Evaluating the relative predictive validity of measures of self-referential processing for depressive symptom severity.

Introduction: The self-referential encoding task (SRET) has a number of implicit measures which are associated with various facets of depression, including depressive symptoms. While some measures have proven robust in predicting depressive symptoms, their effectiveness can vary depending on the methodology used. Hence, understanding the relative contributions of population differences, word lists and calculation methods to these associations with depression, is crucial for translating the SRET into a clinical screening tool. Methods: This study systematically investigated the predictive accuracy of various SRET measures across different samples, including one clinical population matched with healthy controls and two university student populations, exposed to differing word lists. Participants completed the standard SRET and its variations, including Likert scales and matrix formats. Both standard and novel SRET measures were calculated and compared for their relative and incremental contribution to their associations with depression, with mean squared error (MSE) used as the primary metric for measuring predictive accuracy. Results: Results showed that most SRET measures significantly predicted depressive symptoms in clinical populations but not in healthy populations. Notably, models with task modifications, such as Matrix Endorsement Bias and Likert Endorsement Sum Bias, achieved the lowest mean squared error (MSE), indicating better predictive accuracy compared to standard Endorsement Bias measures. Discussion: These findings imply that task modifications such as utilising Likert-response options and the use of longer word lists may enhance the effectiveness of screening methods in both clinical and research settings, potentially improving early detection and intervention for depression.