Countering Islamic Radicalization in Northern Mozambique: Radio Campaigning and Targeted Messages

We evaluate a radio campaign sponsored by local religious organizations that want to detach religion (namely Islam) from the mobilization of young men for violent attacks. Contents are sponsored by the Islamic Council and the Islamic Congress of Mozambique, the main organizations representing Muslims in the country, as well as the Christian Council. The messages are short interventions by the religious leaders of each of the organizations promoting peace and reiterating how their religion is against violence. The messages were the same across the radios with one main distinction. In half of the radios, the campaign only included messages from the Muslim leaders, while in the other half there were messages from all the religions (Muslim and Christian). For the design, we take advantage of the campaign and the random variation in the supporters of the peaceful message and the dispersion of the radio signal across space. We also add to the radio campaign a targeted phone-based campaign, where in some villages citizens receive the same messages as in the radio campaign on their mobile phones. The targeted messages have two objectives. First, to increase exposure of villagers to the peaceful message. Second, to compare the effectiveness of two alternative channels to deliver information campaigns.

2021-09-13

2021-11-12

The main outcomes are: support for extremist ideas, perceived grievances, support and trust, prejudice/acceptance, anti-social behaviour, cooperation and willingness to cooperate, violence and support for peace.

See pre-analysis plan.

For the radio campaign, empirical identification will follow the strategy described in Section 5. We exploit the expansion of the campaign, and quasi-random variation in the topography-corrected signal coverage from the 8 antennas in the province. We will measure radio coverage corrected by the topography of the affected area, and exploit the panel dimension of the dataset by controlling for time-invariant unobservable characteristics at a highly disaggregated level. We will make use of detailed data on the location of radios in the province made available by the provincial authorities and the information on the type of message that was broadcasted (Muslim and Christian or only Muslim). Finally, we will calculate for each village in the sample the maximum quality of the signal by all the antennas.
Quasi-random variation in the message is not limited to topography-corrected variation in the signal coverage, but it also includes random variation in the message broadcast. In half of the radios (randomly selected), the campaign only included messages from the Muslim leaders, while in the other half there were messages from all the religions (Muslim and Christian).
Concerning targeted messages, before implementing the randomization procedure, we group villages into blocks of three according to the Mahalanobis-distance. We use the following information: the sample size at each village (proxy for village size), the maximum signal coverage at each village and the district. For each block we will randomly allocate each village to one of the groups: i) Muslim and Christian message; ii) Muslim message; iii) control.

Please see the attached pdf file for more detail.

Randomization done in office by a computer

Village/community level

Yes

146 villages/communities

1,400

Muslim targeted messages: 56 villages; Muslim+Christian targeted messages: 55; Control: 56