We study anchoring effects in the elicitation of multidimensional beliefs within a single survey task using a representative sample of the German voting-age population. Respondents estimated government-spending levels across several domains (e.g., education, defense, social security), with randomized exposure to different informational anchors in one domain. Anchors significantly influence elicited beliefs in related domains and partially also shift respondents’ policy preferences. While the anchors change absolute estimates, perceived government-spending rankings remain stable. These findings offer methodological guidance for survey design involving multidimensional belief elicitation in information-provision experiments.