Recall, recognition and the measurement of memory for print advertisements: A reassessment

Application of the two-step approach to structural equation modeling to the PARM data studied by Bagozzi and Silk results in quite different conclusions as to the psychometric properties of recall and recognition scores. The data they analyzed are more consistent with a simple, alternative measurement model, which has different implications for the dimensionality questions Bagozzi and Silk raised. After controlling for random error, recall and recognition scores for print ads are highly correlated and yet discriminable. So, while recall and recognition have considerable common variance, the unidimensionality hypothesis is rejected. Either recall or recognition (or both) contains a significant amount of specific variance. Taking reader interest scores into account does not change the conclusion; reader interest scores are best accounted for as a lower reliability indicator of recognition.