MODELLING THE MUSIC SHOW WINS OF THE FOURTH GENERATION OF K-POP IDOLS USING LOGISTIC REGRESSION

• Main Author: Pahlawan, Reza
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84639
• Institution: Institut Teknologi Bandung

K-pop's globalization has made investing in the industry even more appealing. However, along with widespread digitalization and BTS’s massive success, 4th generation idols are becoming more and more competitive when it comes to setting new records, and thus, popularity is becoming more difficult to assess and investment is becoming increasingly risky. On the other hand, winning in a music show is a benchmark of an idol's popularity. This framework can be used to perform a popularity assessment. This research not only aims to conduct regression analysis but also collects data on the music show wins of the fourth generation K-pop idols. To account for more risks, the regression parameters are estimated using the maximum likelihood method and robust estimation proposed by Croux & Haesbroeck (2003) or also known as weighted Bianco-Yohai (WBY). The collected data contains 306 releases in 2018–2022 from 69 fourth generation idols that include winning status; gender; month of release; number of songs in the album; number of views of the MV in 2-4 months after release; number of Melon streams and cumulative unique listeners as of July 3, 2024; number of Spotify streams as of July 4, 2024; and number of album sales on Hanteo. Each of the continuous variables has a right-skewed distribution, many outliers, and is highly correlated with each other. For categorical variables, only gender and winning status have a significant difference in proportion between levels. Male idols dominate in album sales; women in MV views, Melon and Spotify streams, unique listeners, and winning status. More to that, the event of winning is related to the high value of continuous variables. Compared to ML estimates, models with CH estimates have a better fit based on Hosmer-Lemeshow test statistics but have higher standard deviations. Through regression analysis, a logistic regression model with CH estimated parameters, a good data fit, and outstanding discrimination power was chosen. Six unusual covariate patterns were identified and further analyzed. Given their small impact on goodness of fit and the logical explanations behind their observations, the six covariate patterns are still considered in the data for parameter estimations.