Debates about when and whether (partisan) directional motives influence information processing are hard to resolve because rational and motivated learning often look similar. We develop an experimental design to distinguish between these possibilities which focuses on the order in which information is presented. A core tenet of Bayesian updating is that order should not impact final beliefs, but if some information changes the motivation to process other information, order effects may emerge. In our first study, we randomize the partisanship of real endorsements for ballot propositions, as well as whether participants learn about these endorsements before observing other information about the propositions. We find no evidence of motivated information processing across several tests. In a second study, we randomize whether participants themselves argue for or against a proposition, and whether they know this position before observing other information. This produces a strong order effect: being randomized to argue for versus against a position affects beliefs more when it is learned before information about the proposition is provided. We also find suggestive evidence that this order effect is driven by selective attention to information. Overall, our results suggest that motivated reasoning about politics is less prevalent than commonly believed, but may arise primarily when people are in an argumentative mindset.