Controlling Bias in Mail Questionnaires

Abstract In all instances where the mail questionnaire is used, one must be prepared to deal with the problem of bias due to nonresponse. Attention is given to two aspects of the problem: (1) maximizing response by every possible means in order to cut down the size of the non-respondent group whose characteristics and attitudes are unknown; and (2) making allowances or corrections for any bias that may exist in the incomplete returns. In mail follow-ups of veterans who had not responded to the initial questionnaire, personalized salutation and true signature did not lead to significant increases over nonpersonalized forms in rate of response, but the use of special delivery letters markedly increased returns. A multiphasic survey, covering several potentially interesting topics, yielded higher rates of response than a single subject survey of the same population, and also greatly lessened an interest bias in response. Successive waves of response may give an informal basis for estimating bias among those who do not respond even after several follow-ups. The view that mail surveys of a homogeneous population are not seriously affected by bias is refuted by data drawn from such surveys.