Patterns of public opinion recently observed in American politics tempt the conclusion that substantive arguments and evidence are less effective, or ineffective, at changing partisan minds when they overtly contradict cues from in-party leaders. This conclusion follows naturally from theories of partisan motivated reasoning. However, observations of public opinion do not provide the counterfactual outcomes required to draw this conclusion. Here we report a large-scale survey experiment in which we randomized exposure to the policy positions of Donald Trump and Joe Biden, as well as information that overtly contradicts their positions. Our design incorporates 24 policy issues and 48 information treatments. We find that the information does persuade partisans on average, and, critically, fully retains its persuasive force even when paired with countervailing cues from in-party leaders. This result holds across policy issues, demographic subgroups, and one- and two-sided cue environments, and is puzzling for partisan motivated reasoning theory.