Mask Up! Testing Strategies to Increase Mask Use in Uganda

Free mask distribution was paired with additional education and behavioral interventions to encourage the recipients to use masks. These treatments explored three different channels through which mask use may have been low up to date in Uganda and similar contexts: lack of access to masks, lack of information about COVID severity or how masks work, inattention or lack of commitment to wear masks. All individuals over the age of 5 in “treat first” villages were offered a free mask by village health teams (VHTs) and received critical information about masks that adhered to the Ministry of Health guidance about masks. Random groups of recipients also received additional messages that reinforced specific information about mask effectiveness, benefits to oneself and the community that test 4 hypotheses of why mask adoption may be low. Interventions are summarized in the table below.

2021-03-13

2021-03-16

Mask ownership, mask use, COVID-19 / mask knowledge, COVID-19 / mask attitudes, COVID-19 / mask behavior, social distancing, physical health, mental health

In cases where multiple variables will be used to construct primary outcomes, we will create indexes, where all variables used in the index will be coded in the same direction and weighted by their variance.

Trust in institutions, Living standards and economic activities: Total consumption expenditure in the last 7 days, Food security index, Total household earnings in the last 14 days, Total non-agricultural hours worked (in self-employment & employment) in the last 7 days)

The national mask distribution program focused on delivering a cloth mask to all eligible citizens of Uganda who are 6 years or older. Implementation at the local level was under jurisdiction of the different districts. In most districts, resources were scarce and mask distribution was conducted in a bare-bones fashion (i.e. without additional training, education or messaging) and took multiple weeks to roll out. Our intervention built on this foundation by leveraging the reality that not all districts could receive masks at the same time, and that within a district, the last mile of mask distribution would take some time. We worked with the local officials to randomly select the order in which masks would be delivered to the villages in Mbale district, so that a stepped-wedge design could compare “treat first” to “treat later” villages without delaying mask distribution overall.

Individuals in “treat later” villages would receive masks once mask distribution and education on masks is completed in the “treat first” villages, with an expected lag of approximately 3-4 weeks based on the expected roll-out schedule. Treatments were to be cross-randomized. Given budget constraints, we were able to observe mask behavior and conduct phone surveys only in a subset of randomly selected villages in our treatment area (n=90). The remaining 83 villages would be treated according to the schedule, but no data would be collected.

Masks were delivered to Mbale district in early March 2021 and the district quickly began mobilizing “last mile” distribution to all district villages. Ultimately, the district decided that they would be able to deliver masks to all villages on the same weekend, March 13-14 -- much quicker than anticipated. Given this new reality of mask distribution in Mbale and in order to maintain our study’s goal of not delaying mask delivery for study purposes, we aligned our study plan to accommodate the fact that all villages received masks at the same time. Our study still maintained the randomized selection of which villages received VHT training in addition to basic mask delivery, and the additional education/behavioral interventions.

In other words, in this new setup the “pure control” group of 68 villages received free masks at the same time as all other villages, but the VHTs in these villages were not trained by our study team to deliver education about masks or provide any of the additional treatments. The “pure control” group now reflects the free mask distribution program as it was implemented in remaining parts of the country (i.e. status quo of mask distribution campaign), while the villages where VHTs were trained by our study staff received additional experimental treatments. As the final logistical plans for mask distribution were being worked out, officials in one of the subcounties became uncooperative with our study team, including revealing treatment assignments of all VHTs in the subcounty. Therefore, that subcounty was dropped from the study and no endline data collection occurred there.

Randomization done in office by a computer.

Village

Yes

173 villages were randomized

We observed mask behavior in 90 villages and 12 markets. We collected phone survey data from 630 individuals (clustered in 90 villages) at baseline and 641 individuals at endline.

105 villages randomized to get masks first, 68 villages randomized to get masks later.
Cross-randomization of VHTs into treatments:
75 villages get basic education, 15 villages get mask effectiveness education, 15 villages get behavioral nudge, 68 villages get no VHT training (control)

At alpha 0.05 and power set to 0.80, and an intra-cluster correlation of 0.003 corresponding to the correlation in mask wearing self-reports found in an ongoing study in Kenya, we are powered to detect a difference of 4 percentage points in mask wearing using phone surveys and 1 percentage point using observations in the 90 pilot villages.

Mask-Up! Uganda - Pre-Analysis Plan