phone surveys for health: methods and opportunities...sms random digital dial in ghana, kenya,...

32
Phone Surveys for Health: Methods and Opportunities Abigail Greenleaf, PhD, MPH May 5, 2020

Upload: others

Post on 01-Jun-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Phone Surveys for Health: Methods and Opportunities

Abigail Greenleaf, PhD, MPHMay 5, 2020

Page 2: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Overview

1. Cell Phone Ownership in Sub-Saharan Africa (SSA)2. Modes of Data Collection3. Data Collection Goal4. Total Survey Error Framework Applied to Cell Phone Surveys in SSA

• Frame • Sampling• Response • Post- Adjustment

5. Wrap-up6. Questions

Page 3: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

• The first cell phone call in SSA was in 1987 in DRC (formerly Zaire)

• Mobile phone penetration rate: % of unique users within a population

• In 2018, SSA had 44% (or 456 million) unique subscribers

• Looking at penetration by region, growth will be equal but West Africa has highest current penetration (48%) and Central Africa has lowest (40%)Source: GSMA 2019

Cell Phone Ownership in SSA: Past & Present

Presenter
Presentation Notes
Image: GSMA 2019 SSA Report https://www.gsma.com/mobileeconomy/wp-content/uploads/2020/03/GSMA_MobileEconomy2020_SSA_Eng.pdf Africa: The Impact of Mobile Phones. Vodafone, 2005. https://sarpn.org/documents/d0001181/P1309-Vodafone_March2005.pdf GSMA 2019 SSA GSMA 2019 SSA https://www.gsma.com/mobileeconomy/wp-content/uploads/2020/03/GSMA_MobileEconomy2020_SSA_Eng.pdf
Page 4: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

• SSA continues to be the fastest growing region for cell phone ownership, but the Compound Annual Growth Rate (CAGR) is slowing

• Mobile phone penetration: 50% by 2025

• Next 5 years: additional 167 million subscribers with half of new subscribers coming from 5 countries:

1. Nigeria (31 mil) 2. Ethiopia (18 mil)3. DRC (15 mil)4. Tanzania (10 mil)5. Kenya (9 mil)6. Others (84 mil)

Mobile Phone Penetration in SSA, 2018

Source: Greenleaf 2018

Cell Phone Ownership in SSA: Future

Presenter
Presentation Notes
(All info on slides) GSMA SSA 2019 : https://www.gsma.com/mobileeconomy/wp-content/uploads/2020/03/GSMA_MobileEconomy2020_SSA_Eng.pdf Image: created by Abigail Greenleaf and Julien Nobili using GSMA 2018 data
Page 5: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Evolution of Cell Phone Data Collection Studies

• Likely 1st reported cell phone survey in SSA (Dillon, 2010):• Tanzania: Data collection (Sept 2009-Jul 2010) among 195 cotton farmers in 15 villages

sampled from village registry• Face-to-Face Enrollment, Live Interviewer Follow-up (World Bank, 2012-2015):

• Tanzania, Mali, South Sudan, Listening to Africa (Face-to-face [FTF] enrollment, Computer Assisted Telephone Interview [CATI] follow-up)

• Random Digit Dial: • Cote D’Ivoire, 2013: CATI with call center in France (Larmarange, 2016) • Zimbabwe, Afghanistan, Ethiopia, Mozambique: Interactive Voice Response [IVR] (Leo,

2015)• Literature Reviews: Articles from low and middle-income countries

• Population-based cell phone surveys: 11 articles, 9 used CATI (Gibson, 2017)• Mode comparison: 10 articles, (1 article overlap with Gibson) (Greenleaf, 2017)

Presenter
Presentation Notes
Dillon B. Using Mobile Phones to Conduct Research in Developing Countries. Journal of International Development. 2011;24:11. Multiple WB articles: Demombynes G, Gubbins PR. Challenges and Opportunities of Mobile Phone-Based Data Collection Evidence from South Sudan. In: Bank TW, editor. Policy Research Working Paper. Washington, DC: The World Bank; 2013. Etang-Ndip A, Hoogeveen J, Lendorfer J. Socioeconomic Impact of the Crisis in North Mali on Displaced People. Policy Research Working Paper. Washington, DC The World Bank; 2015. p. 30. Croke, K., et al., Collecting high frequency panel data in Africa using mobile phone interviews, in Policy Research Working Paper. 2012, The World Bank: Washington, DC. p. 27. Dabalen A, Etang A, Hoogeveen J, Mushi E, Schipper Y, von Engelhardt J. Mobile Phone Panel Surveys in Developing Countries: A Practical Guide for Microdata Collection. Directions in Development. Washington, DC World Bank 2016. http://documents.worldbank.org/curated/en/877231468391801912/Mobile-phone-panel-surveys-in-developing-countries-a-practical-guide-for-microdata-collection RDD Larmarange J, Kassoum O, Kakou E, Fradier Y, Lazare S. Feasibility and Representativeness of a Random Sample Mobile Phone Survey in Côte d’Ivoire. Population. 2016;71:10. Leo B, Morello R, Mellon J, Peixoto T, Davenport S. Do Mobile Surveys Work in Poor Countries. CGD Working Paper 398. Washington, DC: Center for Global Development 2014. Literature Reviews Gibson DG, Pereira A, Farrenkopf BA, Labrique AB, Pariyo GW, Hyder AA. Mobile Phone Surveys for Collecting Population-Level Estimates in Low- and Middle-Income Countries: A Literature Review. J Med Internet Res. 2017;19:e139. Greenleaf AR, Gibson DG, Khattar C, Labrique AB, Pariyo GW. Building the Evidence Base for Remote Data Collection in Low- and Middle-Income Countries: Comparing Reliability and Accuracy Across Survey Modalities. J Med Internet Res. 2017;19:e140.
Page 6: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

SMS: Short Message Service

Strengths Weaknesses• Software automates data collection so human oversight of data collection is not needed• Respondent can answer when convenient• Message sends when respondent is back in network coverage

• Requires literacy • Texting not a ubiquitous

activity• Keypad-based response

menu can be difficult to navigate or impossible to use if low literacy• Inboxes get full quickly on

phones with limited memory• Limited to 160 characters

• Difficult to judge data quality

• Generational differences in texting: In Nigeria, 89% of cell phone owners, ages 18-34, use SMS compared to 67% of owners, ages 35+ (Pew, 2015)

• SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019)

• Emerging literature: South Africa, Kenya

Receives more SMSOpt-in

Selects preferred language

Introduces survey

Presenter
Presentation Notes
Poushter J, Oates R. Cell Phones in Africa: Communication Lifeline. Washington, DC: Pew Research Center; 2015. Lau, C., et al., In Search of the Optimal Mode for Mobile Phone Surveys in Developing Countries. A Comparison of IVR, SMS, and CATI in Nigeria. Survey Research Methods, 2019. 13: p. 305-318.
Page 7: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

IVR: Interactive Voice Response

• IVR: Respondents listen to a pre-recorded question, then answer by pressing the number corresponding to their answer on their phone keypad

• Hybrid IVR: begins with live interviewer

Strengths Weaknesses• Recorded questions reduce interviewer bias • Software automates data

collection so human oversight of data collection is not needed• Rapid implementation

• Respondent may not be familiar with pre-recorded calls which decreases response rate• Keypad-based response menu can be difficult to navigate or impossible to use if low literacy • Difficult to judge data

quality (i.e. respondents understanding, or if respondents was distracted)

Source: Abigail Greenleaf 2017 (Top)Source: Abdoul Djalil, 2020 (Right)

Presenter
Presentation Notes
Page 8: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

CATI: Computer Assisted Telephone Interview

Strengths Weaknesses• Live operator, familiar

interaction• Good for low literacy

settings• Operators can clarify

questions• Interviewer builds

relationships with respondents thus reducing refusals and attrition

• Resource intensive (interviewers, supervisors)• Interviewer bias

Source: Greenleaf 2017

Presenter
Presentation Notes
Page 9: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Comparison of Modes for Health Surveys

• Representation: • Tanzania (Pariyo, 2019): Female respondents, those with lower levels of schooling and

who are rural residents, have higher participation rates for CATI compared to IVR• Also found in Nigeria (Lau et al.) and among women in Burkina Faso (Greenleaf et al.),

CATI produced more representative sample than IVR • Cost:

• Nigeria (Lau, 2019): For a survey of 3,000 completes, IVR is 43% the cost of CATI, and SMS is 24% the cost of CATI

• Burkina Faso (Greenleaf, 2019.):

CATI Hybid IVRInterview Length 6 minutes 13 minutesCost per completed survey $52 $144

Follow-up to FTF Survey among Women, Burkina Faso, 2017

Source: Greenleaf et al., PMA2020, 2019

Presenter
Presentation Notes
Representation Pariyo, G.W., et al., Does mobile phone survey method matter? Reliability of computer-assisted telephone interviews and interactive voice response non-communicable diseases risk factor surveys in low and middle income countries. PLoS One, 2019. 14(4): p. e0214450. Lau, C., et al., In Search of the Optimal Mode for Mobile Phone Surveys in Developing Countries. A Comparison of IVR, SMS, and CATI in Nigeria. Survey Research Methods, 2019. 13: p. 305-318. Greenleaf, A.R., et al., Comparison of remote data collection modes to monitor family planning progress in Burkina Faso: representativeness, data quality, and cost P.M.a.A.P. Project, Editor. 2019, Bill & Melinda Gates Institute for Population and Reproductive Health, Johns Hopkins Bloomberg School of Public Health: Baltimore, MD. https://www.pma2020.org/sites/default/files/PMA2020_RDC_Phase2_report_06March2019-web.pdf Cost See Lau above See Greenleaf above
Page 10: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Data Collection Goal

Surveillance• Health care workers: e.g. automated calls with

community health care workers in Niger• Populations: Ebola in Liberia, Sierra Leone; chronic

health conditions in Tanzania, Uganda; ‘the cholera phone’ in Bangladesh

Monitoring and Evaluation

Surveys• Sampling: RDD or FTF follow-up

Source: ICAP 2014

Presenter
Presentation Notes
HCW Niger Greenleaf AR, Vogel L. Interactive Voice Response Technology for Data Collection in Sub-Saharan Africa. In: Viamo, editor. Brief Toronto, Canada: Viamo; 2018. p. 6. Ebola Kuehne, A., et al., Mortality, Morbidity and Health-Seeking Behaviour during the Ebola Epidemic 2014-2015 in Monrovia Results from a Mobile Phone Survey. PLoS Negl Trop Dis, 2016. 10(8): p. e0004899. Berman, A., M.E. Figueroa, and J.D. Storey, Use of SMS-Based Surveys in the Rapid Response to the Ebola Outbreak in Liberia: Opening Community Dialogue. J Health Commun, 2017. 22(sup1): p. 15-23 Chronic Health Conditions Gibson DG, Pariyo GW, Wosu AC, et al. Evaluation of Mechanisms to Improve Performance of Mobile Phone Surveys in Low- and Middle-Income Countries: Research Protocol. JMIR Res Protoc. 2017;6:e81. Cholera/ Bangladesh Carstensen, L.S., et al., The Cholera Phone: Diarrheal Disease Surveillance by Mobile Phone in Bangladesh. Am J Trop Med Hyg, 2019. 100(3): p. 510-516.
Page 11: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Total Survey Error

Research Question: Have tigers in Thailand started wearing face masks to reduce

transmission of COVID-19?

Source: Groves, 2009 Source: WWF

Measurement Representation

Survey Statistic

Construct

Measurement

Response

Edited Response

Validity

Measurement Error

Processing Error

Target Population

Sampling Frame

Sample

Respondents

Postsurvey Adjustments

Coverage Error

Sampling Error

Nonresponse Error

Adjustment Error

Presenter
Presentation Notes
Total Survey Error Framework: Groves, Robert M, et al. Survey Methodology. Hoboken, NJ : John Wiley & Sons: 2014. Tiger picture: https://www.worldwildlife.org/species/tiger
Page 12: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Coverage Error

• Target Population • The group about whom you want to make an

inference

• Sample Frame• List of everyone in your target population• If you do not have a cell phone, you cannot be on

the sample frame

• Coverage error• Those included in the target population and

sample frame are different

• Types of Coverage Error • 4 types but undercoverage concerning for cell phone

surveys• Undercoverage: units that should be in the sample

frame are not included

Sample Frame

1

2

3

4

5

6

7

8

9

10

Target Population

Frame

Source: WWF

Page 13: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Coverage Error: Who Owns Cell Phones?

Broadly speaking, owners are:Cell phone ownership data sources:• Surveys:

• PHIA (HH and individual); • Demographic and Health Surveys (HH and female

ownership; • PMA2020 (HH and female ownership);• Afrobarometer (HH ownership)

• Private Sector and Governmental: • GSMA• International Telecommunications Union

Young (young women less likely than men)

Men Educated

Urban

Presenter
Presentation Notes
Surveys https://phia.icap.columbia.edu https://dhsprogram.com https://www.pmadata.org https://www.afrobarometer.org Private Sector & Gov https://www.gsma.com https://www.itu.int/en/Pages/default.aspx
Page 14: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Peer-Reviewed Literature on Phone Ownership

• Mozambique (Velthausz, 2016): • Of the population in four Provinces that used a phone, 68% owned a phone, 32% had

access only.• 90% of follow-up respondents were phone owners

• Among female phone owners in Burkina Faso (Greenleaf, 2018):• 83% reported only one phone number• 14% had two numbers• 3% had three or more phone numbers• The mean cell phone numbers:1.2

• Countries with higher mobile penetration (Ethiopia, Zimbabwe) resulted in less sample distortion and more comparable estimates to recent FTF surveys compared to the countries with lower phone penetration (Afghanistan, Mozambique) (Leo, 2015)

Presenter
Presentation Notes
Velthausz D, Donco R, Skelly H, Eichleay M. Mozambique Mobile Access and Usage Study: Computer-Assisted Telephone Interview (CATI) Survey Results. In: FHI360, editor.2016. p. 66. Greenleaf AR, Ahmed S, Moreau C, Guiella G, Choi Y. Cell Phone Ownership and Modern Contraceptive use in Burkina Faso: Implications for Research and Interventions using Mobile Technology Contraception. 2018. Leo B, Morello R, Mellon J, Peixoto T, Davenport S. Do Mobile Surveys Work in Poor Countries. CGD Working Paper 398. Washington, DC: Center for Global Development 2015. Mentioned during presentation as a resource to look at to assess statistical meaning of people multiple phone numbers Labrique A, Blynn E, Ahmed S, Gibson D, Pariyo G, Hyder AA. Health Surveys Using Mobile Phones in Developing Countries: Automated Active Strata Monitoring and Other Statistical Considerations for Improving Precision and Reducing Biases. J Med Internet Res. 2017;19:e121.
Page 15: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Women and Cell Phone Ownership

• 2 districts in Southern Tanzania (Vasudevan, 2016)• 6% of women reported used a phone daily • 80% of women reported using a phone less than once a week• Urban women more likely (70%) to be sole owner compared to rural women (44%)

• Burkina Faso (Greenleaf, 2018) • 86% of households owned a cell phone and 47% of women reported personal cell phone ownership • Women with secondary education or higher had 4 times the odds of cell phone ownership compared

to women with no formal education • 5 States in Nigeria (Jennings, 2015)

• Women who did not have phone access had half the odds of reporting modern contraceptive use compared to women that had phone access

• GSMA 2020 Gender Gap Report• Barriers to mobile phone ownership for women in SSA

• Affordability• Literacy and Skills • Family does not approve

Presenter
Presentation Notes
Vasudevan, L., et al., Patterns of Mobile Phone Ownership and Use Among Pregnant Women in Southern Tanzania: Cross-Sectional Survey. JMIR Mhealth Uhealth, 2020. 8(4): p. e17122. Greenleaf AR, Ahmed S, Moreau C, Guiella G, Choi Y. Cell Phone Ownership and Modern Contraceptive use in Burkina Faso: Implications for Research and Interventions using Mobile Technology Contraception. 2018. Jennings L, Omoni A, Akerele A, Ibrahim Y, Ekanem E. Disparities in mobile phone access and maternal health service utilization in Nigeria: a population-based survey. Int J Med Inform. 2015;84:341-8. GSMA 2020 Gender Gap Report: https://www.gsma.com/mobilefordevelopment/wp-content/uploads/2020/02/GSMA-The-Mobile-Gender-Gap-Report-2020.pdf
Page 16: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Coverage Error: Conclusions

• Compare the profile of phone owners and non-owners to see if your sample frame reflects your target population

• Look at ownership by gender• Cell phone owners are demographically distinct

from non-owners

• As cell phone ownership increases, so will representativeness of your study

• US started phone surveys in 1970s when ~80% of HH had landlines

• Reaching non-phone owners• FTF enrollment: give a cell phone • Pass the phone to non-phone owner• Half-interval open

Source: ICAP 2014

Page 17: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Sampling

Sampling: the process of selecting a subset of observations from an entire population of interest so that characteristics from the subset (sample) can be used to draw conclusion or making inference about the entire population

Sampled1

2

3

4

5

6

7

8

9

10

Target Population

Frame

Sample

Source: WWF

Page 18: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Random Digit Dial

• Creating or obtaining phone numbers to call • Use prefixes to randomly generate numbers (Gibson, 2017) • Possibly get active numbers from mobile network operator (MNO)• List of respondents who previously completed a survey

• Take into account market share each MNO has and sample accordingly

• MNO matters: having a premium cell phone provider, was significant predictor of participation in Tanzania (Croke, 2012)

• Quotas to ensure representative of hard-to-reach pops • Most RDD studies in SSA to date used IVR • RDD study using CATI to contact women in Burkina Faso

(Greenleaf, 2020)• Of the calls that were picked up 72% were men • To reach 1 rural respondent ages 15-19, ~ 30 picked-up calls Source: Aliou Gadiaga, 2018

Presenter
Presentation Notes
Gibson DG, Pariyo GW, Wosu AC, et al. Evaluation of Mechanisms to Improve Performance of Mobile Phone Surveys in Low- and Middle-Income Countries: Research Protocol. JMIR Res Protoc. 2017;6:e81. Croke, K., et al., Collecting high frequency panel data in Africa using mobile phone interviews, in Policy Research Working Paper. 2012, The World Bank: Washington, DC. p. 27. Greenleaf AR, Gadiaga A, Guiella G, Turke S, Battle N, Ahmed S, Moreau C. Comparability of modern contraceptive use estimates between a face-to-face survey and a cellphone survey among women in Burkina Faso. [available May 13]: PLoS ONE 10.1371/journal.pone.0231819
Page 19: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Follow-up to Face-to-Face Survey

• World Bank gave phones and solar charges to respondents and achieved high response rates during a panel study (monthly contact)

• FTF survey serves as your sample frame

• Often see everyone in frame included

• Differential between those who are eligible (have cell phone number) and consent?

Source: Abigail Greenleaf, 2017

Page 20: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Sampling: Conclusions

RDD • “RDD is not suitable for context in which the target population has low coverage rates and the

aim is to create a representative sample” (Dabalen, 2016)• Quotas extend data collection significantly

• 7 additional days to fill 15-19 rural quota in Burkina Faso

FTF Enrollment • Follow-up as soon as possible after FTF enrollment• Over-sample groups that are harder to contact • Over-sample groups that are less likely to complete questionnaire

Presenter
Presentation Notes
Dabalen A, Etang A, Hoogeveen J, Mushi E, Schipper Y, von Engelhardt J. Mobile Phone Panel Surveys in Developing Countries: A Practical Guide for Microdata Collection. Directions in Development. Washington, DC World Bank 2016.
Page 21: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Non-Response

• Non-response error: Those who respond to the survey are different from those who do not respond

• Non-response in phone surveys generally stems from 3 causes: • Failure to contact sampled

respondents • Refusal to participate • Health or language limitations

Respondents

2

4

6

8

10

Target Population

Frame

Sample

Respondents

Source: WWF

Presenter
Presentation Notes
Bullet 2: Biemer P. Overview of Design Issues: Total Survey Error. In: Marsden PV WJ, editor. Handbook of Survey Research. 2 ed. Bingley, United Kingdom: Emerald Group; 2010. p. 886.
Page 22: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Non-Contact

• Causes of non-contact in cell phone survey in SSA• Removing SIM from phone

• Mozambique: 39% of respondents for 4 provinces who were enrolled FTF and followed up with CATI swapped their SIM Card within the 4-month period, often to achieve better reception (Velthausz, 2016)

• Dead battery• Out of range of network service

• Quicker the follow-up post-contact the higher the retention

• Mozambique: 80% non-contact of random list (Velthausz, 2016)

• Burkina Faso: 1 year, 43% non-contact (Greenleaf, 2019)

Source: Wikimedia, Jorge Barrios

Presenter
Presentation Notes
Velthausz D, Donco R, Skelly H, Eichleay M. Mozambique Mobile Access and Usage Study: Computer-Assisted Telephone Interview (CATI) Survey Results. In: FHI360, editor.2016. p. 66. Greenleaf, A.R., et al., Comparison of remote data collection modes to monitor family planning progress in Burkina Faso: representativeness, data quality, and cost P.M.a.A.P. Project, Editor. 2019, Bill & Melinda Gates Institute for Population and Reproductive Health, Johns Hopkins Bloomberg School of Public Health: Baltimore, MD. https://www.pma2020.org/sites/default/files/PMA2020_RDC_Phase2_report_06March2019-web.pdf
Page 23: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Refusal

• Trying to encourage participation among those that do own a phone (Gibson, 2017)

• Incentive• Amount, timing, structure

• Introduction • Phrasing, voice (gender), modality (CATI or IVR for

IVR)• Modality

• IVR or CATI

• Refusals – usually happens before consent• Burkina Faso follow-up CATI & Hybrid IVR (women only):

8% (Greenleaf, 2019) • Burkina Faso RDD CATI (women only): 3.1% (Greenleaf,

2020 • Ghana RDD IVR: 7% (L’Engle, 2018)• Tanzania CATI: 1.6%; IVR 0.3% (Pariyo, 2019)

Participant Concerns in Burkina Faso RDD

Where are you?

I don’t want to expose my life via telephone

Why is it *me* that you’re calling?

Why do you want to know my age?

Presenter
Presentation Notes
Gibson DG, Pariyo GW, Wosu AC, et al. Evaluation of Mechanisms to Improve Performance of Mobile Phone Surveys in Low- and Middle-Income Countries: Research Protocol. JMIR Res Protoc. 2017;6:e81. Refusals Greenleaf, A.R., et al., Comparison of remote data collection modes to monitor family planning progress in Burkina Faso: representativeness, data quality, and cost P.M.a.A.P. Project, Editor. 2019, Bill & Melinda Gates Institute for Population and Reproductive Health, Johns Hopkins Bloomberg School of Public Health: Baltimore, MD. https://www.pma2020.org/sites/default/files/PMA2020_RDC_Phase2_report_06March2019-web.pdf Greenleaf AR, Gadiaga A, Guiella G, Turke S, Battle N, Ahmed S, Moreau C. Comparability of modern contraceptive use estimates between a face-to-face survey and a cellphone survey among women in Burkina Faso. [available May 13]: PLoS ONE 10.1371/journal.pone.0231819 L'Engle K, Sefa E, Adimazoya EA, et al. Survey research with a random digit dial national mobile phone sample in Ghana: Methods and sample quality. PLoS One. 2018;13:e0190902. Pariyo, G.W., et al., Does mobile phone survey method matter? Reliability of computer-assisted telephone interviews and interactive voice response non-communicable diseases risk factor surveys in low and middle income countries. PLoS One, 2019. 14(4): p. e0214450.
Page 24: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Language Fractionalization

• Order of language options matters: • IVR Pilot in Burkina Faso (Greenleaf, 2019). When

least spoken language (Fulfulde) was listed first, 3 times as many people chose that language

• Results from 2015 IVR study • Zimbabwe: ~40% of the population speaks Shona,

while 76% of the sample selected Shona• Ethiopia: 85% of the sample selected Amharic, but

primary language for only an estimated 29%

Source: African Languages Interpreters

Presenter
Presentation Notes
Greenleaf, A.R., et al., Comparison of remote data collection modes to monitor family planning progress in Burkina Faso: representativeness, data quality, and cost P.M.a.A.P. Project, Editor. 2019, Bill & Melinda Gates Institute for Population and Reproductive Health, Johns Hopkins Bloomberg School of Public Health: Baltimore, MD. https://www.pma2020.org/sites/default/files/PMA2020_RDC_Phase2_report_06March2019-web.pdf Leo B, Morello R, Mellon J, Peixoto T, Davenport S. Do Mobile Surveys Work in Poor Countries. CGD Working Paper 398. Washington, DC: Center for Global Development 2015.
Page 25: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Response Wrap-up

• Challenge of distinguishing coverage error from non-response error • Frame unknown so don’t know if not included in sample or not responding

• Approaches to comparing respondent & non-respondents• DHS, Census, administrative data, other • Early versus late respondents

• Burkina Faso follow-up: survey respondents different from non-respondents for both CATI & Hybrid IVR, with greater sample distortion for Hybrid IVR (Greenleaf, 2019)

• Response Rates• Response rates are low for cell phone surveys

• U.S.: 9% • Response rates are difficult to standardize because the number of non-working phone numbers

is hard to discern (Lau, 2018)• Mode: CATI higher than IVR, Sampling: FTF higher than RDD

Presenter
Presentation Notes
Greenleaf, A.R., et al., Comparison of remote data collection modes to monitor family planning progress in Burkina Faso: representativeness, data quality, and cost P.M.a.A.P. Project, Editor. 2019, Bill & Melinda Gates Institute for Population and Reproductive Health, Johns Hopkins Bloomberg School of Public Health: Baltimore, MD. https://www.pma2020.org/sites/default/files/PMA2020_RDC_Phase2_report_06March2019-web.pdf Lau, C. and N. di Tada, Identifying Non-Working Phone Numbers for Response Rate Calculations in Africa Survey Practice, 2018. 11(2): p. 2.
Page 26: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Non-Response: Conclusions

• A high refusal rate does not necessarily translate to biased estimates if the propensity to respond is not correlated with the outcome of interest

• Refusal should be randomly distributed within your sample for your estimate to be un-biased

• Try to collect or identify data about how your population interacts with their phone

• Ownership vs. access • Ask respondents for back-up phone numbers to

contact them • Frequency of phone use

Source: ICAP 2014

Presenter
Presentation Notes
Non-response error and bias: Johnson TP, Wislar JS. Response Rates and Nonresponse Errors in Surveys. Journal of American Medical Association. 2012;307:2.  Groves RM. Nonresponse Rates and Nonresponse Bias in Household Surveys. Public Opinion Quarterly. 2006;70:20. Non-response in US phone surveys: https://www.aapor.org/Education-Resources/Reports/Cell-Phone-Task-Force-Report/Coverage-and-Sampling-(1).aspx
Page 27: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Post-adjustment Error

When post-survey adjustments to account for representation errors do not improve the accuracy of

the survey estimate

2

6

8

10

Target Population

Frame

Sample

Respondents

Post-adjustmentSource: WWF

Page 28: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Post-adjustment Error in SSA Cell Phone Surveys

• Nigeria: RDD comparing IVR, SMS with DHS• Outcome: voting estimate• Weighting did not improve estimate of voting

• Burkina Faso: CATI RDD among women compared with concurrent in-person survey

• Weights: age, education, residence• Did not improve modern contraceptive estimates

FTF Full sample

FTF Phone owners

RDD Unweighted

RDDWeighted

% using modern contraception

26.0 (22.7 – 29.6)

31.7 (30.0 – 35.6)

40.2(38.2 – 42.2)

38.7 (36.7- 40.8)

Source: Charles Lau, 2019

Source: Abigail Greenleaf, 2020

Presenter
Presentation Notes
Lau, C., et al., In Search of the Optimal Mode for Mobile Phone Surveys in Developing Countries. A Comparison of IVR, SMS, and CATI in Nigeria. Survey Research Methods, 2019. 13: p. 305-318. Greenleaf AR, Gadiaga A, Guiella G, Turke S, Battle N, Ahmed S, Moreau C. Comparability of modern contraceptive use estimates between a face-to-face survey and a cellphone survey among women in Burkina Faso. [available May 13]: PLoS ONE 10.1371/journal.pone.0231819
Page 29: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Representation in cell phone surveys

Cell phone surveys exclude people without cell phones, but often draw inferences to the general population1. Phone owners are male, educated, urban and

young and new ownership growth is slowing2. Phone owners are different from non-owners to

point that weighting hasn’t improved estimates. Due to social connections? Female empowerment? Other?

3. RDD sampling is challenging 4. Data on non-respondents is hard to come by but

refusals can be reduced by following good cell phone survey practice

Source: WWF

Page 30: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Practice Safe Cell Phone Surveys

• Be explicit about who and was not included in your survey! • Try to assess sources of representation error • Stratify results by gender

• Use AAPOR [call] Disposition Codes (AAPOR, 2016)

• CATI, when possible, is preferred to increase representation

• FTF follow-up preferred over RDD

• New frontiers while acknowledging current limitations • Mixed modes • WhatsApp

Presenter
Presentation Notes
AAPOR Disposition Codes https://www.aapor.org/AAPOR_Main/media/publications/Standard-Definitions20169theditionfinal.pdf
Page 31: Phone Surveys for Health: Methods and Opportunities...SMS Random Digital Dial in Ghana, Kenya, Uganda, and Nigeria: response rate < 1% and young men overrepresented (Lau, 2019) •

Recent Resources

• Survey CTOhttps://www.surveycto.com/blog/surveycto-cati/• JPAL https://www.povertyactionlab.org/blog/3-20-20/best-practices-conducting-phone-surveys• World Bankhttps://blogs.worldbank.org/impactevaluations/mobile-phone-surveys-understanding-covid-19-impacts-part-i-sampling-and-mode