Evidence
Structured assessment is backed by decades of psychometric research. Here is what the evidence shows, and how we apply it.
The evidence
Decades of meta-analytic research point in the same direction: structured assessment is among the strongest available predictors of job performance. The word doing the work in that sentence is structured. It is not the presence of an assessment that predicts anything; it is asking every candidate the same things, scoring them against the same criteria, and recording the reasoning.
The unstructured interview fares badly by comparison. It is weakly predictive, highly susceptible to similarity and first-impression bias, and unusually convincing to the person conducting it — a combination that makes it the most trusted weak instrument in professional life.
Situational judgement tests, matched forced-choice formats and work samples show consistently stronger validity across role families, because each asks a candidate to do something rather than describe themselves. Combine them and validity improves again: methods with different weaknesses correct one another, which is the statistical case for triangulation and the reason both of our assessments are multi-method by design.
We are careful about what we claim from this. Published validity belongs to the method, not automatically to our instrument. Our own criterion validity is established at pilot; until then we describe our position as evidence-informed and provisional, and we say the same thing in the reports.
What the evidence supports
Structure beats intuition
Fixed questions and defined scoring criteria outperform free-form conversation, consistently and across role families.
Doing beats describing
Judgement tasks and work samples predict better than self-description, because they are harder to rehearse.
Methods correct each other
Combining formats with different weaknesses raises validity beyond any single method used alone.
Structure reduces bias
A common process and a recorded rationale narrow the space in which unexamined preference operates.
For how we apply these principles in practice, see how it works.
The case for structured assessmentPredictive markers
Research identifies specific behavioural markers that predict performance at work. These markers tend to be quiet — easy to miss in an interview, and powerful in predicting what happens afterwards.
Executive Leadership Assessment
The ELA is built around markers that are hard to perform on demand: how a leader handles a genuine trade-off, how they describe someone who disappointed them, whether they name their own error before anyone asks, and what they do with critical feedback.
None of these are visible in a polished answer about strategy. Together they predict the things senior appointments actually turn on — team stability, decision quality and retention.
These markers are difficult to fake and they predict real-world outcomes. They are the foundation of the ELA. Read more on executive hiring.
Sales Talent Assessment
The STA works the same way for sales roles: how a seller handles rejection, what they protect when they cannot do everything, how they describe a customer, and whether they correct themselves unprompted.
These are the markers that separate a strong interview from a strong quarter. They predict quota attainment, pipeline quality and retention — and they are largely invisible to charisma.
These markers are difficult to fake and they predict real-world sales outcomes. They are the foundation of the STA. Read more on sales hiring.
Methodology
Mint Psychometrics is built on five non-negotiable principles drawn from psychometric research and assessment best practice. They are written into our specifications, and an item or a score that breaks one of them does not ship.
Principle 1
No dimension rests on a single method
Every core dimension draws its evidence from at least three separate parts of the assessment. A score is never the output of one item type, so no single method can distort the picture on its own.
For the ELA that means judgement scenarios, forced-choice blocks, trade-off rankings, written reasoning and a structured verbal assessment. For the STA it means situational judgement, forced-choice patterns, commercial reasoning, ranking tasks, written responses and risk indicators — each core dimension evidenced across at least three of the six sections.
Triangulation is also what allows us to report disagreement between methods rather than hiding it inside an average.
Principle 2
Transparent self-report is one signal among several
Direct self-report is treated as one signal, never a primary source. Every dimension carries at least one non-transparent evidence source, so a candidate who knows what a good answer sounds like still has to demonstrate it.
Judgement scenarios, forced-choice blocks and reasoning items are materially harder to game than a rating scale. In the forced-choice sections, statements within a block are matched for social desirability by an independent rating panel, which removes the obviously attractive option — the single most important defence against impression management.
Self-reported reflection is also held to a standard of specificity: a vague claim about always learning from mistakes scores low, while a concrete instance with a named lesson and a changed behaviour scores high.
Principle 3
Scores support comparison across candidates
Every reported dimension score sits on a common one-to-five scale, so two candidates assessed months apart can still be compared meaningfully.
Profile shape — which dimensions are relatively strongest for an individual — informs interpretation, but ipsative data is never the sole basis for a dimension score. Forced-choice blocks are modelled with a rank-order approach rather than simple summation, precisely because summation produces ipsative distortion.
This is what makes benchmarking and cohort analysis defensible rather than decorative.
Principle 4
Every item maps to observable work behaviour
Every item must map to an observable behaviour in the work it is hiring for. Items that cannot be linked to a realistic scenario are rejected at blueprint stage, before they ever reach a candidate.
This is why the assessments read as work, not as a personality quiz. Scenarios force a choice between legitimate priorities, because that is the shape of the actual job.
It also keeps the assessment defensible: every score traces back to behaviour a manager could recognise on a Tuesday afternoon.
Principle 5
All weights and cut-points are explicitly provisional
Every weight, band and cut-point in our current scoring architecture is an expert-derived prior. They are documented, versioned, and explicitly labelled provisional until Stage 9 recalibration against pilot criterion data.
Some contributions carry a deliberate discount until their model is validated — the forced-choice sections currently enter at a reduced multiplier, and written responses carry a reduced weight until enough double-marked responses exist to equate the prompts.
We would rather tell you a parameter is provisional than present a borrowed number as settled science.
Full methodology detail lives in our design and criteria specifications, which we share with assessment professionals and internal stakeholders on request — see resources below.
Quality
Validity and reliability are non-negotiable. Here is how we protect them, and where we are honest about the work still to do.
Reliability targets
Each dimension carries a defined reliability target before it is allowed into a live form. Single-method targets sit in the region of .75 to .80, and multi-method triangulation lifts the composite target to .85 or the documented figure for that dimension.
In plain language: reliability is the extent to which a score is stable rather than accidental. A reliable score would come out much the same if the candidate sat a parallel form on a different Tuesday. Where a dimension is inherently harder to measure — self-reported learning behaviour is the usual culprit — we set a lower target openly and apply a documented contingency rule instead of pretending otherwise.
Written responses are double-marked on a sample, with an inter-rater agreement target and remedial training if it is missed. Reliability is monitored after launch, not signed off once.
STA dimensions — documented targets
| Dimension | Target | Weight |
|---|---|---|
| Commercial reasoning | Composite ≥ .85 | 13% |
| Achievement drive | Composite ≥ .75 | 15% |
| Prospecting resilience | Composite ≥ .75 | 12% (17–20% hunter/SDR) |
| Customer insight | Composite ≥ .78 | 14% |
| Influence judgement | Composite ≥ .80 | 16% |
| Learning agility | Composite ≥ .72 | 10% |
| Execution discipline | Composite ≥ .78 | 13% |
| Integrity and trust orientation | Composite ≥ .78 | 7% (threshold rule) |
Weights are provisional expert-derived priors, recalibrated at pilot. Integrity and trust orientation carries a low weight but a non-compensatory rule: a low score triggers review regardless of the overall profile.
ELA expectations — evidence sources
| Expectation | Evidenced through |
|---|---|
| Self-awareness | Forced-choice blocks, written responses, verbal assessment |
| Performance orientation | Judgement scenarios, trade-off rankings, verbal assessment |
| Motivation | Forced-choice blocks, written responses, verbal assessment |
| Talent focus | Judgement scenarios, written responses, verbal assessment |
| Future focus | Judgement scenarios, trade-off rankings, verbal assessment |
| Integrity and respect | Risk indicators, judgement scenarios, verbal assessment |
ELA construct mapping is evidence-informed and role-relevant, and explicitly provisional until validated. Per-expectation reliability targets are fixed at pilot against live response data rather than asserted in advance.
Construct, predictive and discriminant
Validity is established along three routes. Construct validity asks whether an instrument measures what it claims. Predictive validity asks whether it forecasts performance at work. Discriminant validity asks whether it avoids measuring things it should not — which is why every dimension in our specifications carries explicit boundaries against its neighbours, so influence judgement is not quietly rescored as commercial reasoning.
Predictive validity against real client outcomes is established at pilot, alongside factor-structure checks and item-level statistics. Options that almost nobody or almost everybody selects are flagged for revision rather than left in place.
Until that work completes, our validity claim is grounded in the research literature and expert consensus, and we label it that way rather than implying a criterion study we have not yet run.
Designed to reduce bias
Content is reviewed for cultural neutrality and accessibility before it goes live. Scenarios use universal business situations rather than region-specific norms or idioms, and differential item functioning analysis runs at pilot and annually afterwards. Where a group difference appears it routes to item review, never to action against a candidate.
Demographic data is captured separately from assessment data and is never an input to scoring or to any model. Adjustments — extended time through pausing the timer, screen reader support, keyboard and tap alternatives — are available on request and are recorded transparently.
The wider point is structural: a common process, a defined rubric and a recorded rationale leave far less room for unexamined preference than a conversation that goes wherever the interviewer takes it.
Full detail sits in our trust and governance centre.
Resources
Our research summaries and methodology documentation. Technical specifications are shared with assessment professionals and client stakeholders under a confidentiality agreement, because they contain the scoring architecture.
Why structured assessment outperforms the traditional interview, and what the evidence does and does not support.
Last updated: In preparation
Coming soon — request early accessHow we design for fairness, reduce bias, review content for cultural neutrality, and support accessibility adjustments.
Last updated: In preparation
Coming soon — request early accessThe full technical specification for the Executive Leadership Assessment: structure, timing engine, scoring pipeline and review workflow.
Last updated: v1.1 · August 2026
Request accessWhat the ELA measures: expectation definitions, talent themes, behavioural anchors, evidence sources and keying rules.
Last updated: v1.2 · August 2026
Request accessThe full technical specification for the Sales Talent Assessment: six-section structure, the forty-minute timing law, scoring pipeline and marking workflow.
Last updated: v2.0 · August 2026
Request accessWhat the STA measures: the eight dimension definitions, discriminant boundaries, reliability targets and provisional weightings.
Last updated: v2.0 · August 2026
Request accessDocuments are branded and versioned, and every release records the parameter versions it was written against. Request access and we will send the current version with its revision date.
Questions
Tell us about the role you are hiring for. We will walk you through the methodology, share an example report, and set out what a pilot would look like.
Talk to us about a role