We went through our ACT Reading questions and gave each wrong answer a name describing what it does to the passage. Then we counted.

Two kinds account for 1,094 of the 2,890 wrong answers we labelled, which is 38 percent.

That denominator is the wrong answers we could actually name. A further 86, which is 3 percent of everything we labelled in this bank, still carry a generic label instead of a pattern, and they are excluded from every figure below rather than counted as a category. That figure was 460 when this was first published; the difference is a second labelling pass, not a change in the questions.

These are our own questions, written to match the current ACT, so this is how our writers build wrong answers. ACT has not published anything comparable.

The largest group states the reverse of the passage

602 of the 2,890 contradict the text. They negate something it affirms, or affirm something it denies.

Many are built out of the passage's own vocabulary, with the nouns and the topic matching and one word carrying a negative the passage did not. A policy that reduced traffic becomes a policy that failed to reduce traffic.

Where a choice is built that way, checking that it is about the correct subject will not separate it from the right answer, because the subject is the same in both. The direction word is what differs.

The second group is word meaning in context

492 of the 2,890 are wrong choices on word-meaning questions, where the choice offers a real meaning of the tested word that is not the meaning it carries in that passage.

That is 17 percent of the wrong answers we labelled, which places the group above scope errors, overstatement and unsupported inference in our bank. We would not have predicted that ordering.

These sit differently from the rest. Most wrong answers we labelled are wrong about the passage. These are correct about the dictionary. "Arrest" really does mean what the police do. It does not mean that when the sentence concerns arresting the growth of a crystal.

ACT states that it tests the meaning of words, phrases and statements in context, and that its Reading questions do not include isolated vocabulary questions. So the tested sentence alone is not always sufficient. A contrast drawn a few lines earlier can be what fixes the meaning.

Cover the word, then compare the direction word

Both checks follow from how the choices above are built.

On a word-meaning question, decide what the passage needs the word to mean before reading the choices, widening past the sentence when the sentence leaves it open. Then read each surviving choice back into the passage.

On the reversal group, once two choices remain, locate the word in each that carries direction or negation. Not, fails, reduces, prevents, increases. Compare it against the passage's own word. That is a one-word comparison rather than a whole-sentence one.

If you test on paper you can annotate the booklet. On the computer-based test you get authorised scratch paper, and the software also provides tools such as highlighting.

The categories are ours

Treating a wrong word sense as its own group, rather than as a kind of scope error, was our decision. Merge the two and the ordering changes, so read categories of similar size as roughly equal rather than ranked.

Six categories cover 73 percent of what we labelled

Set the exact shares aside, because they will move as the labelling improves. A gate recomputes each of them from the question bank on every build, so they cannot quietly go stale on this page.

The durable observation is that the list is short. A small number of moves account for most of the wrong answers, which makes them worth learning to recognise by name.