Validating LIWC Dictionaries

Abstract

This article presents a validation strategy for creating new LIWC (Linguistic Inquiry and Word Count) dictionaries. The context used for this validation study is an analysis of the Israeli–Palestinian rhetoric preceding the 1993 Oslo I accords by Donohue and Druckman (2009). In that study, coders were trained to code six constructs in the texts: looking forward, looking backward, power, affiliation, trust, and mistrust. For the current study, six new dictionaries of words tapping these constructs were developed and evaluated. Several data reduction strategies were used to identify the most relevant words pertaining to the constructs. These reduced dictionaries were correlated with the coder results from the earlier research. The forward-looking construct emerged as a significant correlate with the forward-looking coder rating across all data reduction techniques and the full dictionary list. These findings suggest that both approaches be used. Automated dictionary coding is more efficient; human coding is more sensitive to context.