Estimates of duration of the postpartum variables (breastfeeding, amenorrhea, insusceptibility) can be based on recall of durations or current status. While an appropriate methodology for calculating means from recall of durations is using a life table, the well known issues of substantial heaping on durations which are multiples of 3 or 6 indicate that the results will be biased and inaccurate. The DHS reports therefore use current status rather than recall. They also use medians, which are less subject to the effects of long tails than are means. The prevalence-incidence method of calculating a mean is also a current status measure. However, it is based on a stationary population, meaning that there it assumes a constant numeric flow of births. This assumption is violated in many cases, especially for the ends of the fertile age distribution since there were far fewer births to women now 15-19 three years ago (when they were 12-16) than one year ago (when they were 14-18) and so the means will be wrong. The P/I mean is not so bad for the whole age group 15-49 but is somewhat affected by the few women breastfeeding (amenorrheic) for a long time since it is a mean.

I would use an analysis which looks at whether the woman was currently breastfeeding or amenorrheic or insusceptible at the time of the survey and whether she had begun using contraception since the last birth.

]]>

]]>

Here's some code I wrote to match the table 6.8 in the Ethiopia 2005 data, which is the table on postpartum amenorrhea, insusceptibility and postpartum sexual abstinence. Please note that this table (and code) uses current status data, rather than recall of durations, which Shea mentions in his response.

** Use the KR data file, which contains all births in the 5 years prior to the survey

use "C:\DATA\ETKR51FL.DTA", clear

** Calculate age of the child in months

g agem=v008-b3

** generate sampling weight

g wgt = v005/1000000

**The denominator for the table = women with children <36m old.

**select only 2nd birth for multiple births -- assigned midx=1

ta b0 if agem<36 & (b0==0 | b0==2) [iw=wgt]

**note that this matches the total denominator.

**amenorrheic

g amen=0 if agem<36 & (b0==0 | b0==2)

*if currently pregnant, make not amenorrheic. Select for numerator only information from most recent birth (m6).

recode amen 0=1 if m6==96 & v213!=1 & midx==1

ta amen [iw=wgt]

**abstaining - same denominator and selections as for amenorrheic

g abst=0 if agem<36 & (b0==0 | b0==2)

recode abst 0=1 if m8==96 & v213!=1 & midx==1

ta abst [iw=wgt]

**insuceptible = amenorrheic or abstaining

g insuc = 0 if agem<36 & (b0==0 | b0==2)

recode insuc 0=1 if abst==1 | amen==1

ta insuc [iw=wgt]

**All match ET table 6.8

I hope this helps with your first question.

Cheers,

Sarah]]>

Freddy --The attached programs will calculate the standard mortality rates, and confidence intervals for them, in Stata. The rates agree exactly with the DHS CSPro programs. The confidence intervals are close but do not match exactly.

The programs are just text files but you can insert them into the Stata do-file editor. You will have to change the paths and the file name. This is set up for a Nigeria (NG) survey.

Let me know if you have problems. Cheers--Tom

]]>

For the mean and the median of the three variables, I used the calculations explained in the Guide to DHS statistics pages 74-75-76.

But I did not get the same numbers like the report.

Can someone explain me how to procced after the code given by Sarah B.

Thank.]]>

When you say "For the mean and the median of the three variables, I used the calculations explained in the Guide to DHS statistics pages 74-75-76.

But I did not get the same numbers like the report," could you please tell us exactly what you are doing, perhaps by pasting your code here, and exactly which #s you are trying to match? I'll try to take a look at your code and see why your results are different than the final report.

Cheers,

Sarah]]>

Kindly asking for the stata code for deriving the mean, median as shown in Table 5.6 for the Uganda 2016 udhs. I have used the code above and the figures match though less by a few decimal points. However, the code stops on deriving the totals and not the mean and median.

Any assistance is highly rendered. ]]>