MAST20034 · Critical Thinking with DataMAST20034 · 数据批判性思维（统计素养）
A statistical-literacy subject — argue from data, not compute with it.一门统计素养课：用数据论证，而不是套公式算数。MAST20034 is a School of Mathematics and Statistics subject that trains you to be a critical interpreter and user of data-based evidence — judging the strengths and weaknesses of quantitative arguments, evaluating probabilistic and statistical reasoning, and reporting findings honestly. It is assessed mostly through coursework (regular online quizzes plus short written assignments) and a short-answer exam where you argue a position about data rather than crunch numbers. This guide is built from 74 real MAST20034 course materials in the AskSia Library.
MAST20034 由墨大数学与统计学系开设，训练你成为「数据证据的批判性使用者」：判断量化论证的强弱、评估概率与统计推理是否站得住脚、并诚实地报告结论。它不是湿实验生物课，也不是公式计算课——核心是**对数据下判断、写论证**。考核以平时作业为主（定期在线小测 + 短篇写作作业）外加一场简答论证式期末考。本指南基于AskSia Library 中 74 份真实 MAST20034 课程材料整理而成。
Built from 74 real MAST20034 course materials in the AskSia Library — covering the W1–W12 critical-thinking syllabus (critique vs criticism, study design, confounding, sampling, NHST/CI interpretation, Bradford Hill, meta-analysis) and the short-answer, no-calculation exam style.
基于AskSia Library 中 74 份真实 MAST20034 课程材料整理——覆盖第 1–12 周的批判性思维大纲（批判与指责之别、研究设计、混杂、抽样、假设检验/置信区间的解读、Bradford Hill 因果判据、元分析），以及「简答、不计算」的考试风格。
What MAST20034 is aboutMAST20034 讲什么
MAST20034 Critical Thinking with Data is a 12.5-point, second-year subject offered by the University of Melbourne's School of Mathematics and Statistics (Faculty of Science). It is a statistical-literacy and quantitative-reasoning subject: rather than computing confidence intervals or running hypothesis tests by hand, students learn to think critically about how data is collected, presented and used as evidence. The subject covers identifying the strengths and weaknesses of arguments built on quantitative evidence, evaluating probabilistic and statistical reasoning, recognising bias and confounding, and communicating statistical findings in a principled, honest way. It is available in both Semester 1 and Semester 2, and is also offered as breadth (UNIB10006).
MAST20034《数据批判性思维》是墨尔本大学数学与统计学系（理学院）开设的二年级 12.5 学分课程。它是一门**统计素养 / 量化推理**课，重点不是手算置信区间或做假设检验，而是训练学生批判性地看待「数据如何被收集、呈现、并当作证据使用」。课程内容包括：识别基于量化证据的论证强弱、评估概率与统计推理是否可靠、识别偏倚与混杂、以及有原则、诚实地沟通统计结论。该课在第一、第二学期均开设，并以通识 breadth 形式（UNIB10006）面向全校。
The MAST20034 syllabus, topic by topicMAST20034 大纲 · 逐个主题
Data as evidence数据作为证据
What it means to treat data as evidence for a claim. Introduces the idea that numbers do not 'speak for themselves' and must be interpreted in context.
理解「把数据当作某个论断的证据」意味着什么。核心观念：数字不会自己说话，必须放在语境中解读。
Collecting data: study design数据的收集：研究设计
How data is generated — observational studies vs experiments, sampling, randomisation and controls — and how design choices shape what conclusions are defensible. The course frames every design choice through two independent goals: validity (reducing bias / systematic error, via randomisation, comparison and control/blinding) and precision (reducing variability / random error, via replication, stratification-blocking and balanced group sizes). A key examinable trap: a bigger sample fixes precision but never fixes bias — you can be precisely wrong. Design also has to answer to ethics (consent, fairness, intention, integrity, stewardship) and the three justice lenses (substantive, procedural, distributive).
数据是怎么产生的——观察性研究与实验、抽样、随机化与对照——以及设计上的选择如何决定哪些结论站得住脚。本课用两个相互独立的目标来贯穿所有设计选择：效度（reducing bias，即减少系统误差，靠随机化、对照比较、控制/盲法）与精度（reducing variability，即减少随机误差，靠重复、分层/区组、各组样本量均衡）。一个高频考点陷阱：加大样本只能改善精度，永远修不了偏倚——你可能「精确地错」。设计还要对伦理（consent、fairness、intention、integrity、stewardship 五原则）与三种正义视角（substantive、procedural、distributive）负责。
Bias, confounding and the limits of evidence偏倚、混杂与证据的边界
Identifying selection bias, confounding variables and other threats that make a quantitative argument weaker than it looks. A confounder is a third variable associated with BOTH the exposure (potential cause) and the outcome (effect), distorting their apparent relationship — picture the confounding triangle: exposure → outcome, with the confounder pointing at both. An unmeasured confounder is a lurking variable. Because observational studies cannot randomise away these third factors, confounding is always a live alternative explanation — which is exactly why a tutorial-style task asks you to name the exposure, the outcome, and a plausible confounder (justifying both links), then conclude that the association is not, on its own, evidence of cause.
识别选择偏倚、混杂变量等让量化论证「看起来比实际更强」的问题。混杂变量（confounder）是同时与「暴露/潜在原因」和「结局/效应」相关的第三个变量，会扭曲二者表面上的关系——想象混杂三角：暴露 → 结局，而混杂因素同时指向两者。未被测量的混杂因素就是潜伏变量（lurking variable）。由于观察性研究无法靠随机化消除这些第三因素，混杂始终是一个站得住脚的替代解释——这也正是教程题会要求你指出暴露、结局、以及一个合理的混杂因素（并论证它与两者各自的关联），再得出「关联本身不等于因果」的原因。
Correlation, causation and association相关、因果与关联
Why a correlation is not a cause, what evidence is needed to argue for causation, and how this distinction is routinely abused in reporting. The course's tool for arguing cause from (mostly observational) evidence is the Bradford Hill criteria — strength, consistency, specificity, temporality (cause must precede effect — the key one), biological gradient (dose-response), plausibility, coherence, experiment and analogy. Crucially these are a weight-of-evidence argument, not a tick-box checklist to 'pass'; a randomised controlled trial strengthens a causal claim by removing confounders, when running one is ethical and feasible.
为什么相关不等于因果、要论证因果需要什么样的证据，以及这一区分在媒体报道中如何被滥用。本课用来「从（多为观察性的）证据论证因果」的工具是 Bradford Hill 判据——强度、一致性、特异性、时序性（原因必须先于结果，最关键的一条）、剂量-反应梯度、合理性、一致性（与既有知识吻合）、实验支持、类比。关键在于：这是一套「证据权重」论证，而不是要逐项「打勾通过」的清单；随机对照试验（RCT）能通过消除混杂来加强因果论断——前提是开展它在伦理与现实上可行。
Probabilistic reasoning概率推理
Evaluating arguments that rely on probability — risk, uncertainty, conditional probability and common probabilistic fallacies.
评估依赖概率的论证——风险、不确定性、条件概率，以及常见的概率谬误。
Interpreting statistical analysis解读统计分析
Reading and critiquing the results of statistical analyses (e.g. estimates, intervals, p-values) as a consumer of evidence, focusing on what they do and do not justify.
作为「证据的消费者」去阅读和批判统计分析结果（如估计值、区间、p 值），重点在于它们能证明什么、不能证明什么。
Misleading data and visualisation误导性数据与可视化
How charts, summaries and selective reporting can mislead, and how to spot and counter these distortions.
图表、汇总与选择性报告如何误导人，以及如何识别并反驳这些扭曲。
Drawing defensible conclusions得出站得住脚的结论
Moving from evidence to a conclusion that is appropriately hedged — matching the strength of the claim to the strength of the data.
从证据走向结论，并恰当地保留余地——让论断的强度与数据的强度相匹配。
Principled reporting and communication有原则的报告与沟通
Communicating quantitative findings clearly and honestly to a non-specialist audience, including ethical reporting of evidence.
向非专业读者清晰、诚实地沟通量化发现，包括有伦理地报告证据。
Critical reading of quantitative claims批判性阅读量化论断
Applying the whole toolkit to real reports, news articles and studies — building and defending a critique of a published quantitative argument.
把整套工具用到真实报告、新闻和研究上——对已发表的量化论证构建并捍卫一份批判。
How MAST20034 is assessedMAST20034 怎么考核
Final exam: Yes期末考试：有| Component考核项 | Weight占比 | Note说明 |
|---|---|---|
| Regular online revision quizzes (multiple, across the semester)学期内多次在线复习小测 | coursework | A series of short online quizzes spread roughly fortnightly through the teaching period, checking ongoing understanding of the concepts.学期内大约每两周一次的短在线小测，持续检查对概念的理解。 |
| Short written assignments (several, each up to roughly 800 words)若干短篇写作作业（每篇约 800 字以内） | coursework | Written tasks where you build and defend an argument about a piece of data or a study — argumentation, not computation.写作型作业：针对某组数据或某项研究构建并捍卫一个论证——重点是论证而非计算。 |
| Short-answer final exam (argumentative)简答式期末考（论证型） | exam | An end-of-semester short-answer exam where you argue a position about data rather than calculate statistics. Confirm the exact weight on the handbook assessment page.学期末的简答考试，要求你就数据问题表明并论证立场，而不是做统计计算。具体占分以 handbook 评估页为准。 |
Coursework-heavy: regular online quizzes plus several short written assignments (each ~800 words), finished with a short-answer final exam that asks you to argue from data rather than compute. Confirm exact percentage weights on the handbook assessment page.
以平时分为主：定期在线小测 + 若干短篇写作作业（每篇约 800 字），最后是一场简答论证式期末考（用数据论证、而非计算）。具体百分比权重请以 handbook 评估页为准。
When each MAST20034 task is dueMAST20034 各项考核时间
Test yourself: MAST20034 practice questions自测一下：MAST20034 练习题
- Nothing — any variable with order can be averaged.
- The rating is an ordinal categorical variable; the categories are ordered but the gaps between them are not numerically equal, so a mean is not meaningful — report the median or a bar chart of percentages.
- The rating is nominal, so a median is also forbidden.
- It is fine, but only if there are more than 30 respondents.
- 没问题——只要变量有顺序就能取平均。
- 评分是有序（ordinal）分类变量：类别有先后，但相邻类别之间的「间距」并不数值相等，所以取均值没有意义——应报告中位数或各类别百分比的条形图。
- 评分是名义（nominal）变量，所以连中位数也不能用。
- 没问题，只要受访者超过 30 人即可。
Show answer查看答案
- A larger sample would prove causation.
- Confounding — smoking is plausibly associated with BOTH coffee drinking (the exposure) and lung disease (the outcome), so it can manufacture the association with no causal link from coffee.
- Temporality, because the outcome clearly came first.
- Publication bias in the journal.
- 把样本加大就能证明因果。
- 混杂（confounding）——吸烟很可能同时与「喝咖啡」（暴露）和「肺病」（结局）相关，因此它能在咖啡与肺病之间制造出关联，而咖啡本身并无因果作用。
- 时序性，因为结局明显先发生。
- 期刊的发表偏倚。
Show answer查看答案
- There is a 95% probability the true difference is exactly 1.25.
- p = 0.003 means the null hypothesis is 99.7% likely to be false.
- The interval lies entirely above zero and p is small, so there is strong evidence of a positive difference; the interval gives a range of plausible values for the true difference, while p alone would not convey the effect's size.
- Because no calculator is allowed, you cannot interpret this output at all.
- 真实差值恰好等于 1.25 的概率是 95%。
- p = 0.003 意味着原假设有 99.7% 的可能为假。
- 区间整体位于 0 以上且 p 很小，所以有强证据表明存在正向差异；区间给出真实差值的一段可信取值范围，而单看 p 值无法传达效应的大小。
- 因为不允许用计算器，所以根本无法解读这份输出。
Show answer查看答案
High-value exam questions in MAST20034MAST20034 高频考点 · 考试风格题
A worked MAST20034 problemMAST20034 例题
Signature short-answer: 'Explain convenience sampling and why it's not recommended'代表性简答题：「解释便利抽样，以及为什么不推荐它」
A health team needs to estimate how often adults in a city exercise. To save time, a researcher stands outside a single inner-city gym on a weekday morning and surveys whoever walks past. Explain what kind of sampling this is, and give two reasons it is not recommended for estimating the city-wide figure. (This mirrors the MAST20034 short-answer style: define the concept, then reason — no calculation, no software. Marks are awarded per correct, sufficiently-detailed reason.)
某卫生团队要估计一座城市里成年人锻炼的频率。为了省时间，研究者在一个工作日早晨站在市中心某家健身房门口，对路过的人逐一调查。请说明这属于哪种抽样，并给出两个理由说明：为什么不推荐用它来估计「全城」的指标。（这复刻了 MAST20034 的简答风格：先给概念定义，再讲理由——不计算、不用软件。每个正确且足够具体的理由各得分。）
Step 1 — name the concept precisely: this is convenience sampling — selecting the most easily-accessible units rather than drawing them at random from a proper sampling frame. (A vague 'just picking people' would lose the definition marks.) Step 2 — give two reasons, each as a 'because' that explains a consequence, not just a restatement: (a) it induces sampling bias — gym-goers near an inner-city gym are systematically more active, wealthier and more central than the city's adults, so the estimate is skewed upward; (b) it artificially reduces variability — people sampled this way resemble each other (and the researcher), so the sample understates the real spread across the population. Step 3 — land the course's headline trap: a bigger convenience sample does NOT fix this — taking more people the same biased way just 'repeats the basic mistake on a larger scale'. The fault is the method, not the size; the fix is a random method (e.g. simple random or stratified sampling from a city-wide frame). Note how this reasoning, not any single 'answer', is what earns the marks.
第一步——精确命名概念：这是便利抽样（convenience sampling）——选取最容易接触到的单位，而不是从合适的抽样框中随机抽取。（含糊地说「就是随便找人」会丢掉定义分。）第二步——给两个理由，每个都用「因为」解释后果，而不是把定义换个说法重复：（a）它引入抽样偏倚——市中心健身房附近的人系统性地更爱运动、更富裕、更靠市中心，因此估计会被整体高估；（b）它人为压低了变异性——这样抽到的人彼此相似（也和研究者相似），样本会低估总体真实的离散程度。第三步——点出本课的招牌陷阱：加大便利样本并不能解决问题——用同样有偏的方式多找些人，只是「把根本性的错误放大了一个量级」。问题出在方法，不在样本量；正确做法是改用随机方法（如对全城抽样框做简单随机或分层抽样）。注意：拿分的是这套推理，而不是某个单一「答案」。
MAST20034 glossaryMAST20034 术语表
- Statistical literacy统计素养
- The ability to read, interpret and critically evaluate statistical information presented as evidence.
- 阅读、解读并批判性评估以证据形式呈现的统计信息的能力。
- Confounding variable混杂变量
- A third factor associated with both the supposed cause and effect, distorting an apparent relationship.
- 同时与「原因」和「结果」相关的第三个因素，会扭曲二者表面上的关系。
- Selection bias选择偏倚
- Distortion that arises when the sample studied is not representative of the population a claim is about.
- 当所研究的样本无法代表论断所针对的总体时产生的系统性偏差。
- Correlation vs causation相关 vs 因果
- The principle that an observed association between two variables does not by itself establish that one causes the other.
- 两个变量之间存在关联，并不足以证明其中一个导致了另一个。
- Confidence interval置信区间
- A range of plausible values for an estimate; in this subject the focus is interpreting and critiquing one, not computing it.
- 估计量的一段可信取值范围；本课重点是「解读与批判」一个区间，而不是手算它。
- p-valuep 值
- A measure of how surprising data would be if a null hypothesis held; the course stresses what it does and does not justify.
- 在原假设成立时，数据有多「令人意外」的度量；本课强调它能与不能支持什么结论。
- Observational study vs experiment观察性研究 vs 实验
- Two ways data is generated; experiments with randomisation allow stronger causal claims than observation alone.
- 两种数据产生方式；带随机化的实验比单纯观察能支持更强的因果论断。
- Probabilistic reasoning概率推理
- Reasoning about uncertainty and risk using probability, including spotting common probabilistic fallacies.
- 用概率对不确定性和风险进行推理，包括识别常见的概率谬误。
- Principled reporting有原则的报告
- Communicating quantitative findings clearly, completely and honestly, without overstating the evidence.
- 清晰、完整、诚实地传达量化发现，不夸大证据所能支持的结论。
- Breadth subject (UNIB10006)通识 breadth 课（UNIB10006）
- The same Critical Thinking with Data subject offered as breadth to students across non-science degrees.
- 同一门《数据批判性思维》，以 breadth 通识形式开放给非理科学位的学生（编号 UNIB10006）。
- Critique vs criticism批判 vs 指责
- The course's founding distinction: critique is constructive evaluation that finds the limits of a work and how it could be improved; criticism is mere rejection. The exam rewards critique (good features + a specific improvement), never bare criticism.
- 本课的奠基性区分：critique 是建设性评价，找出一份工作的局限以及如何改进；criticism 只是单纯否定。考试奖励的是 critique（指出优点 + 一个具体改进），而不是空泛的指责。
- Validity vs precision效度 vs 精度
- Two independent goals of study design: validity = reducing bias (systematic error) via randomisation, comparison and control; precision = reducing variability (random error) via replication, stratification and balance. A bigger sample improves precision but not validity.
- 研究设计的两个相互独立的目标：效度 = 通过随机化、对照比较和控制来减少偏倚（系统误差）；精度 = 通过重复、分层和均衡来减少变异性（随机误差）。加大样本能提升精度，却提升不了效度。
- Lurking variable潜伏变量
- An unmeasured confounder — a third variable related to both the exposure and the outcome that was never recorded, so its distorting effect can't even be checked or adjusted for.
- 未被测量的混杂因素——一个同时与暴露和结局相关、却从未被记录下来的第三变量，因此它的扭曲作用连检查或校正都无从下手。
- Bradford Hill criteriaBradford Hill 因果判据
- Nine considerations (strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, analogy) used to argue causation from mostly observational evidence — as a weight-of-evidence argument, not a tick-box checklist.
- 用来从（多为观察性的）证据论证因果的九条考量（强度、一致性、特异性、时序性、剂量-反应梯度、合理性、与既有知识的一致、实验支持、类比）——作为「证据权重」论证，而非逐项打勾的清单。
MAST20034 — common questionsMAST20034 常见问题
How is MAST20034 assessed?MAST20034 怎么考核？
Is MAST20034 a biology subject?MAST20034 是生物课吗？
Do I need to be good at computation or programming?需要很强的计算或编程能力吗？
What's the difference between MAST20034 and UNIB10006?MAST20034 和 UNIB10006 有什么区别？
Is using AskSia allowed under The University of Melbourne's academic integrity policy?墨大的学术诚信政策允许用 AskSia 吗？
Other UniMelb course guides墨大 其他课程指南
AskSia is an independent study aid and is not affiliated with, endorsed by, or sponsored by The University of Melbourne. Course details may change — always confirm against the official handbook. Read about how this guide is built. AskSia 是独立的学习辅助工具，与墨尔本大学没有任何隶属、背书或赞助关系。课程信息可能变动，请始终以官方 handbook 为准。了解本指南的编写方法。