The Preparation Nobody Does
Most candidates prepare for a numerical test by working harder problems. That is usually the wrong investment, because the marks are rarely lost on difficulty.
They are lost on misread questions, mis-extracted figures, and a clock that ran out while three answerable items sat untouched. Five habits address that directly.
Habit 1: Read the Question Before the Data
The instinct is to study the table first. Reverse it — read the question, then go to the table knowing exactly which two or three figures you need.
Employer tests deliberately supply more data than any single question uses. Reading the table first means absorbing material you will not need, at the cost of the seconds you will.
The wording to slow down on
- "Percentage" versus "percentage point" — different answers, and both are always in the options.
- "Of the total" versus "of the regional total" — one word, different denominator.
- "Increase to" versus "increase by" — the difference between the final value and the change.
- Units in the heading or a footnote — thousands, millions, indexed to a base year.
Habit 2: Estimate Before You Calculate
Before touching the arithmetic, decide roughly what the answer should be. This takes about three seconds and does two jobs at once.
It eliminates options — on a well-built multiple-choice item, a rough magnitude usually kills two or three of them outright. And it catches extraction errors, which are otherwise undetectable because the calculation itself was flawless.
The estimation toolkit
- 10% first. Ten per cent of anything is a decimal shift, and every other percentage is built from it.
- Round to convenient numbers. 2,480 is "about 2,500"; that is close enough to choose between options an order of magnitude apart.
- Know the common fractions. A quarter is 25%, a third is about 33%, an eighth is 12.5%. These appear constantly.
Habit 3: Measure Change From the Right Base
Percentage change is the single largest source of lost marks in numerical testing, and the reason is almost never the arithmetic.
The rule is one line: change is always divided by the starting value. A fall from 425 to 340 is 85 ÷ 425 = 20%, not 85 ÷ 340 = 25%.
The percentage point trap
If a share rises from 12% to 15%, it has risen by three percentage points — and by 25%. Both statements are true, they mean different things, and the question is asking for exactly one of them.
Whenever a quantity is itself a percentage, stop and re-read the question. That is the entire tell.
Habit 4: Triage Ruthlessly
Most numerical papers are timed so that the average candidate does not comfortably finish. That is a design choice, not an accident, and it makes item selection part of what is being assessed.
Every item is worth the same. A hard item you spend three minutes on costs you two easy ones — and the score does not know the difference.
A working rule
- Pass one: answer everything that resolves in under a minute.
- Pass two: return to the multi-step items with the time that remains.
- If the paper does not penalise wrong answers, never leave a blank — but confirm that in the instructions rather than assuming it.
Chained items are the exception to fast triage. When four questions hang off one dataset, the reading time is spent once and amortised across all four, which makes the cluster worth entering even if the first item looks slow.
Habit 5: Check the Extraction, Not the Arithmetic
When candidates re-check, they re-check the sum. The sum is almost never the problem.
The problem is the figure that came out of the wrong row, or the wrong year's column, or a subtotal read as a total. Then the arithmetic runs perfectly on the wrong input and produces an answer that feels entirely solid.
So spend the checking second on the source: is this the row I meant, in the year I meant, in the units the heading says?
The Calculator Question
Publishers differ, and the difference is deliberate. Some permit a calculator on the reasoning that real analysts have one; others forbid it to keep arithmetic fluency inside the measurement.
Read the instructions rather than assuming, and practise under whichever condition applies. Preparing with a calculator for a paper that forbids one is a common and entirely avoidable way to lose marks.
If a calculator is allowed
It removes the arithmetic bottleneck and does nothing about the two real ones — choosing the operation and extracting the right figures. Candidates who assume the calculator solves the test tend to be the ones who run out of time anyway.
The Week Before
The highest-return preparation is unglamorous and short.
- Drill the operations to automaticity. Percentages, ratios, reverse percentages, weighted averages. Automatic operations free working memory for the reasoning, which is where the difficulty actually is.
- Do at least two papers fully timed. Untimed practice trains the mathematics and none of the pacing, and pacing is what most candidates lack.
- Read your own wrong answers. Sort them into misread question, wrong extraction, wrong method, ran out of time. Four different problems, four different fixes.
What to skip
Revising anything beyond school arithmetic. Nothing in a graduate numerical paper requires algebra beyond rearranging a simple expression, and none of it requires calculus. Time spent there is time not spent on the pacing that will actually cost you.
On the Day
Numerical performance depends on working memory more than most people expect, and working memory is the first thing sleep deprivation takes.
Have paper, whatever the format. Even where a calculator is permitted, writing down intermediate results externalises the load that would otherwise be displaced by the next lookup.
And if you lose an item, let it go completely. Ruminating on a question you have already left is working-memory capacity spent on a mark you cannot recover, taken from the item currently in front of you.