Words and deeds: the validity of self-report data on alcohol consumption.

Social scientists have known for many years that there may be notable discrepancies, and even an inverse relationship, between reported and actual behavior. Yet this knowledge has had remarkably little effect on the extent to which reliance is placed on the validity of self-report data. Deutscher (1), in a paper entitled Words and Deeds, has reviewed the famous LaPicrc experiment reported in 1934, and the reaction of other sociologists to the problem at that time. He notes LaPiere's contention that, although no one had ever challenged his argument that men say and what they do are not always in concordance, it had no effect on faith in the value of data gathered via the questionnaire. LaPiere's report was widely cited by surveyors, but when it came to the presentation of data, each author promptly ignored the caution and proceeded to assume that his data was indicative of what people would actually do in real-life circumstances (1, p. 30). It seems to us doubtful that LaPicrc would find reason to change his views today, especially were he to examine reports in the alcohol field. In 1974, Pernanen (2) reviewed this literature and showed that, on the average, estimates of derived from surveys were about half the estimates based on sales statistics. Given that in most of the jurisdictions involved the latter are accurate and comprehensive, one may conclude that self-report data on the volume of alcohol consumed are likely to underreport actual behavior substantially. However, most surveyors have either ignored the problem or dismissed it on the grounds that the underreporting is consistent at all levels of so that misclassification of individuals into broad categories of drinkers is unlikely. For example, Fitzgerald and Mulford (3), in their recent comparison of sales and data, note that survey estimates significantly underreport alcohol consumption but argue that if this underreporting is consistent across all users, the rank ordering of users is preserved and the patterns of distribution of derived from sales and data would be comparable (p. 884). However, their contention that the underreporting is in fact consistent because the data of two surveys are reasonably well fitted by a J-curve serves only to confuse the issue. In Table 1, we present two distributions of alcohol purchases in Ontario. One is based on an analysis of a very large sample of sales records for a 1-month