상세 보기
머신러닝을 활용한 공공 커뮤니케이션 캠페인 효과의 인과 추론: 이질적 인과 효과 추정을 위한 인과 포레스트의 적용
- 이병관;
- 김옐레나;
- 마혜현;
- 금세연
초록
Objectives This study investigates the effectiveness of causal forests, a machine learning method for estimating heterogeneous treatment effects, in evaluating public communication campaigns. Unlike traditional evaluations that focus on average treatment effects, this study estimates heterogeneous treatment effects that vary based on audience characteristics and explores their determinants. Methods Data were collected from 1,000 adults who participated in a cross-sectional study assessing the 2015 Tuberculosis Prevention Media Campaign conducted by the Korea Centers for Disease Control and Prevention. Campaign exposure served as the treatment variable, while tuberculosis knowledge and screening intention were the outcome variables. A causal forest was used to estimate the conditional average treatment effect on the treated (CATT). The study also examined treatment effect heterogeneity and the characteristics of high- and low-effect subgroups, validating the estimates by comparing them with previously reported propensity score matching (PSM) results. Results Causal forest estimates revealed positive average treatment effects on both tuberculosis knowledge(CATT = 5.885, 95% CI [3.807, 7.963]) and screening intention (CATT = 0.150, 95% CI [0.061, 0.239]), closely aligning with the previously reported PSM results. However, significant heterogeneity was noted in the CATT estimates, with negative estimates for 3.33% of the exposed group in the knowledge model and 11.09% in the screening intention model. The effect on tuberculosis knowledge varied according to media use frequency and prior tuberculosis-related experience, while the effect on screening intention differed based on age, education level, and family history of tuberculosis. Conclusions This study provides empirical evidence supporting the methodological utility of the machine learning-based causal forest approach in evaluating public communication campaigns. Given the identified distribution and heterogeneity of effects underlying the average treatment effect, we discuss practical implications for resource allocation, audience targeting, and message differentiation, suggesting that causal forests can effectively complement evaluations focused on average treatment effects.
키워드
- 제목
- 머신러닝을 활용한 공공 커뮤니케이션 캠페인 효과의 인과 추론: 이질적 인과 효과 추정을 위한 인과 포레스트의 적용
- 제목 (타언어)
- Causal Inference of Public Communication Campaign Effects Using Machine Learning: Applying Causal Forests to Estimate Heterogeneous Treatment Effects
- 저자
- 이병관; 김옐레나; 마혜현; 금세연
- 발행일
- 2026-09
- 유형
- 정기학술지(Article(Perspective Article포함))
- 저널명
- PR연구 (홍보학연구)
- 권
- 30
- 호
- 3
- 페이지
- 89 ~ 117
- 언어
- KOR
- 출판사
- 한국PR학회
- 발행국가
- 대한민국
- 분량
- 29 페이지
- ISSN
- E 2982-8066
P 2982-754X