Maximizing Confidence in Married Couple Samples

Most probability samples of married couples do not have response rates high enough to justify the use of inferential statistics. Recognizing that response rates are likely to be low this paper describes 4 procedures that can be used in study design during data collection and in analysis to partly compensate for this failure. The study describes a research project on the effects of relative occupational statuses of spouses for dual-earner marriages. 489 couples in Hamilton County Ohio were interviewed in 1982-1983; the estimated response rate was 44%. 1) A nonparticipant questionnaire was mailed to an additional 436 couples and answered by 162 thus bringing the response rate up to 63% for some of the questions. This nonparticipant questionnaire showed that participants were younger and had higher socioeconomic status. The study bias is still probably underestimated because many couples neither participated nor returned questionnaires. 2) When a survey is reasonabley close in time to a census and when a sampling area conforms to a census reporting area bias can be estinated by comparing the 2. 3) Data can be weighted so that the sample more closely resembles the parameters of the population. Weighting is not totally understood and may not correct bias. A sample biased on known parameters is probably also biased on unknown ones. 4) Replication can also generate confidence in findings. While replication is not always interesting it may be necessary.