Child-street migration among HIV-affected families in Kenya: a mediation analysis from cross-sectional data

Figures & data

Figure 1. Percentage of respondents reporting a child lives on street by HIV status of parents.

Figure 2. Path analysis of reporting a family child lives on the streets and HIV-affected households.

Table 1. Univariate and bivariate description of mediator and outcome measures, by reported HIV status.

	Univariate			No reported HIV			Respondent and/or partner reported HIV+			Test statistic (p)
	Mean (%)	95% CI		Mean (%)	95% CI		Mean (%)	95% CI
Wealth index	4.55	4.43	4.68	4.69	4.56	4.81	3.8	3.48	4.13	t: 4.69 (<.001)
Age (years)	38.21	37.61	38.82	37.79	37.2	38.39	40.72	39.11	42.33	t: −3.24 (<.001)
Years of completed school	5.96	5.78	6.14	6.15	5.97	6.33	5.15	4.69	5.61	t: 6.15 (<.001)
Number of children in household	3.55	3.46	3.64	3.42	3.33	3.51	4.14	3.85	4.44	t: 3.68 (<0.001)
Self-reported alcohol use	7.54%	6.25%	8.84%	6.75%	5.54%	7.97%	17.24%	12.00%	22.48%	χ^2: 27.31 (<.001)
Social support (7-point Likert-type scale)	5.42	5.36	5.48	5.5	5.44	5.56	4.87	4.72	5.03	t: 7.05 (<.001)
Overall health (4-point Likert-type scale)	1.85	1.8	1.9	1.97	1.93	2.02	1.15	1	1.3	t: 12.13 (<.001)
Family function (4-point Likert-type scale)	2.66	2.64	2.68	2.68	2.66	2.7	2.56	2.52	2.61	t: 3.67 (<.001)
Violent attitudes toward children (5-point Likert-type scale)	3.59	3.55	3.63	3.63	3.59	3.67	3.33	3.22	3.44	t: 4.96 (<.001)
All biologically related children enrolled in school	32.07%	29.69%	34.52%	33.55%	30.92%	36.25%	14.20%	9.22%	20.54%	χ^2: 25.51 (<.001)
At least one family child lives on the street	6.49%	5.29%	7.69%	4.22%	3.15%	5.28%	22.35%	16.18%	28.51%	χ^2: 98.79 (<.001)
Attributable fraction:							34.57%
No HIV reported	88.52%	86.83%	90.06%
Maternal HIV only	6.03%	4.90%	7.33%
Paternal HIV only	1.80%	1.20%	2.59%
Dual-affected HIV	3.66%	2.78%	4.71%

Notes: Mediator, control and outcome variables stratified by reported HIV status. T-tests used for continuous and χ² used for categorical variables. Attributable fraction is the predicted percent reduction in households reporting at least one child lives on the street if HIV in the population were eliminated.

Table 2. Fixed-effects logistic regression of reporting at least one child currently lives on the streets.

	Child lives on the street
	Model 1			Model 2
	OR	95% CI		OR	95% CI
No reported HIV	1	REF		1	REF
Maternal HIV only	4.31***	2.14	8.68	3.65***	1.69	7.9
Paternal HIV only	4.12*	1.37	12.4	4.26**	1.32	13.77
Dual-affected HIV	5.33***	2.53	11.2	4.11***	1.81	9.33
Wealth index	0.9	0.8	1.02	1	0.88	1.13
Age (years)	1.01	0.99	1.03	1.01	0.99	1.03
Years of completed schooling	0.93	0.86	1.01	0.92	0.85	1
Household children (#)	1.18**	1.04	1.34	1.19*	1.04	1.36
Report alcohol use	1.58*	1	2.53	1.37	0.59	3.15
Social support (7-point scale)				0.8*	0.67	0.96
Good/excellent overall health				0.78*	0.61	1
Full school enrollment of biologically related children				0.5*	0.26	0.97
Wald χ^2 (p)	77.2 (<.001)			82 (<.001)

Note: Bootstrapped confidence intervals (1000 replicates) shown. *p < .05; **p < .01; ***p < .001. Social support taken from Zimet et al. (1988; α = 0.94). Model 1 shows HIV-affected household and control variables. Model 2 includes additional mediator variables.

Table 3. Decomposition of direct, indirect and total effects of HIV-affected household and street-migration of children.

Path notation	HIV-affected -> Child lives on the streets				% Total effect
C				0.23
c′	Direct Effect			0.19	85.59
a1 * b1	Social support	−0.12	−0.1	0.012	5.41
a2 * b2	Good/excellent overall health	−0.2	−0.06	0.012	5.41
a3 * b3	Full school enrollment of biologically related children	−0.12	−0.07	0.008	3.78
Total indirect effect				0.032	14.59
Total effect				0.222

Notes: Decomposed effects use standardized, survey-adjusted coefficients from structural equation modeling. Bootstrapped confidence intervals (1000 reps) were calculated for each pathway to establish statistical significance. Each pathway is significant at p < .05. Conceptual model found in Figure 2.

Zimet, G. D., Dahlem, N. W., Zimet, S. G., & Farley, G. K. (1988). The multidimensional scale of perceived social support. Journal of Personality Assessment, 52(1), 30–41. doi: 10.1207/s15327752jpa5201_2

 Web of Science ®Google Scholar