Question wording effects in the assessment of global self-esteem

Marsh(1996) produced evidence that method effects associated with negativel y worded items might be responsible for the results of earlier factor analy tic studies that reported finding positive and negative self-esteem factors in the Rosenberg Global self-esteem scale (Rosenberg, 1965). He analyzed d ata collected from children using a 7-item self-esteem measure. This report details attempts to replicate Marsh's analysis in data collected from two samples of adults who completed the full 10-item Global Self-Esteem (GSE) s cale. The results reported here are similar to those given by Marsh in so m uch as a correlated uniquenesses model produced a superior fit to the data than the simple one factor model (without correlated uniquenesses) or the o ften reported two factor (positive and negative self-esteem) model. However , whilst Marsh reported that the best fit was produced by allowing negative item uniquenesses to correlate with each other, the model that produced th e best fit to these data was one that contained correlated positive item un iquenesses. Supporting his claim that differential responding to negative a nd positive self-esteem items reflects a method effect associated with read ing Marsh also showed that factors associated with negative and positive it ems were most distinct among children who had poor reading scores. We repor t a similar effect among a sample of older adults where the correlation bet ween these factors was compared across two groups who were selected accordi ng to their scores on a test of verbal reasoning.