大規模多重對照之高效隨機精準配對演算法的實作與應用

Efficient Randomized Algorithms for Large-scaled Exact Matchings with Multiple Controls: Implementation and Applications

許毓軒(Yu-Hsuan Hsu)

指導教授 ： 薛智文

共同指導教授 ： 徐讚昇

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201601445

摘要

匹配是一種常用的統計方式，能從已知的資料中找出因果關係，也被廣泛運用在各種領域。然而，人們通常不會意識到不同的匹配方法有何差異，因此我們在本文中會簡略地介紹現在常用的方法，並提出一種用網路流所完成匹配的新方式。利用網路流可以讓多重匹配可配的比率更多，且捨棄的個體更少，進而導致更精準的結果。在我們的實驗中，此方法約能比傳統貪婪算法多匹配 20% 的個體，達到相對風險標準差小於傳統兩倍的結果。在執行時間的層面，我們也比過去相關方法快了至少 7 倍。同時我們也比較新方法與傳統方法在亂度上的差別，來證明我們仍保有足夠的隨機性。

關鍵字

匹配 ； 抽樣 ； 網路流

並列摘要

Matching is a common statistical method to estimate casual effects from observational data. And it is broadly used in various field. However, people usually are not aware the difference among the matching methods. In this paper, we briefly introduce various methods in matching nowadays. Then provide the new technology which uses network flow. The method has great advantage in ratio matching. The flow matching can support more ratio without dropping too much individuals. This phenomenon can lead to more accurate result. In summary, our method matches 20% more control candidates than the ones found using the traditional greedy method. Furthermore, the standard deviation of the Relative Risk factor found is also twice smaller than. In terms of the amount of improvements obtained in computing speed, our method is at least 7 times faster than a previous comparable study. We also compare with other matching methods in entropy to prove our method has enough randomness.

並列關鍵字

matching ； sample ； flow

參考文獻

[1] Paul R Rosenbaum and Donald B Rubin. The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1):41–55, 1983.

[2] Kevin Arceneaux, Alan S Gerber, and Donald P Green. Comparing experimental and matching methods using a large-scale voter mobilization experiment. Political Analysis, 14(1):37–62, 2006.

[3] Kevin Arceneaux, Alan S Gerber, and Donald P Green. A cautionary note on the use of matching to estimate causal effects: An empirical example comparing matching estimates to an experimental benchmark. Sociological methods & research, 39(2): 256–282, 2010.

[4] Robert J LaLonde. Evaluating the econometric evaluations of training programs with experimental data. The American economic review, pages 604–620, 1986.

[5] Rajeev H Dehejia and Sadek Wahba. Causal effects in nonexperimental studies: Reevaluating the evaluation of training programs. Journal of the American statistical Association, 94(448):1053–1062, 1999.