Does more balanced survey response imply less non-response bias?