Messaging to Prevent the Spread of Covid-19: An Experiment in Rural Uganda

Respondents are randomly assigned to receive one of the following texts:

T0: Expert

To reduce the spread of covid-19, public health experts say that Ugandans should wear a face mask and avoid large gatherings, such as marriage ceremonies, wedding ceremonies, vigils, funerals and prayers in churches and mosques.

T1: Self-interest

Covid-19 can be deadly. Even people who recover from the virus may suffer long-term complications. To reduce the risk that you get sick with covid-19, you should wear a face mask and avoid large gatherings such as marriage ceremonies, vigils, funerals and prayers in churches and mosques.

T2: Punishment

Due to covid-19, wearing a face mask is mandatory and attending large gatherings is illegal. If you attend events such as marriage ceremonies, wedding ceremonies, vigils, funerals or prayers in churches and mosques with more than 10 people, you may be imprisoned for up to 2 months.

T3: Religion

In response to covid-19, many religious leaders including the Archbishop of the Church of Uganda and the Mufti of Uganda are encouraging Ugandans to wear face masks and avoid large gatherings such as wedding ceremonies, vigils, funerals and prayers in churches and mosques. Religious leaders urge you to organize family prayers in your own dwelling to connect with God, since God is everywhere: in our homes and hearts.

The texts are translated into local languages and sent three times over the course of three days. An endline survey is administered by phone beginning four days after the last text is sent. The rollout of the endline survey was randomized in order to estimate treatment effect decay as a function of the length of time between receiving treatment and being surveyed.

2020-07-22

2020-07-24

Our main outcome of interest is citizens’ compliance with public health precautions meant to slow the spread of covid-19. In the endline survey, we ask respondents a series of questions about their behavior related to covid-19. Specifically, we ask how many times respondents have done the following in the last 7 days:

1. Stayed at home all day without leaving the household compound
2. Attended a church or mosque service, or gathered with people from outside the household to pray
3. Gone to a restaurant, kibanda, bar, or café
4. Attended a community meeting
5. Attended a social gathering, like a wedding, funeral, or party

We expect an increase in item 1 when messages effectively persuade individuals to practice stronger forms of social distancing by staying at home. Effective public health messages should cause reductions in responses to items 2-5.
Additionally, we ask respondents how often they practiced physical distancing of 2 meters when in public, washed their hands, and wore a mask or face covering when in public in the last 7 days. We expect that effective forms of public health messaging should increase these behaviors.

We will use these questions to create an index of compliance with public health precautions.

Since admitting to non-compliance with public health precautions could potentially be sensitive for respondents, we also include two list experiments designed to address social desirability bias. The first list experiment includes a sensitive item for whether the respondent attended a public religious gathering in the last 7 days. A separate list experiment includes a sensitive item for whether the respondent has gone out in public without a face mask. Due to concerns about power, these list experiments will serve only as secondary outcomes of interest.

We will also test effects on the following secondary outcomes:

- Knowledge of ways to prevent the spread of covid-19
- Fear of contracting covid-19
- Fear of punishment for violating public health precautions
- Support for public health precautions, specifically the government-imposed national lockdown and the government-imposed mandate that all citizens wear masks when leaving the home

The sample consists of baseline and endline respondents from an existing field experiment on community policing in Uganda. Respondents are either randomly sampled citizens or Local Council leaders (LC1). Details on our sampling frame are in “Community Policing in Uganda” (Blair, Grossman, and Wilke 2020, EGAP Registration ID 20200122AA). To maximize statistical power, we include in our sample respondents who were surveyed at baseline but could not be reached at endline, as well as those who were surveyed at endline as replacements for baseline respondents who could not be reached. In some cases, we do not have a phone number for the respondent, but we do have a phone number for the respondent’s spouse or child. In these cases, we replace the respondent with their spouse or child (in that order). In other cases, we do not have a phone number for the respondent, nor for a spouse or child. We drop these respondents from the sample.

We block randomize by village and phone ownership. Respondents are assigned to one of the four treatment conditions above with equal probability.

We estimate treatment effects using OLS with block fixed effects. In secondary analyses, we also include controls for a select number of endline survey items. While these controls are measured post-treatment, they are either fixed or extremely unlikely to change as a result of treatment in the time between when the last text is sent and when the survey is administered. The controls include party identification, employment status, household access to food, and religiosity.

Finally, we also have pre-treatment survey data on basic demographics and perceptions of community policing. Pre-treatment data will be missing for new respondents that are serving as replacements for another household member. Should this type of attrition be sufficiently low, we will include basic demographic controls (like age, gender, etc.).

Randomization done in office by a computer

Individual

No

N/A

We have phone numbers for 3,246 citizens and 287 LC1s. Some of these respondents will be unreachable.

- Expert treatment: 811 citizens and 72 LC1s
- Self-interest treatment: 812 citizens and 72 LC1s
- Punishment treatment: 811 citizens and 72 LC1s
- Religion treatment: 812 citizens and 71 LC1s