To capture new information in app downloads and average daily users, we calculate Abnormal app downloads (Abnormal daily users) by using the average of the app downloads (average daily users) in the previous 12 months as the benchmark for the expected app usage level as shown in Equation (1). The , can be either Abnormal app downloads or Abnormal daily users, while , can be either App downloads or Average daily users The advantage of using a long-window average as a benchmark is to reduce the impact of transitory fluctuation (Huang 2018). Specifically, Abnormal app downloads (Abnormal daily users) is the natural log transformation of the ratio of App downloads (Average daily users) to the prior-12-month average. We use the natural log transformation to reduce the influence of outliers and improve model fit. As Table 1 shows, the mean of app downloads (and average daily users) is much larger than the median, indicating the app usage data are right-skewed. To mitigate the influence of outliers, it is better to take the log transformation. In our sample, the mean of abnormal app downloads is -0.106 and the mean of abnormal daily users is 0.004. Table