Comparing Support Vector Machines and Neural Networks to Detect Social Desirability Bias Using Person-Fit Indices

• In Event: Machine Learning

Abstract

Since 1933, psychometricians have been concerned about the validity of questionnaires as a result of socially desirable bias (SDB) in response. Several person-fit statistics in literature can detect SDR. Additionally, machine learning classifiers such as support vector machine and neural network, have the potential to distinguish between biased and unbiased responses. This study considers several person-fit indices as features in mentioned classifiers to detect SDB. A comprehensive Monte Carlo simulation study is conducted to evaluate the effect of manipulated factors such as sample size, test length, class separation, and degree of bias on predictive power (i.e., Area Under the Curve, AUC) of mentioned classifiers. The factor mixture model was selected as the population model to generate biased and unbiased items.

Authors

• Sanaz Nazari, University of California - San Diego
  Presenting Author

• Walter L. Leite, University of Florida
  Non-Presenting Author