![Sample our Global Development journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5929/TOC-Subject-Sample-2021-Global-Development.jpg"})
Abstract
Abstract As mobile phone ownership rates have risen in Africa, there is increased interest in using mobile telephony as a data collection platform. This paper draws on two pilot projects that use mobile phone interviews for data collection in Tanzania and South Sudan. In both cases, high frequency panel data have been collected on a wide range of topics in a manner that is cost effective, flexible and rapid. Attrition has been problematic in both surveys, but can be explained by the resource and organisational constraints that both surveys faced. We analyse the drivers of attrition to generate ideas for how to improve performance in future mobile phone surveys.
Résumé Le taux de propriété du portable ayant augmenté en Afrique, on manifeste de plus en plus d'intérêt dans l'usage de cette technologie comme moyen de recueillir des données. Cet article présente deux projets pilotes ayant eu recours au portable pour des fins de cueillette de données par entretien, en Tanzanie et au sud de Soudan. Dans les deux cas, des données sur une grande variété de sujets ont été fréquemment collectées auprès d'un panel, de manière flexible et rapide, avec un bon rapport coût/efficacité. Le taux d'abandon a toutefois posé problème dans les deux enquêtes, mais il peut être attribué à des contraintes organisationnelles et au manque de ressources. Les facteurs qui peuvent expliquer ce haut taux d'abandon sont examinés afin de trouver des solutions pour améliorer la performance des futures enquêtes utilisant le portable.
Acknowledgements
This paper has benefited from support from the PSIA Trust Fund (TF099681), the BNPP Trust Fund, which financed some of the Afrobarometer questions on ownership and uses of mobile phones, and the Demand for Good Governance group.
Notes
1. The lessons learned from these pilots are currently being incorporated into mobile phone panel surveys that are planned in a number of African countries.
3. These figures are from the 2004/2005 DHS, the Citation2007/2008 Tanzania HIV/AIDS and Malaria Indicator Survey; the Citation2010 CitationDHS, and the 2011/2012 HIV/AIDS and Malaria Indicator Survey.
4. A total of 61 per cent send or receive a text message at least once a day and a remarkable 23 per cent send or receive money or pay a bill via mobile phone at least once a day.
5. Blumberg and Luke (Citation2008). For an example of a reputable survey firm applying these approaches see: http://www.ipsos-na.com/download/pr.aspx?id=11397. Note also that the accuracy of landline only polling may be degrading over time, see http://fivethirtyeight.blogs.nytimes.com/2012/11/10/which-polls-fared-best-and-worst-in-the-2012-presidential-race/.
6. All references to dollars in this paper are to US dollars.
7. Information about this project can be found at http://siteresources.worldbank.org/SOCIALPROTECTION/Resources/280558-1138289492561/2158434-1332878435565/RSR-L2L-PPT_19March2012.pdf
8. Since this approach aims to create representative samples of the adult population, it necessitates the random identification of a respondent within the household.
9. Smith et al. (Citation2011) provide an overview of different ways to gather data using mobile phones.
10. Surprisingly, response rates were slightly lower for households that received more compensation.
11. Data and reports from the Tanzania pilot survey are available online and can be downloaded from: www.listeningtodar.org/
12. Remarkably in both Sudan and Tanzania the amount of the reward did not have a discernible impact on response rates (see for evidence from Tanzania).
13. This raises another issue for future research: whether it is possible to track changes in consumption by using poverty mapping techniques (Elbers, Lanjouw, and Lanjouw Citation2003) with a set of correlates that is more sensitive to changes in consumption levels than assets which is currently used in poverty mapping.
15. An example of such an experiment involved a partnership between the Dar es Salaam mobile phone survey, the NGO Helen Keller International (HKI), and the Tanzania Ministry of Health, which tried to maximise the coverage of Vitamin A Supplementation (VAS) days, which are held bi-annually in Dar es Salaam in July 2013. Half of the families in the survey were given an encouragement message about the importance of Vitamin A supplementation and the upcoming event, and the other half a placebo message encouraging other unrelated healthy behaviors.
16. Two of the authors worked at Twaweza at the time.
18. Because of its limitations SMS was not considered.
19. Reports and data produced, including the baseline data, can be obtained from: www.listeningtodar.org
20. Some respondents could not be reached either because their numbers had been captured incorrectly, or because they never seemed to have their phones on.
23. Alderman et al. (Citation2001) suggest that even where attrition is non-random, key parameter estimates are often not affected, using examples from Bolivia, Kenya and South Africa. Fitzgerald, Gottschalk and Moffitt (Citation1998) draw similar conclusions from the US-based Panel Survey on Income Dynamics, as does Falaris (Citation2003) with respect to surveys in Peru, Côte d'Ivoire and Vietnam (cited in Thomas et al. Citation2010).
24. Beegle, De Weerdt, and Dercon (Citation2011) suggest that attrition in developing country settings is likely to be related to unobservable traits, in part because attrition is often linked to migration.
25. This is not dependent on functional form. For example, the results are not substantially different when the dependent variable is a binary variable which takes a value of one when the respondent has participated in more than half of all survey rounds, or when probit is used instead of the OLS linear probability model.
26. In this regression, round 1 is used to determine persistence in non-response. To avoid autocorrelation, this round is omitted from the sum of rounds in which the household participated.
27. Discrepancies are present both with and without reweighting of results to reflect attrition.