Non-response bias in alcohol and drug population surveys.