Kakao platform is one of the most popular social communication mobile apps in the Asian counties market. Under Kakao corp, there are mobile apps such as Kakao Talk, Kakao Story, Kakao Game, etc., which all work together to increase customers’ app engagement and usage. In this data set, we give the information of individual-level weekly panel data of app usage time spent on Android-based mobile devices among 849 users in the time frame of two weeks. By utilizing Panel Difference in Difference Models and the Propensity Score Matching Methods, our team aims to address actionable strategic insights for mobile app platform owners and app developers after a popular app has been released. In specific, our research will focus on how will user’s app usage behavior changes after they adopt a popular app released by a platform.
In this research, we will focus on measuring the effects on 8 different Kakao platforms after adopting the Anipang game
There are 8 Dependent Variables, 4 demographic Independent Variables, Time Dummy variable, Indicator Variable, and Treatment Variable in this data set.
By analyzing the descriptive statistics of this data set, we find the pattern of the demographic information of the participants.
Since this data set is cross multiple periods with the control and treatment effects of the same cross-sectional units, we decided to use the Panel DID Model and Propensity Score Matching to reduce the omitted variable bias and selection bias.
By adjusting different options, such as 1:1, 2:1, 3:1, with or without replacement, and the caliper sizes, we randomly generated 10 matched samples as listed below. Histograms below shown the matched samples before and after the treatment effect are far more alike compared to the raw data.
Tests Results and Findings
We applied both Dummy Variable Regression and FE Estimation on the 10 matched samples on each Dependent Variable. Both of the above methods shown only the Kakao Game app among all 7 used dependent variables, has a significant usage time difference when users adopt the Mobile game Anipang.
Furthermore, the usage time of the Kakao game before and after the treatment has no significant difference among the different age, gender, and education groups. However, the usage time is a significant difference among different income levels. To address this, we designed various representative personas to illustrate the typical app users. Then we gave strategic recommendations based on those different demographics and psychographic groups.
Although we tried to use different methods to eliminate potential selection bias and omitted variable bias, there are still inherent risks in this research.
Below are Appendix and used R codes
Thank you for watching 🙂