Text message surveys enable cost-efficient data collection and can be applied with hard-to-reach populations while other survey modes suffer from undercoverage in regions with low landline and/or internet penetration. Little is known about how to best administer surveys in this mode. We experimentally compared two different designs of automated text message surveys in terms of response rate, nonresponse bias, and participation in a follow-up survey in Egypt. In the single-sitting design, respondents automatically received a text message with a new question once they replied to a question. In the modular design, respondents received a new question each day, regardless of whether they had responded to the previous question. We invited 1,081 Egyptian parents of kindergarten children who owned a mobile phone to participate in a text message survey with eight questions on the nutrition behavior of their children. We found that, compared to the single-sitting design, the modular design yielded a higher number of answered questions but had fewer fully completed questionnaires. We found no nonresponse bias in either group and no difference in the probability of responding to a follow-up survey. Our results will help researchers make design decisions about how to implement text message surveys.
