A notification alone (Stothart et al., 2015). Participants were randomly assigned to receive a call, a text message or no notification on their own phones while doing a sustained attention task. Those who received a notification made more errors than those who did not, even though they did not answer. Press summaries put the size of the effect close to the cost of actually using the phone. We read only the citation and summaries of this paper, not the full text.
A week with alerts on, a week with them off (Kushlev et al., 2016). 221 undergraduates at the University of British Columbia spent one week keeping notification alerts on with the phone within sight or reach, and another week with alerts off and the phone away. They reported higher inattention and hyperactivity symptoms in the alerts-on week. At baseline, students who reported feeling more interrupted by notifications also reported more inattention (a correlation of .31) and hyperactivity (.30). The authors found no direct effect of the manipulation on productivity or psychological well-being, and note that participants knew which condition they were in, so their answers may reflect what they expected.
Batching (Fitz et al., 2019). 237 smartphone owners recruited through an online labor market took part in a two week field experiment with a custom Android app. Notifications arrived as usual, in batches three times a day, in hourly batches, or not at all. Batching three times a day improved stress, productivity and mood, and the authors found the benefit was mediated by less inattention. Hourly batches did not. Getting no notifications did not improve inattention and instead raised anxiety, which the authors linked to fear of missing out.
A day with no alerts (Pielot and Rello, 2017). 30 volunteers turned off notifications on all their devices for 24 hours and compared that day with a normal one. They felt less distracted and more productive. They also felt less able to respond as people expected, worried more about missing information, felt less connected to their social group, and reported checking their phone more (a modest effect, p = .031). Stress did not differ.