The effects of missing data when surveying alcohol habits
AIMS - This study aimed at describing the effects of missing data when surveying alcohol consumption using a Random Digit Dialling procedure. METHODS - Data was part of the Monitor project including repeated monthly data on the alcohol habits in the general Swedish population. Non-respondents during four months were followed up a year later and asked to do a shortened telephone interview and were compared to a concurrent sample of respondents (n=2552 versus n=6005). Further, using a second approach, the monthly levels of non-response was related to the level of measured alcohol use in a time series analysis (n=67500). RESULTS - The results indicated no differences in the level of reported alcohol or tobacco use with except for a slightly higher proportion of alcohol abstainers in the sample of initial non-response. The time series showed no pattern of co-variation between the obtained nonresponse levels and the assessed levels of alcohol or tobacco use. CONCLUSIONS - On the basis of the results it was meaningful to make a distinction between "soft" non-respondents (responding after extensive contacting effort) and "hard" non-respondents (not responding albeit extensive effort) and the results suggest that inclusion of the "soft" non-respondents does not by necessity lead to higher levels of assessed alcohol use.