Introduction The mode of data collection may affect the outcome of a health indicator. This study aimed to examine the consistency between answers to an identical question on chronic morbidity obtained through a face-to-face interview in a health interview survey (HIS) and a self-administered questionnaire from a population census.

Methods During the last quarter of 2001, 2710 people aged 15 years and older participated both to a census and a HIS in Belgium. An individual linkage was performed between the two data sources by using the National Population Registry ID number. Consistency of answers to the question on chronic morbidity was assessed by the κ-statistic.

Results The prevalence of chronic morbidity was 29.4% (HIS) and 26.6% (census). Consistency was relatively poor, with a κ-statistic of 0.56 (95% CI 0.52 to 0.60). The κ-statistic did not differ by gender, but was substantially lower among persons aged 75 years and older (0.44; 95% CI 0.41 to 0.48) than among younger individuals (0.55; 95% CI 0.51 to 0.59). The κ-statistic was also lower for non-Belgians (0.43; 95% CI 0.40 to 0.47) than for Belgians (0.57; 95% CI 0.54 to 0.61). Consistency differed further among educational groups, although no real educational gradient was observed.

Conclusion There was no satisfactory correlation between self-reported chronic morbidity data in the HIS and the population census. The consistency also differed across population subgroups. The mode of data collection appears to impact the estimates. Estimates and sociodemographic determinants of self-reported chronic morbidity should be interpreted cautiously.