The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Unmet Need » Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys (Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys)
Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16217] Mon, 26 November 2018 17:28 Go to next message
cgreenba is currently offline  cgreenba
Messages: 18
Registered: October 2017
Member
Dear all,

I am working on a report looking at demand satisfied among youth across different DHS surveys and am having a difficult time matching the new definition of unmet need and demand satisfied numbers that I am getting in Stata with the estimates in StatCompiler for a few of the surveys that I have included in my analysis: the 2011-12 Honduras DHS, the 2006 Nepal DHS, the 2006 Niger DHS, the 2010 Rwanda DHS, and the 2011 Uganda DHS. For the surveys that did not have the new unmet need variable (v626a), I used either the general or survey specific recoding of unmet need do file produced by Sarah Bradley - attached below - but am still getting different estimates.

Here are the estimates that I get in Stata:
Honduras 2011-12 - Unmet need (all women) 15-19: 4.8%
Honduras 2011-12 - Unmet need (all women) 20-24: 7.6%
Honduras 2011-12 - Unmet need (all women) 15-24: 6.1%
Honduras 2011-12 - Demand satisfied by modern methods (all women) 15-19: 66.8%
Honduras 2011-12 - Demand satisfied by modern methods (all women) 20-24: 75.0%
Honduras 2011-12 - Demand satisfied by modern methods (all women) 15-24: 72.2%
Nepal 2006 - Unmet need (all women) 15-19: 37.8%
Nepal 2006 - Unmet need (all women) 20-24: 33.2%
Nepal 2006 - Unmet need (all women) 15-24: 34.7%
Nepal 2006 - Demand satisfied by modern methods (all women) 15-19: 25.7%
Nepal 2006 - Demand satisfied by modern methods (all women) 20-24: 43.4%
Nepal 2006 - Demand satisfied by modern methods (all women) 15-24: 38.4%
Niger 2006 - Unmet need (all women) 15-19: 6.9%
Niger 2006 - Unmet need (all women) 20-24: 15.4%
Niger 2006 - Unmet need (all women) 15-24: 11.1%
Niger 2006 - Demand satisfied by modern methods (all women) 15-19: 10.6%
Niger 2006 - Demand satisfied by modern methods (all women) 20-24: 16.7%
Niger 2006 - Demand satisfied by modern methods (all women) 15-24: 15.0%
Rwanda 2010 - Unmet need (all women) 15-19: 1.0%
Rwanda 2010 - Unmet need (all women) 20-24: 7.3%
Rwanda 2010 - Unmet need (all women) 15-24: 4.0%
Rwanda 2010 - Demand satisfied by modern methods (all women) 15-19: 62.3%
Rwanda 2010 - Demand satisfied by modern methods (all women) 20-24: 69.1%
Rwanda 2010 - Demand satisfied by modern methods (all women) 15-24: 68.3%
Uganda 2011 - Unmet need (all women) 15-19: 7.8%
Uganda 2011 - Unmet need (all women) 20-24: 25.3%
Uganda 2011 - Unmet need (all women) 15-24: 15.5%
Uganda 2011 - Demand satisfied by modern methods (all women) 15-19: 41.5%
Uganda 2011 - Demand satisfied by modern methods (all women) 20-24: 41.8%
Uganda 2011 - Demand satisfied by modern methods (all women) 15-24: 41.7%

Here are the estimates that StatCompiler lists:
Honduras 2011-12 - Unmet need (all women) 15-19: 6.5%
Honduras 2011-12 - Unmet need (all women) 20-24: 9.7%
Honduras 2011-12 - Unmet need (all women) 15-24: 7.9%
Honduras 2011-12 - Demand satisfied by modern methods (all women) 15-19: 62.0%
Honduras 2011-12 - Demand satisfied by modern methods (all women) 20-24: 72.1%
Honduras 2011-12 - Demand satisfied by modern methods (all women) 15-24: 68.6%
Nepal 2006 - Unmet need (all women) 15-19: 12.2%
Nepal 2006 - Unmet need (all women) 20-24: 26.7%
Nepal 2006 - Unmet need (all women) 15-24: 18.7%
Nepal 2006 - Demand satisfied by modern methods (all women) 15-19: 25.9%
Nepal 2006 - Demand satisfied by modern methods (all women) 20-24: 43.6%
Nepal 2006 - Demand satisfied by modern methods (all women) 15-24: 38.5%
Niger 2006 - Unmet need (all women) 15-19: 6.9%
Niger 2006 - Unmet need (all women) 20-24: 15.2%
Niger 2006 - Unmet need (all women) 15-24: 11.0%
Niger 2006 - Demand satisfied by modern methods (all women) 15-19: 22.4%
Niger 2006 - Demand satisfied by modern methods (all women) 20-24: 35.0%
Niger 2006 - Demand satisfied by modern methods (all women) 15-24: 31.6%
Rwanda 2010 - Unmet need (all women) 15-19: 2.2%
Rwanda 2010 - Unmet need (all women) 20-24: 10.3%
Rwanda 2010 - Unmet need (all women) 15-24: 6.1%
Rwanda 2010 - Demand satisfied by modern methods (all women) 15-19: 45.3%
Rwanda 2010 - Demand satisfied by modern methods (all women) 20-24: 62.1%
Rwanda 2010 - Demand satisfied by modern methods (all women) 15-24: 59.8%
Uganda 2011 - Unmet need (all women) 15-19: 10.5%
Uganda 2011 - Unmet need (all women) 20-24: 27.5%
Uganda 2011 - Unmet need (all women) 15-24: 18.1%
Uganda 2011 - Demand satisfied by modern methods (all women) 15-19: 34.9%
Uganda 2011 - Demand satisfied by modern methods (all women) 20-24: 39.9%
Uganda 2011 - Demand satisfied by modern methods (all women) 15-24: 38.4%

Is anyone aware of why I might be seeing these discrepancies? Is StatCompiler using a different sample, even in the indicator marked all women? Do I need to used a different weighting scheme other than the sample weights provided? I tried to match using the earlier definition of unmet need at as well, but that did not work. Any help or information would be greatly appreciated!

Here is my code in case it helps:
gen indiv_weight = v005/1000000

gen modern_contraception=0
replace modern_contraception=1 if v313==3
replace modern_contraception=. if v313==.

gen any_contraception=0
replace any_contraception=1 if v313==3 | v313==1 | v313==2
replace any_contraception=. if v313==.

gen unmet_need=0
replace unmet_need=1 if v626a==1 | v626a==2
replace unmet_need=. if v626a==.
* OR: gen umet_need=unmettot


gen demand_satisfied=.
replace demand_satisfied=0 if unmet_need==1
replace demand_satisfied=0 if v313==0 | v313==1 | v313==2
replace demand_satisfied=1 if v313==3
replace demand_satisfied=. if unmet_need==0 & any_contraception==0 | unmet_need==. | modern_contraception==.

proportion unmet_need if v013==1 | v013==2 [pweight = indiv_weight]
proportion unmet_need if v013==1 [pweight = indiv_weight]
proportion unmet_need if v013==2 [pweight = indiv_weight]
proportion demand_satisfied if v013==1 | v013==2 [pweight = indiv_weight]
proportion demand_satisfied if v013==1 [pweight = indiv_weight]
proportion demand_satisfied if v013==2 [pweight = indiv_weight]

Thank you so much!

Best regards,
Charlotte
Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16323 is a reply to message #16217] Thu, 13 December 2018 15:19 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
Around the time that the revised unmet need definition was being produced, several recode datasets were produced containing early versions of the revised definition, and classify some non-currently married women differently from the final definition. Principally the difference is that in v626a certain non-currently married women are classified as "not married and not sexually active" before checking whether they are pregnant or postpartum amenorrheic. In the final revised definition, women who are pregnant or postpartum amenorrheic can have an unmet need even if they are currently "not married and not sexually active". For these surveys we recommend using the survey-specific code found at the bottom of the unmet need page on our website. This applies to Honduras 2011-12, Rwanda 2010-11, Uganda 2011, and to other surveys from roughly the same period.

For Nepal 2006 the numbers you quote from the Stata code are for currently married women, not all women. We should not be presenting data for all women in the STATcompiler as the questions needed for unmet need were only asked for currently married women - we will remove this from STATcompiler. The estimate currently in STATcompiler for all women would be an estimate if you assumed that all non-currently married women have no need.

For Niger 2006, when I run the survey-specific Stata code I match the STATcompiler estimates and not the Stata estimates that you provide.


Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16367 is a reply to message #16323] Fri, 28 December 2018 10:02 Go to previous messageGo to next message
cgreenba is currently offline  cgreenba
Messages: 18
Registered: October 2017
Member
Hi Trevor,

I apologize for the delay, but thank you so much for this reply. I really appreciate it.The information that you provided is very helpful. I have just a few follow-up questions if you are able to respond.

1. Is there a list of surveys for which in the v626a variable, certain non-currently married women are classified as "not married and not sexually active" before checking whether they are pregnant or postpartum amenorrheic (as is the case with Honduras 2011-12, Rwanda 2010-11, Uganda 2011)? If I understand correctly, for these surveys, to make the v626a unmet need definition comparable to the v626a variable in other more recent surveys, we need to use the survey-specific code. Is this correct? Is there any documentation about this?

2. You state that you were able to match the unmet need numbers for Niger 2006 when you used the survey-specific code, but this survey is not listed in the .do file as one of the countries that require the survey-specific code. Is there an updated list of surveys for which the survey-specific code should be used as opposed to the general recode?

3. In both the general and survey-specific code for the revised unmet need definition, I noticed at the very end when the binary 'unmettot" variable is generated for the "unmet" variable, the code categorized both 98 "unmarried - EM sample or no data" and 99 "missing" as NOT having unmet need instead of as missing. This appears to match the number that are in StatCompiler, but I'm not sure why those two categories would be counted as not having unmet need rather than missing. To me, the "missing" category indicated that we do not know whether the respondent does or does not have unmet need and the "unmarried - EM sample or no data" category indicated that these women shouldn't be in the sample or weren't asked the unmet need questions. Is there a reason behind this? When calculating overall unmet need, should those categories be included in the denominator when there is not information on whether or not they have unmet need?

Any help on this would be greatly appreciated. Thank you so much!

Best,
Charlotte
Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16583 is a reply to message #16367] Fri, 01 February 2019 10:25 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
Hi Charlotte

Sorry for the delay in replying. In response to your question:
1) I was trying to track down a list of datasets for which you need to the code, but I cannot find a list. To be on the safe side I would say any survey before 2013. I know this affects many surveys around 2010-2012, but I don't have a list of exactly which ones.
2) I always run the survey specific code. It will run for all surveys, even those that don't have survey specific modifications. For Niger 2006 using the survey specific code I match the results in STATcompiler. If you run the general code you should also match STATcompiler. But I was not matching your results.
3) DHS has had a long standing tradition of using a "conservative" approach for indicators whereby we include missing and don't know responses in the denominators of indicators, rather than excluding them. We may be underestimating the indicator, but if you treat them as missing, you may be over estimating the indicator. This is an analysts choice, but for DHS we have used this basic conservative approach for many years.

Trevor
Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16682 is a reply to message #16583] Fri, 15 February 2019 16:06 Go to previous messageGo to next message
cgreenba is currently offline  cgreenba
Messages: 18
Registered: October 2017
Member
Hi Trevor,

Sorry for the additional request, but I have one more question about the unmet need and demand satisfied numbers for these surveys. I have gotten all of the numbers what we have in StatCompiler with one exception. For the Niger 2006 DHS, I was able to match the mCPR and unmet need numbers with StatCompiler, but for some reason, I am still getting the same original numbers for demand satisfied that I was getting earlier. I have tried this multiple times and am unable to match the StatCompiler numbers, even though the other numbers match and I haven't had this issue for any other survey.

You were able to double check these numbers, correct? Is there any reason you can think of that I might be finding this discrepancy? Do you have any ideas for how to address this? Any thoughts would be greatly appreciated! Thank you!

Best,
Charlotte
Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16715 is a reply to message #16682] Fri, 22 February 2019 15:15 Go to previous messageGo to next message
cgreenba is currently offline  cgreenba
Messages: 18
Registered: October 2017
Member
Hi Trevor and all,

I just wanted to give you an update on this email and let you know that I found the discrepancy and why I was unable to match the demand satisfied by modern methods estimates in StatCompiler for Niger 2006. In the v313 and v364 variables, women using LAM were classified as using traditional and NOT modern methods. This is the only survey (DHS IV or later) I looked at for which this was the case. However, in the demand satisfied by modern methods estimate, in order to match the estimate in StatCompiler, LAM needs to be counted in the numerator as a woman having her demand satisfied by a modern method.

I just thought that I would point this out. I'm glad to have resolved the difference!

Thank you!

Best,
Charlotte
Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #16716 is a reply to message #16715] Fri, 22 February 2019 15:20 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
Hi Charlotte

I was trying to figure out the difference. I should have thought of that. There are a few surveys in which LAM was categorized as traditional rather than modern at the time of the survey, but they are all older ones. I'm glad you found it.

Thanks. Trevor
Re: Getting Unmet Need & Demand Satisfied Estimates to Match StatCompiler for 5 Different Surveys [message #17851 is a reply to message #16716] Wed, 26 June 2019 12:36 Go to previous message
cgreenba is currently offline  cgreenba
Messages: 18
Registered: October 2017
Member
Hi Trevor,

I know that this is coming back to you months later, but I want to follow up on the work that I was doing re. family planning indicators for a whole group of countries. After following your instructions carefully form earlier in this thread to use the unmet recode .do file for surveys before 2013 to get them all to match the new definition of unmet need, I had gotten all of our surveys to match the StatCompiler estimates.

Today, I was getting ready to sign off on the revisions to my organization's publication, part of which was done to make sure that FP estimates matched what was in StatCompiler, I happened to open up StatCompiler to look at a few countries, and I found that the numbers I have now don't always match StatCompiler. In fact, it looks like the estimates that are now in StatCompiler match the earlier estimates that I had for the 2010-2012 surveys in which I used v626a without running the recode .do file.

Can you please explain why the numbers in StatCompiler changed and whether the unmet need and demand satisfied estimates that I had produced using the recode file (and that matched the numbers in StatCompiler earlier this year) are correct?

Thank you so much.

Best,
Charlotte
Previous Topic: Denominator for unmet need calculation among unmarried women
Next Topic: Selecting variable and categorization of unmet need
Goto Forum:
  


Current Time: Thu Mar 28 14:38:40 Coordinated Universal Time 2024