Assesing predictors of child stunting in the slums of Mombasa, Kenya using logistic regression and discriminant analysis
Abstract
This study aimed at assessing predictors of child stunting using logistic and
discriminant analysis for functional dependencies of height - for - ages z -
scores in children aged 6 - 59 months drawn from the slums of Mombasa,
Kenya. A total of 192 males and 223 females were studied. The predictor
variables considered in this analysis were: age of the child aged below 5
years, house hold size, number of children aged below 5 years in
households, age in weeks child started receiving food, total number of food
groups consumed, and age of the caregiver as predictors. In either of the
analysis, only age of the child made a significant contribution to prediction
based on Wald criterion (p = .036) for logistic regression, and also based on
the largest loadings (.809) for the discriminant function in the structure
matrix for discriminant analysis. For logistic regression, a test of the full
model against a constant only model was statistically established not to be
significant, indicating that the predictors as a set unreliably distinguished
between the stunted and non stunted (chi square = 6.626, p > .05 with df =
6). Similarly, for discriminant analysis, a strong statistical evidence of
significant differences between means of stunted and non stunted groups
exists for the age of the child (p = .048) producing very high F - value of
3.952. While the log determinants were quite similar in the discriminant
analysis, Box's M = 27.735 with F = 1.805 significant at p = 0.028 < .05
signified that the hypothesis of equality of covariance matrices was violated.
Results of logistic regression indicate 99.5% of the children were correctly
classified into non stunted group, as compared to 99% using discriminant
analysis. Overall 68.7% cases were correctly classified by way of logistic
regression which compares to 68.3% of children correctly classified into
'stunted' or 'not stunted' groups by use of discriminant analysis. There is
thus a strong support for the idea that a child's stunting risk may be raised
by the age of the child. Principally results of Logistic and Discriminant
analysis are consistent in this study
Sponsorhip
University of NairobiPublisher
School of mathematics