Statistics Learning Roadmap Template (Diagnostic and 12-Week Plan)

Statistics Learning Roadmap: [Your course, exam or thesis goal]

A personal curriculum map for learning statistics in three levels (foundations, core inference and models, advanced), with the prerequisite graph, the matching template for every topic, "can do" targets, a 20-question diagnostic that tells you where to start, a 12-week plan with spacing and interleaving built in, a weekly retrieval routine and a progress tracker. It is for students starting or repeating a statistics course, master's and PhD students who need statistics for a thesis, and working analysts filling gaps. It speeds up learning by sending you straight to the topics you do not yet know, in an order where each topic builds on the last, and by scheduling practice the way memory research says works best. The worked examples use fictional data; replace them with your own. Source basis: the GAISE College Report (ASA 2016) recommendations, summarised in our own words, and the learning research of Dunlosky et al. (2013) and Weinstein and Sumeracki (2018).

Document control

How this roadmap speeds up learning

Statistics is a cumulative subject: confidence intervals need standard errors, standard errors need sampling distributions, and regression needs nearly everything before it. Most learners lose time in two ways. They re-read topics they already know because the textbook puts them first, and they skip the few prerequisites that make later topics easy. The diagnostic fixes the first problem, the prerequisite graph fixes the second.

The plan then uses the six learning strategies that Weinstein and Sumeracki (2018) single out from the cognitive psychology research, each built into a column or routine. In the review by Dunlosky et al. (2013), practice testing and distributed (spaced) practice were rated high utility, and interleaving and elaborative interrogation moderate utility:

The concrete examples always use the same fictional settings (Lakemoor University, Brightwell Bakery, Harbour City Library) so that you spend effort on the statistics, not on a new story each time. A seventh ingredient comes from research on worked examples (Renkl and Atkinson 2003): study a solved problem, then a half-solved one, then solve one alone. Every method template and academic/stats-worked-problem follow this fading pattern.

Structure aligned with the GAISE College Report (ASA 2016) and the topic dependencies of a standard two-semester introductory and intermediate statistics sequence.

GAISE College Report recommendations in brief

The American Statistical Association's GAISE College Report (2016) gives six recommendations for teaching introductory statistics. In our words, and with what each means for you as a learner:

1. Teach statistical thinking. Statistics is an investigative process: pose a question, collect or find data, analyse, interpret. Include thinking about several variables at once. For you: always start from the question and the design, never from a formula (academic/stats-worked-problem steps 1 and 2).
2. Focus on conceptual understanding. Knowing why a method works and when it applies matters more than memorising procedures. For you: learn the concept cards (sampling distributions, p-values, intervals) before the method cards, and explain each in plain words.
3. Integrate real data with a context and purpose. For you: use your own course or thesis data in every template as soon as you can.
4. Foster active learning. For you: answer, compute and explain; do not just watch or read.
5. Use technology to explore concepts and analyse data. For you: run the R or Python code in each template, change the numbers and watch the output change; use simulations to see sampling variation.
6. Use assessments to improve and evaluate learning. For you: the diagnostic, the self-tests and the error log are assessments for learning, not grades.

Curriculum map: three levels

Level 1, foundations. The language of data, describing data, probability and the idea that statistics vary from sample to sample.

Level 2, core inference and models. Estimation, testing, the standard tests and simple regression, with effect sizes, power and reporting.

Level 3, advanced. Models for complex designs and outcomes, alternative inference frameworks, and the documents of a full analysis.

The learning tools apply at every level: stats-formula-sheet, stats-error-log, stats-flashcards-spaced-review, stats-worked-problem and stats-practice-which-test.

Prerequisite graph

Arrows point from a prerequisite to the topic that needs it. The graph is simplified: learning tools and reports are left out, and some topics are grouped.

Diagnostic self-assessment

The diagnostic has four blocks of five questions. Each block maps to a part of the curriculum. Answer in about 20 minutes, then score each block from the Answer key (1 point per correct answer) and use the routing table.

Block A: describing data (L1.1 to L1.4)

1. For the data 2, 4, 4, 5, 10, what are the mean and the median?
2. Which is less affected by one very large value: the mean or the median?
3. What is the measurement level of "satisfaction rated 1 = very poor to 5 = very good"?
4. When computing a sample standard deviation, do you divide the sum of squared deviations by $n$ or by $n-1$?
5. A histogram of household income has a long tail to the right. Is the mean usually above or below the median?

Block B: probability and distributions (L1.5 to L1.7)

6. $P(A) = 0.3$, $P(B) = 0.5$, and $A$ and $B$ are independent. What is $P(A \text{ and } B)$?
7. A fair coin is tossed 3 times. What is the probability of exactly 2 heads?
8. About what percentage of a normal distribution lies within 2 standard deviations of the mean?
9. Exam scores are roughly normal with mean 70 and SD 10. What is the z-score of a score of 85?
10. A condition affects 1% of people. A test detects 90% of cases and gives a false positive for 9% of non-cases. If someone tests positive, is the chance they have the condition closer to 10% or to 90%?

Block C: core inference (L2.1 to L2.5)

11. A sample of $n = 36$ has SD $s = 12$. What is the standard error of the mean?
12. A 95% CI for a mean is [4.1, 7.9]. Would a two-sided test of $H_0: \mu = 5$ at $\alpha = 0.05$ reject $H_0$?
13. Complete the sentence correctly: "p = 0.03 means that..."
14. What is a Type I error?
15. The same 20 students are tested before and after a workshop. Which t-test compares the two means?

Block D: models and method choice (L2.6 to L2.11 and Level 3 entry)

16. Three independent groups, numeric outcome, compare means: which method?
17. Describe a correlation of $r = -0.8$ in words.
18. A regression line is $\hat{y} = 20 + 3x$. What is the predicted $y$ at $x = 5$, and what does the 3 mean?
19. The outcome is pass or fail and there are several predictors. Which regression model fits?
20. Two categorical variables summarised in a table of counts: which test checks for an association?

Routing: where to start

Goal-based shortcuts. Exam in four weeks: do the routing, then run weeks from your start point at double pace and keep every review column. Thesis analyst: after routing, add your thesis method (for example L3.6 mixed models) as early as its prerequisites allow, and start academic/stats-analysis-plan in the same week. Working analyst: focus on L2.3, L2.4, L2.12 and L2.14 first: choosing, checking and reporting are the skills that most change the quality of real analyses.

12-week plan

The plan covers Levels 1 and 2 plus a first Level 3 topic. The review column always contains one topic from the previous week and one from at least three weeks earlier (in week 3 the oldest available topic, from week 1): this is spacing. The mixed practice column contains problems from several topics in shuffled order, so that you practise choosing the method rather than repeating one: this is interleaving.

Weekly retrieval routine

Three rules make the routine work. Try before you look: answer from memory first, even if you expect to fail, because an attempt followed by feedback is remembered better than reading the answer. Mix before you master: once two methods are learned, practise them shuffled, because exams and real data never tell you which chapter a problem comes from. Explain as you go: after each answer, say in one sentence why it is right.

Worked examples with fading

Example A (fully worked). A master's student at Lakemoor University (fictional) must analyse survey data from 120 students for a thesis in eight weeks. The outcome is whether a student used the campus tutoring service (yes or no), with several predictors. Diagnostic block scores: A 5, B 4, C 3, D 2.

1. Routing: A and B at least 4, C at most 3, so start at L1.7 (week 4 of the plan).
2. Goal topic: a binary outcome with several predictors needs L3.4 logistic regression. Prerequisites from the graph: L2.11 linear regression, which needs L2.10 correlation, which needs L2.2 hypothesis testing.
3. Compressed plan (8 weeks): week 1 L1.7 and L2.1; week 2 L2.2 and L2.5; week 3 L2.6, L2.7 and L2.4; week 4 L2.10 and L2.11; week 5 L3.4 logistic regression plus academic/stats-analysis-plan for the thesis; week 6 L2.12 and L2.14; week 7 analysis of the thesis data with academic/stats-regression-model-report; week 8 review and writing.
4. Spacing: each week keeps one review topic from the week before and one from three weeks back (for example week 4 reviews L2.7 and L2.1).
5. Interleaving: from week 3, the Friday set mixes t-tests, proportion tests and contingency tables with regression scenarios.
6. Check: re-take the diagnostic in week 4. Block C should now be at least 4; if not, repeat L2.2 before L3.4.

Example B (half worked). An analyst at Brightwell Bakery (fictional) must compare the mean proving time of dough across four flour suppliers and has scores A 5, B 5, C 4, D 2.

1. Routing: all of A, B and C are at least 4 and D is at most 3, so start at [ ] (week [ ] of the plan).
2. Goal topic: comparing means of four independent groups needs [ ]. Prerequisites from the graph: [ ].
3. If the variances differ strongly between suppliers, the template to open is [ ] and the method is [ ].
4. Spacing in the first week: review one topic from [ ] and one from [ ].
5. Interleaving: the Friday set should mix ANOVA scenarios with [ ] and [ ] scenarios.

Exercise C (your turn). Take the diagnostic, write your four block scores, apply the routing table and write your own first four weeks in the progress tracker below. Then answer: which single prerequisite, if missing, would slow you down most, and how will you check it in week 2?

Progress tracker

Diagnostic history (scores out of 5 per block):

The chart shows the example learner's diagnostic profile over the term. Replace the values with your own: a block that does not rise between weeks 1 and 6 tells you where to add spaced review.

Common mistakes when learning statistics

Self-test questions

1. Name the six learning strategies built into this roadmap and one place each appears.
2. Why does the review column of the 12-week plan include a topic from three or more weeks earlier, not only last week's?
3. Using the prerequisite graph, list the chain of topics that leads from probability to Bayesian inference.
4. What is the difference between blocked and interleaved practice, and why is interleaving harder but better for statistics?
5. Your block scores are A 5, B 2, C 4, D 4. Where do you start, and why not at Level 3?

My notes and next steps

My block scores: A [ ], B [ ], C [ ], D [ ]. Start at: [topic ID] in week [ ]. Goal topic: [ID and template].

First action (date and time): [YYYY-MM-DD, 19:00, take the self-test of academic/stats-sampling-distributions].

Confidence in the plan (1-5): [ ]. Open questions for the tutor or supervisor: [ ]. Errors to watch from my log: [link to academic/stats-error-log].

Sources and further reading

• GAISE College Report ASA Revision Committee (2016). Guidelines for Assessment and Instruction in Statistics Education (GAISE) College Report 2016. American Statistical Association.
• Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., and Willingham, D. T. (2013). Improving students' learning with effective learning techniques. Psychological Science in the Public Interest, 14(1), 4–58.
• Weinstein, Y., and Sumeracki, M. (2018). Understanding How We Learn: A Visual Guide (illustrated by O. Caviglioli). Routledge.
• Roediger, H. L., and Karpicke, J. D. (2006). Test-enhanced learning: taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., and Rohrer, D. (2006). Distributed practice in verbal recall tasks: a review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380.
• Rohrer, D., and Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35(6), 481–498.
• Renkl, A., and Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skill acquisition: a cognitive load perspective. Educational Psychologist, 38(1), 15–22.
• NIST/SEMATECH e-Handbook of Statistical Methods, https://www.itl.nist.gov/div898/handbook/ (accessed 2026-10-25). Public domain.
• Further reading (by name only): de Smith, M. J. (2024). Statistical Analysis Handbook, 2024 edition. StatsRef, https://www.statsref.com/; OpenIntro Statistics.

Answer key

Diagnostic.

1. Mean $= 25/5 = 5$; median $= 4$.
2. The median.
3. Ordinal (ordered categories; the gaps between levels are not guaranteed equal).
4. $n-1$.
5. Usually above the median: the long right tail pulls the mean up.
6. $0.3 \times 0.5 = 0.15$.
7. $\binom{3}{2}(0.5)^3 = 3/8 = 0.375$.
8. About 95%.
9. $z = (85-70)/10 = 1.5$.
10. Closer to 10%: $0.009/(0.009 + 0.0891) = 0.092$, about 9%. Most positives come from the large group without the condition.
11. $SE = 12/\sqrt{36} = 2$.
12. No: 5 lies inside the 95% CI, so a two-sided test at $\alpha = 0.05$ does not reject.
13. "...if $H_0$ were true, data at least as extreme as these would occur with probability 0.03." It is not the probability that $H_0$ is true.
14. Rejecting $H_0$ when it is true (a false positive).
15. The paired t-test (on the before-after differences).
16. One-way ANOVA (Welch's ANOVA if variances differ), followed by post hoc comparisons.
17. A strong negative linear association: higher $x$ goes with lower $y$.
18. $\hat{y} = 20 + 3 \times 5 = 35$. Each one-unit increase in $x$ is associated with a 3-unit increase in predicted $y$.
19. Logistic regression.
20. The chi-square test of independence (Fisher's exact test if expected counts are small).

Example B. Start at L2.5 t-tests with L2.3 choosing a test alongside (week 6 of the plan). Goal topic L2.9 one-way ANOVA (academic/stats-one-way-anova); prerequisites L2.5 t-tests and L2.2 hypothesis testing. With unequal variances: the same template, Welch's ANOVA followed by Games-Howell. Spacing: in the first week (plan week 6) the review column gives L2.2 (previous plan week) and L1.4 (three or more weeks back); because this learner skipped those weeks, treat them as light review. Interleaving: mix ANOVA with t-test and nonparametric (Kruskal-Wallis) scenarios.

Self-test.

1. Retrieval practice (self-tests, diagnostic), spacing (review column), interleaving (mixed practice), elaboration (can-do statements, Feynman check), dual coding (Mermaid and Vega-Lite), concrete examples (the shared fictional settings). Worked examples with fading (each template's A, B, C) are the extra seventh ingredient.
2. When the test is weeks or months away, somewhat longer gaps between reviews produce better long-term retention (Cepeda et al. 2006); reviewing only last week's topic would let older topics fade before the exam.
3. Probability and Bayes, then distributions, then estimation and MLE, then Bayesian inference (with a direct arrow from probability to Bayesian inference as well).
4. Blocked practice repeats one problem type; interleaved practice mixes types. Interleaving feels harder because you must first decide which method applies, which is exactly the skill exams and real data demand.
5. Block B is at most 3, so start at the first Level 1 topic of that block, L1.5 probability (week 3). Level 3 methods rest on sampling distributions and probability; skipping them makes every later topic slower.

Revision history