3. Activity Goal
We seek an improved understanding of the relationship
between product availability, population demand, and
use of modern family planning methods
Population
Demand
Product
Availability
CPR
4. Analysis Goal
We seek an improved
understanding of the
relationship between
• Depo-Provera Availability
– Household Distance from SDP
– Stock Reliability
• Household Demand and Use
of Injectable Contraceptives
Distance and
Stock
Reliability
Access and
Product
Availability
Demand
and Use
5. Research Focus
• Is availability of Depo-Provera associated with either
demand for or use of injectable contraceptives
among married women of reproductive age?
Methods
• Data being used
– MEASURE DHS 2010 Malawi Survey
– DELIVER PROJECT LMIS data
• Linking the data
• Regression and Assumptions
7. MEASURE DHS Malawi CPR & Injectable
Contraceptive Use in 2010
0
10
20
30
40
50
1992 2000 2004 2010
Male sterilization
IUD
Implants
Condom
Pill
Female sterilization
Injections
8. Available LMIS Data
The DELIVER Project provided data on Depo-Provera
supplies for virtually all the public facilities in Malawi
providing family planning care.
• Health facility level issues data: 2008-2013
• Health facility stockout data: 2008-2013
• Created Reliability Measure to quantify product
availability
11. Merging Data: DELIVER Supply data with Facility Master
696 Health
Facilities
483 Depo
service
delivery
sites
930 Health
Facilities
Unique
facility
identifier 571 Health
Facilities
Valid GPS
(dropped 234) (dropped 125)
Final Mapped Facility Dataset with Supply Data
• N=571 Total Facilities
• 423 counted as Depo-Provera Service Delivery sites
423 Depo
service
delivery
sites
12. Population-Based Data: Malawi DHS 2010
827 DHS
Clusters:
22,480 Women
849 DHS
Clusters:
23,020
Women
Valid GPS
(dropped 22
clusters)
(dropped
7391 women)
Final Mapped DHS data
• 827 Clusters
• N=15,089 married women 15-49 years old (unwgtd)
Married
Women
827 DHS
Clusters:
15,089 Women
15. What is Kernel Density Estimation (KDE)?
KDE is a technique
employed to distribute
a value associated
with a discrete point
across a plane or
continuous surface.
16. Operationalizing “Access”
Used KDE to create variables measuring access
Access 1 = distance from DHS cluster to a Depo-
Provera service delivery point
Access 2 = distance from DHS cluster to a Depo-
Provera service delivery point + reliability of
Depo-Provera supply
For each variable, created relative measure - quintiles
17. KDE Surface: 10Km radius around all Depo Service
Delivery Sites
KDE Surface: 10Km radius around Depo sites using
weighted variable representing Depo-Provera supply
Kernel Density Estimation: Access
Health Center: Depo Delivery Site
Health Center
Health Center: Depo Delivery Site
Health Center
19. KDE Surface: 10Km radius Depo Service Delivery
Sites, plus 5km buffer around DHS clusters
KDE Surface: 10Km radius around Depo Sites with
supply weight and 5km buffer around clusters
KDE Access & DHS Clusters
Health Center offering family planning
Health Center
DHS Cluster with 5km buffer
Health Center offering family planning
Health Center
DHS Cluster with 5km buffer
20. Research Questions
1. Is access to Depo-Provera
supply positively associated with
use of injectable contraceptives? Injectable Use
Yes
2. Is access to Depo-Provera supply
positively associated with desire to
space and/or limit children?
Demand
Yes
Access to Depo = quintiled variable, ~ 20% per group
Key Outcomes
Key Independent Variable
23. Limitations
• Incomplete data – supply data, GPS data, respondents
• DHS cluster displacement
• Alternative measures of product availability
• Single time point
• Cannot determine causality
• Importance of distance in developing the KDE in terms
of how far people can travel—used a “flat” surface
24. Conclusions
• Access to Depo-Provera is positively associated with
a married woman’s use of an injectable family
planning method in rural areas.
• Access to Depo-Provera (Reliability + Distance) is
associated with an increase in family planning
demand in urban areas.
• In the Central Region, Depo-Provera supply is either
unreliable or suffers from incomplete reporting
• Application and validation of method for linking
survey data with facility data applicable
25. MEASURE Evaluation is a MEASURE project funded by the U.S. Agency
for International Development and implemented by the Carolina
Population Center at the University of North Carolina at Chapel Hill in
partnership with Futures Group International, ICF Macro, John Snow, Inc.,
Management Sciences for Health, and Tulane University.
The USAID | DELIVER PROJECT is funded through contract no. GPO-I-
00-06-00007-00 and is implemented by John Snow, Inc..
Views expressed in this presentation do not necessarily reflect the views
of USAID or the U.S. Government.
27. Stock Reliability Detailed Calculations
• Reliability Score for each facility
– Months with stock on hand at the end of the month have a
value of 1
– Month with no record of stock on hand but which dispense
continuously for 3 months have a value of 1
– A value of 0.5 was assigned to months which were missing a
LMIS record but had more than twice the Average Monthly
Consumption in on-hand in the previous month
– Annual reliability score = average of monthly values
Editor's Notes
Andrew Presents first followed by Martha
This is a core funded activity which seeks to strengthen the statement “No Product No Program” by using data to demonstrate the relationship between an effective supply chain and program outcomes.
Globally, A 1% increase in Average Quarterly Consumption of injectibles per 100 Women of Reproductive Age ( AQC 100WRA ) is associated with a .661% increase in CPR.
1. Differentiate the other Sources
1. Differentiate the other Sources
Months with stock on hand at the end of the month have a value of 1
Month with no record of stock on hand but which dispense continuously for 3 months have a value of 1
A value of 0.5 was assigned to months which were missing a LMIS record but had more than twice the Average Monthly Consumption in on-hand in the previous month
Annual reliability score = average of monthly values
930 = public and private
696 = primarily public, facilities providing patient care
483 = depo delivery points per supply data info
What we learned from previous linking experience:
1. Critical to have a full list of Facilities and their geographic location
2. Link to a service environment rather than the nearest facility.
3. Use of kernel density estimation (KDE) to map this service environment.
Maximum reach of 10 km around each facility
Average KDE value for each cluster as a proxy measure for “access”
Potential error from displacement of DHS clusters
SAND analogy
Use of kernel density estimation (KDE) to map this service environment.
Maximum reach of 10 km around each facility
Average KDE value for each cluster as a proxy measure for “access”
Potential error from displacement of DHS clusters
Nationally this mapping may highlight areas with less reliable depo supply – Central region
L=if supply was reliable everywhere then we’d have this distribution
R=what happens to access when supply is unreliable
Distance: a married women with most access to a facility offering depo services is ~ 5% more likely to use an injectable compared to a woman with least access.
Distance + Supply: a married women with most access to a facility offering a reliable supply of depo is ~ 5% more likely to use an injectable compared to a woman with least access.
Distance: a married women with better access to a facility offering depo services is ~ 3-6% more likely to use an injectable compared to a woman with poor access, and holding all other factors constant.
Distance + Supply: a married women with better access to a facility offering a reliable supply of depo is ~ 4-5% more likely to use an injectable compared to a woman with poor access, and holding all other factors constant.
Note: When model is run for married women in rural areas ONLY, the dip for those with “Most Access” disappears.
Why? Marc??
Note – when this model is run for Rural, it becomes more apparent, and dissipates for rural.
Now switching to our second outcome – demand or our assumption of demand based on a woman’s desire to space or limit children.
Demand:
Yes = want to limit, want > 2yrs
No = want within 2 yrs, undecided, steril, infecund
Note – while there appears to be no pattern here, a pattern emerges for the urban centers
Incomplete data = potential measurement error
DHS cluster displacement = potential spatial error introduced. Effort to minimize error by averaging the KDE value across a 5km buffer around the cluster GPS point.