i have an independent variable as length of breastfeeding in months, with an initial Exclusive period of brestfeeding and afterwards a non-exclusive period of breastfeeding. should i treat them as two different independent variables? as am interested in the effect of the exclusive as well as non exclusive period.
I do not think duration of breastfeeding is in general a suitable measure. It assumes a dose relationship or even a linear relationship, when all the evidence shows that while this may be true for the first few months of life, it is not universally true. For IQ, infections, morbidity, mortality, malnutrition, etc, breastfeeding beyond say 7 or 8 months gives no additional benefit and in some cases is detrimental. The biggest contrast has always been between full breastfeeding versus any alternative.
Thanks Mr Gordon,
since you have a mark-off from 7 or 8 months, will truncated regression do?
If you have sufficient data, one option would be segmented regression.
Another option would be to make a dummy variable (DV). If observation is within exclusive period then DV=0, else DV=1. If the DV is not a significant element in the model, nor does it have significant interaction terms, then delete the variable from the model. Be careful, what you make of this result. It is possible that the break point is not correct, it is possible that the model does not fit the data.
Another option is to run normal regression, and look at the residuals. Color all the residuals from the exclusive period black, and from later red. If all the black dots are in one area and all the red dots in another then there is an effect and you should spend more effort in building a good model.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics