Assorter-Based RLAs for Complex Sainte-Laguë Elections: Reducing sample sizes by trusting manually counted ballots

In several states of Germany, elections for the city council are very complex. They allow voters to cast their vote for one party (so called unchanged ballots) or to distribute their votes to individual candidates from one or several parties (so called changed ballots). While the unchanged ballots are manually tallied by the poll workers, the changed ones are entered into tallying software. We study how to use assorter-based Risk Limiting Audits (RLAs) for such elections. We examine whether we can reduce sample sizes for a given risk limit by focusing the audit on the changed ballots, i.e., not trusting the tallying software.