A survey of political trust has many missing answers, especially among respondents who say they distrust government. Why could analyzing only completed answers distort the estimated average trust?