COMPARISON OF LINEAR AND QUANTILE REGRESSION METHODS FOR DETERMINATION OF FACTORS ASSOCIATED WITH NUTRITIONAL STATUS OF WOMEN OF REPRODUCTIVE AGE IN NIGERIA

Abstract

Mean regression approach explores the average effect effectively of the change in mean BMI, but may not be able to identify how extreme values affect Body Mass Index (BMI). Therefore, mean regression based methods may not be able to answer how the factors may affect large extreme BMI values. The use of quantile regression allowed the impact of the explanatory variable to vary along the whole range of BMI intake. Reproduction is associated with nutritional status because the various roles played by women give rise to serious health problems. While warnings about health penalty of excess weight, less attention seems to be given to the consequences of being underweight. The study aims to evaluate the performance of parametric Ordinary Least Square (OLS) and non parametric regression (quantile function) method to determine factors affecting nutritional status of women of reproductive age in Nigeria. A nationally representative sample of women within reproductive age (15-49 years) within households in communities was obtained from the Nigeria Demographic and Health Survey (NDHS 2013). The Body Mass Index of women defined as weight/height2 was the outcome variable while the explanatory variables include age, family size, total children ever born, marital status, highest level of education, place of residence region, wealth index. OLS regression was used to determine the average effect of factors on DMI while Quantile Regression was used to determine how a particular quantile of the BMI distribution was associated with covariates Analysis was done on STATA version 12 while graphs were done using R version 3.2.0. A total of 31, 828 women were included in the study. The mean age of women was 29 years (SD=7.0), 49.2% had no formal education and 23.5% belonged to the poorest wealth quintile. It was shown that only 16% of variation has been predicted by linear regression. Results of ordinary least square regression analysis show that women's age, number of children, place of residence, level of education, some regions (North East, North West and South West), wealth index (p<0.001) were found to have effects on women's Body Mass Index. While in the 10th quantile, the effect of children ever born, place of residence were not significant. Family size did not contribute significantly to the BMI effect produced. Quantile regression was able to detect the amount of under estimation and over estimation produced by the Ordinary Least Square regression of BMI values. The magnitude of the changes differed depending on the location of the woman in the BMI distribution