12.07.2015 Views

Demographic and Health Surveys Methodology - Measure DHS

Demographic and Health Surveys Methodology - Measure DHS

Demographic and Health Surveys Methodology - Measure DHS

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Respondents with missing reports for liquids <strong>and</strong> foods or who do not know whether the child receivedthe liquids or foods are treated as the liquid or food not given.Notes <strong>and</strong> ConsiderationsAssumptionsMedians <strong>and</strong> means are based on current status of mothers of the births. The distributions of theproportions of births, by month of birth of the child, are analogous to the l x column of the synthetic lifetable. The basic assumption is that there has been no change in the proportions breastfeeding, exclusivelybreastfeeding, <strong>and</strong> predominantly breastfeeding over time so that births that proportions that arecalculated from births, with small times since birth, are the same as those of births that occurred earlier,when those earlier births, had those durations. For the short 36-month period, this assumption is verylikely to be approximately correct. The proportions are assumed to be the proportion ever breastfed at thetime of birth <strong>and</strong> to decrease monotonically with time since birth. The time at which the proportionsdecline to 0.5 is taken as the median. The mean is taken similarly from the l x column of the life table,which in this case is the series of proportions by time since birth. Because the sum of proportions equalsone, there is no need to divide the sum of the proportions times the width of the interval.Because of the relatively small number of births in each month before the interview, the months are firstgrouped <strong>and</strong> then smoothed to reduce r<strong>and</strong>om fluctuations due to sampling variance.MeanTruncated Mean: Because of the limitation to births that occurred within the three years preceding thesurvey, the mean is truncated if there are mothers who are breastfeeding longer than three years after theirlast birth. It is very unlikely that the proportions exclusively or predominantly breastfeeding are more thannegligible after 35 months since birth. Therefore, it is felt that the truncated mean is very close to the fullmean for both of these statistics. On the other h<strong>and</strong>, any breastfeeding can continue for considerablylonger than 35 months so that the mean may be biased somewhat downwards. To get an idea of thepossible extent of this bias, the proportions of children who are breastfeeding, by age of child in months,should be examined.The measures are based on all births that occurred within the three years preceding the interview,including last <strong>and</strong> other births, surviving or not.Decisions on AlternativesCurrent status medians <strong>and</strong> means are used instead of measures based on the recall of durations becauseof the severe heaping (digit preference) on multiples of 3 <strong>and</strong> 6 months in the recall data.In the current calculation of median <strong>and</strong> mean durations, a woman can contribute more than once if shehad more than one birth in the three years preceding the survey. An alternative calculation would be tobase the medians <strong>and</strong> means on women rather than births. In this alternative, each woman is representedonly once, which is equivalent to durations based on only the last birth. Estimates of durations based onlast births (open interval durations) are thought to overestimate average durations of breastfeeding, sincethey almost always exceed those based on recall after other births (closed interval durations).Guide to <strong>DHS</strong> Statistics 126 Updated September 2006

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!