A user table contains duplicate records created when people sign in with multiple email addresses. The analyst samples rows from this table to estimate user-level behavior. What is the best fix before sampling?