Mobile phone and interactive computer interviewing to measure HIV-related risk behaviours: the impacts of data collection methods on research results

'Reporting bias' is particularly relevant to sexual behaviour studies because respondents tend to present socially desirable answers [1,2]. Data collection methods that ensure anonymity provide an important means of overcoming the bias [3,4]. Experiments with various methods have been reported [1,3,5,6]. However, only a few studies have evaluated the outcomes of different modes of data collection [1,5,7]. The study interviewed 2416 18–60-year-old Hong Kong men returning to Hong Kong from Shenzhen in mainland China during April 1997, who were intercepted at the exit of the checkpoint by systematic sampling. Part A of the questionnaire included the respondents' background information and four non-sensitive questions about AIDS-related attitudes. Part B included sensitive questions on sexual risk behaviour (Table 1).Table 1: Frequencies and item response rates of self-reported sensitive sexual behaviours by the three modes of data collection for the Cross-border Travellers Study (part B questions) Part A answers were obtained from face-to-face interviews of all respondents. To compare the efficacy of collecting sensitive data on sexual behaviours by the method adopted by the authors (M1) with two traditional methods (M2 and M3), part B data were collected by these three modes using an identical questionnaire. After part A was completed, respondents were randomly allocated to one of the three methods of administering the part B interview. M1 respondents were provided with a mobile telephone, on which they responded to pre-recorded questions by keying in their responses. This digitized interactive recording system was tailor-made by the local telecommunication utility. If the respondents did not answer within a stipulated time, the system would repeat the question. No personal identification data were collected. M2 respondents were asked to return the completed part B questionnaires in the provided self-sealing envelopes to the interviewers. M3 respondents were given a postage-paid, self-addressed envelope containing the part B questionnaire. They were requested to mail it back to the University. Part A results show that M1, M2 and M3 samples had comparable demographic and travel profiles and a similar knowledge level and perceptions on AIDS. However, M1 respondents systematically reported a higher frequency of risk behaviours, whereas M2 respondents systematically reported a very low level of risk behaviours (Table 1). The frequency of M3 answers tended to be much closer to M1 than to M2 (Table 1). The item non-response rates for M1 respondents were lower than 2% for all part B questions, whereas those for M2 and M3 respondents ranged from 5.1 to 14.7% and 0 to 4.8%, respectively (Table 1). Approximately 35% of the respondents refused to enter the study when they were first approached by interviewers. Almost 100% of those who agreed to participate in the study completed part A. The percentage of part A respondents who went on to complete part B was approximately 90% for both M1 and M2 and 36% for M3. Although the self-administration method (M2) has often been used to minimize reporting bias, it reported much lower levels of risk behaviours and higher item non-response rates. The postal method (M3) was more able to detect sensitive risk behaviours than M2. However, the very low completion rate (36%) of M3 makes it a less credible method. There is no reason to believe that M1 respondents would over-report risk behaviours systematically. It is likely that M1 is better able than M2 to avoid reporting bias. Although the completion rates of M1 and M2 were similar, the higher tendency for reporting bias makes it a less suitable method for studying sensitive behaviours. Another important conclusion is that the nature of data collection is an intervening variable in determining the results of reported risk behaviours. The results of AIDS behaviour studies therefore have to be interpreted with extreme caution. Data from different studies should be compared only if the data collection mode is comparable. In practice, M2, like most other interviewing methods, is `objectively' keeping respondents' data confidential. It seems that it is the `subjective' psychological responses, or the implicit perception of confidentiality, rather than `objective confidentiality', that affects biased reporting behaviours. Joseph T. F. Lau Joe Thomas Joseph L. Y. Liu