MEASUREMENT ERROR AND THE STRUCTURE OF ATTITUDES - ARE POSITIVE AND NEGATIVE JUDGMENTS OPPOSITES

Survey researchers routinely group survey questions into ''batteries'' of items sharing similar wording and response format. In this workshop essay, we demonstrate that this practice may introduce nonrandom response error into the data, rendering invalid standard approaches to reliability assessment and structural equations modeling. Using data on public perceptions of Hispanics collected by the 1991 NES Pilot, we show how nonrandom error may be detected and how LISREL (analysis of covariance structures) may be used to correct its distorting influence.