A novel decision bias, called the evaluation bias (EB), was reported by White, Pothos, and Busemeyer (2014). In a sequence of 2 stimuli of opposite affective valence, evaluating the first stimulus leads to a more contrasting evaluation for the second one, compared with when the first stimulus is just observed. The EB is consistent with a long tradition of constructive influences or decision biases in questionnaire judgments. The prediction of the EB was based on the application of a quantum probability model, taking advantage of the unique role of evaluations in quantum probability. In the present work, we develop the quantum model so as to examine whether similar predictions are possible in the context of real questionnaires, where precise control over the relative valence of stimulus pairs is impossible. It is shown that an EB prediction can be extracted and we test this prediction in an organizational opinion survey, administered to a range of organizations across 4 experiments (total N = 868 and 84 organizations) and with two different languages. In all experiments, there was clear evidence for an EB. We examine the result with the quantum model and Hogarth and Einhorn’s (1992) belief-adjustment model. Both models can broadly capture the empirical findings and so offer promise for providing a formal understanding of constructive influences. (PsycInfo Database Record (c) 2020 APA, all rights reserved)