Research

Three years of the inburgeringsexamen, researched.

Every Staatsexamen NT2 from 2023 to 2025, and the DUO practice exams behind the inburgeringsexamen. 881 questions, coded one item at a time. Here is what the exam does again and again, and what that means for the one you sit.

What we looked at

Complete exam sets, every one with an official answer key or an official marking model. Nothing estimated and nothing recalled from memory: every question was coded on its own fields, and the reading data was then checked again by a second coder briefed to prove the first one wrong.

PartQuestionsExam sets
Reading2889
Listening3059
Writing666
Speaking1429
KNM802
Total88135

Two exams, and we never mix them

This matters more than it looks. The exam you sit at A2 and the Staatsexamen NT2 are set by different bodies, and they behave differently on almost every measure we tested. A number from one says nothing about the other, so every figure on this page names where it came from.

Staatsexamen NT2

B1 and B2

Published by the CvTE, including the marking models. This is where most of the reading, listening, writing and speaking data comes from.

Counted from the exams the CvTE publishes in its NT2 practice environment.

DUO practice material

A2 and KNM

The official practice exams, with the answer key from the results screen. We call it practice material because that is what it is, and never "the inburgering exam".

Counted from the practice exams DUO publishes at inburgeren.nl.

What we found

Each one is a count, not an impression. The sample is on every card, because a percentage without its denominator is an opinion with a number attached.

B1 + B2 ReadingStaatsexamen NT2

One line on your exam tells you to stop reading

34 of 34 in 213 items across 6 exams, B1 and B2

What it means

Above every reading text there is a single line of instruction. Five of them say the same thing. The last one says something else: read the question first, then find the answer in the text. That line appeared on the final text in every exam we looked at, and there the answer sits almost word for word in the text 47% of the time, against 9% everywhere else. The split is the exam board's own, not ours, which is what makes it worth trusting.

The catch

It is NOT easier. Half-correct answers 24% against 12%, right fact wrong person 15% against 4%, and in 13 of 34 items the information sits in two articles at once. The difficulty moves from understanding to locating, so searching faster without checking is how you walk into it. The check: does this article apply to your case, and is there a further condition further down.

A2 ListeningDUO practice material

The first thing you hear is usually not the answer

25 of 75 in 75 items across 3 practice exams

What it means

A speaker names something and then takes it back. "We agreed on the pizzeria, but it is closed. Tomorrow the restaurant is open." A third of the questions we coded work this way. When you are reading you can go back a line; here the correction goes past once and you either catch it or you do not.

The catch

Waiting is not the whole answer either. 21% of the correct answers land in the final quarter, so a fragment that seems finished may not be: the correction can come last as easily as first.

KNMDUO practice material

KNM asks far more about work than about manners

40% in 80 questions across both practice exams

What it means

KNM is 40 questions across eight themes, and the themes are nowhere near the same size. Work and income runs to eight or nine questions and institutions to seven, so together they are about 40% of the exam. Manners and education get three each. That held across both practice exams.

The catch

Studying only the two big blocks is not enough to pass. Together they are about 40%, which leaves 60% elsewhere, and the pass mark is well above that. This tells you where to spend time first, not where to stop.

B1 SpeakingStaatsexamen NT2

Answer beside the question and the whole task scores nothing

0 in 12 official marking models

What it means

Before anything else is marked, the assessor asks one question: does this answer belong to the task that was set? If it does not, nothing else about it is scored. Vocabulary, grammar, pronunciation, all of it stops mattering for that task.

The catch

This is not a reason to write less. The precondition is about answering the question that was set, not about being brief or careful: a short answer that misses the task scores the same zero as a long one.

B1 SpeakingStaatsexamen NT2

A picture on screen means you are not being asked your opinion

4% vs 69% in 87 coded tasks

What it means

Look at your screen before you start talking. Of the tasks that showed a picture, 4% asked what you think. Without a picture, 69% did. The reason is simpler than it sounds: an opinion cannot be drawn, so pictures sit with the tasks that have something to point at.

The catch

The reliable direction is only one way round. A picture means no opinion question 96% of the time; no picture does NOT reliably mean an opinion question, since 31% of those tasks ask for something else.

B1 ReadingStaatsexamen NT2

The advice everyone gives you does not work

4 of 288 in 288 exam questions

What it means

"Pick the option with words you recognise from the text" is the most repeated piece of exam advice there is. We checked it on 288 questions. In four of them was there exactly one option with word overlap. The rest of the time it points at two or three at once, because the commonest wrong answer is one that quotes the text and does not answer the question.

The catch

Word overlap is not evidence of a wrong answer either. Both the key and the distractors use words from the text, about 86% against 95%, so overlap tells you nothing in either direction.

What we are not going to claim

Every pattern below looked convincing and then failed a check. Five of them are here, from reading, listening and speaking. We are listing them because the checking is what makes everything above worth reading, and because you will find some of these repeated elsewhere as fact.

Answer B is more often correct

It held across 176 items and survived a correction for multiple testing. Then we tested it on exams we had not used to find it: 31.7%, which is chance. And even if it had held, always answering B scores 10 to 19 points against a pass mark of 24. Not a pass in any of the six exams.

Pick the option that repeats words from the text

Both the right answer and the wrong ones use words from the text, around 86% against 95%. A feature that appears in nearly every option cannot tell them apart. This is the most repeated piece of exam advice we know of, and it is a compass with no needle.

Difficulty climbs steadily from A2 to B1 to B2

A2 differs from B1, but B1 against B2 comes out at p=0.18 and the individual exams overlap. There are two groups here, not a ladder, and the middle step we first reported does not exist.

Count the pictures to know how much to say

Within a task type the correlation is about zero. What looked like a rule was pictures being attached to the kind of task that is short anyway.

The neutral-sounding option wins

Three coders reported this independently, which is exactly why it needed checking. Only 13 of 19 target questions even have such an option, and it is wrong about half the time.

That is the standard everything above was held to. A pattern that only shows up in the exams you used to find it is not a pattern, and we would rather tell you that than sell it. The same goes for what we build: see how the course is built on top of it, or go straight to the inburgeringscursus itself.

By level

How each part is built, in numbers: how many questions, how long you get, what the pass mark is, and which kinds of question make up the paper.

Practise the exam
we measured.

Every part, in the format the real exam uses. One full practice exam per part is free.

Exam Research: 3 Years of Dutch Exams, 881 Questions Coded