Using Mobile Phones Phone ownership is highly non-random and is rarely observed among the very poor. Therefore, to Collect Panel Data we sampled from a list of cotton farmers in the official village registry. While none of our sample By Brian Dillon villages was on the electric grid, some source of power was available everywhere. We signed a In 2009 and 2010, a small research team contract with a “charging station” in each village, conducted a study in rural areas of western paying for survey participants to receive one free Tanzania entitled Research on Expectations in charge during the two days prior to each Agricultural Production (REAP). The primary aim scheduled call. We also transferred credit to each of REAP was to gather high frequency phone after each interview, both to prevent the quantitative data on agricultural inputs and cancellation of project SIM cards, and to farmers’ subjective expectations for uncertain compensate respondents for their time outcomes such as weather, prices, and crop yields. Instead of embedding enumerators in Average interview time across the 14 rounds of survey villages for an extended period of time, the survey was 27 minutes. Questionnaires the REAP team used mobile phones to conduct included pre-coded, quantitative questions on surveys on a high frequency basis. subjective expectations, labor, sales and purchases, agricultural inputs, demographic The REAP Survey changes, weather, and numerous other topics. The REAP study involved a baseline in-person Some data were gathered every 3-6 weeks, survey in the summer of 2009, 14 phone survey others less frequently. rounds over the next 10 months, and a follow-up in-person survey in the summer of 2010. During We reached an average of eight respondents on scoping visits, the team confirmed that a network the scheduled day. A host of small obstacles signal from one of the major Tanzanian carriers prevented interviews from taking place as was available throughout the study area. The scheduled, such as illness, family events, network sample consisted of 195 randomly chosen cotton outages, and phone problems. Despite these farmers in 15 villages in the Mwanza and challenges, virtually all respondents who were Shinyanga regions. We provided a phone to all not interviewed on schedule were interviewed in respondents, regardless of whether they already the ensuing few days. Community members owned one. Phones were distributed in the sometimes assisted us by contacting missing households, after completion of the baseline respondents and arranging interviews. interview. Respondents also received laminated sheets with the village-specific call schedule and Infrastructure Matters contact numbers for the research team. With 15 The limitations of the mobile network may villages receiving calls on a Monday-Friday present the most definitive challenge to the schedule, each village had a calling day once feasibility of phone-based data collection. every three weeks. Inconsistent mobile network coverage introduces bias at the village selection stage. This is In rural areas of developing countries it is important for many research questions, since unlikely that researchers could justifiably sample network access is likely to be correlated with from an available list of mobile phone users. distance from major towns, road quality, water Living Standards Measurement Study Brief Series www.worldbank.org/lsms-isa supply, average wealth, and other relevant Data Quality characteristics. Researchers who find that In a phone survey, multiple languages can easily network shortcomings preclude sampling from be accommodated by staffing the project with at the original population of interest may need to least one enumerator who speaks each local reconsider the idea of conducting a phone language. Likewise, if enumerators directly enter survey. We were fortunate that while none of the data into a computer during the interview, REAP villages were connected to the electrical supervisors and researchers can check grid, each village possessed some source of questionnaires in real time. This allows for both power. However, a pre-existing source of the detection of problems as well as the electricity is not requisite for participation in a identification of new questions of interest, while phone survey; if necessary, researchers could the survey is still in the field. Finally, if the establish charging stations specifically to support respondent is not able to talk when called, a the research or distribute small solar chargers phone survey can be rescheduled more easily (which cost less than $10) to each participating than a face-to-face survey, reducing the necessity household. of surveying of anxious, hurried respondents. Reduced Costs Some issues such as gender discrimination, Table 1 shows some of the basic elements of the domestic violence, or corruption may be difficult REAP budget. Once the survey was operational, to study via phone, because the interviewer the marginal cost of gathering data by phone cannot ensure confidentiality. On the contrary, was only a small fraction of what it would have the nature of a phone interview may enhance been to conduct every round face-to-face. That confidentiality, if only the respondent is able to said, the cost savings from a phone survey are hear the interviewer, and questions require a only substantial if the project calls for the “yes”, “no”, or otherwise innocuous response. collection of panel data over relatively short time horizons. Furthermore, some field costs cannot Replacing Phones be avoided, due to the necessity of conducting Over 10 months, about 8% of sample households baseline interviews and distributing phones. reported a lost or broken phone. Replacement of survey materials introduced an element of moral REAP Survey Costs hazard, which required careful management. To Cost deter sales of the project phones, we told Budget item (USD) respondents at baseline that we could exchange Price per phone $20 malfunctioning phones and batteries for new ones, but we could not replace items that were SIM card $0.38 lost. In a few cases we violated this policy, usually Credit transfer per survey round $0.76 for someone living in an isolated area. Battery charge per respondent per round Nonetheless, it was clear that some of the “lost” $0.15 phones were in fact sold. From a research Average cost of phone interview $6.98 perspective, this was not problematic as long as Average cost of face-to-face interview $97 respondents continued to participate in the survey by using a different phone. Avoiding Attrition This brief is based on: Dillon, Brian (2010). Using Attrition is clearly affected by the frequency of Mobile Phones to Collect Panel Data in interviewing. The optimal lag between calls Developing Countries. LSMS-ISA Working Paper, depends on the research content, the length of Washington, D.C.: The World Bank. the project and the length of each interview, as The full paper is available for download at well as any constraints to the budget. REAP www.worldbank.org/lsms-isa. enjoyed surprisingly low rates of attrition: an average of 191.2 out of 195 respondents were For more information, please visit: interviewed in each round. It helped that www.worldbank.org/lsms-isa participants were very willing to help find missing Or contact: respondents. Low rates of attrition and non- response were also due to the direct benefits of Brian Dillon, Cornell University participation, as many respondents looked bmd28@cornell.edu forward to the free battery charge and the credit transfer that they received for each interview. Living Standards Measurement Study Brief Series www.worldbank.org/lsms-isa