Social Desirability Bias in the Reporting of Alcohol Consumption: A Randomized Trial

To investigate reporting of alcohol consumption, we manipulated the contexts of questions in ways designed to induce social desirability bias.

We undertook a two-arm, parallel-group, individually randomized trial at an Australian public university. Students were recruited by email to a web-based "Research Project on Student Health Behavior." Respondents answered nine questions about their physical activity, diet, and smoking. They were unknowingly randomized to a group presented with either (A) three questions about their alcohol consumption or (B) seven questions about their alcohol dependence and problems (under a prominent header labeled "Alcohol Use Disorders Identification Test"), followed by the same three alcohol consumption questions from (A).

A total of 3,594 students (mean age = 27, SD = 10) responded and were randomized: 1,778 to Group A and 1,816 to Group B. Outcome measures were the number of days they drank alcohol, the typical number of drinks they consumed per drinking day, and the number of days they consumed six or more drinks. The primary analysis included participants with any alcohol consumption in the preceding 4 weeks (1,304 in Group A; 1,340 in Group B) using between-group, two-tailed t tests.

In Groups A and B, respectively, means (and SDs) of the number of days drinking were 5.89 (5.92) versus 6.06 (6.12), p = .49; typical number of drinks per drinking day: 4.02 (3.87) versus 3.82 (3.76), p = .17; and number of days consuming six or more drinks: 1.69 (2.94) versus 1.67 (3.25), p = .56.

We could not reject the null hypothesis because earlier questions about alcohol dependence and problems showed no sign of biasing the respondents' subsequent reports of alcohol consumption. These data support the validity of university students' reporting of alcohol consumption in web-based studies.