Cell Phone Surveys among Women of Reproductive Age in Burkina Faso: Identifying Sources of Error
Greenleaf, Abigail Regan
MetadataShow full item record
The rapid growth of cell phone ownership in low and middle income countries (LMIC) provides an opportunity to frequently collect data at low cost by calling respondents remotely. However, little is known about the validity of reproductive health estimates derived from cell phone surveys in LMIC – either collected through computer assisted telephone interviews (CATI) or interactive voice response (IVR). This dissertation identifies sources of frame and non-response error using four datasets collected between 2016 and 2018 among women of reproductive age in Burkina Faso. In Aim 1, we used data from the Performance Monitoring and Accountability 2020 (PMA2020) Survey Round 4, a nationally representative population-based FTF survey of women of reproductive age conducted in 2016. For Aim 2, we conducted a follow-up cell phone survey of cell phone owners identified in Round 4. These women were randomized to receive either a CATI or IVR survey that was introduced with a human operator. Finally, in Aim 3 we used data from a national probability FTF survey (PMA2020 Round 5) and a CATI phone survey selected through RDD. We found 47% of women in Burkina Faso owned a cell phone based on PMA2020 Round 4 survey data, with greater ownership among women with secondary education or higher (Odds Ratio (OR) = 4.3, 95% confidence interval 2.9 – 6.2), women who resided in urban areas, women from wealthier households (highest wealth quintile), and women who were over the age of 19. When examining non-response patterns in the second aim, we found the survey response rate was higher among women randomized to CATI (50%) versus Hybrid IVR (19%). Cell phone owners who consented to our follow-up cell phone survey were more likely to reside in an urban area, were more educated and were more likely to be over the age of 20 years old compared to women who did not consent to participate (were not contacted or refused participation). Finally, the results of Aim 3 showed that the estimate of modern contraceptive use generated by the CATI survey with RDD sampling was not equivalent but higher to the FTF PMA2020 Round 5 (R5) survey estimate (RDD: 40%, FTF: 26%), even after applying post-stratification weights (RDD weighted modern contraceptive prevalence estimate: 39%). The odds of modern contraceptive use remained significantly higher among RDD respondents versus FTF respondents, even after adjustments for additional covariates. This dissertation identified sources of error for both follow-up and RDD cell phone surveys in Burkina Faso suggesting that over-estimation of modern contraceptive use in phone surveys stems from frame bias rather than non-response bias and potentially measurement bias.