이중차분법(DID), 통제집단합성법(SCM), 통제집단합성- 이중차분법(SDID)의 비교 연구: 새벽배송 진입효과 분석 (A Comparative Study of Difference-in-Differences (DID), Synthetic Control Method (SCM), Synthetic Differencein- Differences (SDID): Effects of Early-Morning Delivery)

서지정보

· 발행기관 : 한국유통학회
· 수록지 정보 : 유통연구 / 29권 / 4호 / 121 ~ 148페이지
· 저자명 : 서진형, 정환, 조춘한, 김주영

초록

인과추론 연구는 특정 사건의 처치 또는 개입이 발생했을 때 나타나는 효과 혹은 변화를 분석하는 방법들로서, 경제학, 사회과학, 의료 연구 등 다양한 분야에서 활용된다. 윤리적·실무적인 제약으로 인해 실험 연구를 진행하기 어려운 관찰연구상황에서,정확한 인과관계를 추정하기 위해 이런 다양한 방법들이 개발되었다. 본 연구는 인과추론을 위한 대표적인 준실험적 연구 방법인 이중차분법(DID)과, 이를 기반으로 발전된 통제집단합성법(SCM), 통제집단합성-이중차분법(SDID)을 비교하여, 각 방법을적용하기 적합한 상황에 대해 논의한다. 세 가지 방법의 공통된 전제는 실험군과 대조군이 특정 사건의 처치 전 기간 동안 동일한 추세를 보인다는 평행추세 가정인데, 이는 처치 외의 모든 환경이 동일하다는 것을 의미한다. SCM은 특정 사건 이전 기간 동안 실험군과 대조군의 차이를 최소화하도록 대조군에 가중치를 부여하여, 실험군과 유사한 합성대조군을 생성하며, SDID는SCM과 유사하게 대조군에 가중치를 주어 합성대조군을 생성하는 과정에서, 특정 사건의 처리 후 기간에 대해 처리 전 기간과의차이가 최소화되도록 시간 가중치도 고려하게 된다. 즉 세 가지 방법 모두, 실험군이 특정 사건의 처치를 받지 않았을 경우의 반사실적 상황을 가정하여, 실험군과 대조군 혹은 합성대조군들 간의 차이를 통해 처치 효과를 확인한다. 본 연구의 분석 결과, 특정 사건의 처치 전 기간 동안 평행추세가 유사한 경우에는 DID를 적용하는 것이 적절했으며, 처치 전 기간의 평행추세가 유사하지 않거나 대조군 수가 적은 경우에는 SCM과 SDID를 적용하는 것이 더 적절한 것으로 나타났다. 특히, 처치 전 기간이 길수록SDID를 적용하는 것이 더 효과적이었으며, 이는 SDID가 대조군과 시간별 가중치를 함께 적용하여, 처리 전 기간의 평행추세뿐만 아니라 시간적 변동성에도 더 유연하게 대응할 수 있기 때문이다. 결과적으로, 연구 상황에 따라 적절한 방법을 선택하는 것이 현실적인 상황에서 특정 사건의 처치 효과를 정확하게 추정하는 데 중요함을 확인하였다.

영어초록

Purpose: The purpose of this study is to compare and analyze the strengths and weaknesses of three widely usedquasi-experimental methods in causal inference research: Difference-in-Differences (DID), Synthetic ControlMethod (SCM), and Synthetic Difference-in-Differences (SDID). The study specifically examines the impact ofEarly-Morning delivery services on existing offline commercial districts as a case study to empirically evaluatethe accuracy and applicability of each method.
Research design, data, and methodology: To compare the three methodologies DID, SCM, and SDID this studyanalyzed the impact of Early-Morning delivery services on offline supermarket sales. The treatment groupconsisted of the Seo-gu and Yuseong-gu districts in Daejeon, while the control group included selected districts inGwangju and Busan. Using daily credit card panel data, the analysis covered sales changes from May 2, 2019,two years before the entry of dawn delivery services, to October 31, 2021, six months after their introduction.
The DID method compared post-treatment sales changes between the treatment and control groups. The SCMmethod created a synthetic control group by assigning weights to the control group to closely match the treatmentgroup during the pre-treatment period, then compared post-treatment sales changes between the treatment andsynthetic control groups. Similarly, the SDID method generated a synthetic control group with the weights ofSCM and the time weights to minimize differences between the pre- and post-treatment periods, and thencompared post-treatment sales changes between the treatment and synthetic control groups. This approachallowed for the analysis of treatment effects based on differences observed after the intervention among thetreatment, control, and synthetic control groups.
For analyses based on the number of control groups, selections were made in both similar and dissimilar orders by comparing the pre-treatment trends of Seo-gu and Yuseong-gu in Daejeon with those of potential controlgroups using Euclidean distance. The comparative analysis based on the pre-entry period fixed the post-treatmentperiod to 6 months after the entry of the dawn delivery service, while the pre-treatment period was divided into 2years, 1 year, 6 months, and 3 months before the entry. This setup allowed the study to examine cases where thepre-treatment period was longer than, equal to, or shorter than the post-treatment period.
Results: The analysis results indicated that the introduction of Early-Morning delivery services had a negativeimpact on offline supermarket sales across all three models: DID, SCM, and SDID. However, the SCM and SDIDmodels, by constructing synthetic control groups, maximized the similarity between the treatment and controlgroups, resulting in more reliable counterfactual estimates and providing more stable estimates than the DIDmodel. This is because DID may produce biased results if the pre-trends of the treatment and control groups arenot similar. Specifically, the SCM model delivered stable results even when the number of control groups wassmall and when the parallel trend assumption was not met. The SDID model produced more stable results as thepre-intervention period extended.
Conclusions: In conclusion, the DID model is appropriate when the parallel trend assumption is satisfied;however, when this assumption is not met, the SCM or SDID models are more suitable. SCM and SDID areparticularly effective in scenarios where the parallel trend is not satisfied or the number of control groups is small.
Additionally, the SDID model has proven to be useful for analyzing long-term data. The study highlights that theperformance of each method can vary depending on the situation, emphasizing the importance of selecting theappropriate methodology based on the specific circumstances of the analysis. Consequently, these modelscontribute to more accurate causal inference in policy evaluation and can aid in decision-making and strategydevelopment across various fields.

참고자료

· 없음