> The researchers studied 35 million users of the Pandora streaming music service during the period from June 2014 to March 2016. Because of the large sample si

> The researchers studied 35 million users of the Pandora streaming music service during the period from June 2014 to March 2016. Because of the large sample size, all the findings in the study are statistically significant at the p < .01 level, and I won’t even bother discussing the statistical analyses, which you can find in the full paper. > 18.3 million users were assigned to a control group who used the normal Pandora service with its normal amount of advertising: 3.6 ads per hours of listening. > 1.8 million users were assigned to a low-ad group who were served 25% less advertising than the control group (2.7 ads per hour). > 1.8 million users were assigned to a high-ad group who were served 38% more advertising than the control group (5.0 ads per hour). >(13 million users were assigned to intermediate treatment groups that I will not discuss further.) The results are clear, though not as striking as I would have expected: after 1.5 years of being exposed to the experimental conditions, people did use the service more, the fewer ads they were served. At the end of the experiment: > The low-ad group listened for 1.7% more hours weekly than the control group. > The high-ad group listened for 2.8% fewer hours weekly than the control group.