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