Inferring Grandiose Narcissism From Text: LIWC Versus Machine Learning

Abstract

People have long used language to infer associates’ personality. In quantitative research, the relationship is often analyzed by looking at correlations between a psychological construct and the Linguistic Inquiry and Word Count (LIWC)—a program that tabulates word frequencies. We compare LIWC to a machine learning (ML) language model on the task of predicting grandiose narcissism (valid N = 471).We use the ML model discussed in Cutler and Kulis and formulate it as an extension of LIWC. With a strict validation scheme, the LIWC prediction was not more accurate than chance. The ML representation did moderately better ( R2 = .043). This indicates that the ML model was able to preserve personality information where LIWC failed to do so, suggesting that precautions are warranted for social-personality research that relies solely on LIWC.