Decentralization and Subnational Management in LAC: Lessons from experience
National and subnational experience with estimating the ... · National and subnational experience...
Transcript of National and subnational experience with estimating the ... · National and subnational experience...
National and subnational experience with estimating the extent and trend in
completeness of registration of deaths in South Africa
Rob Dorrington Centre for Actuarial Research
University of Cape Town
Analytical methods to evaluate the completeness and quality of birth registration
Session 4 of the UN Expert Group Meeting on the methodology and lessons learner to evaluate the completeness and quality of vital statistics data from civil
registration
Background
Population 51.8 million in 2011
VR data are incomplete – slightly more than 500,000 deaths in 2011
Significant HIV/AIDS epidemic + significant treatment from 2004/5
Surveys: DHS 1998 and 2003 (unsuccessful); GHS (limited information, Udjo 1997; 2005); 2007 Community Survey (not very reliable); HSRC 2012 (not publicly available)
3 post-apartheid censuses (1996, 2001 & 2011) not without problems (Moultrie & Dorrington 2004) – only 2001 and 2011 asked households to report on deaths
Most recent reliable estimates of IMR & U5MR from 1998 DHS and 1996 census
2
Rapid mortality surveillance (RMS)
Hiatus in processing VR data while rising AIDS epidemic
Population register (PR) comprising all registered births and adults with IDs
RMS originally set up to ‘prove’ that people were dying of AIDS, and track the impact on mortality
More recently tracking impact of interventions (ART)
Extrapolate trend in PR vs VR estimate future VR from PR
Completeness of VR deaths 15+ from a trend in estimates of completeness in intercensal periods
Completeness of VR deaths <5 was interpolated/extrapolated from estimates of infant and under-five mortality from census and survey data
3
Ratio of VR deaths to estimate of complete deaths (15+) by duration since year of death: 2008-2014
5
RMS NNR, IMR & U5MR vs other methods/data
6
0
10
20
30
40
50
60
70
80
90
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
De
ath
s p
er
1 0
00
live
bir
ths
NMR (VR) IMR (VR) U5MR (VR)
NMR (DHIS) IMR (RMS) U5MR (RMS)
IMR (direct) NMR (direct VR) NMR (direct DHIS)
Some observations about the assumption of constant completeness by age
Also, urban completeness > rural completeness and age distribution differs in the two areas violation of assumption
With deaths reported by households
Where there are many one or two-person households
under-reporting of deaths at the old ages (particularly for women?)
For young adults the opposite might happen – a death being reported by more than one household
8
Estimating of mortality of sub-national areal populations
A challenge since
VR data – lack of correspondence between place of death and place of usual residence (which is usually not captured) and need reliable estimates of sub-national migration
Household deaths – often violate the assumption of constant completeness by age and are for a short period creating uncertainty about the estimates
Thus Dorrington and Timaeus (2015) suggest combining the two sources, by
using VR to estimate the true numbers of deaths nationally by age for each sex (and possibly other divisions that could affect reporting of household data)
Assuming all households by sex (and other divisions) are equally likely to report the deaths at each age
9
SA: 45q15 vs projection model estimates
10
WC
EC
NC
FS
KZ
NW
GT
MP
LP
WC
ECNC
FSKZ
NW
GT
MP
LP
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70%
45q
15 f
rom
ho
useh
old
death
s
45q15 from model
2001
Males
Females
45o
NationalWC
EC
NC
FSKZ
NW
GT
MPLP
WC
EC
NC
FS
KZNW
GT
MP
LP
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70%
45q
15 f
rom
ho
useh
old
death
s
45q15 from model
2011
Males
Females
45o
National
Brazil: vs household deaths alone
11
N NE
SE
S
MW
NNE
SES
MW
0%
5%
10%
15%
20%
25%
30%
0% 5% 10% 15% 20% 25%
Qu
eiro
z
Dorrington&Timaeus
45q15
45o
Males
Females
N
NE
SE
S
MW
N
NE
SE
S
MW
75%
80%
85%
90%
95%
100%
105%
110%
75% 80% 85% 90% 95% 100% 105% 110%Q
uei
roz
Dorrington&Timaeus
Completeness
45o
Males
Females
Brazil: vs VR data alone
12
N NESE
S
MW
NNE SES MW
0%
5%
10%
15%
20%
25%
0% 5% 10% 15% 20% 25%
Qu
eiro
z
Dorrington&Timaeus
45q15
45o
Series2
Series3
N NE
SE SMW
N
NE
SE S
MW
70%
75%
80%
85%
90%
95%
100%
105%
70% 75% 80% 85% 90% 95% 100% 105%Q
uei
roz
Dorrington&Timaeus
Completeness
45o
Males
Females
Discussion
RMS
Even an incomplete population register can be useful
Consistency of trend over time as important than attempting to increase completeness or release data more rapidly at the expense of seriously disrupting a well-established trend
When completeness high direct estimates of IMR are not too inaccurate
Assumption of constant completeness by age
Only true above age 30 in South Africa
Likely to be a problem in significantly heterogeneous populations (e.g. urban vs rural)
Household deaths likely to be problematic unless stable multigenerational families
16
Discussion
Sub-national populations
Method is promising, but no gold standard to check
Works better if can identify factors linked to whether households report deaths
Mortality at old ages
Age problems (year of birth heaping) can be entrenched in the official data, limiting usefulness of NEG methods
NEG-GAM has potential but may not eliminate the age exaggeration in the death data (particularly if incentives to exaggerate age)
17