SMS Campaign to Increase COVID-19 Vaccine Take-up in Tanzania

We propose sending SMS messages of varying content in a randomized placebo-controlled experiment. Our experiment will be based on two (2) treatment arms which will receive two (2) different text message content and one placebo group which will receive SMS content on some other non-public health issue unrelated to COVID-19.

Treatment 1 (T1): We call this treatment the EMPATHY treatment, where mobile phone subscribers in this arm will receive an SMS message emphasizing one’s increased potential contribution to harming others through exposing them to COVID-19, in the absence of vaccination.

Treatment 2 (T2): We call this treatment the SELF - INTEREST treatment, where mobile phone subscribers in this arm will receive an SMS message on adverse long-term effects of COVID-19 infection.

Placebo group (C), where mobile phone subscribers in this arm will receive an SMS message on some other non-public health issue unrelated to COVID-19.

2022-11-30

2023-04-30

a. Self-reported vaccination status. This will be assessed by a follow up survey question “Have you received the full dose of any COVID-19 vaccine?”

b. Household members’ vaccination status. This will be assessed by a follow up survey question “How many of your household members have received the full dose of any COVID-19 vaccine?”

c. Objectively verified vaccination status: “Please provide either your NIDA (national ID), passport number, voter ID, or other ID number or your full name that you used when you registered and received the COVID-19 vaccine.”

d. Self-reported intent to vaccinate. This will be assessed by a follow up survey question “Do you intend, in the next 12 months, to receive the full dose of any COVID-19 vaccine?”
e. Household members’ intent to vaccinate. This will be assessed by a follow up survey question “How many of your household members have expressed intentions, in the next 12 months, to receive the full dose of any COVID-19 vaccine?”
f. Subjective trust on COVID-19 vaccines: From a scale of 1 to 10 where 10 is complete trust and 1 is no trust, how much do you trust the COVID-19 vaccines?
g. Self-reported compliance on wearing masks. This will be assessed by a follow up survey question “In the last 14 days, how often did you wear a mask whenever you were in public spaces with other people?


Heterogeneity Analysis.

According to Africa CDC vaccine hesitancy survey of March 2021, those who are more skeptical towards COVID-19 vaccines tend to be female, young people, those who are unemployed and those living in cities and therefore, with respect to our treatments we expect female, young individuals, unemployed and those living in cities to react less to our treatments. Moreover, with respect to our EMPATHY and SELF-INTEREST treatments, we expect women to be more empathetic and have differential discount factors of time relative to men (Falk, et al., 2018). Similarly, because younger people have more time relative to older people, we expect their preferences for time but also others to differ. Finally, individuals who are unemployed and/or rural residents will have relative lower incomes than the employed/urban dwellers and we also know that incomes matter for risk preferences. Hence, we anticipate sex, age, employment status and urban residency will display heterogeneous effects. To capture this, we will run the analysis by regressing outcomes on an interaction between the treatment dummy and the characteristic.

We also going to do a “change-from-baseline” analysis for no more than 4,766 respondents for the 6 outcomes we currently have some data on for each of the six (6) outcomes along with covariates included as well.

We propose to use a randomized placebo-controlled experiment. There will be two treatments (T2 and T2) and placebo group. Randomization will ensure the balance of observable and non-observable characteristics between control and experiment groups and therefore any difference between them in our outcomes of interest will be due to our proposed intervention.

Randomization will be done in office by a computer

Mobile phone subscribers

No

None

Our targeted sample is 12,000 mobile phone subscribers. We plan to send the intervention to 17,209 nationally representative sample. If we do not reach this target, we will supplement our sample with an additional booster sample of 8,564 respondents. However, the booster sample may not be as representative of the national population as our initial sample of 17,209 nationally representative respondents.

Our analysis unit is individual phone subscribers from the IPSOS database, with a target sample size of 12,000 for our study which is distributed as 4,500; 4,500; and 3,000 for T1; T2; and Placebo, respectively. To achieve this sample size, we plan to recruit 17,209 respondents, who will be nationally representative and distributed in the same ratio as our target sample. Specifically, we aim to recruit 6,454, 6,453, and 4,302 respondents for T1, T2, and Placebo, respectively.

If we did not reach the targeted sample, we will supplement our sample with additional booster sample of 8565 respondents. This will be distributed in the same ratio as above across our treatments arms.

Introduction Throughout the following power calculations of the pairwise comparisons between T1 and Placebo; T2 and Placebo; and T1 vs T2 we begin by setting the sample sizes as to 4,500; 4,500; and 3,000 for T1; T2; and Placebo, respectively. We also begin by setting alpha = 0.05 and power = 0.8. In accounting for multiple comparison bias, we divide alpha = 0.05/7 = 0.00714286 for our seven (7) outcomes. Outcome 1: Actual Vaccination Rates We begin by first outlining the assumptions on baseline values of two outcome variables. The first is on actual vaccination rates in Tanzania, which we set at 7%. We set it at 7% because Our World In Data’s latest data shows that as a percentage of the adult population, only about 6.6% of adult Tanzanians are vaccinated as of this writing. Power calculations for a two-sample proportions chi2 test shows that the MDE for the actual vaccination rate outcome given an assumed 7% rate of vaccinations among adult Tanzanians is 1.79% points (T1/T2 vs Placebo). This MDE increases to 2.29% points when accounting for multiple comparisons across our 7 outcomes (alpha = 0.05/7 = 0.00714286). Comparing T1 to T2 assuming a baseline of 8.06% with alpha = 0.05, the MDE is 1.68 % points, which increases to 2.14 % points. When accounting for multiple comparisons (alpha = 0.05/7 = 0.00714286). Outcome 2: Vaccine Hesitancy/Confidence Next, we continue onto the vaccine hesitancy/confidence outcome. We set the baseline percentage of vaccine hesitancy/confidence for Tanzania at 41%. This is informed by Africa’s CDC February 2021 (Figure 2.3.2 on page 10 [pdf page 11]) on survey results from across 15 African countries, excluding Tanzania, which has the Democratic Republic of Congo (DRC) as hosting the most vaccine hesitant population where only about 59% of respondents expressed willingness to accept a new COVID19 vaccine. Please note that surveying likely occurred before any public news of actual successfully-tested COVID19 vaccines being made available to the wider public. Nevertheless, we assume a baseline level of 41% as our baseline on vaccine hesitancy/confidence, which is likely a lower limit to actual COVID19 vaccine hesitancy/confidence in Tanzania were it to be measured today. Power calculations for a two-sample proportions chi2 test shows that the MDE for the COVID19 vaccine hesitancy outcome given an assumed 41% vaccine hesitancy/confidence among adult Tanzanians, is 3.27% points (T1/T2 vs Placebo). This MDE increases to 4.12% points when accounting for multiple comparisons across our 7 outcomes (alpha = 0.05/7 = 0.00714286). For T1 vs T2, with alpha = 0.05 with a baseline of 44% the MDE is 2.94 % points, which after accounting for multiple comparisons (alpha = 0.5/7 = 0.00714286) increases to 3.71% points.