The Diagnostic Approach: Why Most Candidates Make the Same Mistakes
Numerical reasoning test performance rarely hinges on raw mathematical ability. Instead, the gap between competent candidates and those who fail comes down to systematic error patterns. Most mistakes fall into four predictable categories: reading and interpretation errors (misunderstanding the data), calculation setup errors (solving the wrong problem), time management errors (rushing or stalling at the wrong moments), and pattern recognition errors (jumping to conclusions without checking context). The key to improvement is not grinding harder, it's categorizing where you're actually failing, then addressing that specific weakness. Candidates who improve fastest spend more time understanding why they got something wrong than doing more practice tests.
Reading and Interpretation Mistakes
Before you calculate anything, you must read the data correctly. This is where most candidates stumble hardest under time pressure.
Misreading row and column headers: In a typical numerical reasoning test, a table might show quarterly revenue broken down by region or product line. Candidates scanning quickly will confuse which column represents which quarter, or transpose rows. The fix: before reading any number, point to the headers (mentally or physically) and say them aloud. "This column is Q3. This row is Europe." Takes five seconds, prevents catastrophe.
Missing or misinterpreting units: A dataset reports figures in £k (thousands). A question asks for the answer in £m (millions). Candidates calculate correctly but forget to convert, or convert in the wrong direction. This is especially common in Saville and Cubiks formats. The fix: always extract and write down the units from the source data, then write the units requested in the question. Compare them side by side before answering.
Confusing time periods: "Revenue in Q3 2023" versus "revenue in Q3 2024" look identical while scanning. SHL tests frequently include year-over-year comparisons to catch this. The fix: circle or underline the year and period in the question before you look at the data. Use that as your anchor.
Misunderstanding absolute versus percentage values: A table shows both absolute numbers and percentage changes. The question asks for one; candidates calculate the other. For example, "market share increased by 5%" (absolute percentage points) is different from "market share increased by 5 percent of its previous value" (relative percentage). The fix: distinguish these explicitly. If the question says "increased by 5%," write "5% of previous value" next to it before you touch the calculator.
Calculation Setup Mistakes
Even candidates who read correctly often set up the calculation wrong. They solve a valid problem, just not the one being asked.
Applying percentage growth backwards: A candidate sees that revenue was £100k last year and grew by 20%. They calculate 100 × 0.20 = 20 and think the answer is £20k. But the question asked for the new total, not the growth amount. Worse: some candidates calculate 100 ÷ 1.20 and get £83k, inverting the logic entirely. The fix: write out the formula before plugging in numbers. "New value = old value × (1 + growth rate)." Then substitute.
Forgetting or misplacing denominators in ratios: A question asks "What is the ratio of product A to product B?" Candidates calculate A ÷ B correctly but then misread the answer format. Some tests want the ratio expressed as "A:B," others want the decimal. Candidates also mix up which number goes in the numerator. The fix: state the ratio aloud: "A to B" or "A is the numerator." Write the ratio in words before you write it in numbers.
Confusing average with median when data is skewed: A dataset has five values: 10, 12, 14, 16, 500. The mean is 110.4; the median is 14. A question asks for "the typical value," and candidates assume it means mean. They don't notice the outlier. The fix: glance at the data distribution first. If you see a number wildly larger or smaller than the others, suspect a skew. Calculate both if the question is ambiguous, then choose based on context.
Using the wrong base year for percentage changes: Year 1 revenue was £50k. Year 2 revenue was £70k. Year 3 revenue was £84k. A question asks "What is the total percentage growth from Year 1 to Year 3?" Candidates sometimes calculate the growth from Year 1 to Year 2, plus the growth from Year 2 to Year 3, and add them (30% + 20% = 50%). But compounding is not addition. The correct calculation is (84 − 50) ÷ 50 = 68%. The fix: identify the starting year and ending year explicitly. Write them down. Plug only those two values into the percentage change formula.
Time Management Mistakes
Speed under pressure exposes weaknesses. Poor time allocation costs more points than slow calculation.
Spending too long on hard questions: A candidate encounters a question with complex wording and a five-part calculation chain. They lock onto it, spending three minutes trying every approach. Meanwhile, they skip six easier questions that would have taken 45 seconds each. The fix: set a hard time limit per question (usually 90 seconds for numerical reasoning). When you hit it, make your best guess and move on. A wrong answer you can re-examine later is better than a skipped answer worth no points.
Not skipping questions strategically: Related to the above: many candidates assume they must solve every question in order. But the test doesn't penalize skipping. Some questions rely on a table or context that confuses you; others are simpler. The fix: on your first pass, flag hard questions with a mental note. Finish all the medium and easy ones first. Return to hard ones with whatever time remains. You'll often find that context from later questions illuminates an earlier one.
Second-guessing correct answers out of anxiety: A candidate answers a question, feels uncertain because of the time pressure, changes their answer at the last moment, and picks the wrong one. Anxiety, not logic, drove the change. The fix: your first instinct on a calculation is usually right. If you re-solve and get a different answer, solve a third time using a different method (e.g., estimation or working backwards). If two out of three matches, stick with that. If all three differ, you're rushing; mark it and move on.
Misreading due to anxiety-driven scanning: Under time pressure, candidates skim questions instead of reading them. They miss words like "not," "except," "increased by more than," or specific constraints. The question becomes a different problem than what was asked. The fix: read the question once slowly. If you're unsure, read it a second time aloud (or in your head if testing in a shared space). Two careful reads is faster than one careless read followed by recalculation.
Pattern Recognition Mistakes
Many numerical reasoning tests include trends, forecasts, and data interpretation. Candidates often jump to conclusions without checking the pattern holds.
Extrapolating linear trends from non-linear data: A graph shows profit growing 10%, 15%, 18% over three years. A candidate assumes it will grow 21% in year four (adding 3% each time). But the data is accelerating, growth might reach 20% or 25%, not a steady increase. Or the opposite: the data might plateau or drop. The fix: before forecasting, ask whether the data is accelerating, decelerating, cyclical, or random. If a pattern isn't obvious, don't assume linearity.
Missing seasonality or cyclical patterns: Sales data shows a consistent dip every Q1 and a spike every Q4. A candidate calculates an average or trend line without accounting for this. They forecast Q1 sales as higher than they'll actually be. SHL and Saville tests frequently include seasonal data to catch this. The fix: look at multiple years if the data is available. If you see a pattern repeated, it's likely seasonal. Adjust your forecast accordingly.
Ignoring context and industry-specific conventions: In manufacturing, a "defect rate" might be expressed as defects per million units. In finance, profit margins are typically expressed as a percentage of revenue. A candidate who doesn't know (or forgets) the convention will misinterpret the data. For example, a defect rate of 50 might be excellent (50 per million) or catastrophic (50%), depending on context. The fix: read any context provided in the question stem. If a metric is described, note its units and typical range. If something seems wildly off (a company with 500% profit margin), re-check the context.
The Self-Diagnostic Drill
Theory only works if you apply it. Here's a framework to accelerate improvement. Take a practice test under realistic conditions: timed, no interruptions, same format as your real test (SHL, Saville, or Cubiks). Don't retake it immediately. Instead, review every question you got wrong. For each error, categorize it into one of the four buckets: reading and interpretation, calculation setup, time management, or pattern recognition. Write the category next to your answer. Tally which category accounts for most of your errors.
If reading and interpretation dominates, spend your next two weeks drilling reading. Slow down. Re-read questions. Write down headers and units. If calculation setup is the problem, practice setting up formulas before solving. Write out the logic in words first, then numbers. If time management, take a second practice test and enforce the 90-second rule strictly. Skip liberally. If pattern recognition is weak, study datasets and make predictions before reading the question, then check your intuition against the correct answer.
Most candidates improve fastest by narrowing their focus to one weakness at a time. A 20% improvement in your weakest area usually beats a 2% improvement across all areas. After two weeks of targeted work on your category, retake a fresh practice test. Tally errors again. Your category distribution should shift. Repeat until reading/calculation/pattern recognition issues are rare, and time management is automatic.
For structured preparation with practice tests, diagnostic feedback, and progress tracking, explore our numerical reasoning test resources to refine your strategy.