The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Nutrition and Anthropometry » incorrect stunting rates
incorrect stunting rates [message #14227] Wed, 07 March 2018 08:45 Go to next message
margovg is currently offline  margovg
Messages: 3
Registered: March 2018
Location: Amsterdam
Member
Hello,

Im trying to calculate the prevalence of stunting using the Burundi 2010 DHS dataset.
I have used the PR file and - presumably- the correct survey settings, the overall percentage that I have calculated is correct (57,7% stunted), however, when I try to calculate the sex specific prevalence I'm of by a bit.



I am using R:

# making a variable for stunting
data2$stunt <- ifelse(data2$hc70 < -200,'stunted','not_stunted')

#survey settings

int.design <- function (){
data.w <<-
svydesign(
id = ~ hv001 ,
data = data ,
weight = ~ hv005 ,
strata = ~ hv022)
}

int.design()

> svyby(~stunt, ~hv104, data.w, svymean)
hv104 stuntnot_stunted stuntstunted se.stuntnot_stunted se.stuntstunted
male male 0.3750139 0.6249861 0.01372043 0.01372043
female female 0.4729965 0.5270035 0.01436399 0.01436399


According to the DHS report 62,1% of boys is stunted and 53,1% of the girls. As you can see from the output above, my percentages come down to 52,7% for girls and 62,5% for boys.

I have deleted from my dataset:
- Children with missing data for variable HC70
- Children with HC70>9000
- Children who did not sleep in the hh the night before the survey(hv103==1)

Can anybody tell me what I doing wrong, and how I can get the correct percentages?

Kind regards,
Margo



[Updated on: Wed, 07 March 2018 08:45]

Report message to a moderator

Re: incorrect stunting rates [message #14233 is a reply to message #14227] Wed, 07 March 2018 15:09 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
I'm matching the numbers from the report with the following code:
library(foreign)
library(survey)
# switch to directory where my data exists
setwd("C:/Data/DHS_stata/")

# read the PR dataset
data <- read.dta("BUPR61FL.dta", convert.factors = TRUE)

# subset for children measured with valid measures who are de facto
data2 <- data[!is.na(data$hc70) & data$hc70<9000 & data$hv103==1,]

# making a variable for stunting
data2$stunt <- ifelse(data2$hc70 < -200,'stunted','not_stunted')

# create weight variable
data2$wt <- data2$hv005/1000000

# survey settings
data.w <- svydesign(id = ~ hv001, data = data2, weight = ~ wt, strata = ~ hv022)

# tabulate stunting by sex
svyby(~stunt, ~hv104, data.w, svymean)

        hv104 stuntnot_stunted stuntstunted se.stuntnot_stunted se.stuntstunted
male     male        0.3790181    0.6209819          0.01352742      0.01352742
female female        0.4687265    0.5312735          0.01374611      0.01374611

Re: incorrect stunting rates [message #14243 is a reply to message #14233] Sun, 11 March 2018 11:38 Go to previous messageGo to next message
Nicholus Tint Zaw is currently offline  Nicholus Tint Zaw
Messages: 13
Registered: March 2018
Location: Myanmar
Member

H, I am Nicholus from Myanmar and I am also experiencing same issued as margovg did before.

I am now working on child nutrition status data analysis and used KR at first and use following stat code to identify the mean HAZ score and stunting rate of children. But my result was quite different in term of denominator with the number from DHS Myanmar report. (4,213 in my analysis and 4,089 in DHS report)

[global
global kr_child D:\SPA SCI\Technical Knowledge\GOVERMENT INFO\DHS 2015-16\dataset\MM_2015-16_DHS_05022017_255_106315\mmkr71dt\ MMKR71FL.dta

global pr_person D:\SPA SCI\Technical Knowledge\GOVERMENT INFO\DHS 2015-16\dataset\MM_2015-16_DHS_05022017_255_106315\mmpr71dt\ MMPR71FL.dta

*----------------------------------------------------------- --------------------

use "${kr_child}", clear


** Calculate Stunting Rate **
** Use DHS HAZ Variable **

tab hw70, m

gen haz_dhs_revised = hw70
replace haz_dhs_revised = .n if hw70 >= 9996
replace haz_dhs_revised = hw70/100 if !mi(haz_dhs_revised)
tab haz_dhs_revised, m

gen stunt = (haz_dhs_revised < -2)
replace stunt = .n if mi(haz_dhs_revised )
tab stunt, m

** Construct WEIGHT var using Women Insidividual Sample Weight **
gen wgt = v005/1000000
tab wgt, m

lookfor sampling // search for samping unit var


** Generate Stunting Rate & HAZ Mean Score by Weighted Data **
svyset, clear
svyset [pw = wgt], psu(v021) strata (v022)
svy: mean haz_dhs_revised stunt
svyset, clear][/code]

Then, when I found this post in user forum, I revised my code and dataset usage with PR file (as follow). But, still got the different figure compare with DHS report and this time the denominator discrepancy became more larger as my analysis figure become 4,640.

[code][/clear

use "${pr_person}", clear

** Check with DHS Calculated HAZ Variable **

tab hc70, m

gen haz_dhs_revised = hc70
replace haz_dhs_revised = .n if hc70 >= 9996
replace haz_dhs_revised = hc70/100 if !mi(haz_dhs_revised)
replace haz_dhs_revised = .n if hv103 != 1
tab haz_dhs_revised, m

tab hv103, m // should be 1

** Construct WEIGHT var using Women Insidividual Sample Weight **
gen wgt = hv005/1000000
tab wgt, m

lookfor sampling // search for samping unit var

gen stunt = (haz_dhs_revised < -2 & hv103 == 1)
replace stunt = .n if hv103 != 1 | mi(haz_dhs_revised)
tab stunt, m

svyset, clear
svyset [pw = wgt], psu(hv021) strata (hv022)
svy: mean haz_dhs_revised stunt
svyset, clear

]

So, I was wondering there is anyone who can review my code and provide guidance to fix this issue. Thanks in advance.

Best regards,
Nicholus


nicholus

[Updated on: Sun, 11 March 2018 11:39]

Report message to a moderator

Re: incorrect stunting rates [message #14268 is a reply to message #14243] Thu, 15 March 2018 11:16 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
In the output you get, the number of observations is 4640, but this is the unweighted number of cases. If you look at the Population Size, this is the weighted number of cases and should match the 4089 in the report.
Re: incorrect stunting rates [message #14269 is a reply to message #14268] Thu, 15 March 2018 11:25 Go to previous messageGo to next message
Nicholus Tint Zaw is currently offline  Nicholus Tint Zaw
Messages: 13
Registered: March 2018
Location: Myanmar
Member

Yes, the weighted population size was identical with DHS publication report.
So, for further analysis, I should check with that weight population size and DHS report instead of # of observation. right?

Thanks a lot for your reply.

best regards,
Nicholus


nicholus
Re: incorrect stunting rates [message #14270 is a reply to message #14269] Thu, 15 March 2018 11:26 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
Yes, the DHS report presents the weighted number of cases which are given as the weighted population size in the Stata output, and not the unweighted number of observations
Re: incorrect stunting rates [message #14272 is a reply to message #14270] Fri, 16 March 2018 04:43 Go to previous messageGo to next message
Nicholus Tint Zaw is currently offline  Nicholus Tint Zaw
Messages: 13
Registered: March 2018
Location: Myanmar
Member

Thanks a lot.

By the way, one last question remained. Why we used PR data file instead of KR for this child stunting indicator.
To my knowledge, KR is the children dataset file. right?

best regards,
Nicholus


nicholus
Re: incorrect stunting rates [message #14275 is a reply to message #14272] Fri, 16 March 2018 09:54 Go to previous messageGo to next message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 787
Registered: January 2013
Senior Member
KR is the children's recode file for children of interviewed women. The PR file includes all people who are usually resident in the household (de jure) or stayed in the household the prior night (de facto). By analyzing the children based on the PR file we include children who do not live with their mother or whose mother has died, which would be excluded if you used the KR file.
Re: incorrect stunting rates [message #14276 is a reply to message #14275] Fri, 16 March 2018 10:49 Go to previous messageGo to next message
Nicholus Tint Zaw is currently offline  Nicholus Tint Zaw
Messages: 13
Registered: March 2018
Location: Myanmar
Member

Thanks for your kind explanation. Now, it is very clear to me.

nicholus
Re: incorrect stunting rates [message #17709 is a reply to message #14275] Wed, 08 May 2019 11:31 Go to previous message
anneclaireclaire is currently offline  anneclaireclaire
Messages: 11
Registered: May 2014
Location: Italy
Member
Dear Trevor, I am trying to match the stunting numbers for the RW ( Rwanda) DHS 2014-2015 but it does not work . Can you please help me . I am using the following code
Thanks in advance
library(haven)
library(data.table)
RWPR70FL <- as.data.table(read_dta("~/Data/RW/2015_DHS/RWPR70FL.DTA", convert.factors = TRUE)))

RWPR70_2 <- RWPR70FL[!is.na(hc70) & hc70<9000 & hv103==1,]


RWPR70_2$stunt2 <- cut(RWPR70_2$hc70,
breaks=c(-Inf, -200, Inf),
labels=c("stunted","not sunted"))

RWPR70_2[,wt:= hv005/1000000]
RWPR70_2[, district:=as.factor(shdistrict)]

RWPR70.w2 <- svydesign(id = ~ hv001, data = RWPR70_2, strata = ~ hv022, weights=~wt)
svyby(~factor(stunt2), RWPR70_2$district, RWPR70.w, svymean, verbose=TRUE)
Previous Topic: DHS of India 2015/16 and 2005/06
Next Topic: Consent for additional tests
Goto Forum:
  


Current Time: Thu Mar 28 16:18:19 Coordinated Universal Time 2024