Published 1 May 2026 · Updated 3 September 2026 · 16 min read · Cross-provider numerical reasoning hub
| What it measures | Ability to interpret data and perform calculations under time pressure |
|---|---|
| Common formats | Tables, charts, financial reports, multi-step calculations |
| Used by | Every major aptitude provider (Cut-e, Saville, Cubiks, Korn Ferry, Sova, etc.) |
Numerical reasoning is the most common cognitive test in graduate and professional recruitment. Almost every major aptitude provider offers a numerical reasoning module — Cut-e (scales eql / lst / nmg), Saville (Numerical Analysis), Cubiks/Talogy (Logiks Numerical), Korn Ferry (Elements / Aspects Numerical), Sova, Pearson, Mercer Mettl, AMCAT. The format and timing differ, but the core skill is identical: extract relevant data from a chart or table, perform calculations, select the correct answer under time pressure.
Two people can both be told they have a numerical reasoning test on Thursday and sit two mechanically different papers. One gets five answer options and a single clock running across the whole paper. The other gets a drop-down list of twenty or more values, a separate timer on every single question, and no way to carry unused seconds forward. The invitation email reads the same in both cases.
That gap matters because most of what candidates actually bring into the room is strategy, not arithmetic. Back-solving from the options, ruling out answers that are the wrong order of magnitude, and banking time on the easy items so you can spend it on a hard one are all sensible tactics, and each is deliberately disabled by a design choice some publishers make. Which version you are facing is decided by six mechanical variables rather than by the topic of the charts.
Practising for a real assessment? The Numerical Reasoning Practice Pack runs these question types at full length — 5 timed tests, 130 questions, a worked explanation on every one.
The most consequential difference is how you enter an answer. A conventional item gives four or five options, which means you can substitute candidate answers back into the stem, and you can discard anything obviously the wrong size. A long drop-down list, sometimes twenty to thirty closely spaced values, is built to remove that. They sit near enough together that magnitude checks tell you nothing, and there are too many to test one by one. You must compute the answer properly, then find it. Free numeric entry goes further: nothing to match against, so rounding instructions and units become part of the item.
A whole-test clock over a fixed number of items lets you skip, flag and return, so a hard item costs you only the time you choose to give it. A per-question timer does the opposite. Each item gets its own allowance, and unused time is forfeited rather than banked, so finishing an easy item early buys you nothing on the next one. An indefinite bank against a clock reverses the pressure again: the pool is longer than anyone finishes, and your score depends on how far you get accurately, so speed is measured directly.
A single chart with one question is the easy case. Multi-panel dashboards are not, because the answer requires a figure from one panel and a rate or weighting from another, and the panels often use different units or periods. Tabular data with footnotes is the most commonly mishandled: a footnote often restates a figure, excludes an item, or changes the basis of a column, and ignoring it produces a clean-looking wrong answer that is usually sitting in the options. Interactive tables you filter or sort yourself add a step where the mistake happens before any arithmetic.
Some formats attach one data set to one question; others attach three or four questions to the same set, which changes the economics of reading. Where the set serves one question, read only what the question needs. Where it serves four, spending 30 seconds understanding the column headings, units and footnotes up front is repaid across every item that follows.
Fixed forms give everyone the same items in the same order. Adaptive forms choose the next item from your answers so far, with two practical effects: the test should feel persistently hard, because it is converging on your level, and you usually cannot revise an earlier answer.
An on-screen four-function calculator and a permitted physical scientific calculator are not the same test. The on-screen version typically has no memory keys, no bracket handling and no percentage key, so a chained calculation must be held in your head or written down in stages, and every stage risks a transcription error. A scientific calculator with memory and brackets removes most of that load, which means the same item is measuring something different depending on the tool.
Two scoring models are in use and they pull in opposite directions. Under raw scoring, your score is the number of correct answers, wrong answers cost nothing, and unanswered items score zero. Under formula scoring, sometimes described as negative marking, marks are deducted for wrong answers and nothing is deducted for items left blank.
That distinction decides your endgame. Under raw scoring, a blank is strictly worse than a guess, so every item should carry an answer before time runs out. If the scheme deducts for a wrong answer and nothing for a blank, then working one more item properly beats guessing at three. An uninformed guess has negative expected value, and only a guess narrowed to two plausible options is worth making. Check the instructions screen for this before item one; it is nearly always stated there, and it is the only piece of information that changes what you do in the final minute.
The timing model interacts with this. With a whole-test clock, triage is available: sweep for the items you can finish, then return. With a per-question timer, triage is impossible, so the discipline is different. Decide early whether an item is going to come out, and if it is not, commit to your best answer and move on, because the time is gone either way. With an indefinite bank, there is no finishing line to sprint for, so the correct approach is a steady accurate pace rather than an acceleration at the end. On an adaptive form, early items carry disproportionate weight in locating your level, so slowing down for the first handful is usually a good trade.
Sitting it soon? TestSolve reads each question off your screen and shows the worked answer, over 90% correct on our own practice tests.
Stimulus. A table shows Northern region service revenue: Q1 €18.0 m, Q2 €21.0 m, Q3 €24.0 m, Q4 €27.0 m. A footnote reads: "Q3 and Q4 include €3.0 m from the Baltic contract, which is reported separately from Q1 next year." A second panel gives average exchange rates of 0.85 GBP per euro in Q2 and 0.90 GBP per euro in Q4.
Question. On a like-for-like basis excluding the Baltic contract, what was the percentage change in Northern region revenue between Q2 and Q4, measured in pounds?
Working. Q2 contains no Baltic revenue, so like-for-like Q2 is €21.0 m, which is 21.0 × 0.85 = £17.85 m. Q4 like-for-like is 27.0 − 3.0 = €24.0 m, which is 24.0 × 0.90 = £21.60 m. The change is 21.60 − 17.85 = £3.75 m on a base of £17.85 m, giving 3.75 ÷ 17.85 = 21.0 per cent.
What the wrong options are built from. 14.3 per cent applies the footnote correctly but never converts, measuring the change in euros. 36.1 per cent converts correctly but ignores the footnote, comparing £24.30 m with £17.85 m. 28.6 per cent ignores both, taking 6 over 21 in euros. 17.4 per cent uses the right two figures but divides by the later value, putting the base at the wrong end. Each distractor is one specific rule misapplied, which is why a plausible-looking answer is rarely a safe one.
Find out the response mode before you practise, because it determines which skills are worth drilling. If the answer comes from a long drop-down, back-solving practice is wasted effort and you should instead work on carrying calculations through to a precise figure and rounding exactly as instructed. If it is four or five options, practise the reverse: estimate first, discard the impossible, and only compute precisely when two options survive.
Rehearse with the calculator you will actually get. If it is an on-screen four-function tool, practise breaking a chained calculation into written intermediate steps, and get used to having no bracket or memory keys. Write intermediate values down even when you think you can hold them.
Drill footnote handling as its own exercise. Take any table, read the footnotes before the question, and write down in one line what each one changes about which column. This is where most avoidable marks go.
Practise percentage change with an unfamiliar base, percentage change of a percentage, ratios expressed as "x to y" rather than fractions, and reading values off axes with unlabelled gridlines. Finally, settle your guessing rule in advance by confirming the scoring model, so that decision is already made when the clock is short.
It removes two strategies at once. With four or five options you can substitute answers back into the question, and discard anything of the wrong magnitude. A drop-down carrying twenty or more closely spaced values defeats both: testing them is too slow, and they are too similar to separate by estimation. You must produce the answer independently, then locate it.
Nothing useful. The unused time is forfeited rather than added to your allowance for the next item. That is the key difference from a whole-test clock, where time saved on easy items is genuinely available later. Under a per-question timer the only thing speed buys you is a calmer transition, so accuracy within each allowance matters more than rushing.
It depends entirely on the scoring model, which the instructions screen normally states. If the score is simply the number of correct answers, answer everything, because a blank can never beat a guess. If marks are deducted for wrong answers and not for blanks, an uninformed guess has negative expected value, and only a guess narrowed to two options is worth making.
Both fixed and adaptive forms are in common use, and the invitation rarely says which. A useful signal is whether you can navigate backwards. Adaptive forms almost always lock each answer once submitted, because the next item was chosen from it. If the test feels relentlessly difficult throughout and you cannot revisit anything, you are probably on an adaptive form.
Read the instructions rather than assuming. Some versions supply an on-screen four-function calculator and prohibit anything else; some permit your own scientific calculator; a few expect mental arithmetic with rough figures. The three produce genuinely different tests, because memory and bracket keys change how much of a chained calculation you have to hold yourself.
That depends on stimulus reuse. Where one data set serves a single question, read only what the question needs and go straight to the relevant row or series. Where the same set serves three or four questions, read the headings, units, scale factors and footnotes first, because that effort is recovered on every later item.