Sampling Considerations, Including Using a PDA with GPS to choose a sample and conduct a survey.

1
Sampling Considerations, Including Using a PDA
with GPS to choose a sample and conduct a survey.

• Statistical and Spatial Considerations
• Allen Hightower

2
Sampling Design Considerations 1

• Precision at the national and first level
  administrative units.
• The ability to evaluate large scale programs
• National in scope
• First admin level
• Not the second admin level!
• The ability of the survey to give answers to your
  tabulation plan (mock report).

3
Sampling Design Considerations 2

• Appropriate sampling frame
• Is your target population correct?
• Survey the whole country or only areas with
  malaria risk?
• Cost benefit /logistics
• Should you include large/remote areas with little
  population?
• Eg, NE province in Kenya large, arid area
• Political Considerations
• Program evaluation
• Ethnic groups of interest

4
Sampling Design Considerations 3

• Usually use a multi stage survey with a
  combination of the below components
• Simple Random Sample
• Best precision, valid
• Systematic sample.actually is a cluster sample.
• Cluster Sample
• Less precision, valid. Operationally easier to
  implement.
• Multi-stage clustering potential analytic issues
• Stratified sample to increase precision. Often
  used to define domains of interest.

5
Sampling Design Considerations 4

• Be sure that your sampling plan will cover
  domains/target populations of interest
• Rural/urban 75/25?
• Malaria endemic zones
• Pregnant women This was the hard one for Kenya
• Achieved by increasing cluster size
• CU5s

6
Sampling Design Considerations 5

• What do you do if your sampling frame is out of
  date or does not exist?
• Work with the CSO to develop a new one.
• Work with CSO to improve the existing one.
• Two tracks
• Paper the household listing system
• PDA Use GPS Survey

7
Why Use GPS Survey?

• CSO maps often out of date
• Peri urban areas
• Urban areas
• All of them?
• GPS survey creates an up to date sampling frame
  (ie, HH list)
• GPS survey creates a statistically valid sample

8
Why Use GPS Survey?

• GPS survey can be used in cluster or simple
  random samples at the last stage.
• GPS Survey can provide information on the
  interviewer not possible with paper-based surveys.

9
PDA data collection How safe?

• Data saved field-by-field
• Battery Loss non-issue.
• Memory is nonvolatile like a digitall camera.
• Battery life gtgt16 hrs
• Better than laptop/tablet computers
• Big screens, HDs use power!
• No moving parts (ie, better than a laptop)
• Encryption/PW at OS level
• PDAs we use are very, very sturdy.
• Have never had one stolen/lost.
• CDC Kenya collected gt 12 million pieces of paper
  electronically with PDAs last year.

10
PDA data collection Can Paper do this?

• Live skip patterns less training
• Only see the right questions.
• QC at Data Collection point
• Autoword completion less typing
• Turn off unneeded features (games/buttons)
• Know where interviewer is/should be
• Know when the interview started/stopped
• Very, very rapid data availability

11
Concept of PDA-GPS system

• Interview teams take nothing more than PDAs with
  attached GPS units into a cluster
• Map HHs in the cluster
• Choose a statistically valid sample
• Conduct Software Guided Interview w/PDA
• Data Rapidly Available for Reporting/Spatial
  Analyses

12
PDA-GPS mapping system

• Simple, fast
• Most clusters mapped in 4 hrs or less
• Village Name (auto carry)
• HH number (auto increment)
• Comment (e.g., the house in the middle, family
  name)
• GPS gets you close, Comment provides final ID
  info
• Allows Multiple mappers

13
GPS Survey ProgramGPS readings

• Program checks the CF-GPS unit for a valid
  reading
• Gives color coded GPS validity
• Green good precision
• Yellow marginal
• Red poor
• Random number assigned to HH when data are saved.

14
(No Transcript)
15
Sampling Design

• It is your decision!
• The system only affects how households or
  compounds are sampled within a cluster.

16
GPS Survey-Sampling

• You tell the PDA program
• How many HHs to sample from the EA.
• How many alternates HHs?
• SRS or Cluster Sample
• Parameters set up in XML file
• You get
• a random sample where each HH has a known
  probablity of selection.
• A File of the GPS data for all mapped HHs and
  features of interest
• Denotes which HHs are in the sample

17
GPS Navigation Software

• Guides the interviewer back the sample HH
• List of HH IDs with distance and compass
  direction /pointer
• Also a map of HHs
• Can mark HH off list
• Can divide a list

18
GPS Navigation Software

• List of HH IDs with distance and compass
  direction/pointer
• Can mark HH off list
• Can divide a list to allow multiple interviewers
  to split work.

19
GPS Navigation Software

• Also a map of HHs
• Can show HH ID info

20
More to this than technology!!!

• This is a system that uses technology.
• Software development and testing.
• Training, training.
• Stay in touch with the data collectors in the
  field regularly.

21
Training!!!

• Can train people to use a lousy system
• Poor training can ruin a perfect system
• Learn how to solve common problems
• Tip sheets/manuals
• Training schedules/exercises
• Stay in touch with the data collectors in the
  field regularly.

22
Advantages of GPS-based Simple Random Sample

• Gold Standard Approach
• SRS better precision than other approaches.
• Unbiased (accurate)
• Any HH in village can be a part of sample
• You know the probability of selection for a HH.
• IE, a probability sample.
• Spatial Analyses

23
Why is Statistical Validity Important?

• For any survey, even those not using
  statistically valid methods, you can compute
  means, proportions, std errors.
• However, to know the statistical distribution and
  compute confidence intervals for the means and
  proportions, you MUST use a random sampling
  process.
• Or you must pretend that you used a random
  sampling process.

24
How much work to map cluster?

• 4 hours to map the EA, typically
• Some take 2 days!
• 1 to 3 minutes between GPS readings (between
  HHs)
• Each interviewer mapped between 30 to 45 HHs per
  EA

25
GPS can be used for quality control. Here we
caught someone standing in one place. All GPS
readings were within 60m. Sept 2005 Bad (Red)
Green remapping of village!
26
GPS can be used for quality control. In the
Kenya MIS, we had an extra cluster mysteriously
appear.
27
As we looked, the phantom cluster consisted of
points that belonged to other 4 clusters. All
pts were very close together. Turns out, the team
had stopped at a meeting place to complete their
quota. Our system has an audit trail, taking a
GPS reading each time a household record is
changed.
28
Can Produce a Map of HHs in each EA,Showing
which were in the Survey

• Survey Program produces Database file with
• Lat/Lons for all HHs in EA
• Indicator for whether HH was included in sample
• Lat/Lons for all other features of interest
• Easily Imported into other GIS products

29
Standardized data collection tools

• MISs in Zambia, Kenya, Ethiopia
• Bednet Ownership/Usage surveys in a number of
  countries.
• Comparable methods, systems
• Extremely cost effective if systems and pdas are
  used multiple times.
• Not sure it is worth it to do just once

30
Thank you!