor). The work will also provide the basic criteria necessary for proper implementation of the EFA methods (Thompson, 2004), for example the choice of a consistent approach with any sample normality violations, with an example of how the PCA leads to a significant distortion of the results. Our results show that other extraction procedures (Principal Axis Factoring, Maximum Likelihood Estimation) are superior to methods of PCA in terms of efficiency in the distribution of the variance in relation to the various factors extracted. our scales is desired that are valid and reliable: Do you have confidence in the effectiveness of a barometer that measures the fever?
Original languageFrench
Pages24-24
Number of pages1
Publication statusPublished - 2014

Cite this