Discuss various matters and issues related to conducting a statistical survey – Notes

Section 1 & 2: Data Sources (Primary vs. Secondary Data)

第 1 及 2 節:數據來源(一手數據對比二手數據)

  • Primary Data: First-hand data gathered directly for a specific investigation.
  • 一手數據(Primary Data): 為特定統計調查目的而親自直接收集的原始數據。
  • Secondary Data: Pre-existing data collected by an outside entity and reused for another purpose.
  • 二手數據(Secondary Data): 由其他機構或人士為其他目的而收集,其後被轉移作新用途的現成數據。

Example 1: Local Transit & Octopus Card Balances

範例 1:本地交通與八達通餘額

  • Scenario: An IVE Engineering student wants to investigate the average balance left on adult Octopus cards by commuters exiting Admiralty MTR station during the morning peak hour.
  • 情境: 一名 IVE 工程系學生希望調查早上繁忙時間在金鐘港鐵站出閘時,上班族成人八達通內的平均剩餘金額。
  • Primary Data: The student stands near the turnstiles at Admiralty station and surveys willing commuters directly, recording their exact card balances on a log sheet.
  • 一手數據: 該學生站在金鐘站出閘機旁,直接向願意受訪的乘客進行調查,並在記錄表記錄他們準確的八達通餘額。
  • Secondary Data: The student uses aggregated quarterly smart-card balance reports published on data.gov.hk by the Hong Kong Transport Department.
  • 二手數據: 該學生使用香港運輸署在 data.gov.hk 發佈的季度電子儲值卡綜合統計報告。
  • Analysis of Trade-offs:
  • 權衡分析:
  • Secondary Advantage: Instant access, zero financial cost, and covers thousands of data points across the whole territory without field work.
  • 二手數據的優點: 可即時獲取、零金錢成本,且無需實地調查即可涵蓋全港數千甚至數萬個數據點。
  • Secondary Disadvantage: The public dataset reports general territory-wide figures and does not isolate Admiralty station commuters at peak rush hour.
  • 二手數據的缺點: 政府公開數據集只呈報全港整體宏觀數字,無法特別細分出金鐘站在早上繁忙時間出閘的特定乘客群。

Example 2: Mobile Fantasy RPG Server Drop Rates

範例 2:手機奇幻角色扮演遊戲的伺服器抽卡掉落率

  • Scenario: A student club analyzes whether a limited-edition anime avatar banner in a mobile RPG has a true rare item drop rate of 1.5%.
  • 情境: 一個學生學會正在分析某款手機奇幻 RPG 遊戲中的限定動漫造型卡池,其稀有道具的實際掉落率是否真的為 1.5%。
  • Primary Data: The club pools in-game summoning crystals and records 500 live banner pulls to log the exact outcomes directly.
  • 一手數據: 該學會集中遊戲內的召喚水晶,親自實測抽取 500 次卡池,並將每次抽取的真實結果直接記錄下來。
  • Secondary Data: The club retrieves the published drop-rate table provided by the game developers on the game’s official website.
  • 二手數據: 該學會直接從遊戲官方網站上下載遊戲開發商公開發佈的機率掉落表。
  • Analysis of Trade-offs:
  • 權衡分析:
  • Secondary Advantage: Readily available with no need to spend play hours or premium in-game currency.
  • 二手數據的優點: 現成可用,完全無需花費遊戲時間或課金購買召喚水晶。
  • Secondary Disadvantage: Secondary publisher data assumes the back-end code runs without bugs; it may not reflect the live observed server performance.
  • 二手數據的缺點: 發行商的數據假設伺服器後台程式碼運作完全沒有程式錯誤(Bug),可能無法反映遊戲伺服器現場實測的真實表現。

Section 3 & 4: Data Collection Methods

第 3 及 4 節:數據收集方法

Choosing an appropriate collection method depends on accuracy requirements, budget, and accessibility. 選擇合適的數據收集方法取決於對精確度的要求、預算以及可接觸性。

Overview Comparison Table

總覽比較表

Collection Method / 收集方法Context Scenario / 應用情境Key Advantage / 主要優點Key Disadvantage / 主要缺點
Direct Observation


直接觀察法
Counting the number of double-decker buses crossing the Cross-Harbour Tunnel during rush hour.

在繁忙時間點算通過紅磡海底隧道的雙層巴士數量。
Captures actual behavior in real-time with zero disturbance to drivers.

能實時記錄實際行為,而且完全不會干擾司機。
Cannot capture underlying reasons, opinions, or route preferences.

無法獲知背後原因、個人意見或路線偏好。
Personal Interview


個人面談訪問
Speaking with competitive esports players at an arcade in Mong Kok about control schemes.

在旺角遊戲機中心訪問電競選手對操控配置的看法。
Allows deep exploration of motives, expressions, and immediate follow-ups.

可深入探討動機,觀察表情反應,並即時跟進提問。
High time investment and potential respondent bias.

耗費大量時間,且可能產生受訪者偏見(如為了面子而誇大)。
Telephone Interview


電話訪問
Surveying local residents about municipal recycling habits across 18 districts.

就全港 18 區的都市廢物回收習慣向居民進行電話調查。
Broad geographical coverage and fast turnarounds.

地理涵蓋面廣闊,能在短時間內收回大量數據。
High rejection rate and restricted to short, basic questions.

拒訪率高,且只能局限於簡短且基礎的題目。
Experiment


實驗法
Testing battery drain on a new handheld gaming console under varying frame rates.

在不同螢幕更新率(fps)下測試新款掌上遊戲機的耗電量。
Provides objective, precise measurement under controlled parameters.

能在嚴格控制的實驗參數下提供客觀且精準的量度結果。
Expensive setup; cannot directly measure user satisfaction.

設備成本昂貴;無法直接量度玩家的主觀滿意度。
Questionnaire


問卷調查
Distributing a digital feedback form to visitors at an anime convention at the HKCEC.

於香港會議展覽中心舉辦的動漫電玩節向參觀者派發電子意見表。
Reaches hundreds of respondents simultaneously at low cost.

成本低廉,能同一時間收集數百名受訪者的回覆。
Prone to low completion rates and missing answers.

回收率或完成率通常較低,容易出現漏答情況。

Example 1: Direct Observation vs. Personal Interview

範例 1:直接觀察法對比個人面談訪問

  • Direct Observation (Tsim Sha Tsui Luggage Flow):
  • 直接觀察法(尖沙咀旅客行李流動):
    • Implementation: An observer counts the number of visitors wheeling large suitcases out of the Hong Kong West Kowloon High Speed Rail Station during a 2-hour window.
    • 實施方式: 調查員在香港西九龍高鐵站出口,於 2 小時的時段內點算拉著大型行李箱出站的乘客人數。
    • Pros: Captures real behavior accurately with zero disturbance to travelers.
    • 優點: 真實且準確地記錄旅客行為,完全不會耽誤旅客行程。
    • Cons: Cannot capture traveler motivations, destinations, or spending plans.
    • 缺點: 無法得知旅客的訪港目的、前往的目的地或預計消費預算。
  • Personal Interview (Hong Kong Comic Festival):
  • 個人面談訪問(香港動漫電玩節):
    • Implementation: An interviewer approaches attendees queuing at the Hong Kong Convention and Exhibition Centre to ask in-depth questions regarding their monthly budget for limited-edition anime merchandise.
    • 實施方式: 調查員親自接觸在香港會議展覽中心排隊入場的動漫迷,深入訪談他們每個月購買限量版動漫周邊的預算分配。
    • Pros: High response rate and allows probing complex buying motivations directly.
    • 優點: 回應率高,且調查員能因應受訪者的回答深入追問複雜的消費動機。
    • Cons: Time-consuming, geographically restricted to the hall entrance, and prone to prestige bias where respondents exaggerate their spending.
    • 缺點: 極花時間、調查地點僅局限於展館入口,且受訪者可能因虛榮感而誇大自己的消費金額。

Example 2: Telephone Interview vs. Experiment

範例 2:電話訪問對比實驗法

  • Telephone Interview (Local District Council Public Sentiment):
  • 電話訪問(觀塘區公共交通民意調查):
    • Implementation: A research assistant dials landline numbers across Kwun Tong to ask if residents support new bus routes along Anderson Road.
    • 實施方式: 研究助理隨機撥打觀塘區住宅固網電話,詢問居民是否支持安達臣道發展區新增巴士專線。
    • Pros: Reaches a broad demographic quickly without sending staff door-to-door.
    • 優點: 能迅速接觸廣泛的居民群體,無需派員逐家逐戶拍門。
    • Cons: High rejection rate; questions must remain short and binary (e.g., Yes/No).
    • 缺點: 拒絕率非常高;題目必須保持極度簡短,大多只能設置是非題(例如:贊成/不贊成)。
  • Experiment (Esports Monitor Response Times):
  • 實驗法(電競螢幕反應時間測試):
    • Implementation: An IVE IT student sets up three gaming monitors under standardized lab conditions, logging frame-latency values using high-speed optical sensors across 100 benchmark trials.
    • 實施方式: 一名 IVE 資訊科技系學生在標準化的實驗室環境下架設三款電競螢幕,利用高速光學傳感器進行 100 次基準測試以記錄畫面延遲數據。
    • Pros: High experimental control, eliminating human bias completely.
    • 優點: 實驗環境控制高度嚴謹,能徹底排除人為主觀偏見。
    • Cons: Costly specialized equipment required; does not capture subjective player preference.
    • 缺點: 需要昂貴的專業量測儀器;且無法直接反映玩家在實際遊戲操作時的主觀手感與偏好。

Section 5, 6 & 7: Questionnaire Design & Layout

第 5、6 及 7 節:問卷設計原則與版面佈局

High-quality questionnaires avoid ambiguity, double questions, loaded language, and unstated assumptions. 一份優質的問卷必須避免模稜兩可、一題兩問、引導性字眼以及預設立場。

Example 1: Fixing Ambiguity, Double Negatives, and Calculations

範例 1:修正含糊不清、雙重否定及運算回憶問題

  • Defective Question 1 (Calculation & Memory Recall):
  • 有缺陷的問題 1(強制計算與記憶回憶):

“How many total Hong Kong dollars have you spent on convenience store snacks and MTR fares combined over the last 180 days?”

「請計算在過去 180 天內,你在便利店零食與港鐵車費上的總開銷是多少港幣?」

  • Issue: Forces heavy mental arithmetic and impossible memory recall.
  • 問題點: 強迫受訪者進行繁瑣的心算,且要求回憶長達半年的細碎支出,根本不可能準確回答。
  • Improved Revision:
  • 改善後的設計:

“In the past 7 days, how much did you spend on public transport?

「在過去 7 天內,你在公共交通工具上的總開銷大約是多少港幣?

  •  Under HK$100 / 少於 HK$100
  •  HK$101–HK$200 / HK$101 至 HK$200
  •  HK$201–HK$300 / HK$201 至 HK$300
  •  Over HK$300″ / HK$300 以上」
  • Defective Question 2 (Double Negative & Ambiguity):
  • 有缺陷的問題 2(雙重否定與模稜兩可):

“Do you not disagree that local retail discounts are rarely unhelpful?”

「你難道不同意本地零售優惠很少是沒有幫助的嗎?」

  • Issue: Double negatives cause confusion and ambiguous interpretation.
  • 問題點: 句子包含雙重否定,令受訪者思維混亂,答案難以準確解讀。
  • Improved Revision:
  • 改善後的設計:

“To what extent do you find local retail discounts helpful?

「你認為本地零售店提供的折扣優惠對你的生活有幫助嗎?

  •  Very Helpful / 非常有幫助
  •  Helpful / 有幫助
  •  Neutral / 普通 / 無意見
  •  Unhelpful / 沒有幫助
  •  Very Unhelpful” / 完全沒有幫助」

Example 2: Eliminating Leading Phrasing and Jargon

範例 2:剔除引導性用詞及專業黑話

  • Defective Question 1 (Leading & Loaded Language):
  • 有缺陷的問題 1(帶有偏見與強烈情緒的引導性問題):

“Wouldn’t you agree that spending five hours grinding daily quests in mobile games is an absolute waste of productive time?”

「你難道不認同在手機遊戲花 5 個小時刷日常任務,完全是浪費生產力時間的行為嗎?」

  • Issue: Leading wording steers the respondent toward a specific value judgment.
  • 問題點: 帶有情緒化用詞,帶有先入為主的道德批判,誘導受訪者作出負面選擇。
  • Improved Revision:
  • 改善後的設計:

“On average, how many hours do you spend on daily game tasks each day?

「你平均每天花多少時間完成手機遊戲的日常任務?

  •  Less than 1 hour / 少於 1 小時
  •  1–2 hours / 1 至 2 小時
  •  3–4 hours / 3 至 4 小時
  •  5 hours or more” / 5 小時或以上」
  • Defective Question 2 (Unexplained Acronym & Forced Open End):
  • 有缺陷的問題 2(未說明的英文縮寫與強制填寫長篇問答):

“State your current monthly DPS output and give a detailed paragraph explaining your stat allocation choices.”

「請寫出你每月的輸出 DPS,並寫一段詳細文字解釋你的屬性點分配理由。」

  • Issue: Contains unexplained specialized gaming terminology (Damage Per Second) and an open-ended essay prompt that lowers completion rates.
  • 問題點: 包含行外人不理解的遊戲專業術語(每秒傷害輸出),且要求開放式填寫文章,會大幅降低填寫意願。
  • Improved Revision:
  • 改善後的設計:

“Which primary combat attribute does your main game character focus on?

「你的主要遊戲角色主要偏重哪一項戰鬥屬性?

  •  Physical Attack / 物理攻擊
  •  Magic Attack / 魔法攻擊
  •  Defense / Support / 防禦或輔助
  •  Agility / Speed” / 敏捷或速度」

Section 8: Population and Sample

第 8 節:總體與樣本

  • Population (NN): All subjects possessing a common characteristic that is being studied.
  • 總體(Population, NN): 具有正在被研究的共同特徵之所有研究對象的集合。
  • Sample (nn): A representative subset or portion selected from the population.
  • 樣本(Sample, nn): 從總體中選取出來進行實際調查的代表性子集或一部分。

Example 1: IVE Canteen Meal Preferences

範例 1:IVE 學生飯堂午餐偏好調查

  • Scenario: The student union wants to assess whether the Chai Wan campus canteen should introduce more plant-based meal sets.
  • 情境: 學生會希望評估柴灣分校的學生飯堂是否應該推出更多素食健康套餐。
  • Target Population (NN): All full-time and part-time students currently enrolled at the IVE Chai Wan campus.
  • 目標總體(NN): 目前在 IVE 柴灣分校註冊就讀的所有全日制及兼讀制學生。
  • Target Sample (nn): 300 enrolled students selected to answer the lunch evaluation survey.
  • 目標樣本(nn): 從全體學生中抽樣選出並回答午餐評估問卷的 300 名在讀學生。

Example 2: Mobile RPG Guild Level Progression

範例 2:手機 RPG 公會等級進度研究

  • Scenario: A guild analyst wants to calculate the mean character level across a fantasy server to determine end-game raid tuning.
  • 情境: 一名公會數據分析師想計算伺服器內玩家角色的平均等級,以決定高難度公會副本的強度調整。
  • Target Population (NN): All 45,000 active player characters created on the Hong Kong regional server.
  • 目標總體(NN): 在香港伺服器上建立的全部 45,000 個活躍玩家角色。
  • Target Sample (nn): 500 characters chosen across various factions whose profiles are evaluated.
  • 目標樣本(nn): 從各大陣營中抽選出並用作分析數值檔案的 500 個代表角色。

Section 9 & 10: Probability Sampling Methods

第 9 及 10 節:概率抽樣方法

Every item in probability sampling has a known, non-zero probability of being selected, which eliminates selection bias. 在概率抽樣中,總體內的每一個單位都有一個明確已知且非零的被選中概率,能客觀消除挑選偏差。

1. Simple Random Sampling (SRS) via Random Number Table

1. 簡單隨機抽樣(利用隨機數字表)

Sampling in which every subject has an equal chance of being chosen. 每個單位均有均等的被選取機會。

Table Segment / 隨機數字表節錄:
Row 1:  7  3  1  8  0  4  9  2  1  5  0  8  3  3  7  1
Row 2:  0  2  4  1  2  9  8  5  3  1  4  0  7  6  2  0

Worked Exercise (Guild Squad Selection): 逐步運算範例(公會小隊長抽選):

A guild master selects n=3n=3 squad leaders out of N=40N=40 members labeled 0101 to 4040. Starting from Row 1, Column 1, read two-digit numbers horizontally. 公會會長需從編號為 0101 至 4040 的 4040 名會員中,抽選出 n=3n=3 名小隊長。從第 1 行、第 1 列開始,橫向讀取兩位數字:

  1. Determine required digits: Since the population is 4040, 2-digit numbers are required (0101 to 4040).
  • 確認數字位數: 由於總體為 4040 人,因此需要兩位數編碼(0101 至 4040)。
  1. Extract numbers sequentially:
  • 依序讀取數字:
    • 7373 (Out of range >40>40, skip)
    • 7373(超出編號範圍 >40>40,跳過)
    • 1818 (Selected 1)
    • 1818(選中第 1 位)
    • 0404 (Selected 2)
    • 0404(選中第 2 位)
    • 9292 (Out of range >40>40, skip)
    • 9292(超出編號範圍 >40>40,跳過)
    • 1515 (Selected 3)
    • 1515(選中第 3 位)
  1. Conclusion: Selected members are 18, 04, and 15.
  • 結論: 選中的會員編號為 18、04 以及 15。

2. Systematic Sampling

2. 系統隨機抽樣

Data is obtained by selecting every k-thk-th object after a random starting point: 先隨機選定一個起點,然後每隔固定的第 kk 個單位抽取一個樣本:

Sampling Interval (k)=Population Size (N)Sample Size (n)Sampling Interval (k)=Sample Size (n)Population Size (N)​

抽樣間距 (k)=總體大小 (N)樣本大小 (n)抽樣間距 (k)=樣本大小 (n)總體大小 (N)​

Hong Kong Retail Scenario: 香港零售實景範例:

An electronics boutique in Sham Shui Po audits n=30n=30 customers from N=600N=600 Saturday shoppers. 深水埗一間電腦零售店打算在星期六的 N=600N=600 名進店顧客中,抽取 n=30n=30 名進行售後保養滿意度調查。

  1. Calculate the interval:
  • 計算間距 kk:

k=60030=20k=30600​=20

  1. Execute selection: If customer 77 is randomly picked first, subsequent customers are:
  • 執行抽樣: 若隨機抽中的第一個號碼為 7 號,後續選出的顧客編號為:

{7,27,47,67,87,…,587}{7,27,47,67,87,…,587}

  1. Methodological Note: Ensure the interval doesn’t match store flash sales to avoid periodicity bias.
  • 方法學注意: 確保抽樣間距不會剛好撞上店舖每小時的限時特賣時段,以防止出現週期性偏差。

3. Stratified Sampling

3. 分層隨機抽樣

The population is split into groups (strata) sharing common traits, and proportional samples are drawn from each. 將總體按照特定特徵分成若干組別(層),然後按各層比例抽取樣本。

Sample from Stratum=n×Stratum SizeTotal Population SizeSample from Stratum=n×Total Population SizeStratum Size​

某層的應抽樣本數=n×該層所屬總數總體總數某層的應抽樣本數=n×總體總數該層所屬總數​

IVE Departmental Guild Scenario: IVE 學科公會範例:

An IVE Gaming Society with N=1,500N=1,500 students needs a focus sample of n=50n=50: 一個擁有 N=1,500N=1,500 名會員的 IVE 電競學會計劃抽樣 n=50n=50 名學生進行焦點訪談:

  • Warriors (Engineering / 工程系): 600600 members
  • Mages (IT / 資訊科技系): 450450 members
  • Healers (Health & Life Sciences / 健康及生命科學系): 450450 members
  1. Calculate proportional allocations:
  • 計算各學科應選人數:

Warriors (Engineering)=50×6001500=50×0.40=20Warriors (Engineering)=50×1500600​=50×0.40=20

Mages (IT)=50×4501500=50×0.30=15Mages (IT)=50×1500450​=50×0.30=15

Healers (Health & Life Sciences)=50×4501500=50×0.30=15Healers (Health & Life Sciences)=50×1500450​=50×0.30=15

  1. Verify sum:
  • 驗證總數:

20+15+15=50(Correct / 正確)20+15+15=50(Correct / 正確)


Section 11: Non-Probability Sampling

第 11 節:非概率抽樣方法

Units are selected based on accessibility or subjective discretion rather than defined random chances. 樣本並非按隨機概率產生,而是根據便利性、接觸機會或主觀專業判斷而選定。

Example 1: Judgement Sampling

範例 1:判斷抽樣法

  • Scenario A (Retail Store Placement): A bubble tea brand selects two specific branches in Mong Kok and Causeway Bay to trial a new drink because the marketing director judges that these specific districts attract the trendiest youth.
  • 情境 A(零售選址試銷): 某連鎖珍珠奶茶品牌挑選了旺角和銅鑼灣兩間分店試銷新款特飲,因為市場部主管憑專業經驗判斷這兩個核心商圈最能吸引追求潮流的年青人。
  • Scenario B (Esports Team Tryout): An esports coach picks 5 veteran players for a new patch trial based purely on observed adaptability and competitive rank.
  • 情境 B(電競戰隊挑選): 電競教練直接挑選 5 名資深選手參與新遊戲版本的平衡性封測,完全基於教練對選手反應能力與比賽段位的主觀專業判斷。
  • Evaluation: Relies on expert knowledge and can be set up fast, but cannot calculate formal statistical confidence intervals.
  • 評估: 依賴專家經驗且設定迅速,但無法計算客觀的統計誤差邊界(Sampling Error)。

Example 2: Quota Sampling

範例 2:配額抽樣法

  • Scenario A (MTR Station Exit Survey): A surveyor outside Sha Tin station is instructed to interview exactly 30 adult and 30 student Octopus holders, approaching whoever seems friendly.
  • 情境 A(港鐵站外調查): 調查員在沙田站外被要求必須訪問 30 位成人八達通持有人及 30 位學生八達通持有人,調查員會在現場挑選看起來面善且不趕時間的途人進行訪問。
  • Scenario B (Gaming Cafe Feedback): A lounge manager sets a quota of 20 PC gamers and 20 console gamers, leaving staff to select players resting at the service counter.
  • 情境 B(網吧意見調查): 網吧主管設定了必須收集 20 名 PC 玩家及 20 名家用主機玩家意見的配額,前線員工因此直接找坐在水吧櫃台旁休息的顧客填表。
  • Evaluation: Guarantees proportional representation across targets without a registry, but surveyor selection choices introduce human bias.
  • 評估: 無需完整名冊即可滿足特定族群的樣本配額比例,但訪員挑選對象時的主觀便利性會引入人為偏差。

Let’s do a quick check to see if you have mastered this: If the IVE Student Union wants to survey 100 students about the campus canteen, and they divide students by department (Engineering, IT, Business) to make sure each group is represented properly, which sampling method are they using? Share your thoughts and tell me why you chose that method! 現在考考大家是否已經完全掌握:如果 IVE 學生會想調查 100 名學生對飯堂的滿意度,並先將學生按學科(工程系、資訊科技系、工商管理系)分組,再按比例抽選學生,這屬於哪種抽樣方法?請分享你的答案和分析理由!

Leave a Reply