3, I fit separate models for rounds with good recommendations and poor recommendations using winner agreement as the dependent variable. Figure 3 visualizes these models. When receiving good recommendations, the difficulty of the decision had almost no influence on the probability of agreeing with the DSS in either condition. When receiving a poor recommendation, subjects were slightly more likely to agree with the DSS when the decision was more difficult, and this was true in both conditions. Because the estimates for customization and non-customization are effectively parallel for both types of recommendations, it does not appear that the difficulty of the decision moderates customization bias, al- though it does seem that when receiving poor recommendations, users of any type of DSS will be more likely to trust it when the decision is difficult than when they have an easier decision.