Implications of the HIV testing protocol for refusal bias in seroprevalence surveys

BMC Public Health

Table 4 Binary and multinomial logistic regressions predicting refusal of testing for HIV (Zewditu Memorial Hospital, Addis Ababa, 2003–04)

	Binary logistic regression predicting refusal (exp(b) or odds ratios)		Multinomial logistic regression predicting refusal (exp(b) or relative risk ratios)
	B & C versus A		B versus A	C versus A	B versus A	C versus A
	Model 1	Model 2	Model 3		Model 4
Likelihood of infection	1.01**	1.01**	1.02**	1.01**	1.01**	1.01**
Counselor (vs #1)
Counselor 2	0.07**	0.07**	0.12**	0.05**	0.06**	0.09**
Counselor 3	0.46	-	0.09**	0.71	-	-
Counselor 5	0.50	0.33	1.58	-	1.48	-
Counselor 6	-	-	-	-	-	-
Counselor 7	0.01**	0.01**	0.01**	0.02**	0.00**	0.01**
Counselor 8	0.46**	0.44	0.56	0.42**	0.26	0.68
Counselor 9	0.65	0.56	0.43	0.78	0.38	0.76
Study month (vs period prior to ART)	0.82**	0.81**	0.78**	0.84**	0.73**	0.88*
Ward (vs ER)
GY		1.23			0.44	2.68**
IM		1.42			0.56	2.73**
SU		0.83			0.60	1.04
Male		1.96**			1.68**	2.30**
Age		1.05			1.13**	1.02
Age squared		.999*			.998**	1.00
Education (vs no schooling)
Grade 1–6		1.33			1.16	1.57
Grade 7–12		1.27			0.79	1.95**
> 12^th grade		1.70*			0.82	2.85**
Marital status (vs never married)
Married		1.44*			1.15	1.71**
Sep/Div/Wid		1.92*			1.78	2.03
N	1544	1357"	1546		1359
LR chi² (df)	354.57(8)	364.68(18)	406.49(18)		453.04(38)
Prob > chi²	0.00	0.00	0.00		0.00
Pseudo R²	0.21	0.25	0.18		0.24
Log likelihood	-680.93	-554.13	-917.05		-732.04

Notes:
* p ≤ .10; ** p ≤ .05
See Table 3 and the notes to that table for a definition of the explanatory variables. Other variables that were controlled for, but omitted in the final models because they lack statistical significance are: birth region (Addis Ababa versus other); religion (Orthodox Christian versus other); a squared term for likelihood of infection; an interaction between the likelihood of infection and study month; an interaction between birth region and sex. Because education and marital status were only introduced as additional variables in the second month of the study, models two and four are based on fewer cases.