美國資訊工程研究所申請經驗分享
前陣子因為參與了我的好朋友哈佛姐 Alicia Yang 的 YouTube 頻道，拍了兩集影片之後 (同場加映 Female Engineers Working in Silicon Valley / 矽谷科技公司大哉問)，陸續收到一些詢問博士班申請問題的來信，兩年前也經歷過碩士班申請，剛好也結束了 Project Tyra 發起的海外研究所申請指導計畫，指導了兩位預計申請碩士班的導生，藉這個機會紀錄導生們問的問題以及分享自己過去的經驗與想法。
簡短自我介紹：我目前是 UC Berkeley Eletrical Engineering & Computer Science 博士生，在申請博士之前我是在 UC Berkeley School of Information 唸碩士，唸碩士前工作了一段時間，大學則是畢業於國立臺灣大學。
Photo by Jorge Cham on PHD Comics
目錄
Preliminary 暖身
研究所申請文件
如何累積研究經驗
研究所財務規劃
選校 / 選指導教授
結語
Preliminary 暖身
在開始查任何資料、動筆寫任何文件之前，我覺得很重要的一件事情是靜下心來，思考並詢問自己以下幾個問題：
我為什麼想要出國讀研究所，我的動機是什麼？
讀研究所可以讓我得到什麼樣的經驗，能夠帶給我什麼，我又想要從中得到哪些？
我未來的目標是什麼，讀完研究所的下一步是什麼？
研究所可以幫助我達到我想要的目標嗎？有沒有其他更適合的方法？
我有哪些限制或是哪些現實上的考量？
畢竟碩士一到兩年、博士四到六年，研究所永遠不會是你的終點站。或許有幾個問題你還沒有明確的答案，但是如果這些問題你完全都沒有思考過的話，那我會建議你先想想這些問題，再開始執行後續的計畫。
研究所申請文件
碩士班跟博士班的申請文件其實差不多，不過博士班的申請文件更加重視研究經驗，所以如果是申請研究導向的碩士班或是博士班的話，每一份文件都應該優先強調與「研究」相關的內容。如果是想要申請就業導向的碩士班，網路上有很多人的分享，可以再多閱讀其他人的經驗，這邊主要以博士班申請文件為重點。
推薦大家在動筆之前先閱讀過 Prof. Philip Guo 的文章：A Five-Minute Guide to Ph.D. Program Applications，你會更知道 admission commitee 的篩選標準，畢竟要打出一手好牌，還是得先深入了解遊戲規則。
履歷表 Academic CV
履歷表主要就是事實呈現 (what, when, where, how)，不過學術用的履歷跟找工作的履歷還是有些微差距。學術用的 CV 比較沒有一頁內的限制，不過盡量也不要超過兩頁，CV 的內容主要有以下這些：
Education
Research Experience：這邊的內容盡量要跟 Statement of Purpose 的內容相呼應，也應該要跟你的推薦信有相關聯。
Publication：如果你有論文是以第一作者的名義發表，可以特別強調，因為第一作者是整篇論文貢獻度最高的人，也就代表你的研究實力很堅強。
Non-research Work Experience
Teaching Experience
Honors & Awards：如果有研究相關的獲獎經驗可以特別強調，例如論文獲獎。
Services or Leadership Experience
Statement of Purpose (Research Statement)
Statement of Purpose (SoP) 最主要的內容就是你過去的修課及研究經驗、做了哪些事、學到哪些事情、你未來想要做的研究方向以及為什麼想要做這個題目。內容話我覺得盡量以不超過兩頁為限，過去的經驗占大約 2/3 的篇幅，然後剩下的 1/3 詳述申請博士班的動機、未來想做的研究方向以及想要和哪些教授合作等等。
一個常見的誤區就是在 SoP 中單純重複履歷或是成績單上面有的內容，沒有提供更多的資訊。如果你的 SoP 上寫的內容，從履歷或是成績單上就可以得知，通常表示你的 SoP 寫得不夠深入。舉一位我的導生的 SoP 為例，他在 SoP 的第二段把過去一些修過的課以及課堂內容都描述了一遍，這些資訊其實從他的成績單上面就可以得知了。我覺得比較好的寫法，是挑選一到兩個你印象深刻並且有實際產出的課程 (如果是其中一位推薦人的課更好)，然後深入描述具體內容，例如課堂上做的專案後來發表成了一篇論文，課堂上的內容啟發了你對某個領域的興趣，所以你後來加入了 OO 教授的研究室或是主動爭取跟授課教授做獨立研究等。
推薦信 Recommendation Letter
推薦信是另外一項很重要的文件，因為它提供了從第三者的角度評鑑申請者的機會。話雖這麼說，我的三位博士班推薦人都是自己親自寫推薦信的，沒有要求我打草稿，所以我也不知道實際上發生了什麼事 (攤手 🤷♀️)。不過博士班申請推薦人會希望能親自寫推薦信也非常合理，畢竟招收一個博士對學校或是未來的指導教授來說是一個很大的投資，對學生來說也是一個很重大的決定。
雖然我的推薦人都是親自寫推薦信，還是需要提早提供他們一些參考資訊，我自己提供給推薦人的資料包含以下：條列做過的事情以及重點 (列點式就好，不用打草稿), 完稿的 Statement of Purpose, 履歷, 預計要申請的學校/科系/截止日期。建議提早跟你的推薦人聯絡，確認他們的意願、能夠提供幾封推薦信、是否能夠幫你說好話等等，根據我的其中一位推薦人說，必須要有至少兩位能夠極力推薦你，也就是他們可以在推薦信裡面很肯定地寫: this student is excellent!
Standard Test Score
這部分就是非常制式化的各種成績單包含大學到研究所的成績單 (transcripts), GRE/GMAT 成績以及 TOFEL/IELTS 成績。
Personal Statement or Supplemental Questions
有些學校還會另外要求 Personal Statement，例如 UC Berkeley，或是另外要求你回答一些問題，有些學校也會讓你上傳額外的 supplemental material。每個學校要求的內容都不太相同，我自己是先完成 UC Berkeley 要求的 Personal Statement，然後再以這份文件為草稿，修改成其他學校要求的內容。Personal Statement 的內容，每個學校也不盡相同，建議去學校的網站詳讀，這邊附上 UC Berkeley 的 tips and advice。
Personal Statement 是另外一個讓學校了解你這個「人」的機會，因為上述的申請文件都只能呈現出自己的專業能力，但是其他跟你的申請專業比較沒有直接相關的能力或是人格特值，就比較難從以上的文件裡面了解，所以 Personal Statement 是一個讓你自由說故事的地方，也因為每個人都有不同的故事可以說，內容也會因人而異。我自己主要以我過去的社群經營、志工參與、大型會議籌備以及推廣人才培育、職場多樣性的內容為主，因為這部分是我在自己的專業以及工作之外，很重視也付出很多時間的地方。
順便藉機工商一些我參與過的社群組織們：
Taiwanese in Data Science / Women in Data Science Taipei
Taiwan Data Science Meetup / 臺灣資料科學協會
Women in Machine Learning
Delta Analytics
ccClub
如何累積研究經驗
研究經驗的累積通常都需要一年以上的時間才能比較看得出成果，所以我也建議大家如果有相關的機會就盡量爭取，盡量提早開始累積，以下提供一些我自己做過或是知道的機會給大家參考：
課堂期末專案
課堂期末專案如果做得好是有機會可以寫成論文發表的，尤其是那門課要求期末報告以論文格式撰寫的話，非常一石二鳥，做完期末專案順便投稿。我碩士班的一個期末專案居然就這樣讓我投稿上了，也讓我有機會能到義大利參加會議 (順便觀光)。課程期末專案有時候也可以繼續延伸，變成未來的學士、碩士論文或是變成專題研究題目等。
主動幫忙碩士、博士或是博後
如果一開始沒有想法或是不知道要怎麼開始，也可以主動幫忙其他碩博士生的研究專案。UC Berkeley 大學部有 undergraduate research apprenticeship program，讓大學部的學生也可以有做研究的機會。臺灣比較沒有類似的計畫，至少我在臺大的時候好像沒有類似這樣的全校性計畫，不過就算沒有類似的計畫，也可以主動寄信給碩博士生們或是主動聯繫教授，我在博班的第一年也有收到大學部的同學主動寄信詢問是否能幫忙我的 project。
Independent Research Units
UC Berkeley 有 independent research units，類似臺灣的專題研究課，是有算學分的但是需要找指導教授，有些學院專題研究課是必修，如果不是必修也可以主動詢問系上的教授，是一個非常好累積經驗以及讓教授認識你的機會。我在 UC Berkeley 碩士班的時候有跟 EECS 的教授做 independent research，在臺大也有幫忙教授做研究，不過當時在臺大沒有類似相對應的學分可以拿，所以我大學做的研究後來是改成申請科技部大專學生研究計畫 (詳細介紹在下一項)。
科技部大專學生研究計畫
這是科技部的計畫，補助大學部或專科學生的研究計畫 (非碩博士生)，需要找指導教授，我自己大學的時候也有申請到。我覺得這是一個很棒的機會，除了有一些補助，讓自己更有動力做研究之外，也可以放在履歷上面，畢竟能在大學的時候就申請到研究計畫補助，代表你是很有潛力的。
研究助理或研究相關的實習
有些教授、實驗室或是中研院都會徵研究助理或是暑期實習，大家可以積極申請。我大學的時候有在中研究的經濟研究所當研究助理，唸碩士前有在臺大資工系以及 University of Southern California 當研究助理，念碩士的時候也有在學校內當研究助理 (Graduate Student Researcher)，每一份工作都學到了非常多東西，包括做研究的方法，怎麼寫論文，還有認識了很多很厲害的同儕跟前輩們，現在各個已是經濟博士、資工博士、助理教授、research scientist, economist 等等，覺得每一個人都超厲害，當時每天都在擔心自己扯別人後腿。
當研究助理或是實習可以非常快速扎實的累積研究相關的經驗，如果有成果也能夠發表論文，幸運的話還能夠以第一作者發表，合作的教授也比較能寫出有內容的推薦信，這些事蹟以及經驗對於博士班的申請都是非常強而有力的。
不過做研究要有成果也並非很容易的事，中間會碰到許多的困難以及未知，會遇到摸索撞牆期，實驗失敗也是家常便飯，除了堅持下去之外，也可以從中檢視自己是否真的喜歡「做研究」這件事，自己未來是否還能夠忍受這些困難，不然就算申請上研究所，未來也很有可能讀得很不開心。
交換學生、訪問學生、教育部或國家補助的培育計畫
學校的交換或是訪問學生計畫可以讓你有機會到國外的大學進修，可以趁在國外大學的期間，利用以上的幾種方式讓外國教授認識你，若是有計劃申請當地的學校，能夠有一封當地教授的推薦信，幫助會非常大。我自己大學的時候，因為經濟能力不允許加上未來想申請碩士班，所以沒有考慮交換學生，不過我後來非常幸運的申請上了教育部人文社會科學學術人才跨國培育計畫，也是因為這個計畫，讓我更加確定未來出國唸研究所的目標，雖然一開始我是想做跟經濟相關的研究，而且當時還誇下海口說：「我絕對不可能唸博士班啦」，下場就是被自己賞了一巴掌，完全應證墨菲定律 😂。
研究所財務規劃
碩士班經費
碩士班通常都是自費，有少部分的學校會給部份或是全額獎學金，不過獎學金通常就是可遇不可求。經費的部分，若家裡沒有辦法支持也不要太早放棄，其實還是有辦法的，不過過程中心理壓力一定會比較大，需要謹慎計畫開銷，以及積極開源節流，我自己家庭經濟狀況也無法提供支持，以下分享我的做法：
工作 2 年的存款：工作後存的錢大部分都拿去還大學的就學貸款了，幸好大學時有獎學金補助讓我可以少還一點錢，不過自有存款其實還完學貸就所剩不多了，約莫 10 萬台幣。
教育部或台北市留學貸款：留學貸款 100 萬台幣，碩士兩年，分兩次一年撥款 50 萬。需要注意的是如果大學或是碩士期間有就學貸款，需要先清償就學貸款才能再申請留學貸款。我自己是申請教育部的留學貸款，如果跟我一樣家裡有低收入戶證明 (或家庭年收入低於某個數字)，就學期間 (寬限期) 內的利息由政府補助。
校內研究助理或是教學助理：留學生在學校內是可以工作的，如果有研究或是教學助理的機會一定要爭取，達到一定時數以上的工作通常會減免一定比例的學費，工作的時間也會另外算薪水。我碩士兩年都有工作，第一年是研究助理然後第二年是教學助理，省下了非常多的學費。第一年的工作會比較難找，因為人不在當地，所以要非常積極地打聽消息跟申請，但是也要有心理準備開學後會非常的忙，需要兼顧課業及工作還需要準備實習面試或是找正職工作等。
暑期實習 (三個月正職薪水)：撇除一年的碩士，通常一年半或兩年的碩士都有機會能夠在暑假去業界實習，多賺 3 個月的正職月薪。
私人借款：為了保險起見，一位我非常敬重而且幫助我非常多的前輩有私下借給我一小筆金額，不過我完全沒有動用到這筆款項。
財務這部分是我在申請研究所的時候覺得最挫折的地方，因為有些情況不是自己多努力一點就能夠解決的。一開始我的家人沒有能力擔任我的留學貸款擔保人，我的經濟狀況也沒有辦法滿足學校所需的財力證明，也因此我需要學會如何向他人尋求幫助，最後真的非常幸運身邊有好朋友以及貴人前輩們提供我非常多的幫助，才讓我順利解決財務這個關卡。我知道有些人可能會因為自身的經濟狀況而猶豫，也有些人可能會像我一樣不喜歡麻煩他人或向人求助，不過適時的求助也是非常重要的。別人會幫助我們也是因為相信我們，所以我們也要相信自己，未來一定有能力可以回報，並且成為一個也有能力幫助後輩的人。
博士班經費
博士班就跟碩士班非常的不一樣，博士班除了不用付學費以外，通常也是有薪水或是獎學金的，以下是博士生幾種常見的 funding 來源：
Fellowship & Scholarship
學校的或是系上的獎學金, fellowship
美國公民可以申請的 fellowship (留學生不符合資格), e.g. NSF, DoE
業界的 fellowship, e.g. Google PhD fellowship
臺灣的獎學金, e.g. 公費留學, Fullbright, 教育部與世界百大合作設置獎學金 （我申請的那一年剛好新增了 UC Berkeley, 但是我沒有注意到，知道後已經超過申請期限，捶心肝 🥺）
Research Assistantship (在 UC Berkeley 叫 Graduate Student Researcher, GSR)：你的薪水是由某個研究計畫的經費來的，這個計畫有可能來自某個實驗室、教授的研究計畫、與業界合作的計畫或是國家實驗室 (UC Berkeley 有 Lawrence Berkeley National Lab, LBNL)，實際情況以及工作內容通常都要跟指導教授談過才會比較清楚。
Teaching Assistantship (在 UC Berkeley 叫 Graduate Student Instructor, GSI)：顧名思義就是教學助理，認真把助教當好就可以了，除非你遇到授課老師希望你幫忙翻新教材或是那堂課助教少學生多，那工作量就會增加非常多。
Research or Conference Travel Grants: UC Berkeley 有一些給研究生申請的 research grants (研究經費) 或是 conference travel grants，可以拿來用在研究相關的花費上，例如研討會註冊費及車馬費、問卷調查回饋金、購買實驗器材或是運算設備等，如果沒有自己的一些小額研究經費，通常就需要請指導教授用實驗室的研究經費付，所以有時候有一些自己的經費可以付也是比較有彈性。
Summer Internship：有些人暑假也會選擇去業界實習，不過這要跟指導教授溝通，因為實習不一定會跟你的研究題目有相關或是指導教授希望你暑假專心做研究等等。
選校 / 選指導教授
碩士班：挑選選校
在選校這點，我的導生們都非常認真，選校的表單做得非常的詳細，請教很多過去學長姐的經驗，也上網查了非常多資料，不過有一點我覺得大家會稍微忽略掉的就是參考學校網站。我在申請碩士班的時候最主要的參考來源是系所的網站，了解學校提供的課程，系上教授的研究領域，以及整體著重的面向。例如有些科系強調培養跨領域人才，有些科系偏好有工作經驗的學生，而有些科系可能更強調專業科目或是要求須具備一定程度以上的技術背景。這些資訊通常都能夠從學校的網站裡找到，閱讀這些資訊也能夠幫助你判斷自己的經驗與背景是否和這個 program 相契合。申請學校是一個雙向的過程，第一是這個 program 能不能幫助你達到想要的目標，第二是你的背景是否大致符合這個 program 想要招收的學生的樣貌，通常仔細閱讀完學校的網站後，你應該可以大致感覺的出來，幫助你挑選最適合自己的 program。
博士班：挑選指導教授
博士班和碩士班最不一樣的地方，就是博士班申請通常是綁定指導教授 (faculty/research advisor)，也因此整個申請過程中最重要的，就是了解哪些教授做的領域以及題目，是跟自己未來想做的研究領域相關。這件事情非常的耗時但卻是非常重要的環節，了解一位教授的研究領域比較快的方式，就是到每一位教授的網站或是 Google Scholar，閱讀他們近幾年發表的論文，另外也可以看他們尚未畢業的博士生們正在做的題目，綜合一下大概就可以略知一二。
我身邊的博士生朋友們大多都同意指導教授是決定你博士班唸的開不開心最重要的因素，當然也有很多其他的因素，不過慎選指導教授還是非常重要的。在選擇指導教授這方面，除了研究方向外，管理風格也是另外一項蠻重要的指標。博士班畢竟是一個 4-6 年的過程，也因次你能不能和你的指導教授相處得好、能不能一起合作這麼長的一段時間、實驗室的氛圍如何、實驗室的運作風格等等，也都會間接影響你未來的研究生活 (我個人還很在意天氣，因為我需要陽光)。
要了解一位教授或是實驗室的管理或運作風格通常比較困難，畢竟從網路上比較難得出任何結論，比較有效的方式是直接詢問這些教授的學生，如果有認識當然是最容易詢問（套八卦），如果沒有認識也沒關係，可以有禮貌的寄信給幾位你覺得比較親切的研究生，跟他們約個 30 分鐘的時間，我相信大部分的研究生都會非常願意撥空跟你聊一聊的。
如果你已經錄取了某些學校，那麼我會非常強烈建議你參加學校舉辦的 visity day。Visity day 通常是一到兩天，大部分的學校都會補助交通以及住宿費 (通常會有上限，所以國際生可能就只能 cover 一部分)。這個活動就是系上傾全力跟你推銷他們的時間，很多實驗室都會有 open house 讓大家參觀，也會安排很多正在就讀的博士生們給新生認識，讓新生有時間可以問問題，總之有非常多的活動跟機會可以了解你未來的工作以及學習環境。
結語
申請學校的過程有非常多的不確定因素以及個人考量，有很大部分也來自於運氣，會不會申請上某間學校也跟當年其他申請者的條件，指導教授是否有計劃收新學生等因素有關，這當中有很多都是我們無法控制的，但有件事是我們能夠做的，就是對未來保持彈性，繼續朝自己想要的方向以及目標前進。我的指導教授之前和我分享過，他有一位學生想加入他的實驗室，但是一開始沒有被 UC Berkeley 錄取，後來他先到另一所錄取他的學校讀了一年博士班後，第二年又再申請了一次，也順利的成為了我指導教授的學生。最後，分享一句某位教授說的："It might be random sometimes, but it cannot be random all the time."
如果你順利申請上博士班，恭喜你/妳！同時也代表你成功解鎖了更多困難的挑戰，未來一定會遇到很多挫折或失敗想放棄的時候，鼓勵你 (也鼓勵未來的我自己)：對未來保持彈性，繼續朝自己想要的方向以及目標前進。如果在這個過程中找到更適合自己的方向，就算博士班沒有讀完其實也無所謂，當然也不要輕易放棄就是了，預祝大家研究順利！
附上一些我讀過也覺得很棒的參考連結：
UC Berkeley Fall 2020 新開的課，我自己非常喜歡，討論了很多在博士班會遇到的問題：Research Culture and Community Norms
Prof. Matt Might's blog (see Graduate School section)
PhD Comics (放鬆舒壓用)
Notes on Science Research Writing
This note is summarized from the book Science Research Writing: For Non-Native Speakers of English by Hilary Glasman-Deal. I would recommend the book for those who are new to science research writing or want to get better at it, especially for beginners.
Scientific Publication Structure
- Abstract
- Introduction
- Methodology
- Results
- Conclusion/Discussion
- References
Introduction Section
The best time to write the Introduction section is after you have written, or drafted, the Methodology and Results sections. In the introduction you start out by being fairly general and gradually narrow your focus, whereas the opposite is true in the Discussion/Conclusion.
Grammar and writing skills
Tense pairs
- Present Simple tense is used in science writing to state accepted facts and truths.
- Besides time of the verb, i.e. when it happened, another key difference between Present Simple/Present Perfect is the relevance of the event to the situation NOW.
Signalling language
- Connect one sentence or idea to the next so that your reader is carried carefully from one piece of information to the next. It also forces you to develop your ideas logically.
- Overlap sentences, meaning to repeat something from the previous sentence.
- Use a pronoun (it, they) or pro-form (this method, these systems) to glue the sentences together.
- Don't finish the sentence at all, but join it to the next sentence with a semicolon or a relative clause (a which clause). This works when two sentences are very closely related and one of them is quite short.
- Use a signalling sentence connector (therefore, however) to indicate the relationship between one sentence and the next, or one part of a sentence and the next.
- Functions of signalling language: cause (because), result (hence), contrast/difference (whereas), unexpectedness (although), addition (in addition).
Passive/Active
- You can use we to refer to your research group or team, but do not use it to refer to people or humanity in general.
- Use a construction with It (It is know/thought that...) if you are referring to people in general.
- It is also common to use the passive instead of we, especially in the Methodology/Results sections (was measured, was added, etc.).
- If you are writing as an individual, write in the passive from and avoid using I.
- Use words like here and in this study to refer to your own work.
- The agent (the person who performed the action of the verb) is often not mentioned in the passive form. If this causes confusion, use dummy subject (This article, The present paper) to take the place of I or we rather than the agentless passive (x is presented).
Paragraphing
- A paragraph in academic writing often starts with a topic sentence, which gives the main idea of the paragraph, and tells the reader what the paragraph is about.
- When the topic or idea moves too far away from the first sentence, begin a new paragraph.
- To create paragraphs that have a logical and coherent structure, write down each idea/concept that you want to talk about, check that they are in a logical order and then list what you want to say about each, using bullet points.
The model of writing introduction
Part 1
- Establish the importance of the research topic.
- Provide general background information or facts.
- Do the above two things again, but in a more specific/detailed way.
- Define the terminology in the title/key words.
- Describe the general problem area or the current research focus of the field.
Part 2
- Provide a transition between the general problem area and the literature review.
- Provide a brief overview of key research projects, previous and/or current research and contributions.
Part 3
- Locate a gap in the research.
- Describe the problem you will address.
- Present a prediction to be tested.
Part 4
- Describe the paper itself.
- Give details about the methodology reported in the paper.
- Announce the findings of the paper.
Methodology Section
Grammar and writing skills
Passive and tense pairs
- Use the Present Simple tense to describe what is normally done or to describe a standard piece of equipment used in the research.
- Use the Past Simple tense to describe what you did yourself.
- Use the agentless passives for both cases above.
- To make your own contribution clear and easy to identify, mark it with words -- by adding phrases like In this study or In our experiments and by identifying the procedure used by other researchers with careful references at the appropriate place in the sentence like In Brown (1999).
Use of A and THE
- Use the if or when you and your reader both know which thing/person you mean.
- Use the if there is only one possible referent.
- Use a if it doesn't matter or you don't know or your reader doesn't know which thing/person you are referring to.
- The, a, or plural can all be used generically, i.e. when expressing a general truth.
- A is used before consonant sounds, while an is used before vowel sounds. Sounds not spelling, so write an MRI scan because the letter "M" is pronounced "em", but a UV light because the letter "U" is pronounced "yoo".
Adverbs and adverb location
- Adverb location may cause invisible errors, which are easy to make and hard to detect.
- Adverbs needing prepositions can be ambiguous (Look at that dog with one eye can either mean USING one eye or HAVING one eye).
- Adverbs may attach themselves to unexpected parts of a sentence.
- Be careful if more than one adverbs are used. It is generally better to avoid adverb clusters and rewrite the information in a different order.
- If your adverb relates to the whole sentence (i.e. clearly, last January, as a result) then consider putting the adverb at the front of the sentence.
- Consider breaking the sentence down into units, each with its own adverb.
The model of writing methodology
Part 1
- Provide a general introduction and overview of the materials/methods.
- Restate the purpose of the work.
- Give the source of materials/equipment used.
- Supply essential background information.
Part 2
- Provide specific and precise details about materials/methods (i.e. quantities, temperatures, duration, sequence, conditions, locations, sizes).
- Justify choices made.
- Indicate that appropriate care was taken.
Part 3
- Relate materials/methods to other studies.
Part 4
- Indicate where problems occurred.
Results Section
Grammar and writing skills
Sequence:
- Time sequence means how long each step took and where it occurred in the sequence.
- 8 groups of sequences:
- before the beginning: beforehand
- at the beginning/first step: at first
- indicate steps/order: next
- after a short while: shortly after
- at a late/later stage; after a while/longer period: later on
- one period/period occurring almost or exactly at the same time as another (possible causal relationships): as soon as
- at the end/last step, after the end: lastly
- after the end: afterwards
Frequency and Quantity
- Results do not speak for themselves. Your reader needs to know what the numbers or quantities mean in order to understand them.
- The results 23% can be communicated as a strong result (in as many as 23% of cases) or a weak result (in only 23% of cases). If you just write: As can be seen in Fig. 1, the effect occurred in 23% of cases, you have not added anything to what the reader can see for themselves.
- One way to communicate your interpretation of the results is to use the frequency language, e.g. The effect was seen frequently or The effect was seen occasionally.
- Beware that frequency language is often used in a subjective way, ranging from 100% sure to 0% sure.
- Another way to communicate your comments on the numbers is to use quantity language. It can be used to replace numbers (many) or to comment on numbers (as many as 45).
- 5 groups of quantity language:
- increase the size/quantity: considerable
- reduce the size/quantity: marginal
- emphasize how big/small/high/low the size/quantity is: far (above/below)
- the size/quantity is similar/close to another: approximately
- a reluctance to commit oneself to an interpretation of the size/quantity: fairly
Causality
- In some of the verbs/phrases, the position of the cause and the effect fixed (cause produced effect, effect originated in cause). In others, such as x is linked to y, it depends on what the writer wishes to say.
- Some verbs/phrases communicate a clear/strong causal connection (cause, produce, be due to), some refer to a partial cause (be a factor in, contribute to), some refer to initial or first cause in a causal chain (originate in, initiate*), and others communicate a weak causal connection* (be related to, link*).
- To be a cause or a result of something implies that other factors were also involved, whereas to be the cause of or the result of something implies that it is the only cause or result.
- x results from y means x is a consequence of y; whereas x result in y means y is a consequence of x.
- You may also add frequency, quantity qualifier or model verb (may, could) to soften a causal statement, but if you add too many, the sentence may not mean very much at all.
The model of writing results
Always start by providing an overview. This enables you to show your reader the "wall" before you begin to describe the "bricks". It is more "reader-friendly" to start with some introductory material. This type of general overview may need to be repeated when you move from one set of results to another.
Make sure that you understand the difference between the explanation of a result (why it occurred as it did), the evaluation of a result (what the numbers mean), and the implication of a result (what the result suggests or implies). Your explanations in this section should be limited to fairly direct comments about your results. Leave the broader explanations and implications in the Discussion/Conclusion section.
Part 1
- Revisit the research aim/existing research obtained by other researchers.
- Revisit/expand your own methodology and adds more information about it.
- Provide general overview/statement of results to begin a new paragraph.
Part 2
- Comment on the results then invite the reader to look at results/figures/tables, etc.
- Provide specific/key results in detail, with or without explanations, using language that comments on the results.
- Compare with results in other research, using subjective, evaluative language.
- Compare with model predictions.
Part 3
- Mention problems in the results and use quantity language (e.g. slightly) to minimize its significance.
- If you can, suggest possible reasons for the problem and/or offer a solution or a way forward.
- Mention and acknowledge the problems or difficulties you encountered in the Methodology and Results section; it isn't appropriate to mention them for the first time in the Discussion/Conclusion.
Part 4
- Provide possible implications and applications of the work/results.
- Begin to open out the focus on the paper and transition away from the central "reporting" section towards the conclusion.
Reading in Progress
back to note / back to all posts
Rethinking The Real-name Policy on Facebook
The Platform Naming Policy
As online platforms become prevalent in our daily lives, the properties, affordances and the dynamics of these platforms introduce different behaviors that we may encounter in our physical lives. While the affordances may not directly determine the behaviors of the users on the platform, they can shape the social practices that the users develop under such affordances [1]. One such affordance is the identity and reputation policy of a platform. Overall, there are 3 main policies for the identities of the users, real-name, pseudonymity and anonymity. We next analyze what social practices were developed by users under each naming policy.
The biggest social media platform, Facebook, is known for the real-name policy on their user profiles and requires phone numbers and emails to register for an account. In a real-name policy platform, users’ identities are easily traced back to their real lives in the offline space. Because users are participating in the platform using their real identities, the platform can be treated as an extension of the users’ physical real lives. And thus, users usually maintain their common social norms and practices from their daily life. However, this real-name policy also make it challenging for minorities. For example, users with rare names usually are suspended by the platform and legitimate dissidents or advocates of controversial issues may risk exposing their true identity. These minorities are usually forced to falsify their profile to be deemed “normal” or to protect themselves.
Platforms with pseudonymity naming system like Reddit allow users to register using a nickname without their real names. One does not need to provide phone numbers or emails to verify their account. However, even though the accounts are not linked to an offline identity, the history and statistics of the account are traceable and may be visible to other users. The pseudonymity nature of the platform also reduce one’s accountability for his or her words or actions. The unsavory subreddit “Jailbait”, where users share involuntary, sexually suggestive poses of young girls, has grown largely because men felt protected by the pseudonymity and are free from scorn and delegitimization they would have had from a physical world ([4] p.165). The affordance of pseudonymity and scalability allow for supporters of a certain value or interest, whether legitimate or not, to form and engage without risking their real identity.
Yet other platforms maintain or encourage complete anonymity, such as 4chan.com. Posts on these platform are unable to be traced to any users online and offline, allowing any participant to carry out their behaviors without any accountability. Anonymous sites differs from pseudonymous sites as there is no symbolic profile associated with an user and no history can be traced. On the pseudonymous sites, content moderation and antiharassment regulations can be enforced on the basis of the accounts, but this cannot be done on the anonymous sites. For these completely anonymous sites, it is safe to assume that the identity of any user will not be traced and hence the users are free from the consequences of their actions.
Alternatives of Real-name Policy for Anti-social Behaviors
Will Facebook be able to handle safety concerns and anti-social behaviors without the realname policy? Some alternative approaches can be made to combat the inappropriate behaviors with pseudonyms. Facebook relies on users to report inappropriate content and has deployed some algorithms detecting irritating content. To defer anti-social behaviors, training programs can be imposed on the users when they first register an account or before posting. If the policies are made explicit and the users are told to contribute positively and act responsibly before using the platform, the training program may deter some users from destructive actions and help prevent some undesirable behaviors. Wikipedia has employed this strategy when a user’s edit is reverted. Users will get a more comprehensive guideline and some tips from the editors. The guidelines serve as a training for new users to contribute to Wikipedia [2].
Another strategy to promote positive behaviors is a reward system, signaling individuals’ acceptances within a community and recognizing their contributions. Various platforms have the reward systems, Reddit’s karma points and upvotes, Wikipedia’s barnstarts for promoting the value of actions such as social and emotional support of the community that are less visible yet critical to the platform [3]. The social reputation mechanism is used to motivate and encourage positive behaviors. We can also use the same strategy to penalize the social reputation of those who act destructively by issuing penalties to those users’ accounts. Just as rewards incentivize contributions, penalties visible to the public and the communities may discourage inappropriate behaviors. In sum, alternative approaches can include but not limited to the following: a more rigorous content moderation algorithm, semi-automated reviewing systems for moderators employed by Facebook and volunteered from the users, reward and penalty system associated with the pseudonyms visible to the public.
References
[1] Danah Boyd. Social network sites as networked publics: Affordances, dynamics, and implications. A networked self: Identity, community, and culture on social network sites, 01 2010.
[2] R. Stuart Geiger and David Ribes. The work of sustaining order in wikipedia: The banning of a vandal. In Proceedings of the 2010 ACM Conference on Computer Supported Cooperative Work, CSCW ’10, pages 117–126, New York, NY, USA, 2010. ACM.
[3] Travis Kriplean, Ivan Beschastnikh, and David W. McDonald. Articulations of wikiwork: uncovering valued work in wikipedia through barnstars. 01 2008.
[4] Zeynep Tufekci. Twitter and Tear Gas: The Power and Fragility of Networked Protest. Yale University Press, New Haven, CT, USA, 2017
This post is written when I was taking the class “Social Issues of Information” at UC Berkeley.
back to note / back to all posts
Truckers and Electronic Surveillance
Are Truck Drivers’ Complaints Valid?
In December, 2017 a new Electronic Logging Device (ELD) rule was enforced on truck drivers. This new rule requires drivers to digitally track the hours they work. The so-called electronic logging device (ELD) is a “flash drive that plugs into a truck engine control module to track things like whether the engine is running, the odometer, GPS location, and so on” [5]. With the ELD, the regulators hope to gain more control over illegal, overtime driving and prevent drivers from falsifying the records, a common situation for the old pen and paper system. However, the enforcement irritated the drivers. The main concern that the truckers and the Owner-Operator Independent Drivers Association have raised is the inability of the device to account for frequent unexpected situations.
The truckers complain that they have little control over the schedule, the weather, the traffic, and the parking space, etc. None of the situations above can be reflected in the tracked data, yet they influence the amount of hours logged. The tracked data can only reflect the hours that the engine is running, the distance traveled, the location or maybe the trajectory of the trip. Thus the numeric system and sometimes ill-represented tracked data can be unfair and ineffective. With the old pen and paper system, the uncontrollable circumstances can be factored into the records by “cheating the logbook” as a work-around. With the ELD, there seems to be no effective way to reflect their actual working conditions. Uber and Lyft’s drivers have faced the same challenges posted by data-driven evaluations, the passenger-driver rating system. The numeric metrics can sometimes be an inaccurate measure of the services quality as passengers misattribute situations such as system faults over which the drivers have no control to the ratings [4].
Can Some of the Problems be Solved?
The change from a logbook to a tracking device is not sufficient enough for the regulators to gain their desired benefit from the new surveillance system. As pointed out in Beyond the Productivity Paradox, “technology is only one components of an IT investment; there are usually large expenditures on training, process redesign and other organizational changes accompanying a systems investment” [2]. In order to improve the situation, we focus our attention on the ELD data collection process with the help of implicit interaction framework. The thesis of the framework is to study the human-human interaction of a situation or a series of actions involved, and then translate the interactions to the appropriate human-system interactions [3].
The framework characterizes the space of interactions along the dimensions of attentions demand and initiative. On the attention demand axis, the tracked data that are currently being collected are mostly background information about the vehicle such as the distance traveled, the engine running-hours. The lack of foreground information about the truckers, the traffic condition, and the other uncontrollable factors has been a major issue of the regulation. To alleviate this major problem, one can instruct the device to explicitly collect these foreground data.
Next, we turn to the initiative axis. Should the system be proactive or reactive? Clearly, having a device asking for various foreground information constantly is impractical. But neither is having the truckers manually labeling what happened for a several-hours trip data everyday. A simple plug-in flash drive is evidently not enough to account for the task and a more sophisticated logging device is required. In addition to the presumed information, the system also needs a mechanism for the drivers to report any additional uncontrollable factors.
Will Self-driving Truck Replace The Drivers Soon?
Given the current development of self-driving cars, one may want to argue that truck drivers may soon be automated out the job. To fully automate the job, the system must support decision making for various situations that a driver can handle independently. The system will be required to handle smoothly some nuanced and contextualized tasks such handing accidents, interacting with other drivers, and reporting damage or lost, etc. Trips involving complicated conditions such as traffic jams and navigating in a city require social interactions between the drivers and other drivers or the clients. To argue that drivers can be replaced by the system, one must demonstrate the ability of the system to handle all situations that a driver may encounter. However, the current state of technology can only demonstrate self-driving cars given a simple environment without interacting with other drivers or clients. From Ackerman’s [1] point of view, this divide between what we know the fully automated self-driving truck system must support socially and what the current self-driving system can support technically is the social-technical gap. It is unlikely that the drivers will be automated soon given a high complexity of their job involved with interacting with various stakeholders and handling unexpected situations on the road.
References
[1] Mark S. Ackerman. The intellectual challenge of cscw: The gap between social requirements and technical feasibility. Human–Computer Interaction, 15(2-3):179–203, 2000.
[2] Erik Brynjolfsson and Lorin M. Hitt. Beyond the productivity paradox: Computers are the catalyst for bigger changes. Communications of the ACM, 1998.
[3] Wendy Ju and Larry Leifer. The design of implicit interactions: Making interactive systems less obnoxious. Design Issues, 24:72–84, 2008.
[4] Min Kyung Lee, Daniel Kusbit, Evan Metsky, and Laura Dabbish. Working with machines: The impact of algorithmic and data-driven management on human workers. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, CHI ’15, pages 1603–1612, New York, NY, USA, 2015. ACM.
[5] Nick Stockton. Truckers take on trump over electronic surveillance rules. 2018.