Why You Fail Practice Tests but Pass the Real Exam (and Vice Versa)

Two things happen often enough that both are worth explaining. Someone grinds through a practice bank, sits at 65%, books the exam in a panic, and passes comfortably. Someone else runs 90% for a fortnight, walks in relaxed, and fails.

Neither outcome is mysterious. A practice score and an exam result measure different things, and the gap between them has several separate causes.

Is a practice test score the same as an exam score?

No, and on most certification exams it is not even the same kind of number.

Practice apps report percent correct. You answered 78 of 100, so you see 78%. Most certification bodies report a scaled score instead. CompTIA passes Security+ at 750 on a range that runs from 100 to 900. That 750 is not 83%, and CompTIA does not publish how many questions you need to get right to reach it.

Scaled scores are not the only alternative to a percentage. The ASVAB reports an AFQT percentile benchmarked against a national sample of 18 to 23 year olds from a 1997 norming study, so an AFQT of 65 places you above 65% of that reference group rather than telling you how many questions you answered correctly. Two exams, two different kinds of number, and neither is a percentage of the questions you answered correctly.

Scaling exists because questions are not interchangeable. A harder item carries more weight than an easy one, exam forms vary, and live exams usually include unscored trial questions that count toward nothing. Percent correct throws all of that away.

A practice bank cannot tell you your scaled score. Not approximately, not with a clever formula. The cut score, the item difficulties and the scaling model are the testing body's property and are not published. Any tool that shows you a predicted exam score or a pass probability is modelling its own question bank and presenting the output as though it were about the exam.

Why are practice questions harder or easier than the real exam?

Because nobody calibrated them against the real thing.

A real exam goes through standard setting. Subject matter experts review each item, the item gets trialled on live candidates, its difficulty is measured, and the cut score is set against a defined standard of competence. That process is expensive and slow, and it is why exam bodies charge what they charge.

A third-party question bank has none of that. Its difficulty is whatever its authors picked. Some banks are deliberately written harder than the exam on the theory that over-preparing is safer. Others drift easier because writing genuinely tricky items is hard work. Both produce scores that do not map onto the exam, in opposite directions.

So the two failure modes have the same root. Scoring 65% on a bank pitched above the exam and then passing is the system working as designed. A 90% on a soft bank, followed by a fail, is the same mechanism running the other way.

Does percent correct mean anything on an adaptive exam?

On a computer adaptive test, not really. The CAT-ASVAB and the MBLEx both work this way.

An adaptive test picks each question based on how you have answered so far. Get one right and the next is harder. Get one wrong and the next is easier. The test is searching for the level where you answer around half the items correctly, so a well-functioning adaptive exam should feel hard the whole way through. Candidates read that difficulty as failure when it is the algorithm doing its job.

It also means your percent correct is close to meaningless as a measure. Two candidates can both answer 55% of their items correctly and land far apart, because they saw questions of completely different difficulty. What the test reports is an ability estimate, not a tally.

Any static practice bank misses this entirely. It cannot scale to you, so it cannot reproduce the experience or the score.

Why do practice conditions flatter you?

Most people practise in conditions they will not get on exam day. You study at home, at your own pace, in short sessions, often with notes nearby, and you can retry a question you fluffed.

Koriat and Bjork (2005) documented the underlying error. People judge their knowledge using how the material feels while they are studying it, and studying happens with the answer available. They found predictions of 75.7 against actual recall of 60.3 on items where study conditions activated a link that would not be there at test. Their caveat matters: the effect is selective rather than universal, and for many items in the same study the predictions were accurate. It appears specifically where the conditions of practice flatter you in a way the exam will not.

Practice conditions flatter you in exactly that way. The fix is unglamorous. Take at least some sessions full length, timed, with no notes and no retries, because a practice score collected under easy conditions is measuring the conditions.

Does exam anxiety change your score?

It is associated with lower performance, and the association is well established even though the causal story is not simple.

Von der Embse and colleagues (2018) reviewed 30 years of test anxiety research and found anxiety negatively related to a broad range of outcomes, including standardised tests, university entrance exams and grade point average. The evidence is correlational. Anxiety and poorer performance travel together, and the research cannot cleanly separate anxiety depressing a score from a shaky grasp of the material producing both.

For the practice-versus-exam gap, the practical version is narrower and safer to state. Your practice sessions have no consequences and your exam does. If you have only ever practised in the low-stakes version, exam day introduces a variable you have never tested.

Why would you pass after failing practice tests?

The reverse case is common and usually has a dull explanation.

Your bank was pitched harder than the exam. Or it tested obscure corners of the blueprint while the exam concentrated on the core. Or you improved between your last practice session and exam day and never re-measured, which happens constantly to people who stop taking full sessions in the final week. Or your weak domain was a small slice of the real exam and a large slice of the bank you were using.

None of those means you got lucky. They mean the bank was a biased sample of the thing you were actually being tested on.

What should you use practice scores for?

Three things, none of which is predicting your result.

Use the trend, not the level. A score moving from 58% to 71% to 77% on the same bank tells you something real. Any one of those numbers on its own does not.

The domain breakdown is the most useful output a practice bank produces. It tells you where to spend the next week, and unlike the composite it is not distorted by the bank's overall difficulty.

Then go question by question. Which ones did you get wrong, and did you misread the stem, miss the concept, or run out of time? Three different problems with three different fixes, and the percentage hides all of them.

For judging whether you are ready to book the exam at all, the percentage is the wrong instrument, and I have written separately about what actually signals readiness. Getting the material into memory so it survives to exam day is a different problem again, and the mechanism there is spacing your reviews out.