How We Calculate Our Benchmark Data
Where the data comes from
The figures we publish are drawn from EmployTest’s own assessment platform. Our benchmark dataset covers approximately 56,900 completed test administrations across 396 assessment titles, submitted between August 2021 and August 2026 by job applicants at the request of the employers who tested them. This dataset already reflects the account exclusions described below.
What the sample represents
These are job applicants who reached the assessment stage of a hiring process at an organization that chose to use skills testing.
They are not a random or representative sample of the U.S. workforce, of job applicants generally, or of employees in any occupation. Our client base skews toward small and mid-sized employers, and toward roles where measurable skills matter to the hiring decision.
Results should be read as a picture of applicants tested through EmployTest — not as a population estimate.
Minimum sample size. Unless otherwise noted, we do not publish a figure based on fewer than 100 test administrations. At smaller samples, the margin of error becomes wide enough that a published average could be misleading.
Score scale. Some assessments report a percentage of items answered correctly. Others report typing speed, keystroke rate, or a percentile rank. Percent-correct benchmarks are calculated only from assessments scored on that scale.
Client-specific assessments. Some assessments are built for a single organization. These are excluded from published benchmarks, since a figure drawn from one employer describes that employer rather than the applicant population.
Distribution checks. We exclude assessments where nearly all applicants score at or near the maximum. A benchmark built on a test that almost everyone passes reports the design of the test rather than the ability of the applicants, and gives a hiring manager nothing to work with.
How we calculate
Administrations, not individuals. Each figure is based on completed test administrations. An applicant who tests more than once is counted each time. We do not link records across tests, so we do not publish figures describing how an individual performed across multiple assessments.
Averages and medians. We report both. Where a distribution is skewed, the median is the better description of a typical applicant, and the gap between the two is itself informative.
Ranges. “Middle 50%” refers to the span between the 25th and 75th percentile scores — the range containing half of all applicants, excluding the top and bottom quarters.
Threshold figures. When we report the share of applicants scoring above or below a given percentage, that percentage is a reference point chosen for familiarity. It is not a passing score, and we do not set passing scores. See below.
Timed and untimed versions. Reported separately for individual assessments. Time limits reduce scores, and more so on harder material, so combining the two would distort both. Where we report a figure aggregated across many assessments, timed and untimed versions are pooled, and we note this alongside the figure.
Software versions. Where the same assessment exists for multiple software versions, we combine them into a single benchmark and note this alongside the figure. Where content differs meaningfully between versions, we report them separately.
Completion times. Median elapsed time from start to submission, excluding sessions under one minute or over three hours as likely abandoned or interrupted.
For Methodology for figures, click here.
On cut scores
We do not set passing scores, and we do not recommend one.
A cut score is a judgment about what a specific job requires. It depends on how central the skill is to the role, what a mistake costs, whether the employer intends to train, and how deep the applicant pool is. Those are decisions for the employer who knows the job — and a threshold not tied to actual job requirements is difficult to defend if challenged.
Where we report the share of applicants scoring above a given percentage, we are describing what a threshold would cost in candidates. That is an input to a hiring decision, not the decision.
Privacy
All published figures are aggregated. We publish no individual scores, no personally identifying information, and no figures attributable to any single client organization. No individual applicant or employer can be identified from anything we publish.
Corrections and updates
We recalculate published figures when we refresh the underlying data, and we note the extract date on every benchmark page. If we discover an error in a published figure, we correct it and note the change rather than replacing it silently.
A note on our interest
EmployTest sells the assessments this data comes from. We publish it because we think it is useful, and we document our methods here so you can judge it accordingly.
Page last updated: August 10, 2026 .