Persistent prejudice: De-biasing and the demand for biased news

Last registered on April 02, 2024

View Trial History

Pre-Trial

Trial Information

General Information

Title

Persistent prejudice: De-biasing and the demand for biased news

RCT ID

AEARCTR-0005468

Initial registration date

February 19, 2020

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published

February 19, 2020, 3:04 PM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated

April 02, 2024, 8:00 PM EDT

Last updated is the most recent time when changes to the trial's registration were published.

Locations

Country

India

Region

Primary Investigator

Name

Christopher Blattman

Affiliation

University of Chicago

Contact Primary Investigator

Other Primary Investigator(s)

PI Name

Chinmaya Kumar

PI Affiliation

University of Chicago

Contact Investigator

PI Name

Leonardo Bursztyn

PI Affiliation

University of Chicago

Contact Investigator

PI Name

Christopher Blattman

PI Affiliation

University of Chicago

Contact Investigator

Additional Trial Information

Status

Completed

Start date

2020-02-22

End date

2022-12-31

Keywords

Other

Additional Keywords

Motivated Beliefs, Media Bias, Selective exposure, Religious Polarization, Fake News, Experiments, News Consumption, Information, India

JEL code(s)

D91, D83, C91, N35

Secondary IDs

Prior work

This trial does not extend or rely on any prior RCTs.

Abstract

Identity groups often hold incorrect and biased beliefs about competing groups. Examples include Democrats and Republicans in the US, or Israelis and Arabs in the Middle East. In India, the setting for this study, Hindu nationalists commonly believe that Muslims are untrustworthy, or that the Muslim population is growing so fast that their population will overtake Hindus. These beliefs may be persistent and difficult to correct. Why is that? One reason, we hypothesize, is that people exposed to information counter to their group identity may work to rebias themselves by increasing their selective exposure--their consumption of biased news and information. We design an experiment in which we randomly provide Hindu respondents with information to correct a biased belief about Muslims, using informational videos. We first confirm that the videos shift their beliefs during the experiment. We then examine their demand for new information—having been de-biased, are they more likely to seek information from a biased source?

External Link(s)

Registration Citation

Citation

Blattman, Christopher et al. 2024. "Persistent prejudice: De-biasing and the demand for biased news." AEA RCT Registry. April 02. https://doi.org/10.1257/rct.5468-4.1

Sponsors & Partners

Experimental Details

Interventions

Intervention(s)

Intervention Start Date

2020-06-30

Intervention End Date

2021-02-28

Primary Outcomes

Primary Outcomes (end points)

Demand for biased vs. unbiased news for a long-term news subscription service

Primary Outcomes (explanation)

Demand for biased news in a long-term news subscription service (selective exposure). This takes the form of an offer of a free news clipping service—video clippings and alerts via WhatsApp. Subjects may choose to receive news clippings from a list of prominent entertainment and news journalists with commonly known political affiliations (left, centrist, or right wing).

Secondary Outcomes

Secondary Outcomes (end points)

Change in posterior beliefs for topics related to the treatment videos, as well as spillovers to associated beliefs.

Secondary Outcomes (explanation)

Participants are asked to estimate statistics related to the treatment videos, and incentivized for correct answers.

Experimental Design

Study participants will be recruited through an online survey firm targeting Indians in Hindi-speaking states. Our sample of interest is individuals with strong anti-Muslim sentiments. We use several attitudinal questions to create an index of anti-Muslim bias ranging from 0-12. We aim to recruit 1,000 respondents with a bias score of 8 or more. Given the difficulty of recruiting this sample, it is possible that we will not be able to reach our target N. As such, we will pool data from our small pilot study in our main specification (N = 42). If after pooling we are unable to reach our target sample size, we will include respondents with slightly lower bias scores.

After collecting simple demographic questions and passing attention-check/screening questions, participants are randomized into seeing either the two treatment videos or two placebo videos on trust or population.

- Trust Treatment: The trust game is briefly explained, and participants are told that Muslims return all money twice as often as Hindus.
- Trust Placebo: The trust game is briefly explained, and participants are told nothing about the behavior of Hindus or Muslims.
- Population Treatment: The video debunks the myth that the Muslim population could overtake the Hindu population in India.
- Population Placebo: Participants are provided with information about Buddhist population trends in India.

Then, the posterior belief outcomes are elicited (a secondary outcome).

We measure demand for selective exposure -- news subscription for journalists with biased vs. unbiased stances (primary outcome).

We hypothesize that:
- After treatment, if participants exhibit motivated reasoning (in the sense of experiencing disutility from revising their negative views about Muslims), then they will be more likely to select into the biased/pro-Hindu news subscription, or at least no more likely to select away from it.
- People will immediately update their posteriors to reflect the information they have been given in the treatment (secondary outcome)

In addition to this primary analysis, we will also collect data on a random sample of respondents with a bias score below this cutoff, and examine treatment effects by level of bias (i.e. heterogeneity analysis by bias).

Experimental Design Details

Randomization Method

Randomization done by Qualtrics

Randomization Unit

Treatments are randomized at the individual level

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

1,000 high-bias individuals depending on recruitment costs, no clusters, plus a similar number of lower-bias respondents. (The specific number is impossible to prespecify because we select a random sample of respondents, and can only specify probabilities of selection into the full online experiment, not a target sample size).

Sample size: planned number of observations

1,000 high-bias individuals depending on recruitment costs, plus a similar number of lower-bias respondents.

Sample size (or number of clusters) by treatment arms

500 high-bias individuals per arm

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

We expect to detect an 8 percentage point treatment effect for our primary outcome, using a power level of 0.8, a 5% significance level and a sample size of 1,000 individuals (500 per arm).

Supporting Documents and Materials

IRB