COVID-19, fake news and religious tensions: experimental evidence from India

We will address three research questions (RQ):
RQ1: How effective are doctors’ messages to counter misinformation about ways to prevent COVID-19?
RQ2: How does religion identity moderate the processing of new information?
RQ3: Can higher financial rewards lead to higher uptake of messages?

Our intervention makes use of technologies of information through mobile phones (i.e. Whastapp chatbot and voice messages).

A main intervention with variants will be evaluated. We will send messages from different doctors working in renowned health centres in the study area, debunking fake news and reminding the audience about the proven ways to protect against COVID-19. This is called the “Doctor” intervention.

We will send these messages in the form of short videos (approximately 2 minutes) through a Whatsapp chatbot . Given that 50% of the slum households do not have a smartphone with Whastapp, we will send the audio of the videos to study households without Whastapp via voice messages. Each video (and audio) starts with a short clip of a citizen from Uttar Pradesh introducing the doctors’ messages.

The variants that we introduce are the following. First, instead of doctors’ messages debunking fake news about COVID-19, some households will receive messages debunking fake news about Bollywood stars. This is what we call the “Control” intervention, which will allow us to disentangle the effects of our intervention from receiving a message through mobile phone technologies.

Second, we will vary the identity of the messenger by changing the religion of the citizen in the introductory clip. Half of the households will receive a “Hindu citizen” intervention and the other half a “Muslim citizen” intervention. We vary the identity by changing the looks, name and surname, body language and phrases used by the citizen. We hold constant any other factors affecting the identity of the messenger by using the same citizen.

Third, in order to incentivise households to watch the video and listen to the audio, we will give participants the chance to enter a lottery. Half of the households will enter a lottery for Rs. 5,000 and others for half. We call the former the “High incentive” intervention, and the latter the “Low incentive”. This will allow us to analyse the effect of a higher expected payoff on the uptake of messages.

2020-09-26

2020-10-26

RQ1: Knowledge about COVID-19 prevention
RQ2: Religion bias
RQ3: Exposure to intervention

1. RQ1: Knowledge about COVID-19 prevention: measured as the extent to which the participant agrees with statements on different (confirmed and not confirmed) ways to prevent COVID-19.
2. RQ2: Religion bias: elicited by randomly varying the names of citizens that agree with different statements (not confirmed ways) about how to prevent COVID-19 and asking participants the extent to which they also agree with the statements. We use names that clearly convey the gender and religion of the citizen. For this, we use the most common names in our household Census for each identity.
3. RQ3: Exposure to intervention: number of participants that watched the video and listened to the audio; extent to which participants recall receiving a video through Whatsapp or voice message related to COVID-19.

1. Acquiring and spreading information
2. Risk perception
3. Trust
4. Attitudes towards vaccination
5. Complying with policy guidelines

1. Acquiring and spreading information: the extent of discussion about COVID-19 with other people; time spent in acquiring COVID-19 related information; knowledge about COVID-19 symptoms.
2. Risk perception: the extent to which respondents believe a member of the household can get COVID-19; how anxious they feel about the pandemic.
3. Trust: the extent to which participants trust doctors and people from other religion, in comparison to people in their State in general.
4. Attitudes towards vaccination: the extent to which the participant is willing to vaccinate (having to pay or not) when the vaccine for COVID-19 becomes available; attitude towards people that don’t want to vaccinate.
5. Complying with policy guidelines: behaviour related to better hygiene and physical distance

We address the research questions using a field experiment through mobiles phones in 142 slums in the cities of Lucknow and Kanpur, Uttar Pradesh. We have a 2 x 2 x 2 cross-randomized design, resulting in 8 treatment arms.

We address the research questions using a field experiment through mobiles phones in 142 slums in the cities of Lucknow and Kanpur, Uttar Pradesh.

We randomly allocate households (one or more mobile phones in each household) to receive one of the variations of the following three treatments:

T1: Doctor messages vs. Control messages

T2: Hindu citizen vs. Muslim citizen

T3: High incentive vs. Low incentive

We end up having a 2 x 2 x 2 cross-randomized design, resulting in 8 treatment arms.

To allocate households to the treatment arms, we stratify the sample by religion (Hindu or other) and by city of study (Lucknow or Kanpur).

The distribution of households across treatment arms is as follows:
- 498 households allocated to the “Doctor-High Incentive-Hindu” treatment arms
- 505 households allocated to the “Doctor-Low Incentive-Hindu” treatment arms
- 507 households allocated to the “Doctor-High Incentive-Muslim” treatment arms
- 492 households allocated to the “Doctor-Low Incentive-Muslim” treatment arms
- 505 households allocated to the “Control-High Incentive-Hindu” treatment arms
- 479 households allocated to the “Control-Low Incentive-Hindu” treatment arms
- 503 households allocated to the “Control-High Incentive-Muslim” treatment arms
- 502 households allocated to the “Control-Low Incentive-Muslim” treatment arm

The statistical software Stata, and specifically the random number generator, will be used to apply this procedure.

Randomisation into the experimental arms is conducted at the household level given that the intervention is directed one-to-one through mobile phones. Randomising at the household level allows us to take advantage of greater variation in response to the intervention within slums.

No

We will not conduct a clustered RCT, but our households are part of 142 slums.

3,991 households, with a mean of 28 households per slum.

~500 units by treatment arm.

Within each slum, we sample up to 60 households, aiming for an average of 30 households per slum. Our sampling procedure is informed by the power calculation used in the initial study registered in the AEA Registry Number AEARCTR-0003087.

Pre-Analysis Plan