Adil Omar Bahathiq
Reliable appraisal of risk factor exposure is a prerequisite for efficient prevention. A street survey conducted in Jeddah, Saudi Arabia, included questions for measuring the prevalence of tobacco smoking. When entering the data, occurrence of long sex specific sequences of questionnaires suggested non-random sampling. The objective of this paper was to propose a method for diagnosing non-randomness, and measuring its impact on the prevalence of smoking and on the quality of collected data. The distribution of the length of observed sequences of questionnaires collected among male respondents was compared with expected length under the hypothesis of random selection. For groups of sequences with significant difference between observed and expected length, two specific categories were set up according to the sign of the difference. A third category included sequences without significant length difference. The proportion of smokers and of missing information was compared between the low and the long length categories. Adjusted differences of these proportions was obtained by a logistic regression model. Given that short sequences were underrepresented and long sequences were overrepresented, there was a sampling bias towards long sequences. Adjusted difference for the prevalence of tobacco smoking was 8.7% (95% confidence interval (CI95)-0.3%; 18.7%). For missing data, it was 3.9% (CI95) 0.3%; 8.4%). Adequate sampling design, and sex specific training of interviewers were recommended for reducing biases associated with non-randomness due to the selection procedure. Obtained impact of non-randomness might suggest to adjust the crude prevalence of smoking obtained in this survey.
English 
Spanish 
Chinese 
Russian 
French 
Japanese 
Portuguese 
Hindi