The Psychometric Properties of the GSS Wordsum Vocabulary Test

Social scientists in many disciplines have used the General Social Survey’s ten-item Wordsum vocabulary test to study the causes and consequences of vocabulary knowledge and related constructs. In adding up the number of correct answers to yield a test score, researchers have implicitly assumed that the ten items all reflect a single, underlying construct and that each item deserves equal weight when generating the total score. In this paper, we report evidence suggesting that extracting the unique variance associated with each word and measuring the latent construct only with the variance shared among all indicators strengthens the validity of the index. We also report evidence suggesting that Wordsum could be improved by adding words of moderate difficulty to accompany the existing questions that are either quite easy or quite difficult. Previous studies that used Wordsum should be revisited in light of these findings, because their results might change when a more optimal analytic method is used.