Page 154 - Contributed Paper Session (CPS) - Volume 6

CPS1852 Leonard KA
and child factors has to take into account this hierarchical structure of the data.
A natural approach is to apply multilevel models (Fitzmaurice, Laird, and Ware;
2004), with the child as the first level and the mother as the second level.
This paper uses multilevel models to examine the relationship between
birth order and birth weight using the 2016 Uganda DHS data. Studies of the
factors influencing birth weight variously use birth weight in kilogrammes (e.g.
Diamond et al., 2001; Côté et al.,2003) or low birth weight (<2.5 Kg) (e.g.
Gathimba et al., 2017; Ngwira 2015)as the dependent variable. In this paper
we use both measures.

2. Methodology
The data used are from the 2016 Uganda Demographic and Health Surveys
which collected information on a nationally representative sample of women
in child-bearing age (15-49) (Uganda Bureau of Statistics (UBOS) and ICF,
2017). The survey collected a large number of indicators for the respondent,
her partner, the household she resides in, and her children who were born
within the five years preceding the survey. This study is based on 10,429
children whose weights at birth are available. These belong to 7562 women.

Two models were fitted:
(i) a multilevel linear regression model for birth weight in kilograms,

(ii) a multilevel binary logistic regression model for the binary outcome
(low birth weight),

In both (1) and (2) yij is the weight of the jth child of the ith woman, xtij is
a row of covariates, β is the vector of coefficients, ui are the mother level
random effects, and εij are the residual errors. Both models were fitted with
stata version 13 (StataCorp, 2013).

3. Result
Table 1 below gives the number of children, birth weight and
percentage with low birth weight by the child and mother characteristics
143 | I S I W S C 2 0 1 9