Response to: “A comment on J. Akin et al., ‘The determinants of breastfeeding in Sri Lanka’”

Abstract

Commenting on the authors paper Determinants of Breastfeeding in Sri Smith reanalyzed the World Fertility Survey (WFS) data for Sri Lanka using different estimation procedures variables and sample sizes. The results agree with respect to the impacts of parity education and residence on breastfeeding duration. Smith stated that the only variables for which results differed were working away from home contraceptive use and mothers age. He did however think that the authors results on working away from home were better because of use of a more sensitive indicator. The authors examined the use of pills on breastfeeding finding generally significant and negative results at each age cutoff. Smith on the other hand included other modern birth control methods and so it is irrelavant to compare the results in contraceptive use. The authors are in some disagreement with Smiths argument regarding the effect of mothers age on breastfeeding. That both parity and mothers age are positively related to breastfeeding in simple 1-to-1 relationships certainly does not indicate that both are positively related to the outcome variable in a correctly specified complex relationship. In the opinion of the authors the 1 major objective of multivariate analysis is to sort out which simple correlations are spurious and which hold up in a fully specified model. Smiths argument appears to be that multivariate results are to be ignored if independent variables are at all correlated with each other (multicollinear). He argued that multicollinearity causes the authors variable both to change signs and to become statistically significant in the wrong sign. Smith used inappropriate methodology to test the significance of variables in proportional hazards model. Despite the differences in the methodologies variable definitions and sample sizes there is a great degree of similarity.