We conducted a citizen forecast for the German Federal Election of 2021 by fielding a survey that asks respondents about the expected vote share for the parties on the federal level and for the district candidates. The survey included an between-subject experiment in which we showed respondents either information from the most recent poll about party support, information about the last district results, or both and asked them again about their expected district candidate results. We analyze whether the information increases predictive accuracy and certainty in citizen forecasts of district-level results. We further study whether the information aligns respondents' forecasts with proportional or unit swing forecasting models and whether citizen forecasts can be used to improve swing models.