Evidence questions are where "I got it down to two" goes to die. All four choices sound defensible, the final two both sound true, and you pick the one that felt smarter — exactly the feeling the test engineered. The fix is not reading harder. It's changing the order you do things in.
The mistake underneath most evidence misses
Command of Evidence hands you a claim and asks which choice most directly supports, weakens, or completes it — from a passage, or from a graph or table on quantitative items. The key word is directly: the test rewards proof of one exact claim, not true statements. Read the choices before deciding what the proof must say, and they'll decide for you. That's anchoring — the behavioral leak that runs much of the reading section — and evidence questions are its natural habitat.
Step 1 — Decide first: answer the claim yourself
Before you look at a single choice:
- Restate the claim in your own words — the claim, not the topic.
- Say what the proof must do: support, weaken, or fill the blank with a specific value.
- Point at the proof: the sentence type (textual) or exact data point (quantitative) the right answer must contain.
A biologist argues that the decline of a bat colony near a growing town is caused by artificial light at night, not by habitat loss. Which finding, if true, would most directly support the biologist's claim?
Decide first: the proof must tie the decline to light while habitat stays out of it — light up, habitat unchanged, bats down. Commit before any choice touches you.
Cognitive science has a name for why this works — the generation effect: producing an answer beats recognizing one. It's Predict-Then-Peek aimed at evidence questions; the full habit is its own guide.
Step 2 — Kill-Three: make the choices apply for the job
Now read the choices — not as options to like, but as applicants to fire. Three of the four fail for a nameable reason. Name the failures; the survivor gets the job. (Expect the right answer to feel flat — traps are built to feel insightful.)
The bat colony applicants:
- "Bat populations nationwide have fallen over the past decade." Fired — nothing about light, this colony, or cause. Too broad: wrong scope.
- "The colony's numbers were stable for thirty years before the decline began." Fired — accurate, on-topic, and about when, not why. True-but-irrelevant.
- "The bats roost in dead trees, which the town removes as it expands." Fired — that supports the rival explanation, habitat loss.
- "The steepest declines occurred on blocks where new streetlights were installed and tree cover stayed constant." Hired — light varies, habitat held constant, decline follows. Step 1's job description.
Three named kills, one survivor. The trap families:
Trap 1 — True-but-irrelevant (right passage, wrong claim)
The choice is accurate, sometimes practically a quote — and it supports a claim you weren't asked about. The most dangerous of the three: nothing about it is false, so your truth-detector waves it through.
A historian claims that a nineteenth-century canal company failed because railroads undercut its shipping prices.
Trap choice: "The canal took eight years longer to build than its engineers projected." Vivid, plausible, about the canal's troubles — and it answers "was the canal a difficult project?" instead of "did railroads kill it financially?" The kill question: true, maybe — but does it support this claim?
Trap 2 — Data-misread (right chart, wrong number)
Quantitative only. The reasoning is fine; the number comes from the wrong cell — wrong row, wrong season, misread axis. It survives a glance because the number really is on the chart.
Decide-first, quantitative version: guess the number the claim needs before you read the data, then read only that data point. A two-season gardening experiment:
| Season 1 | Season 2 | |
|---|---|---|
| Drip-irrigated plots | 42 kg | 51 kg |
| Hand-watered plots | 44 kg | 39 kg |
Claim to complete: "In the second season, the drip-irrigated plots outproduced the hand-watered plots, yielding ___"
The claim lives entirely in the Season 2 column: you need 51 and 39, nothing else. The applicants:
- "51 kg versus 44 kg" — right drip number, wrong season for the hand figure. Data-misread. Fired.
- "44 kg versus 42 kg" — Season 1 numbers, which run the other way. Fired.
- "42 kg versus 39 kg" — real numbers, transposed across seasons. Fired.
- "51 kg versus 39 kg" — the two cells you pre-read. Hired.
The cousin: "hand-watered plots produced 44 kg in the first season" — a perfect reading of a cell the claim never asked about. True-but-irrelevant in a lab coat.
Trap 3 — Wrong scope (right topic, wrong size)
The choice covers the wrong amount: one detail sold as the whole story (too narrow), or a claim so general it fits any passage (too broad). On-topic content, wrong coverage. The counter: state the size of the claim first. If the claim is about a playwright's early comedies, praise for "the wit that runs through her entire career" is too big; one pun in one scene is too small. Match the size or fire it.
The self-test: point at the proof
Run this on your next ten evidence questions, textual and quantitative mixed. Before reading any choice, write the proof target — the sentence type or the exact cell the right answer must contain. Then answer normally and sort your misses:
- Couldn't point before choosing → the leak is Step 1: you're browsing uncommitted. Drill decide-first until pointing is automatic.
- Pointed right, picked a trap anyway → the leak is Step 2: run the kill list by name — off-claim? wrong number? wrong size? Naming the trap is what makes it visible next time.
Ten questions, one written target each — you'll know which half of the method to drill. If evidence is one of your fattest miss buckets, good news: it's among the most method-fixable question types (sorting misses into buckets is covered in the plateau guide).