DATA1001: pass the exams, not just read the notes
Your complete guide to University of Sydney's foundations of data science unit. See where the marks are, work real practice questions, and study with an AI tutor that knows DATA1001.
Sia generates DATA1001 practice questions, walks through the linear model and chance variability: the box model step by step, and quizzes you on the material the exam weights most heavily.
Find what is wrong
A DATA1001 student tests whether caffeine changes endurance. Nine elite cyclists each ride to exhaustion twice, once with no caffeine and once with a 13 mg/kg caffeine dose, so the two readings on each row are the SAME cyclist. The student wants a single test of whether caffeine changes the mean time to exhaustion, and writes:
no_caf <- c(45.1, 52.0, 38.7, 60.2, 47.9, 41.3, 55.6, 49.0, 43.8)
with_caf<- c(58.3, 70.5, 46.5, 79.1, 66.4, 37.6, 69.5, 59.3, 36.2)
t.test(no_caf, with_caf) # two-sample t-test
The study design is paired (each cyclist is their own control), but this call throws the pairing away. Which single change makes it the correct test?The design is paired: each row is one cyclist measured twice (no caffeine and with caffeine). A two-sample test treats the 18 numbers as two independent groups and ignores that they are linked cyclist by cyclist.
In R that is exactly t.test(no_caf, with_caf, paired = TRUE). Pairing removes the between-cyclist variation, so the standard error is smaller and the test has more power to detect a real effect.
var.equal = TRUE only changes the two-sample test to a pooled version; it is still unpaired. Swapping the argument order flips the sign of the difference but does not pair the data. A one-sample test against 45 answers a different question (is with-caffeine endurance 45?), not whether caffeine changed each cyclist's own time.
The trap: Running an unpaired two-sample t.test on data that is actually paired. It is not a syntax error, so R returns a clean answer, but it throws away the pairing, inflates the standard error and loses power. Whenever the two columns are repeated measurements on the same units (before and after, left and right, twin pairs), use paired = TRUE. classic slip!
One exam decides 60% of your grade. Covers the whole unit; it is the universal backstop because the quiz better-mark and project progress-mark principles can roll weaker components up into it. This whole page is built around that.
Overview
What DATA1001 is, and where it sits
DATA1001 (Foundations of Data Science) is the University of Sydney's first-year introduction to statistical thinking and computation, taught in R, RStudio and Quarto. Its tagline is 'learn how to problem solve with data': you build two skills in parallel, statistical thinking and computational skill, so you can become a data storyteller who supports evidence-based decisions. The unit runs as four modules that build strictly in sequence, Exploring Data, Modelling Data, Sampling Data and Decisions with Data, and almost every exam question walks that same pipeline on a fresh dataset.
The conceptual spine is the box model and a single inference engine. From the sampling module onward, the course models any chance process as drawing tickets from a box (observed = expected + chance error), and nearly every test you meet (proportion test, z-test, t-test, chi-square, test for the regression slope) is the same calculation, (OV minus EV) over SE, with only the expected value, the standard error and the reference curve changing. Master that one skeleton plus the HATPC scaffold (Hypotheses, Assumptions, Test statistic, P-value, Conclusion) and the back half of the unit collapses into one repeated pattern.
DATA1001 is the assumed-knowledge statistics foundation for data science, science and many quantitative degrees, and it pairs naturally with the business-analytics and data-engineering units later in a degree. It also has an advanced twin, DATA1901, which shares the same lectures, topics and final exam and differs only in harder workshop and project data, so if you are in DATA1901 the conceptual core here is identical, you simply go deeper.
Official outline: sydney.edu.au · DATA1001 outline. Always treat the official outline and the exam timetable as authoritative.
Difficulty & time commitment
Is DATA1001 hard, and how much time does it take?
DATA1001 is manageable if you keep a weekly rhythm and treat the back half as the main event. Across student reviews the pattern is consistent: it starts gently and steepens, and the heaviest assessment is the part that separates grades.
A read across student reviews and course feedback. See what students say ↓
The difficulty curve and the assessment weighting point the same way: the back half is harder and worth more. Front-loading effort there is the highest-return decision in the unit.
Is this unit for you
Who tends to do well, and who tends to struggle
You will likely do well if
- You internalise the one inference engine early: see that the proportion test, z-test, t-test and slope test are all (OV minus EV) over SE, with only the expected value, the standard error and the reference curve (Normal versus t) changing.
- You always write out HATPC (Hypotheses, Assumptions, Test statistic, P-value, Conclusion) and split the conclusion into a statistical layer and a scientific in-context layer, because graders reward that scaffold explicitly.
- You can interpret rather than just compute: read a study, pick the right method, state what a p-value and a confidence interval do and do not mean, and write a clear recommendation for a client.
- You treat the two projects as exam practice with a longer deadline, drilling EDA, the regression line, choosing summaries and checking assumptions in Quarto.
You may struggle if
- You memorise formulas instead of understanding the box model and the (OV minus EV) over SE engine; the exam is 'words not formulae' and rewards reasoning, not recall.
- You misread p-values and confidence intervals (treating p as the chance the null is true, or a 95% CI as a 95% probability the parameter is in this interval); these traps are tested almost verbatim.
- You leave the back half (the box model, the CLT, sampling and the four hypothesis tests) to cram, even though Modules 3 and 4 carry most of the conceptual weight and the whole 60% final.
- You lean on a calculator, since the final exam allows none, so any method you cannot run by hand and explain in words will cost you.
- Build one master diagram of the four-module pipeline (describe, model, quantify chance, decide) and one table of the tests showing each test's EV, SE and reference curve, so the whole inference half is one page.
- Drill HATPC on the real practice and sample exams under timed, closed-book, no-calculator conditions; practise the full reasoning and the in-context conclusion, not just choosing a letter.
- Over-practise the interpretation traps the unit examines directly: correlation versus causation, design versus analysis, p-value pitfalls, CI randomness, SD versus SE, and using counts not proportions for chi-square.
- Make Project 2 your showcase: a clean reproducible Quarto report with the right summaries, a justified regression and assumption checks, and a concrete client recommendation, because the progress-mark principle lets a strong Project 2 lift Project 1 too.
Syllabus
The 11 topics, topic by topic
The exam-weight marker on each topic shows where the marks concentrate. The amber topics carry the highest exam weight.
T1 · Design of experiments
Module 1: Exploring DataWhy statistics matters (ethics, privacy, big data), variable types, and the central idea that study design drives the conclusion you may draw: observational studies cannot establish causation, only randomised controlled experiments can. Confounding and the read-causation-from-observational-data trap (LO9).
T2 · Data and graphical summaries
Module 1: Exploring DataProducing, interpreting and comparing graphical summaries in base R and ggplot: histograms (area, shape, skew), boxplots and the five-number summary, the 1.5 times IQR outlier rule, and matching the right plot to the variable type.
T3 · Numerical summaries
Module 1: Exploring DataCentre (mean versus the robust median), spread (SD as RMS distance, variance, IQR, range), percentiles, and how R stores quantitative versus qualitative variables. When to prefer the median and IQR for skewed data.
T4 · The Normal model
Module 2: Modelling DataThe Normal (Gaussian) model, standard units (z-scores), the 68-95-99.7 rule, and measurement error (measurement = exact value + chance error). pnorm for areas, qnorm for percentiles. All models are wrong, but some are useful.
T5 · The linear model
Module 2: Modelling DataCorrelation r, the least-squares regression line, the signature SD-line versus regression-line distinction, regression to the mean and the regression fallacy, r-squared, and reading residual plots for a good fit. cor() and lm().
T6 · Understanding chance
Module 3: Sampling DataProbability, conditional probability, the addition and multiplication rules, independence versus mutual exclusivity, and the binomial distribution (mean np, SD sqrt of np(1 minus p)). The box model is introduced. dbinom, pbinom, sample, set.seed.
T7 · Chance variability: the box model
Module 3: Sampling DataObserved = expected + chance error, the Law of Large Numbers (proportions stabilise, absolute deviations grow), EV and SE for a sample sum and mean, SD versus SE, and the Central Limit Theorem that makes the sampling distribution Normal even when the data are not.
T8 · Sample surveys and confidence intervals
Module 3: Sampling DataParameter versus statistic, sampling methods, sources of bias (a big biased sample is still biased), modelling proportions with a 0-1 box, the finite-population correction, confidence intervals, and the heavily-tested CI interpretation trap. Bootstrapping.
T9 · Hypothesis testing (HATPC)
Module 4: Decisions with DataThe HATPC scaffold, null versus alternative, the universal test statistic (OV minus EV) over SE, the p-value and its pitfalls (p is not the chance H0 is true), Type I and Type II errors, and the proportion test worked end to end (LO7, LO8).
T10 · Tests for a mean
Module 4: Decisions with DataZ-test for a mean (sigma known), one-sample t-test, paired t-test (reduce to differences) and two-sample pooled t-test, t versus Normal, and checking assumptions with QQ-plots, histograms and Shapiro-Wilk. t.test() for one-sample, paired and two-sample.
T11 · Tests for a relationship
Module 4: Decisions with DataChi-square goodness-of-fit and chi-square test for independence (always use raw counts, not proportions), and the test for the slope of the regression line read straight off summary(lm). Right-tailed chi-square and the LINE assumptions for regression.
How it's assessed
Assessment structure
| Component | Weight | Format & timing |
|---|---|---|
| Evaluate quizzes | 5% | Weekly online quizzes (due Sunday nights), including the Week 3 Early Feedback Task; the Early Feedback Task plus the best 8 of the remaining 10 quizzes count. Weekly across the semester, except Week 7 (project week). Better-mark principle: if your final-exam percentage beats your quiz percentage, the exam percentage replaces your quiz score, so the exam is also a safety net here. |
| Workshop participation | 5% | Attendance and participation across all workshops, particularly the coding milestones and project work. All teaching weeks. Special consideration available: a missed workshop can be pushed onto the final exam weight (for example a missed lab adds 1%, making the exam worth 61%). |
| Project 1 | 10% | Group reproducible data report in Quarto (.html plus .qmd, around 650 words) built on USyd census and survey data, with a short presentation in the Week 7 workshop. Due in the mid-semester break (aim for Week 6); presented in the Week 7 workshop. Progress-mark principle: a better Project 2 Part 2 mark replaces your Project 1 mark. Group work, so no special consideration; 5% per day late penalty. |
| Project 2 | 20% | Individual reproducible report in two parts: Part 1 is EDA (2 to 3 visualisations plus interpretation), Part 2 is a full report with a concrete recommendation for a client, on datasets such as CO2, Melanoma or NYC Crime. Part 1 due around Week 9, Part 2 due around Week 11 (dates subject to change). Special consideration available (simple 5-day or up to 7-day extension); an unsubmitted Part 2 can be carried onto the final exam weight (in the extreme pushing the exam to 77%). |
| Final exam | 60% | One in-person, supervised, closed-book written exam: 2 hours 10 minutes including 10 minutes reading. 30 multiple-choice questions (75% of the exam) plus 1 extended-answer question with several parts (25% of the exam). No reference materials and no calculator permitted. Formal exam period (date and time on the official exam timetable). Covers the whole unit; it is the universal backstop because the quiz better-mark and project progress-mark principles can roll weaker components up into it. |
- Pass on a weighted average of at least 50%. No single-component hurdle is stated in the unit materials reviewed, and the better-mark and progress-mark principles make a 50% pass unusually reachable.
- The final is the dominant single mark (60%) and also the safety net for almost everything else, because a strong exam percentage can replace weak quizzes and a strong Project 2 can replace Project 1. Questions are conceptual and interpretive, not a coding exam: they walk the four-module pipeline (describe, model, quantify chance, test) on a fresh dataset, and the same (OV minus EV) over SE engine drives most inference questions.
- Calculator policy: Final exam: closed book, no reference materials and no calculator permitted. The unit is designed around 'words not formulae', so the exam tests reading a study, choosing the method, running the logic and interpreting in context rather than heavy calculation.
This is an exam-cram unit. With the exams at 60% of the grade and the final exam alone at 60%, your result is overwhelmingly decided by how well you perform under time pressure. Covers the whole unit; it is the universal backstop because the quiz better-mark and project progress-mark principles can roll weaker components up into it.
Final exam timing: approx Nov 2026 (S2 2026 offering, in the formal exam period; confirm against the official exam timetable). Confirm the exact date and venue on the official exam timetable.
How to actually pass it
A weekly rhythm, two checklists, and the traps to avoid
The unit rewards consistency over cramming, and practice over re-reading. Here is the loop that works, then what to have nailed before each exam.
The weekly loop
Before the mid-semester checklist
- Lock down Modules 1 and 2 early: study design and the causation rule, graphical and numerical summaries, the Normal model and z-scores, and the linear model (correlation, the regression line, SD-line versus regression-line, regression to the mean).
- Practise reading and interpreting plots fast: histogram shape and skew, boxplot outliers by the 1.5 times IQR rule, and residual plots for a good linear fit.
- Get Project 1 done well in Quarto (reproducible .qmd to .html, embed-resources set, a clear interpretation), since it doubles as exam practice on EDA and summaries.
- Use the Week 3 Early Feedback Task to calibrate before the heavier sampling and inference material arrives.
Before the final heaviest topics
- Master the one inference engine: be able to write any test as (OV minus EV) over SE and state its EV, SE and reference curve from memory (proportion, z, one-sample t, paired t, two-sample t, chi-square, slope test).
- Run every test through HATPC and rehearse the in-context conclusion, since the extended-answer question rewards the full reasoning.
- Drill the interpretation traps the exam tests verbatim: p-value pitfalls, CI randomness, SD versus SE, correlation versus causation, design before analysis, and counts (not proportions) for chi-square.
- Sit the real sample exam timed, closed-book and with no calculator, in the actual format (30 MCQ then one multi-part extended answer), and review method not just the answer key.
- Treat the 60% final as the universal backstop: because a strong exam can replace weak quizzes and a strong Project 2 can replace Project 1, a confident exam protects the whole grade.
The mistakes that cost marks
Running an unpaired test on paired data. When two columns are repeated measurements on the same units (before and after, left and right), use t.test(x, y, paired = TRUE). An unpaired two-sample test is not a syntax error, so R returns a clean answer, but it throws away the pairing, inflates the SE and loses power. This is the single most common t-test mistake in the unit.
Misreading the p-value and the confidence interval. A p-value is not the probability the null is true, and a 95% confidence interval does not mean a 95% probability the parameter is in this interval (the parameter is fixed, the interval is random). The unit tests these almost verbatim; 'retain the null' also never means 'the null is true'.
Reading causation from observational data. Only randomised assignment licenses a causal claim. Design drives the conclusion you may draw, and a confounder can fully explain an observed association. This is the most-tested LO9 critique and a favourite extended-answer trap.
Treating DATA1001 as a coding exam. You learn R in the milestones and projects, but the closed-book, no-calculator final tests whether you can read a study, choose the right method, run the logic and interpret in context. Memorising code instead of understanding the box model and the inference engine misreads where the marks are.
Teaching team
Who teaches DATA1001
The bios below are factual. The star ratings are not ours: they are impressions from students who have taken the unit, so you can hear from people who sat in the lectures.
Dr Yeeka Yau
Coordinates DATA1001/1901 and lectures Modules 1 and 2 (Exploring Data and Modelling Data). Staff profile
A/Prof Ellis Patrick
Coordinates DATA1001/1901 and lectures Modules 3 and 4 (Sampling Data and Decisions with Data). Staff profile
Teaching team as listed in the unit materials reviewed. AskSia does not rate lecturers; star ratings are submitted by students who have taken DATA1001.
Formula & concept sheet
The vocabulary and formulas you must own
- Standard units (z-score)
- z = (x minus mean) / SD: how many SDs a value sits above or below the mean. Unit-free, so it lets you compare across variables and read Normal-curve areas.
- 68-95-99.7 rule
- For Normal data, about 68% of values lie within 1 SD of the mean, about 95% within 2 SD (more precisely z = 1.96) and about 99.7% within 3 SD. Check the histogram is roughly Normal before using it.
- Correlation and the regression line
- r measures the strength and direction of a linear association, with minus 1 <= r <= 1. The least-squares line has slope b1 = r times (sy / sx) and passes through (mean x, mean y); it is flatter than the SD line whenever the absolute value of r is below 1.
- Regression to the mean
- Because the regression slope is r times (sy / sx) with the absolute value of r below 1, a point k SDs above average in × is predicted to be only r times k SDs above average in y, closer to the mean. Attributing this to a cause is the regression fallacy.
- Box model
- Model a chance process as drawing tickets from a box. Observed value = expected value + chance error. Specify the tickets, how many of each, and the number of draws (with or without replacement).
- EV and SE for a sum and a mean
- For n draws: the sample sum has EV = n times the box mean and SE = sqrt(n) times the box SD; the sample mean has EV = the box mean and SE = box SD / sqrt(n). SE shrinks like sqrt(n) for the mean (quadruple n to halve SE) but grows like sqrt(n) for the sum.
- SD versus SE
- SD describes the spread of the data or the box; SE describes the spread of a statistic. The SE of the sample mean equals SD / sqrt(n). Confusing the two is a recurring exam error.
- Central Limit Theorem
- For large n, the probability histogram of the sample sum or mean follows the Normal curve regardless of the box's shape, so the sampling distribution is approximately N(EV, SE squared). This is why z and t tests and CIs work even when the data are not Normal.
- Standard error of a proportion
- Coding success as 1 and failure as 0, the box mean is p and the box SD is sqrt(p(1 minus p)), so SE of the sample proportion = sqrt(p(1 minus p) / n).
- Confidence interval
- estimate plus or minus z-star times SE. For 95%, z-star = 1.96; for 99%, z-star = 2.58. Correct reading: if you repeated the sampling many times, about 95% of such intervals would contain the true parameter; it is the interval that is random, not the parameter.
- The universal test statistic
- test statistic = (Observed value minus Expected value) / Standard error = (OV minus EV) / SE: how many SEs the data sit from what the null predicts. Proportion, z, t and slope tests are all this one engine with different EV, SE and reference curve.
- HATPC
- The course's exam scaffold for every test: Hypotheses (null with =, alternative with >, < or not-equal), Assumptions (stated and justified), Test statistic ((OV minus EV)/SE), P-value (chance of a result this extreme if the null is true; double it for two-sided), Conclusion (statistical versus alpha, then scientific in context).
- Paired versus two-sample t-test
- Paired t-test: two readings on the same unit; reduce to the differences d and run a one-sample t on d against 0 with df = n minus 1. Two-sample t-test: two independent groups, pooled SE with df = n1 + n2 minus 2. Using an unpaired test on paired data loses power.
- Chi-square and the slope test
- Chi-square = sum of (Observed minus Expected) squared / Expected, always on raw counts and right-tailed: goodness-of-fit (df = k minus 1) or independence (df = (r minus 1)(c minus 1)). Test for the slope: t = beta1-hat / SE(beta1-hat) against t with df = n minus 2, read off summary(lm).
Common acronyms: EDA · SD · IQR · EV · SE · OV · CLT · LLN · CI · HATPC · LINE · QQ · LO.
What students say
What students actually say about DATA1001
Recurring themes from student reviews, paraphrased in our own words.
- Officially described as challenging but very rewarding; it ramps up after the mid-semester break as the box model, sampling and hypothesis testing arrive.
- Manageable for students who keep up weekly, because best-8-of-quizzes, a better-mark principle and a progress-mark principle make the grade forgiving as long as the final goes well.
- Reportedly more conceptual and interpretive than students expect: a 'words not formulae' exam with no calculator, so understanding beats memorising.
- Students lean on the one repeated engine (OV minus EV) over SE and the HATPC scaffold, and build a single master sheet of the tests with their EV, SE and reference curve.
- The two Quarto projects are treated as exam practice, since the EDA, regression and interpretation skills they drill are exactly what the exam rewards.
- Demand for clear, worked walkthroughs of the high-leverage inference topics (the box model and CLT, the four hypothesis tests, and the p-value and CI interpretation traps) under timed, no-calculator conditions.
Recurring student opinions, paraphrased and aggregated, not official course information.
Where it fits
Prerequisites, related units & why it matters
No formal prerequisites and no coding experience assumed; R is taught from scratch through the Coding Milestones. DATA1001 is the standard-stream foundation; DATA1901 is the advanced twin (same lectures, topics and final exam, harder project and workshop data), so you take one stream, not both for credit.
Your DATA1001 study toolkit
Study the unit with Sia, not just read about it
Each tool already knows DATA1001: your syllabus, your texts, and where the marks are. Grouped by how you study, from first contact to exam week.
FAQ
Frequently asked questions
Is DATA1001 hard?
It is moderate for a first-year unit. The official line is 'challenging but very rewarding', and it ramps up after the mid-semester break as the box model, sampling and hypothesis testing arrive. What keeps it moderate rather than hard is the assessment design: best-8-of-quizzes, a better-mark principle where a strong exam replaces weak quizzes, and a progress-mark principle where Project 2 can replace Project 1, so a 50% pass is very reachable. The single 60% final is the part to take seriously.
How is DATA1001 assessed?
Evaluate quizzes 5% (Early Feedback Task plus best 8 of 10), workshop participation 5%, Project 1 (group Quarto report) 10%, Project 2 (individual report in two parts) 20%, and a final exam worth 60%. You pass on a weighted average of at least 50%, with no single-component hurdle stated in the materials reviewed.
What is the final exam format?
One in-person, supervised, closed-book written exam of 2 hours 10 minutes (including 10 minutes reading time): 30 multiple-choice questions worth 75% of the exam, plus 1 extended-answer question with several parts worth 25%. No reference materials and no calculator are permitted. It covers the whole unit and is conceptual and interpretive rather than a coding exam.
Do I need coding experience or to know R beforehand?
No. The unit assumes no coding background and teaches R, RStudio and Quarto from scratch through the weekly Coding Milestones. You learn enough R to produce the reproducible reports and to recognise the key function for each method (for example cor, lm, pnorm, t.test, chisq.test), but the exam tests understanding and interpretation, not writing code under exam conditions.
How much maths is involved?
It is moderately quantitative but deliberately 'words not formulae, thinking not memorisation'. Most inference reduces to one engine, (OV minus EV) over SE, plus the box model, the Normal model and the regression line, all at HSC arithmetic level rather than calculus. Note the final exam allows no calculator, so the questions are built around reasoning and interpretation rather than long computation.
What is the difference between DATA1001 and DATA1901?
DATA1901 is the advanced version of this unit. It shares the same Canvas site, lectures, topics and final exam as DATA1001 and is differentiated only in the workshops and projects, where DATA1901 students work with harder research data and are expected to add extension components. If you are in DATA1901 the conceptual core in this guide is identical; you simply go deeper. You take one stream, not both.
Study DATA1001 with Sia
Work through the linear model, chance variability: the box model, sample surveys and the rest of the unit with a tutor that knows it and quizzes you on the topics the assessments weight most heavily.
Start studying with Sia