Predicting unemployment durations and the risk of long-term unemployment (LTU) is key for labor market policy planning. When unemployed individuals do not find a job at early stages of their unemployment spell individual and societal costs are potentially very high. It is therefore of crucial importance to identify those at risk of facing LTU. Empirical LTU predictions based on classical administrative data suffer from the shortcoming that these data ignore many dimensions of the job seeker's profile, like personality traits, search behavior and other "behavioral" variables. This paper addresses the question to which degree information on such additional variables is able to improve the fit of LTU predictions. We exploit a unique dataset that features a vast variety of such information for a representative cohort of German unemployed. We assess seven blocks of information by iteratively adding and testing them against the baseline specification using classical (administrative only) variables. Our results show that the addition of information on individual expectations, family background, job search behavior, personality traits and life satisfaction significantly improves the fit of the LTU prediction model. The predictive power of information on expectations, job search behavior and life satisfaction is even higher for a shorter-run outcome.