Page 135 - IJPS-2-2

Priyanka Dixit and Laxmi Kant Dwivedi

Continued
95% C.I. for EXP(B)
Exp(B)
Lower Upper
No. of sons greater than daughters 1.11 0.84 1.47
No. of sons less than daughters 1.09 0.83 1.44
Equal sons and daughters 1.06 0.80 1.40
Ever had terminated pregnancy

No a
Yes 1.34* 1.24 1.44
Child age (in months)
a
0 - 23
24 - 35 0.89* 0.83 0.96
36 - 47 0.73* 0.68 0.78
48 - 59 0.67* 0.63 0.72
Birth order
a
nd
2 order birth
st
1 order birth 1.92* 1.79 2.05
3 order birth 0.74* 0.68 0.80
rd
3+ order birth 0.67* 0.62 0.74
Constant 0.06* 0.04 0.08
Random Effects Coefficient (Standard Error)
Level 3 (PSU Level) 1.119 (0.045)
Level 2 (Mother Level) 0.557 (0.043)
a
Note: Reference category.*p<0.05

Table S2. Random-parameter estimates for the three-level (child, mother, and PSU) univariate multilevel regression model for the utilization of
institutional delivery, NFHS-III, 2005–2006.
Parameter Parameter estimates in unconstrained model (S.E.) Parameter estimates in constrained model (S.E.)
2
PSU level σ 3.43 (0.13) 1.12 (0.05)
ϑ
2
Mother level σ 9.63 (0.14) 0.56 (0.04)
u
2
Pregnancy level σ 0.16 (0.002) 1.00
e

when we converted the constrained variance to unconstrained, the random effects at PSU and mother
levels significantly increased. Therefore, a statistical model is needed that does not assume condi-
tional independence between the pregnancies to the same mother. Thus a multivariate response mod-
el was used to address the problem of dependency. We developed a three-level multivariate regres-
sion model which includes the numbers of pregnancies during the last five years at Level 1, used as
multiple outcomes nested within mothers at Level 2, who in turn are nested within their local com-
munities (PSU) at Level 3. By treating multiple pregnancies (outcome) within the multivariate re-
sponse model, we were able to estimate the covariance between three pregnancies nested within
mothers, as well as the variance for each outcome in a simultaneous manner.

3.3 Results from Multivariate Multilevel Models
Table 2 presents the fixed estimates of the full model. There were similarities and differences in the
socioeconomic patterning of the place of deliveries for all three births. For the most recent and
second to last pregnancy, it was found that mothers in five regions were more likely to deliver

International Journal of Population Studies | 2016, Volume 2, Issue 2 129