Chapter 2 — Exercises

This is the chapter whose exercises matter most. Part C in particular is not busywork — it's the skill the rest of the book assumes you have.

Difficulty legend: ⭐ basic recall · ⭐⭐ applied · ⭐⭐⭐ synthesis and judgment · ⭐⭐⭐⭐ extension

Selected answers in Appendix J. The full Claim Filter and abstract-reading checklist live in Appendix D.

Part A — Conceptual Understanding ⭐

A1. List the eight rungs of the evidence ladder in order, from weakest to strongest as a basis for changing your diet.

A2. State the three questions that let you place a study on the ladder in about ten seconds.

A3. Define healthy-user bias in one sentence, without using the word "healthy."

A4. Explain why increasing the sample size of an observational study does not fix confounding.

A5. What is residual confounding, and what are the two distinct reasons it survives statistical adjustment?

A6. Give two examples of reverse causation in nutrition, other than the breakfast and diet soda examples used in the chapter.

A7. What is a surrogate endpoint? Name one surrogate the chapter describes as relatively trustworthy and one it describes as weak, and say what makes the difference.

A8. Explain the difference between relative risk and absolute risk using the processed meat example. Show the arithmetic.

A9. What does IARC Group 1 actually mean? What does it not mean?

A10. What is Mendelian randomization, and why does it sidestep healthy-user bias?

A11. The chapter lists five features of convincing evidence. Name them.

A12. What happened with beta-carotene, and why is that story described as the most important one in modern nutrition science?

Part B — Applied Analysis ⭐⭐

B1. For each headline, identify the most likely study design, the rung, and the single biggest threat to its conclusion:

• a) "People who eat nuts daily live longer, 30-year study of 120,000 finds"
• b) "Compound in broccoli halts tumour growth"
• c) "Fasting for 14 hours improves blood sugar, trial of 26 adults shows"
• d) "Children who eat family dinners have better mental health"

B2. A study reports that people who drink three or more cups of coffee daily have 15% lower all-cause mortality, adjusted for smoking, BMI, alcohol, exercise, and income. Write out the four distinct objections from §2.2–§2.4 that apply, in the order you'd raise them.

B3. Rewrite each malformed question as a well-formed one, specifying the comparator, the population, the duration, and the outcome:

• a) "Is dairy bad for you?"
• b) "Are potatoes healthy?"
• c) "Should I cut out gluten?"
• d) "Is coconut oil good for you?"

B4. A supplement's marketing says it is "clinically proven to reduce inflammation." Using §2.7, explain in three sentences why this phrasing is close to meaningless as a health claim, even if the underlying study was well conducted.

B5. The baseline lifetime risk of a certain cancer is 1.2%. A study reports a relative risk of 1.40 for a dietary exposure. Calculate the absolute risk and the difference. Express it as a natural frequency ("about X extra cases per 1,000 people"). Then say whether you'd change your diet, and what other information you'd want first.

B6. Return to the constructed abstract in §2.9. Suppose the researchers had pre-registered flow-mediated dilation as their single primary outcome, and reported the other nine as clearly labelled secondary outcomes, and had not run the post-hoc subgroup analysis. How much would that change your assessment, and why?

B7. Explain why the PREDIMED retraction-and-republication episode can be used to argue both "you can't trust nutrition trials" and "the system works." Which use is more defensible?

B8. A cohort study finds that vegetarians have lower rates of heart disease. Name four non-dietary characteristics that differ systematically between vegetarians and non-vegetarians in most Western populations, and explain how each could produce the association independently of diet.

Part C — Skills & Practice ⭐⭐–⭐⭐⭐

C1. Run the Claim Filter, all six questions. Take three nutrition claims — from your Belief Inventory, from your feed, from a family argument — and run each fully through the six questions. Write the answers down. Note specifically which questions you cannot answer from the information available. That's the finding.

C2. Ninety seconds with a real abstract. Go to PubMed (pubmed.ncbi.nlm.nih.gov). Search a nutrition topic you care about. Open the first three abstracts. For each, in ninety seconds or less, record: species · design · sample size · duration · what was measured (disease or marker) · rung · one red flag from the §2.9 table. Then compare your assessment to the paper's own conclusion sentence. How often is the conclusion broader than the results?

C3. The confounder sprint. Set a timer for fifteen seconds. Name as many plausible confounders as you can for: "People who take a daily multivitamin have lower mortality." Then do it again for "People who eat organic food have lower cancer rates." Then again for "People who drink kombucha report better digestion." If you can consistently generate four or more in fifteen seconds, you have internalized §2.2.

C4. Convert three headlines to absolute risk. Find three real health headlines reporting a percentage change in risk. For each, find the baseline rate (search the condition plus "lifetime risk" or check a national cancer or heart association site). Calculate the absolute change. Express it per 1,000 people. Note how many of the three articles gave you the baseline.

C5. Find the substitution. Find one study or article recommending reducing a food. Determine — from the paper if you can reach it, from reasoning if you can't — what replaced it in the population studied. If the source doesn't say, write "not specified," and note how confident the article was despite that.

C6. Argue the other side. Take a nutrition position you hold confidently. Build the strongest case against it using only the tools in this chapter: what design supports your position? What's its biggest weakness? Is there convergence, or are you leaning on one type of evidence? Would Mendelian randomization, if it existed for this question, be likely to agree with you?

C7. Build your card. Make the physical Claim Filter card from the Project Checkpoint. Carry it or pin it somewhere. This sounds trivial and it's the exercise most likely to still be affecting you in a year.

Part D — Synthesis & Critical Thinking ⭐⭐⭐

D1. The chapter argues that "only RCTs count" is probably false, partly because the evidence on smoking was never randomized. Is that a fair rebuttal? Smoking's effect size is enormous; most nutrition effects are small. Does the smoking analogy actually support observational nutrition research, or does it undermine it?

D2. Given everything in §2.2–§2.4, on what basis should a government issue dietary guidelines at all? They must recommend something, and waiting for definitive evidence means waiting forever. Construct a defensible decision rule for acting under this much uncertainty.

D3. The multiverse problem (§2.3) says that hundreds of defensible analyses of the same dataset can produce results ranging from harmful to protective, and a published paper reports one point in that space. If that's true, what — if anything — does any single observational nutrition paper tell us? Argue a position, and address the strongest objection to it.

D4. Is it ethical for a university press office to write a release that is technically accurate but shifts emphasis, knowing the predictable downstream distortion? Who, along the pipeline from Chapter 1, bears the most responsibility — and does responsibility track influence, intent, or reach?

D5. The chapter claims over-skepticism is "credulity with extra steps." Defend the opposite: make the strongest case that, given nutrition's track record of reversals, near-total agnosticism about specific dietary claims is the epistemically correct position. Then say what's wrong with it.

Part M — Mixed & Interleaved Practice ⭐⭐–⭐⭐⭐

M1. (With Chapter 1.) Chapter 1 described a five-step pipeline from finding to belief. Map each step onto the ladder from §2.1: at which rung did the finding start, and what rung does the audience believe it's at by the time it reaches them? Where in the pipeline does the perceived rung jump most?

M2. (With Chapter 1.) Chapter 1 said "misleading-but-true is more dangerous than false." Use §2.6 to give the sharpest possible example of a statement that is entirely true and profoundly misleading.

M3. (With Chapter 1.) Chapter 1's four questions and Chapter 2's Claim Filter overlap. Identify what the Filter adds, and explain why "what kind of study?" wasn't in Chapter 1's version.

M4. (With Chapter 1.) Return to the Case Study 1 pair about saturated fat, where a meta-analysis of cohorts and a substitution-based guideline disagreed. Using the ladder, say which rung each sits on — and explain why the higher rung isn't automatically the winner here.

M5. (Revisit at Chapter 16.) The beta-carotene story, the vitamin E story, and the hormone therapy story all share one shape. Write that shape as a general template, in one sentence, with blanks. You'll fill it in three more times before Chapter 16 ends.

M6. (Revisit at Chapter 12.) The chapter mentions that Mendelian randomization has disagreed with cohort data for alcohol. Predict, before reading Chapter 12, what that disagreement is likely to be about and which direction it points. Seal your prediction.

Part E — Research & Extension ⭐⭐⭐⭐

E1. Find the published accounts of the ATBC and CARET trials, or authoritative summaries from the National Cancer Institute. Read what the investigators themselves said about why the result surprised them. Then find a recent supplement being marketed on antioxidant claims and assess whether the reasoning being used is different in kind from the pre-1994 beta-carotene reasoning, or only in detail.

E2. Locate a published specification-curve or multiverse analysis in nutrition or epidemiology. Look at the distribution of results across analytical choices. Then find a single published paper on the same question and consider where in that distribution it likely sits — and whether you could have known.

E3. Read the PREDIMED retraction notice and the republished paper's description of what was corrected. Write a page on what the episode should and should not change about how you weight that trial. Extension: find at least one commentator who used the episode to argue that Mediterranean-diet evidence is worthless, and evaluate whether their argument survives §2.10.

E4. Pick a nutrition question where Mendelian randomization studies exist (alcohol, vitamin D, BMI, and homocysteine are all good candidates). Compare what MR concluded with what the observational literature concluded. Where they diverge, work out which assumptions of MR could be violated — this is the exercise that separates people who've read about MR from people who understand it.