Skip to main content
Back to Blogs
Guide9 min read

Numerical Reasoning Data Interpretation: Charts, Tables, Ratios and Change

Where numerical reasoning marks are lost: two-source questions, units and scale traps, percentage change against percentage points, ratios and distractors.

The arithmetic on a numerical reasoning test is school level and almost everyone can do it. The marks still go missing, and they go missing before any calculation starts: in the wrong row, the wrong column, the units line nobody read, the second table nobody opened. This article is about reading the stimulus. The arithmetic shortcuts are covered separately, and they only pay off once this part is under control.

Four things to read before you read a number
The title of the exhibit, the units line, the axis or column headers, and any footnote marked with an asterisk. That is ten seconds, and on a dense exhibit it is the highest return ten seconds on the paper. Every one of those four has a matching wrong answer sitting in the option list, placed there because item writers know they are the four things candidates skip.

Questions that need two sources

Test publishers stopped building single-lookup items years ago, because a single lookup measures eyesight. What they build instead is an item whose answer requires one figure from one exhibit and one from another, and the difficulty is not the combination, it is noticing that a combination is needed at all.

The tell is in the stem, and it is reliable. Look for a noun phrase containing per, as a proportion of, compared with, relative to, or any unit built from two other units.

  • Revenue per employee needs a revenue table and a headcount table. The headcount is usually in a different exhibit, often in a different unit, and sometimes only available as a percentage of a total given elsewhere.
  • Cost as a share of turnover needs the cost line and the turnover line, and those are routinely split across an income exhibit and a segment exhibit.
  • Growth in market share needs the company figure and the market total, for two periods. That is four lookups, and the four are rarely adjacent.
  • Anything with a currency conversion needs the rate, which lives in a footnote or a small side table roughly as often as it lives in the main exhibit.

When you see one of those phrases, count the lookups on your fingers before you start. If the count is four and you have only found two places to get them, you have not found the second exhibit yet, and going ahead with two numbers is how you produce an answer that is confidently wrong and present in the option list.

The single most expensive habit on a data-heavy paper is starting to calculate as soon as you have found a number that looks relevant. Find every number the question needs first, then calculate once.

Formats differ in how much they punish this. Aon cut-e spreads the source material across tabbed documents, so a missing second source costs you a tab switch you may not have budgeted for, and the test is negatively marked. Talent Q Elements puts a per-question clock on each item, so a missed source is not a slow item, it is an item you run out of time on. SHL Verify gives you one long clock across eighteen questions, which is more forgiving of one slow item and much less forgiving of a habit of them.

Units, scale and base years

Scale errors are the most common way to arrive at a wrong answer that is also one of the five options. They come in four shapes.

Thousands against millions

An exhibit headed "$m" sitting next to one headed "$000s" is a standard construction, not an accident. If the question combines them, one of the options is your answer multiplied or divided by a thousand. There is no way to catch this after the fact, because the number looks perfectly reasonable. The only defence is reading the header before you start.

Indexed series

A row labelled "index, 2020 = 100" contains no amounts at all. A value of 118 is not 118 of anything, it is 18 per cent above the base year. Two indexed series with different base years cannot be compared directly, and a question that asks which grew faster is answerable from the index while a question that asks which is larger is not. Where the question asks for an absolute figure and the exhibit only gives you an index, the answer is usually that a second exhibit holds the base-year amount.

Percentages of percentages

Europe is 40 per cent of group revenue, and the Netherlands is 25 per cent of European revenue. The Netherlands is therefore 10 per cent of group revenue, not 65 per cent and not 25 per cent. The rule is that nested percentages multiply, and the trap is that the two percentages in the stem have different denominators while looking identical on the page. Whenever you see a percentage, say what it is a percentage of before you use it.

Per-unit and per-period figures

Monthly against annual, per store against total, per head against per household. A table of monthly averages and a question about a year differ by a factor of twelve, and the factor of twelve is in the options. This is the same error as thousands against millions wearing different clothes, and it is caught the same way.

Change against percentage points

This is the most reliably exploited distinction in numerical reasoning, and it is worth being able to state precisely rather than approximately.

Percentage change
The size of a movement relative to where it started
Difference divided by the original value, then times 100.
A margin moving from 4 per cent to 6 per cent is a 50 per cent increase, because 2 divided by 4 is one half.
Asked for by phrases like "increased by", "grew by", "fell by what percentage".
The denominator is always the earlier or stated original figure, never the later one.
Percentage point change
The arithmetic gap between two figures already expressed as percentages
Later value minus earlier value. No division at all.
The same margin moving from 4 per cent to 6 per cent is a rise of 2 percentage points.
Asked for by "percentage points", "pp", or a stem about a share, rate or margin moving.
Only meaningful when both figures are percentages of the same base.

Both numbers will be in the option list on any item built around a rate or a margin, and they will not be adjacent, so a glance at the options will not warn you. Two further wrinkles are worth drilling:

  • Percentage changes are not symmetric. Going from 80 to 100 is a 25 per cent rise; going from 100 back to 80 is a 20 per cent fall. An item asking about a fall and a recovery is testing exactly this.
  • Successive percentage changes do not add. A 10 per cent rise followed by a 10 per cent fall leaves you at 99, not 100. Over three periods the compounded answer and the added answer are far enough apart to be separate options, and the added one is the distractor.
  • A percentage of a total can fall while the amount rises.If the total grows faster than the segment, the segment's share drops even though it sold more. Stems that say "share" and stems that say "sales" are different questions.

Ratios, shares and rates

Ratio items look easy and are the second largest source of avoidable losses, because three different objects get written in ways that look alike.

  • A ratio is not a fraction of the total. If A to B is 3 to 5, then A is three eighths of the pair, not three fifths. The number of parts is the sum, and the first move on any ratio item is to write down the total number of parts.
  • Order matters and the stem controls it."The ratio of staff to managers" and "the ratio of managers to staff" are reciprocals. Where the exhibit presents one and the question asks for the other, the reciprocal is one of the options.
  • Ratios across different totals are not comparable as amounts. Two regions can share the same 2 to 1 split while one is four times the size of the other. A question about which region has more of something is not answerable from the ratios alone, and that is frequently the point of the item.
  • Scaling a ratio means finding one part first. A 3 to 5 to 8 split of 96,000 has sixteen parts, so one part is 6,000. Working each share as a fraction of the total is slower and generates rounding errors that then fail to match any option exactly, which costs you a second pass.

Korn Ferry Talent Q Elements deserves a specific note here. Its numerical items offer a long option list rather than four or five choices, which is a deliberate design decision: with that many options, an answer you arrived at by a plausible wrong route is still likely to be on the list, and guessing is close to worthless. On that format, checking which denominator the stem wanted is not caution, it is the whole game.

What charts hide that tables do not

A table gives you the number. A chart gives you a picture of the number, and the gap between those is where chart-specific items live.

Stacked bars
Read the segment, not the top of it
On a stacked bar, the value of a segment is its height, not the gridline it reaches. The second segment of a bar running from 40 to 70 is worth 30. Reading 70 is the single most common chart error, and 70 will be an option.
Twin axes
Two series, two different scales
A combination chart with bars on the left axis and a line on the right is showing quantities that cannot be compared by looking. Where a line sits above a bar it may still be the smaller figure. Check which axis each series belongs to before reading either.
Truncated scales
A bar twice as tall is not twice as big
When an axis starts at 90 rather than 0, a bar that looks double its neighbour may be a 5 per cent difference. Questions on these charts are usually asking for a percentage change precisely because the visual impression is so misleading.
Pie charts
Shares of a total that moves
Two pies for two years show shares, not amounts, and the totals behind them are usually different and stated in the caption. Comparing slice to slice across the two pies without applying the totals answers a question nobody asked.
Cumulative lines
A line that never falls is a running total
Cumulative series only ever rise or flatten. The value for a single period is the gap between two points, not a point. Flat stretches mean zero in that period, not a steady amount.
Category order
The x axis is not always time or evenly spaced
Charts sorted by size rather than by date, or with gaps in the years, invite you to read a trend that is not there. Read the category labels one by one on any chart where you are about to describe a direction.

How the wrong answers were built

Distractors on a professionally written numerical item are not random numbers. Each one is the output of a specific plausible misstep, which is useful to you in two ways: it tells you that an answer matching an option is not evidence you are right, and it lets you work backwards from a surprising option list to the mistake you were about to make.

Family 1
Right sum, wrong row
The correct method applied to the adjacent period, region or product line. Produced by reading across a dense table without tracking the row.
Family 2
Out by a factor of a thousand
Built from mixing an exhibit in thousands with one in millions, or a monthly figure with an annual question.
Family 3
Inverted denominator
Change over the new value instead of the original, or B over A instead of A over B. Always present on percentage change and ratio items.
Family 4
Subtracted back
On reverse percentage items: taking 20 per cent off 96 rather than dividing by 0.8. Gives 76.8 where the answer is 120.
Family 5
Points instead of per cent
The percentage point gap offered alongside the percentage change, or the other way around. Decided entirely by the wording of the stem.
Family 6
One source only
The answer you get from the first exhibit when the question needed two. The most dangerous family, because nothing about it feels incomplete.
Use the option spread as a diagnostic
If two options differ by exactly a factor of ten, a hundred or a thousand, the item is testing units, so go back and read the header. If two options are close together and the rest are far away, the item is testing which of two denominators the stem wanted, so reread the stem rather than recalculating. If the options are widely spread, estimation will settle it and exact arithmetic is wasted time.

The formats you will actually meet

The same reading skills are tested through quite different plumbing, and the plumbing changes your tactics more than the content does.

  • SHL Verify Numerical. Around eighteen questions on one clock of roughly seventeen to twenty-five minutes, tables and charts, calculator permitted, nothing deducted for a wrong answer. Guess rather than leave anything blank.
  • Aon cut-e scales numerical. The odd one out: instead of picking a value, you judge statements as true, false or cannot say against data spread over tabs, with marks deducted for wrong answers. Read the statement first and open only the tab that could settle it.
  • Talent Q Elements Numerical. Adaptive, individually timed questions, with a longer clock on the first question attached to a new table because that one includes reading it. Use that first clock to read the exhibit properly; it is being given to you on purpose.
  • Kenexa, now hosted by SHL. Twenty five-option questions in twenty minutes, business finance at school level, calculator allowed. One clock across the set, so slow early items take time from later ones.
  • Saville Swift Analysis. Eight minutes for the numerical section, which is short enough that reading efficiency dominates and the arithmetic barely matters.

A practice routine that targets reading rather than sums

Do
Work ten items with the calculator closed and your only job being to list which figures the question needs and where each lives. Do not solve them. This isolates the skill that is actually failing.
On every item you get wrong, classify the error into one of the six distractor families above. Almost everyone has one dominant family, and finding yours is worth more than another fifty mixed questions.
Say the units out loud as you take each figure: "eight hundred and forty, thousands of euros, for 2023". Errors that survive being spoken are rare.
Practise on the specific delivery format you have been invited to, especially if it is tabbed or per-question timed, because navigation cost is part of the test.
Do not
Do not study the exhibit before reading the question. A dense table holds four times more information than any single item needs, and reading it first is paying for data you will not use.
Do not treat an answer that matches an option as confirmation. Five of the options were built from wrong methods, and they were built from the wrong methods candidates actually use.
Do not recalculate when you are unsure. Reread the stem instead. The error is in the reading far more often than in the sum.
Do not carry the Aon skip habit into a test with no penalty, or you will hand back marks that were free.

Once the reading is reliable, the arithmetic is the small half, and the six shortcuts worth learning will make it smaller still. If you are not yet certain which vendor sent your invitation, work that out first, because the marking rule and the delivery format decide how you should be playing the clock. Timed practice in each of these formats is under numerical reasoning.

Practise on the stimulus, not just the sums

Every drill here uses the real formats: dense tables, tabbed documents, long option lists and the clocks the actual tests run.

Practise Numerical Reasoning