Jiewen Liu's ARIA project:听Multiple causes: blessing or curse? A Monte Carlo analysis and real data application of the 鈥渄eco founder鈥� method of Wang & Blei
听
From May to August, I finished my ARIA research project under the supervision of Dr. Erica Moodie in the department of epidemiology, biostatistics and occupational health. In this project, I carried out a data analysis on Alberta wildfires using a novel and powerful statistical approach called G-dWOLS, which is devised by my supervisor. My aim was to measure di铿�erent forest 铿乺e suppression methods鈥� e铿�ectiveness, or interventions, precisely and causally, so as to develop a strategy of 铿乺e suppression that is tailored to an individual 铿乺e characteristics. In addition to the optimal intervention suggested by G-dWOLS method, the analysis reveals something detailed about how each 铿乺e suppression method has an impact on an individual 铿乺e, which inspires us to come up with possible explanations of previous (counter-intuitive) results suggested by others鈥� works.
Through my ARIA project, I was fortunately given a chance to explore the field I am interested in. I am very glad to have had an early look into my future path and unquestionably it is my most unforgettable and valuable experience this summer. What I have learned really goes far beyond my expectations. First of all, to conduct the research and go forward, I needed to understand G-dWOLS methodology. It is a method that is invented to casually quantify the treatment effect. Thus, I read some papers on causal inference and learned some classical frameworks and the rationale behind them. Then, I proceeded to grasp the background knowledge in G-dWOLS method, and got a good understanding of how this works on data. Finally, I completed code implementation and performed the G-dWOLS analysis on Alberta鈥檚 wildfires using R, a language and environment for statistical computing and graphics. Besides that, I also learned how to utilize a variety of R鈥檚 functions to carry out data analysis and how to wrap up my findings and present them in a formal way (writing up a paper).
A highlight comes from my unintentional look into the wildfire context. There have been some works done in the same dataset. The result given by G-dWOLS analysis agree with the previous works too. All analyses indicate that air tankers, which are believed to be the most aggressive intervention, are not found to be the most effective, and they are thought to be counter-intuitive. When I reached this step, I also stopped here and struggled to think what was really going on. One day, inspired by the mechanism of how a particular intervention affects an individual fire reflected in the G-dWOLS analysis, I traced back to the officially issued Alberta鈥檚 wildfire review and found some clues like restriction on resources, limited visibility for aerial attack, etc. that might account for this reverse in the different interventions鈥� power. To be more specific, restrictions on resources may play an important role in affecting the interventions鈥� effectiveness, which is cemented by the extra information uncovered by G-dWOLS analysis. The analysis shows air tankers give poor performance in comparison to other interventions when it comes to big fires and when there are many fire sites simultaneously. The wildfire review implies that air tankers are a scare resource and the enforcement of them on many situations may indeed lead to foreseeable worse outcomes due to fewer resources available for air tankers deployed nearby. My accidental findings, to a certain extent, contribute to possible explanations on the counter-intuitive result and enriched the content of my analysis report. Hopefully, we might be approaching the truth in some ways.
Doing research drives me out of my comfort zone. At the beginning of my research, I spent lots of time comprehending the knowledge of biostatistics. The G-dWOLS method is from the biostatistics field, but biostatistics is not treated as an undergraduate topic and it hardly appears in undergraduate courses, which means in many ways I had to start from scratch and read the specific part of the material that I had difficulty in understanding over and over again. Though feeling despondent sometimes, with persistence and Erica鈥檚 help, I eventually made it and built up more confidence to take on new challenges.
To sum up, ARIA is an amazing project. I had a glimpse of what my academic path would be like and obtained many insights. Without a doubt, it will facilitate greatly my uptake in my further study in this field. Having gained a strong start through ARIA, I acquired many essential skills and more possibilities. I will continue to make efforts and dive into the kaleidoscopic world of biostatistics. I would like to thank Mr. Harry Samuel for his generous support of my ARIA project.
