The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Pooling men, women and household DHS from Haiti
Pooling men, women and household DHS from Haiti [message #11101] Wed, 26 October 2016 18:35 Go to next message
acolombo is currently offline  acolombo
Messages: 5
Registered: September 2016
Member
Dear all,

I would like to pool three different waves of DHS carried in Haiti in 2000, 2005/06 and 2012. The main goal is to investigate the determinants of employment in the island using a probit model. In particular, I would like to understand how the likelihood of being employed is affected by: gender; level of education attained; place of residence (metro-area, other urban areas and rural areas); number of children; size of the household. Scott and Rodella (2016) already carried a similar analysis by pooling surveys coming from a difference source.

I dived into this thread all day and after having read several post (and numerous aspirins), I would like to sum up what I have learnt and to check whether I am on the right track about procedure to follow on Stata14.

  1. "Denormalizing" weights: for each men (HTMR61DT) and women (HTIR61DT) survey and wave, I would have to:
    • multiply weights (mv005) and (v005) by the ratio of total men/female 15-49 population divided by the size of the male/female sample. In my case, the best I can get is estimation of male/female population aging 15-64 for every year. Is it enough?
    • create a cluster including the survey id
    • create a strata that also includes the survey (identified by v000) in the group command

      Just for the sake of clarity, using as an example the 2012 women survey, the two previous steps would be implemented by coding the following:
      replace v005=v005/1000000
      gen v005_new=v005*wpop1564_2012/wsize1564_2012
      egen v001r = group(v000 v001) // cluster also includes the survey in the group command
      egen strata=group(v000 v025 sregnew) // strata also includes the survey (identified by v000) in the group command
      svyset v001r [pw=v005_new], strata(strata) singleunit(centered)
      
    • In order to have a full overview over the labor market, I would like to append the men survey and the women surveys. I would end up with two weight variables: v005_new and mv005_new. Should I just "merge" them or should I further distinguish the weights coming from the men surveys from the weights from the women surveys? How would I do that?
    • I would like then to include in this new dataset the respective household characteristics for every eligible men and women. Can I merge them without risk? Or should I take into account the weight assigned to the eligible individuals in the household dataset (HTPR61DT)?
    • I repeat the above procedure for all the waves. At the end I should I have three dataset with eligible men, women, and their respective household characteristics for three years
  2. Append the three dataset from each wave, but:
    1. should I have only one variable with the weights I derived in the previous steps? Or should I further manipulate them so that weights from different surveys can be distinguished from eachother?
    2. I will have only one variable distinguishing each strata (metroarea/urban, metroarea/rural, grandeanse/rural, grandeanse/urban, and so on and so forth) for each wave.

    I would like to remark, though, that I have a different number of strata depending on the wave:
    1. In 2000 there are 19 strata (9 districts*2 along the urban rural dimension, and one strata metroarea-urban)
    2. In 2005/06 there are 21 strata (10 disticts*2, plus the metro area), since one district split in two in 2003
    3. In 2012 there are 23 strata (10 districts without camps*2, metro area without camp, camps/rural, and camps/urban), since camps were built after the 2010 earthquake and the survey aims at investigating the living conditions of those households relocated after the disaster.

    Is this a problem? If yes, how can I handle it?
  3. Finally, there is no unanimous consensus on whether I should re-normalize the weights or not. What would you suggest? How would you proceed?
  4. If everything I wrote above is correct, I should be able to carry my analysis by simply using the svy commands together with the new weight and strata variables.


What do you think? Thank you in advance for your help. I remain at your disposal if I was not clear enough and you need further details. Also, if you bumped into this post, have the same issue of mine, either you found a solution or not (or you think you have found it), please do not hesitate to contact me!

Andrea
Re: Pooling men, women and household DHS from Haiti [message #11507 is a reply to message #11101] Wed, 04 January 2017 14:16 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 802
Registered: January 2013
Senior Member
Your thinking as you laid it out seems valid. A few notes, though:
1a) For the de-normalizing, you can find estimates of population age 15-49 from the UN World Population prospects in the following likns: https://esa.un.org/unpd/wpp/DVD/Files/1_Indicators%20 (Standard)/EXCEL_FILES/1_Population/WPP2015_POP_F08_2_TOTAL_ POPULATION_BY_BROAD_AGE_GROUP_MALE.XLS and https://esa.un.org/unpd/wpp/DVD/Files/1_Indicators%20 (Standard)/EXCEL_FILES/1_Population/WPP2015_POP_F08_3_TOTAL_ POPULATION_BY_BROAD_AGE_GROUP_FEMALE.XLS
1b) Merging men and women - I would generate a variable for sex for the men's and women's files, and then you can combine things like the weights into a single variable, rather than two separate variables (e.g. by renaming the men's variable before merging the two datasets).
1c) Once you have combined the women's and men's data, you can then merge the PR dataset to the combined data using the household ID and the line numbers.
2a) In your appended dataset, I would create a variable for the phase (or, in this case just use V000 as it is unique for each phase - not that for other countries they may not be unique as v000 tells you the recode structure being used and that might be the same for two separate surveys)
2b) Your strata variable should probably include the survey year or phase so that you have unique strata for each year/phase.
3) Unless there is a compelling reason to re-normalize the weights I would not bother re-normalizing. DHS has, by convention, always normalized the weights such that the total weighted N matches the total unweighted N, but there is no statistical reason for doing this.
Re: Pooling men, women and household DHS from Haiti [message #19258 is a reply to message #11101] Sun, 17 May 2020 03:39 Go to previous message
geass is currently offline  geass
Messages: 4
Registered: June 2018
Member
Hello,

Did you get any methods to control for the household size which comes from HR dataset?
I have the same problem. I want to use the two years pooledIR dataset in my research. I need to add a weight to account for the pooled data.
But I need to control for the household size which comes from HR dataset, I have no idea about the weight.
Can you give me some advice?
Previous Topic: Weight Variable
Next Topic: Sample weights and stratification - Nigeria 2008 and 2018
Goto Forum:
  


Current Time: Tue Nov 5 15:27:11 Coordinated Universal Time 2024