Home » Topics » Nutrition and Anthropometry » DHS Burundi 1987
 DHS Burundi 1987 [message #3202] |
Thu, 06 November 2014 09:15  |
tnkunzimana
Messages: 14
Registered: November 2014
Location: Italy
|
Member |
|
|
Dear all,
I want to use DHS of 1987 for anthropometric measurements.
Who can help for the stata code for results by province or by region?
As I have a dataset on Height and Weight Scores - WHO Child Growth Standards (BUHW01FL.dta)with 'hwcaseid' var., how can merge this dataset with the one of the household raw?
Many thanks for your help.
Tharcisse
Tharcisse Nkunzimana
Agricultural Economist
Scientific Officer/Food and Nutrition Security
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: DHS Burundi 1987 [message #20176 is a reply to message #20175] |
Mon, 05 October 2020 18:17  |
Trevor-DHS
Messages: 805
Registered: January 2013
|
Senior Member |
|
|
I think you are misunderstanding the data. In all of the surveys that you are using the anthropometry data were collected for children listed in the household schedule (the PR file). For the earlier surveys (2000 and 2005), for which you need to use the HW files, the WHO growth standards did not exist at the time of those surveys, and so these HW files were constructed later. These HW data files are designed to be used with the PR file, and include the hc7* series of variables. These variables already exist if the PR file for the later surveys (2011 and 2016). If you wish to analyze anthropometry we recommend using the PR file.
The KR files include the hw7* series of variables which are a copy of the hc7* series of variables from the PR file, but restricted just to children of interviewed women. Thus analyzing anthropometry using the KR file will give you somewhat biased estimates as it is only based on children of women interviewed, and thus likely excludes orphaned and fostered children. Thus we recommend using the PR file.
Why do you need to restrict your analysis just to children of interviewed women?
If you do need to restrict your analysis and need to use the KR file, then, yes, you can rename hc7* to hw7* when you merge the data.
|
|
|
|
|
|
| Re: DHS Burundi 1987 [message #20182 is a reply to message #20180] |
Tue, 06 October 2020 13:10 |
Trevor-DHS
Messages: 805
Registered: January 2013
|
Senior Member |
|
|
Thanks - that makes sense. I just wanted to make sure that you were using the KR file for a good reason.
Weighting is applied when you analyze the data after you have pooled the data. Note that you cannot just combine the datasets and use them as they are with the weights given for each survey as the probability of selection of a woman in each survey varies. You need to adjust the weighting to take into account that the probability of selection differed. The sample sizes for the surveys were:
Year : Sample : Total women 15-49* : Probability of selection
2000 : 15367 : 14,619,000 : 1 in 951.324
2005 : 14070 : 16,861,000 : 1 in 1198.365
2011 : 16515 : 20,659,000 : 1 in 1250.923
2016 : 15683 : 24,949,000 : 1 in 1590.831
See other posts on the forum about pooling.
Using gen wgt = v005/1000000 is not applying the weights, just calculating a weight variable to be used later (note that it uses v005, not v001). The weights are applied later when analyzing the data, and should be adjusted to take into account the probabilities of selection as above.
|
|
|
|
| Re: DHS Burundi 1987 [message #20187 is a reply to message #20182] |
Tue, 06 October 2020 21:39 |
id709nvz
Messages: 43
Registered: March 2019
|
Member |
|
|
Dear,
thanks for your insight and helpful comments. After going through the forum Q&A I follow the following weighting. Could you please evaluate it?
*After opening KR file of 2000 survey
gen wgt=v005/1000000
gen weight=(wgt*FEMALE POPULATION)/FEMALE SAMPLE -For KR 2000; FEMALE POPULATION=14619 and FEMALE POPULATION=15367
gen survey=1
*Same procedure for 2005, 2011 and 2016. Except that I used different total number of female population, female sample and with gen survey 2, 3, and 4 respectively. Then I appended the data starting from the latest survey (2016). Afterward, I did the following.
egen cluster=group(survey v021) *V021 is PSU and survey identifies year of survey (coded 1, 2, 3 and 4)
egen stratum=group(survey v022) *v022 is sample strata for sampling error
svyset cluster [pw=wgt], strata(stratum) singleunit(centered)
Is it correct to use: tab x1 x2 [iweight=wgt] or tab x1 x2 [iweight=weight]
how about svy: reg Y x1 x2 (Will this understand the complex nature of the survey).
Added to this: svy is not supported with ivreg2 stata command is there an alternative to this? I simply used ivreg2 with cluster option at v001.
Thank you in advance for any comments.
Kind regards!
|
|
|
|
| Re: DHS Burundi 1987 [message #20193 is a reply to message #20187] |
Wed, 07 October 2020 00:31 |
Trevor-DHS
Messages: 805
Registered: January 2013
|
Senior Member |
|
|
Yes, this mostly looks good. A couple of notes:
1) For 2000, FEMALE POPULATION=14619 and FEMALE SAMPLE=15367. I assume this is just a typo in the message.
2) For the svyset it should use [pw=weight]. wgt is for use with a single survey, weight would be used for the pooled dataset.
3) For tabulating the combined pooled file, you should use tab x1 x2 [iweight=weight], not wgt
4) The svy: reg will use whatever you have defined in the svyset, so that should be weight for the pooled dataset, or wgt if you are using only a single survey.
5) I don't know ivreg2, so I can't advise on this, but typically not taking the complex sampling into account (with svy) means that significance tests and confidence intervals are not accurate.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Goto Forum:
Current Time: Tue Mar 4 18:26:58 Coordinated Universal Time 2025
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
