A numerical reasoning test is an aptitude test of how well someone can interpret and use numerical information. Candidates typically get a table, chart or short business scenario and have to answer questions about it: what percentage changed, which region grew fastest, what the total cost would be. The arithmetic is rarely hard; the test is about finding the right figures and reasoning with them accurately under time pressure.
For employers, the useful questions are whether the role needs one, which test to pick, where to set the bar, and how to use it without shutting out good candidates unfairly.
What does a numerical reasoning test measure?
Most numerical reasoning tests measure some mix of:
- Data interpretation: reading values from tables and charts correctly
- Percentages and ratios: changes, shares, conversions
- Rates and proportions: per-unit costs, growth rates, averages
- Multi-step reasoning: combining figures from more than one source to answer a question
- Speed and accuracy under time pressure
They are different from a maths test. A maths test checks whether someone knows methods; a numerical reasoning test checks whether they can apply simple methods to unfamiliar information quickly and correctly, which is closer to what most office jobs ask.
Example question (illustrative)
A region sold 4,000 units in Q1 and 5,000 in Q2. Unit price fell from 20 to 18. By how much did revenue change?
Q1 revenue is 80,000; Q2 is 90,000; revenue rose by 10,000, or 12.5%. The skill being tested is noticing that both volume and price changed.
Which roles need a numerical reasoning test?
Start from the job, not from the test catalogue. The US Office of Personnel Management (OPM) describes job analysis as the foundation for all assessment and selection decisions: you should be able to point to the tasks that make the test relevant.
| Role type | Typical number work | Numerical reasoning test? |
|---|---|---|
| Finance, accounting, audit | Reconciling, analysing variances, forecasting | Yes, often alongside a role-specific task |
| Data, business and operations analysts | Reading dashboards, comparing trends, sizing problems | Yes |
| Graduate and management trainee schemes | Varied, often includes analysis | Common, as one part of a wider process |
| Sales and account management | Pricing, discounts, pipeline maths | Sometimes; a pricing scenario may be better |
| Retail and hospitality supervisors | Stock, rotas, cash handling | A short practical task is usually better |
| Creative, care and most frontline roles | Little | Usually no |
If numbers are a minor part of the job, a general numerical test adds a hurdle that predicts little and can screen out people unfairly.
Get used to being assessed before it counts
Run an AI job simulation of the real work and get a scored report on where you stand. 3-day free trial, cancel anytime.
How do you choose a numerical reasoning test?
Questions to ask a vendor:
- What level is it pitched at? Tests for graduates and managers are harder than tests for operational roles. A test that is too hard for the role rejects people who would do the job well.
- What content does it use? Business-style tables and charts suit office roles; some tests use more abstract number series, which are further from most jobs.
- What evidence is there that scores relate to job performance? Ask for validation evidence for roles like yours, not just general claims.
- What comparison group are scores reported against? Many tests report a percentile against a norm group. Check that the group resembles your applicants.
- What adverse impact data do they have? Ask how different groups score and what the vendor recommends.
- What accessibility options and adjustments are supported? Extra time, screen-reader compatibility, alternative formats.
- How is it delivered and supervised? Online, unsupervised tests are convenient; consider how you will confirm results for shortlisted candidates.
For a closer look at one widely used test, see our guide to the SHL numerical reasoning test. For a broader view of ability testing, see cognitive ability tests for hiring.
How should you set the cut-off score?
A cut-off decides who goes forward, so it needs more thought than "top half" or "60%".
- Match it to the role. Decide what level of numerical work someone needs to do the job adequately from day one.
- Benchmark with current staff. If you can, ask people who already do the job well to take the test. A bar most of them would fail is too high.
- Consider a band, not a hard line. Treat scores just below the line as "review" rather than automatic rejections, especially when other evidence is strong.
- Watch the effect on different groups. The EEOC's guidance says that if a test disproportionately screens out a protected group, employers should look for an equally effective alternative with less adverse impact. Our explainer on the four-fifths rule shows a common way to check.
- Review after a hiring cycle. Compare test scores with how hires actually perform and adjust.
How to run a numerical reasoning test fairly
This is general information, not legal advice.
- Tell candidates in advance. Say in the advert or invitation that there is a numerical test, roughly what it covers and how long they will have. Point them to the publisher's practice material.
- Offer adjustments. In the UK, Acas includes job applicants in the reasonable adjustment duty and gives extra time on written or reading tests as an example. In the US, the EEOC says employers must provide reasonable accommodation in test administration unless it would cause undue hardship, so that results reflect the skill being measured, not the impairment.
- Treat everyone the same. The EEOC says employers cannot adjust scores or use different cut-offs based on protected characteristics.
- Do not let one test decide. Combine the score with structured interviews and a realistic task. A strong candidate can have a bad day on a timed test.
- Validate for each use. The EEOC's best practice is to validate tests for the specific positions and purposes they are used for; vendor documentation helps, but the employer stays responsible.
Numerical reasoning test or a job task?
A generic numerical test tells you how someone handles generic numbers quickly. A task built from the job tells you how they handle your numbers: a variance analysis with your kind of data, a pricing decision, a stock reorder. Many employers use both: a short ability test early to manage volume, then a realistic task for the shortlist. Our guides to work sample tests and how to assess numerical reasoning cover the second approach.
Checklist
- [ ] Number tasks in the role listed from a job analysis
- [ ] Test level and content matched to those tasks
- [ ] Vendor validation and adverse impact evidence reviewed
- [ ] Cut-off set from the role and checked against current staff
- [ ] Candidates told about the test in advance, with practice material
- [ ] Adjustments offered to everyone
- [ ] Test combined with interviews and, ideally, a job task
- [ ] Results reviewed against job performance after a cycle
FAQs
What is a good score on a numerical reasoning test?
There is no universal good score. Most tests report a percentile against a comparison group, so a score only means something next to the group and the role. Set your own bar from the job and from how current good performers score.
Are numerical reasoning tests fair?
They can be, if the role genuinely needs the skill, the test is pitched at the right level, adjustments are offered and the results are monitored for adverse impact. Using a hard test for a role with little number work is where unfairness usually creeps in.
Should candidates be allowed a calculator?
Follow the publisher's instructions for the test you use. If you design your own task, allow whatever tools the person would use on the job.
