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

Introduction Increased mobile phone subscribership in low- and middle-income countries (LMICs) provides novel opportunities to track population health. The objective of this study was to examine reliability of data in comparing participant responses collected using two mobile phone survey (MPS) delivery modalities, computer assisted telephone interviews (CATI) and interactive voice response (IVR) in Bangladesh (BGD) and Tanzania (TZA). Methods Using a cross-over design, we used random digit dialing (RDD) to call randomly generated mobile phone numbers and recruit survey participants to receive either a CATI or IVR survey on non-communicable disease (NCD) risk factors, followed 7 days later by the survey mode not received during first contact; either IVR or CATI. Respondents who received the first survey were designated as first contact (FC) and those who consented to being called a second time and subsequently answered the call were designated as follow-up (FU). We used the same questionnaire for both contacts, with response options modified to suit the delivery mode. Reliability of responses was analyzed using the Cohen’s kappa statistic for percent agreement between two modes. Results Self-reported data on demographic characteristics and NCD behavioral risk factors were collected from 482 (CATI-FC) and 653 (IVR-FC) age-eligible and consenting respondents in BGD, and from 387 (CATI-FC) and 674 (IVR-FC) respondents in TZA respectively. Survey follow-up rates were 30.7% (n = 482) for IVR-FU and 53.8% (n = 653) for CATI-FU in BGD; and 42.4% (n = 387) for IVR-FU and 49.9% (n = 674) for CATI-FU in TZA respectively. Overall, there was high consistency between delivery modalities for alcohol consumption in the past 30 days in both countries (kappa = 0.64 for CATI!IVR (BGD), kappa = 0.54 for IVR!CATI (BGD); kappa = 0.66 for CATI!IVR (TZA), kappa = 0.76 for IVR!CATI (TZA)), and current smoking (kappa = 0.68 for CATI!IVR (BGD), kappa = 0.69 for IVR!CATI (BGD); kappa = 0.39 for CATI!IVR (TZA), kappa = 0.50 for IVR!CATI (TZA)). There was moderate to substantial consistency in both countries for history of checking for hypertension and diabetes with kappa statistics ranging from 0.43 to 0.67. There was generally lower consistency in both countries for physical activity (vigorous and moderate) with kappa statistics ranging from 0.10 to 0.41, weekly fruit and vegetable with kappa ranging from 0.08 to 0.45, consumption of foods high in salt and efforts to limit salt with kappa generally below 0.3. Conclusions The study found that when respondents are re-interviewed, the reliability of answers to most demographic and NCD variables is similar whether starting with CATI or IVR. The study underscores the need for caution when selecting questions for mobile phone surveys. Careful design can help ensure clarity of questions to minimize cognitive burden for respondents, many of whom may not have prior experience in taking automated surveys. Further research should explore possible differences and determinants of survey reliability between delivery modes and ideally compare both IVR and CATI surveys to in-person face-to-face interviews. In addition, research is needed to better understand factors that influence survey cooperation, completion, refusal and attrition rates across populations and contexts.