The AssessAll Blog
Assessment science, decoded weekly.
Measurement, hiring, proctoring, and the future of skills — written for people who make talent decisions.
77% of India's Web Traffic Is Mobile. Five Myths About Assessing Candidates on Phones.
Across 3.5 million applicants, mobile GMA scores ran substantially lower - and in a five-assessment study, measurement invariance held while mobile SJT means sat half a standard deviation below desktop. Five myths about device effects, tested against the evidence.
Read articleA 15-Item Adaptive Test Runs on a 95-Item Bank. That Ratio Is the Whole Decision.
Computerized adaptive testing reaches a given precision in fewer questions - but only off a large pre-calibrated bank. What CAT-ASVAB, NCLEX and PISA 2022 reveal about the real cost, and when a fixed form is the better call.
Work Sample Validity Fell From .54 to .33. The Test Didn't Change - the Correction Did.
Sackett et al. (2022) found range-restriction corrections had inflated selection validity estimates by .10-.20 points. What the correction does, where the overcorrection came from, the revised numbers, and six questions to ask before accepting a validity coefficient.
Applicants Score Half a Standard Deviation Higher on Conscientiousness. Here Is How to Audit Faking on Your Own Assessment.
Meta-analysis puts applicant score inflation at d = .52 on conscientiousness and .50 on emotional stability, and in one experiment Likert validity fell from .15 to .01 once the ideal-employee factor was removed. A seven-step audit you can run on your own applicant data.
Extra Time Is Not a Favour: A Practitioner FAQ on Assessment Accommodations
Extra time helps everyone, and candidates with disabilities somewhat more - it is not an advantage. Nine practitioner questions on assessment accommodations: what the evidence shows, why score flagging ended in 2000, and what the ADA, Equality Act 2010, EU law and India's RPwD rules each require.
A Percentile Is Not a Standard: Norm-Referenced vs Criterion-Referenced Scoring
Norm-referenced scores rank candidates against each other; criterion-referenced scores test them against a defined standard. A comparison of when each is correct, what the Standards and the Uniform Guidelines require, and why percentile cuts quietly expire.
Same Cut Score, Different Questions: A Practitioner FAQ on Test Forms, Item Banks and Score Comparability
Randomising questions is not the same as having multiple test forms. A practitioner FAQ on equating, item bank size, exposure control, retest effects and the quarterly checks that keep scores comparable.
31.5% of Distractors Are Dead Weight. Your Expert Reviewers Catch Two-Thirds of Them.
Most assessment questions are never checked against response data. Evidence from 2,948 distractors, an 80-year meta-analysis and 2026 AI item-generation research on the four item statistics that find broken questions expert review misses.
Reliability vs Validity: Your Assessment Can Be Perfectly Consistent and Completely Wrong
Reliability tells you whether a score repeats; validity tells you whether it means what you claim. A practitioner comparison of the two checks, with the maths on standard error of measurement, attenuation ceilings and what validity evidence actually requires.
70% Is Not a Cut Score: How to Set a Passing Mark You Can Defend
A cut score is a policy judgment, not a property of the test. A step-by-step guide to setting a defensible passing mark: modified Angoff panels, selection-ratio reality checks, measurement error at the boundary, and an adverse-impact check before launch.
No Single Right Answer, Real Predictive Power: The Science of Situational Judgment Tests
Five decades of research and a 2025 review of 524 studies show SJTs predict job performance with smaller group differences than cognitive tests. What they measure, why instruction wording matters, and how AI-scored open-response formats change the economics.
Can You Trust an AI-Graded Answer? What 2026's Scoring Studies Actually Show
Three 2026 studies settle the AI grading question: hybrid LLM scoring now matches trained human raters on open-response tests and beats them on consistency. What agreement benchmarks, bias checks, and architectures to demand before trusting machine-scored assessments.
Half the Questions, Twice the Signal: How Adaptive Testing Works
Computerized adaptive testing reaches the same measurement precision as fixed tests with roughly half the questions. How CAT engines work, what the evidence shows, and why test length is quietly killing your hiring funnel.
How Biased Is AI Assessment, Really? What 2026's Landmark Audits Reveal
A Stanford-led audit of 4 million applications and 150+ independent bias audits give the first real answer on AI assessment bias: fairer than humans on average, yet failing in one in ten jobs. How to audit your own stack.
AI Proctoring in 2026: What Integrity Bands Actually Measure
Remote assessment is only as good as its integrity signal. Here is how modern AI proctoring works — identity baselines, face detection, screen monitoring — and why a High/Medium/Low band beats a binary "cheated / didn't".
Want this thinking applied to your organisation?
Talk to the team behind the platform.
solutions@bodhih.com