공익광고 효과추정을 위한 베이지안 구조 시계열 모델의 적용: 텔레비전 금연 캠페인의 인과효과 추정

초록

The Bayesian structural time series (BSTS) model is a new statistical analysis method using machine learning, that can simultaneously perform time series prediction and causal impact estimation. In particular, it is necessary to pay attention to its utilization in that it can overcome the methodological limitations of evaluating the effectiveness of mass media campaigns nationwide. The purpose of this study is to explore the usefulness of this new method in evaluating the effectiveness of media campaigns by applying BSTS to the 2014 television campaign for smoking cessation. To analyze the impact of the anti-smoking advertising executed from June to December 2014 on tobacco consumption in South Korea, tobacco sales data from January 2014 to July 2015 and potential control series that could affect tobacco sales were used. The causal impact of the anti-smoking advertising was estimated based on the predicted counterfactual response through the BSTS consisting of time series components and regression components. Analysis of the BSTS model leads us to conclude that the empirical evidence for the statistically significant causal impact of the anti-smoking campaign on people’s tobacco consumption could not be found. Based on the derived results, the usefulness of the BSTS methodology in measuring advertising effects was discussed.

키워드