In the Belgian HIS, information is gathered on the health status, life style and medical consumption of a representative sample of the population through a face to face interview and a self-administered questionnaire. The national population register is used as the sampling frame. Interviewers are recruited all over the country and have to pass through a standard training program. As in many other surveys unit non-reponse is an important concern. Compared to other national health interview surveys, unit non-response in the Belgian HIS is quite high. Moreover, it seems to increase. Between 1997 and 2008 the household refusal rate, calculated as the number of households who refused participation among those who were contacted, has risen from 39% to 45%. It is generally agreed that the motivation, training and follow up of the interviewers may have an important impact on the non-response. The impact of the interviewer on the response-rate can be explored by assessing to which extent response rates vary between interviewers after controlling for respondents' characteristics which may be related to non-response. The aim of this study was to verify if household's non-response varies significantly among interviewers and if there is a link between interviewer's characteristics and non-response. This was investigated separately for contact failure and refusal rate.Data were used from the HIS 2004, in which 10.013 households were selected for participation and 264 interviewers carried out the fieldwork. Information on age, sex and nationality of the household's reference person, household composition and place of residence was obtained from the national population registry. Information on the housing type and characteristics of the residential area was collected during the fieldwork. The household database was merged with data on the interviewer's gender, age and educational attainment. Non-reponse was explored by using multilevel logistic regression models with interviewer's characteristics as dependent variables, household characteristics as covariates and household's neighbourhood and interviewer as random effects. After adjustment for household characteristics and neighbourhood the proportion of non participating households is lower among interviewers belonging to the older age groups. Compared to the reference group of youngest interviewers (<45 years) the odds ratio for refusal is 0,67 (95% CI 0,60-0,75) for interviewers between 45 and 64 years and 0,46 (95% CI 0,35-0,61) for interviewers of 65+ years. The odds ratios for contact failure are respectively 0,69 (95% CI 0,59-0,81) and 0,62 (95% CI 0,41-0,92). Interviewers with tertiary education have significantly lower refusal rates than interviewers with only secondary education (OR 0,88; 95% CI 0,79-0,99). The variation of contact failure and refusal rates by interviewer remains significant after adjustment for household and interviewer characteristics.Our results indicate that in the Belgian HIS, non-response varies substantially among interviewers. Neither characteristics of the household, nor characteristics of the interviewer can explain this variation completely, but it is striking that older and high educated interviewers score better. If adjustment is made for characteristics that have an important impact on the response rate (such as the type of household, the neighbourhood, etc.) we would not expect large variations in response rate by interviewer. Large variations may indicate that there is a difference in the approach interviewers follow to contact the selected households and convince them to participate. The response rate in the Belgian HIS could probably be improved by putting more effort in the recruitment, training and follow up of interviewers, targeting specifically young interviewers and interviewers with a lower education level. Focusing the recruitment of interviewers on higher educated persons in the age groups above 45 years may also be a strategy to improve the response rate.
