Abstract
TWO METHODS OF ADJUSTING FOR NON-PARTICIPATION BIAS IN ROADSIDE BREATH TESTING SURVEYS ARE COMPARED. METHOD 1 USES THE ASSUMPTION THAT THE PROBABILITY OF MEASURED BAC LEVEL CONDITIONAL ON A PRIORI JUDGEMENT OF DRINKING IS THE SAME FOR RESPONDENTS AND NONRESPONDENTS. METHOD 2 USES THE ASSUMPTION THAT THE PROBABILITY OF A PRIORI JUDGEMENT OF DRINKING CONDITIONAL ON BAC LEVEL IS THE SAME FOR RESPONDENTS AND NONRESPONDENTS. METHOD 2 YIELDS SUBSTANTIALLY LARGER ESTIMATED PROBABILITIES OF HIGH BAC FOR NONRESPONDENTS. BOTH METHODS CAN BE SUPPORTED BY SOUND LOGICAL ARGUMENTS. METHOD 2 IS SHOWN TO YIELD NEGATIVE PROBABILITIES, WHEN APPLIED TO A LARGE ROADSIDE SURVEY SAMPLE, AND IS THEREBY REJECTED.(Author/publisher).