Skip to main content
ANGLIA ONLINE BLOG

One step at a time

new Anglia placement test

How We Built the New Anglia Placement Test

A guide, not a judgement

Anglia has launched a new free online placement test. It takes about 35 minutes; it is free, and at the end it gives each learner their result, the CEFR equivalent and how they did in each skill. Your result is the Anglia level to work towards: the exam to prepare for next.

There is no pass and no fail. That is deliberate. Every learner is somewhere on the ladder. The only question a placement test should answer is which rung to start on.

This article explains how we built it, what we had to give up to keep it short, and the problems we had to solve along the way. We have tried to be honest about the limits, because a test you understand is a test you can use well.

Building on what came before

Our previous test served Anglia well for many years. In designing its successor, our aim was to improve how the test handles the borders between levels. Moving up from one stage to the next used to require a score close to full marks, so a learner sitting near a border could be placed more cautiously than they needed to be.

We wanted the new test to give those learners a fairer chance to show what they can do, so that teachers can enter a class for the exam it is ready for.

Placing a learner at the wrong level always has a cost. Too low wastes their time and tells them they are weaker than they are. Too high can mean entering an exam they are not ready for. Our first decision was that the new test would aim squarely at the right level, and lean only very slightly low where it is unsure.

What 35 minutes can buy, and what it cannot

A test is a sample. We ask a learner a small number of questions and use the answers to say something about everything we did not ask. The more questions we ask, the more stable the result becomes.

So why not make the test longer? Because of the room it has to live in. A typical lesson is 45 to 50 minutes. Students arrive, settle, find a device, log in and listen to instructions. Thirty-five minutes is what is usually left. Young learners also lose concentration long before an adult does, and a tired candidate gives you bad data, not more data.

The honest consequence is this. A short test can place a learner in the right region of the ladder with reasonable confidence. It cannot always separate two neighbouring levels. We decided to accept that and to say so, rather than to pretend a 35 minute test is more exact than it is. Results that sit close to a boundary are flagged for the teacher, so you know when the test itself is unsure.

How the test works

The test adapts. It is built in three short stages, and each stage uses what came before.

  1. A short routing section with questions spread right across the range, from the very first steps to the most advanced level.
  2. A targeting section, chosen by how the first stage went, which looks at one part of the ladder in more detail.
  3. A boundary section, which concentrates on the one or two level borders that actually matter for that candidate.

It works a little like an eye test. The optician does not show you every lens in the drawer. They narrow the field quickly, then spend the time on the fine decision at the end.

The gain is comfort as well as accuracy. A beginner never has to sit through thirty questions they cannot read. An advanced learner is not kept busy with easy ones. All four auto-marked skills are sampled at every stage: grammar, vocabulary, reading and listening. A short writing task can be switched on by the teacher, but a human has to mark it, so it never changes the automatic level.

The compromises, and why we made them that way

A bad start should not decide everything. An adaptive test has one obvious weakness. If a candidate has a poor first few minutes, perhaps from nerves or a noisy room, the test sends them down a path that caps how high they can finish. We found this in our own first design and rebuilt it. The test now has escape routes in both directions, and the final level is weighted across all three stages rather than resting on the last one alone.

Multiple choice can be guessed. With four options, pure guessing earns marks. We worked out that guessing floor for each section and set the level boundaries well above it. If a candidate's score in a section is no better than chance, we stop treating it as evidence rather than letting it pull the result around.

The two errors are not equally costly. Placing someone too low is disappointing but easy to correct in class. Placing someone too high means they enter a real exam, pay a fee and may not succeed. So we lean very slightly low. If our result looks one level below what you see every week in class, you are probably right and we are probably being cautious.

There is no speaking. We cannot assess speaking fairly, free of charge, for thousands of candidates online. Rather than fake it, we leave it out and say so. For some learners, speaking is their strongest skill and the test will not see it.

The questions are public. A free online test can be sat by anybody, including someone who wants to copy the questions. We answer that with size rather than secrecy. Questions are drawn at random from much larger pools, so two students sitting side by side meet different questions built to the same specification.

The complicated bits

Some of the hardest problems were not about English at all.

How many times may a learner hear the audio? In an Anglia exam, listening material is played twice. Online, a candidate can click replay as often as they like, which turns a listening test into a patience test. We had to build our own audio player to hold the limit at two hearings.

Accessibility can fight validity. We are building to WCAG 2.2 AA, which usually means offering a text alternative to sound. For a listening question, a transcript is not an alternative. It converts the question into a reading question and measures something else. So we keep the controls that genuinely help, such as pause and volume, remove the ones that break the test, and record openly why no transcript is offered for listening items.

A level is not an age. Anglia's four lowest levels exist in both young learner and adult versions. A forty-five year old beginner and a nine year old beginner may sit at the same point on the ladder, but sending the adult to an exam written for children would be wrong. The test therefore recommends an exam using the level and the age band together.

The bottom of the ladder has no CEFR words. The CEFR starts at A1, which is already a long way for a young child. Anglia's first levels sit below it. Describing them only as "below A1" sounds like a deficit, so we are careful about the wording learners see. Their result should read as a starting point, not as a shortfall.

One hard word can break an easy question. Every item is checked against the wordlists for its level, and against the Oxford 3000 and 5000 above that. A grammar question can be perfectly pitched and still fail because a single word in one of the wrong answers is three levels too hard.

Machines find faults that careful readers miss. Our question bank was written and reviewed by experienced people. Automatic checks across all 595 items still found problems no item-by-item review could see. The clearest was the position of the correct answer: more than half the multiple choice questions had their key as option B. A test-wise candidate who simply picked B would have scored above their real level. We also found reading questions that had drifted away from their passage, and anagram-style questions stored with the letters already in the right order. None of these are visible when you look at one question at a time. They are only visible when you count.

No pass, no fail

Nobody fails a placement test. The report does not use the words pass or fail anywhere, and there is no overall percentage for students to compare with each other.

What the report gives you

How to read it

An Anglia level

The best single estimate of where the learner is working now

The CEFR equivalent

For comparing with other frameworks and for parents

A skill profile

Grammar, vocabulary, reading and listening, shown separately

A recommended exam

The level and age band together, so adults are not sent to children's papers

A borderline note

Shown when the learner sits close to a boundary and the test is unsure

This is the Anglia idea of one step at a time, put into a test. The ladder has nine levels because small steps are achievable steps. A learner who is told "you are here, and the next step is this one" has something they can do on Monday morning. A learner who is told they have failed has nothing.

The skill profile is often more useful than the level. A learner who is strong in reading and weak in listening needs different lessons from one whose profile is flat, even when both are placed at the same level.

Please check our answer against yours

This is the most important part of the article. The test is one source of evidence about a learner. You are another, and a better one.

You see your students every week. You know who was ill, who has just arrived, who writes well but freezes when asked to speak, and who was sitting next to a friend during the test. Thirty-five minutes of clicking cannot know any of that.

We would ask you to use three sources together:

  1. The test result, including the skill profile and any borderline note.
  2. Your own assessment from class work, homework and speaking.
  3. The learner's own view of what they find easy and hard.

Where all three agree, enter the candidate with confidence. Where they disagree, your judgement should win. If the test says Elementary and everything you have seen says Junior, enter them for Junior. We would far rather you overrule us than follow a result you do not believe.

Two practical notes. A low listening score is sometimes a headphone problem or a noisy room rather than a weakness, so it is worth checking before you plan around it. And a very uneven profile is worth a second look, because it often points at something about the test day rather than the learner.

What happens next

The test is not finished, and a test of this kind never really is.

In the first months after launch, students whose CEFR levels are already known are also sitting the test, so that we can compare what the test says with what we know to be true. Those results will be used to adjust the level boundaries. This is the right way round: set the boundaries from evidence, not from a tidy percentage.

After launch we will watch every question. Two numbers matter. The first is how many candidates answer it correctly, which tells us whether it sits at the level we claimed. The second is whether stronger candidates get it right more often than weaker ones, which tells us whether it is measuring English at all. Questions that fail on either count will be retired and replaced. Cut scores are stored as settings that can be changed and versioned, not buried in code, precisely so that this can happen.

Every question has been reviewed and approved by Anglia's Chief Examiner. Even so, the first version will not be the best version.

This is where we would like your help. If a result does not match the student in front of you, please tell us. Tell us the level we gave, the level you would have given, and anything you noticed about the test day. Patterns in that feedback are worth more to us than any amount of internal review. The same goes for anything confusing in the wording, the timing or the screens.

We would rather hear that we got it wrong early than find out slowly.

This email address is being protected from spambots. You need JavaScript enabled to view it.

Or email This email address is being protected from spambots. You need JavaScript enabled to view it.. More about Anglia's levels and exams at www.anglia.org.