Mixing mixed-mode designs in a national health interview survey: a pilot study to assess the impact on the self-administered questionnaire non-response

Table 3 Multilevel model for item non-response in SAQ (n = 21,763)

	Model 1 (Empty model)	Model 2	Model 3^a
	OR (95% CI)	OR (95% CI)	OR (95% CI)
Fixed effects at item level
Mode: P&P		2.25 (1.44–3.52) *	2.13 (1.33–3.43) *
Fixed effects at respondent level
Age			1.00 (0.98–1.02)
Men			0.81 (0.52–1.28)
Higher education level ^b			0.79 (0.47–1.33)
Country of birth: non-Belgium			1.00 (0.55–1.81)
Employment status (ref: non-employed) ^c
employed			0.91 (0.48–1.71)
student			0.84 (0.31–2.26)
Household size: 1+			0.97 (0.59–1.58)
Random parameters
Variance between respondents (SE)	1.92 (0.28) *	1.70 (0.25) *	1.66 (0.25) *
ICC _respondents	0.37	0.34	0.34
PCV		11.4%	13.6%
Model fit: −2 Log Likelihood	7840.89	7829.08 **	7826.66

Values based on SAS Proc Glimmix; Estimation method = Laplace
^*p value < 0.05; ** = likelihood ratio test significant; ICC = intra-class coefficient of correlation; PCV = proportional change in variance; SAQ = self-administered; P&P = paper-and-pencil
The PCV expresses the change in the respondents’ level variance between the initial model and the model with more terms.
^a Best fitting model
^b For students the education level equals the highest education level in the household. If this was not possible it was based on their current field of study
^c “Employed” refers to respondents who had a paid job for at least one hour a week at the moment of the interview, “Unemployed” refers to respondents without a paid job at the moment of the interview, “Student” refers to respondents who were (fulltime or part-time) enrolled in the regular education system at the moment of the interview. Students who had a (part-time) paid job were classified as students

ISSN: 1471-2288