Survey SolutionsCOMPUTER-ASSISTED PERSONAL INTERVIEWING
Michael Lokshin
World Bank
The problem:Average interval between surveys (years)
2
3
4
5
6
7
9
0 2 4 6 8 10
LAC
ECA
World
EAP
SA
SSA
MENA
The problem:Percentage of missing or invalid answers
(questions on learning and literacy)
0% 2% 4% 6% 8% 10% 12% 14% 16% 18%
Malawi
Uganda
Ghana
Nepal
Madagascar
Burundi
Ethiopia
Laos
Only 11 out of 55 countries in Africa had comparable poverty data for2005. The situation has not changed much since.
• 2011: Comprehensive Assessment of CAPI software products is commissioned to University of Maryland by the WB.
… no existing software provides exactly the right mix of features necessary for the sort of surveys conducted by World Bank and its clients.
• 02.2012: LSMS and Computational Tools teams of Research Department of the World Bank with support from Global Strategy of FAO started development.
• 09.2013: First public version of Survey Solutions is released
• 05.2015: Survey Solutions 4.4 is released
Background
Pick 3: Fast, High Quality, Cheap
Ensure data quality and comparability
Improve timeliness of data collection
Cost-effective, sustainable solution for NSOs
From data collection to policy: 1-2 years
From data collection to policy: 4-5 months
Data analysis
Policy decisions
Data collection
Data entry
Data cleaning
Data processing
CAPI
Results:
• Simple yet flexible system for the non-expert users. Typical clients – National Stat Offices
• Functionality for
– data capturing
– survey management
– data management
• Tablet-based with ability to display and navigate through multi-level large questionnaires.
• Support of panel surveys and complex validation algorithms.
• Cost effective system that can be used and supported by NSOs without external TA.
CAPI System Requirements
• CSPro – free, closed source. resides in Bureau of Census USA. Funded by USAID. Windows and Android
• ODK – free, open source software funded by USAID; UWashington, UC Berkeley, UC Davis. A– SurveyCTO, license fee
• Blaise – Statistics Netherlands, license fee. W
• SurveyBe – closed source, license fee. W
• SurveyToGo – closed source, license fee. W
Other CAPI products
𝐷𝑄 = 𝛼𝐷𝐶 ∙ 𝛽𝐷𝑀 ∙ 𝛾𝑆𝑀
DQ – data quality
DC – data capturing
DM – data management
SM – survey management
Survey Solutions : α ≈ β ≈ γ
• Pros:
– A flexible system that can
accommodate complicated
surveys.
• Cons:
– Custom solutions cannot be
easily modified for different
environments
– High costs
– High requirements for the users.
– Questionable sustainability in
the high turnover environments
typical for NSO.
CAPI: Expert-centered approach
• Pros:
– High levels of standardization, solutions could be easily adapted for different surveys in different countries.
– Lower requirements for the users, fast learning curve
– High sustainability
• Cons:
– Less flexible compared to custom solutions of expert-centered software.
CAPI: User-centered approach
• Sustainable, low-cost system for NSO
• Simple, flexible interface for questionnaire development
and testing.
• Tablet interface allows easy navigation through complex
questionnaires.
• Standardized survey management protocol based on the
best practices of data collection
• Intuitive, informative survey status reporting, survey
maps.
• Yet, powerful language for data validation and control of
questionnaire flow.
Survey Solutions: hybrid approach
Survey Solutions
Competition
• Large questionnaires: several thousand questions
• All standard types of questions: – text, numeric, date; multi-choice; dynamic lists;
– Linked questions: Whose cow is it? User can select from the list of household members.
– GPS location; Time; Barcode; Binary files (pictures)
– HQ/Supervisor-filled questions
• Rosters: can be generated from:– Fixed lists; dynamic lists; numeric; multi-choice questions.
– Nested rosters with unlimited degree of nesting
• Interviewer Comments on a question and interview
• Question Instructions
Survey Solutions: data capturing
GPS in action
Plots can be easily and precisely measured by simply walking around the area with the tablets
Ability to measure distances to banks, schools, roads, wells. Maps used for optimal routing
Sensors
Soil Fertility sensor
Water quality testing
Multi-purpose sensor
Vision tests
Portable scales
Wearable devices
Para Data: Adptive Survey Design
• Improve data quality by correcting survey process during the field operation.
• System records all events with timestamps that happen on a tablet: Data entry, Data correction, Responsibility changes, etc.
• Analysis of time per interview, time per question, section. • Changes in productivity over time, for different
interviewers, teams• Quality control, monitoring and evaluation
Prices of tablets start from $60: cheaper than paper questionnaires
Questionnaire delivery, Changes or corrections at no cost
Free software but not free survey
Software might be free but the TA support is very expensive:
Competitor #1: Proposal for a survey in Sri Lanka, 24,000 hhs –$340KCompetitor #2: Actual cost of a survey in Uganda, 10,000 hhs -$350K
Survey Solutions: Cost varies from about $5,000 to 0,000s for large surveys.
• Designer: online tool for questionnaire creation and validation @ solutions.worldbank.org
• Tester: Android app connected to Designer to test questionnaires in real time. @ Google play
• HQ: online tool for centralized survey management, validation, data aggregation and reporting.
• Supervisor: online/offline tool to manage process of data collection on a team supervisor level.
• Tablet CAPI: Android app for data capturing on a tablet.
Survey Solutions: system components
SURVEY PIPELINE: 50+ SURVEYS IN 30+ COUNTRIES
• Africa: Malawi, Uganda (4), Tanzania, Togo, Benin, Madagascar (3), Niger, Cote D’Ivoire (4), Zambia (3), South Africa, Ghana, South Sudan, Mozambique, Nigeria
• MENA: Djibouti (4), Kuwait, Morocco, Tunisia• SAR: Bhutan (3), India, Pakistan• EAP: Thailand (2), Myanmar (7), Pacific Islands (4)• ECA: Armenia, Azerbaijan, Kirgizstan• LAC: St. Lucia (4), Paraguay, Belize
• Survey types: LSMS, HBS, LFS, Enterprise Survey, EDU, Health
• Largest survey: Malawi - 200,000 households• Largest questionnaire: Tanzania – 3,000 questions
Enumerators repeat interviews if errors are detected
Interviewers synchronize theirdevices and upload completedquestionnaires
Supervisors monitor the submissions
Questionnaires with no errors are uploaded to the central server
Supervisors assign households to individual interviewers
WiFi/USB
Researchers design questionnaires using visual tools and upload them to the central server
Interviewers visithouseholds and collect data
HQ distributes the sample lists across teams of enumerators
Internet
WiFi/USB
Internet
Main Website: www.worldbank.org/capi
Online questionnaire designer: solutions.worldbank.org
Youtube channel: Survey Solutions
Real time status of interviews
Map of the survey
Monitor the survey by checking the GPS locationof where and when the interview took place.