Most technical interviews measure two things: whether the candidate can code, and whether they can code while a stranger stares at them. Those are not the same skill, and a controlled study at NC State showed just how badly the second one contaminates the first. Being watched cut candidate performance by more than half.
Key Takeaways
- In a randomized study at NC State, coding under observation (whiteboard, watched) cut performance by more than half versus a private setting (Behroozi et al.).
- Live coding is fast but anxiety-heavy, which distorts the signal.
- Take-home tasks measure real code quality but can be abused with long unpaid asks for a low pass rate.
- The format is itself a signal candidates judge you on when deciding whether to accept.
What the Research Found
Researchers at NC State ran a randomized controlled trial comparing developers solving problems on a whiteboard while watched against developers solving equivalent problems privately. Performance dropped by more than half simply from being observed (Behroozi et al.). Read that as a measurement problem: a chunk of what a whiteboard interview measures is stress tolerance and performance-under-observation, not engineering ability. You can reject a strong engineer who freezes on a whiteboard and hire a mediocre one who's comfortable performing.
That doesn't make live coding useless, but the classic "watch them code on a whiteboard" format is a noisy instrument, and the noise correlates with things you don't care about (social confidence) instead of things you do (can they build).
Comparing the Formats
| Live coding | Take-home | |
|---|---|---|
| Measures | Thinking out loud, comfort under pressure | Real code quality, architecture, tests |
| Time | 60–90 min | Days, in their environment |
| Main risk | Anxiety distorts signal (halved performance) | Big unpaid time-ask, low pass rate |
| Best for | Communication, collaboration | How they actually build |
Each format measures something real and misses something real. Live coding shows you how someone thinks and collaborates in the moment, but taxes it with performance anxiety. A take-home shows you how they actually write code, structure a solution, and test it, but asks for hours of unpaid work, and a bloated take-home (15 hours for a slim chance of passing) is both a bad signal and a candidate-experience disaster (industry analysis of the two formats).
A Concrete Version
Two candidates for a senior role. In the whiteboard round, Candidate A is relaxed, narrates smoothly, and gets a "strong yes." Candidate B, who has shipped far more serious systems, blanks under the stare and gets a "no." Now give both a scoped, three-hour take-home in their own editor: B returns clean, well-tested, well-structured code; A returns something that works but is sloppy. The NC State result is exactly this gap, and the whiteboard round had it backwards.
The Honest Counterpoint
Take-homes aren't a free win, and "just do take-homes" ignores their real problems. They favor candidates with spare time (bad for parents and people already working full-time), they can be gamed with AI or a friend, and an unbounded take-home is exploitative. Live conversation also catches things a take-home can't: how someone reasons, communicates, and responds to a curveball. The answer isn't one format. It's a humane mix: a short, scoped, respectful take-home or a collaborative (not adversarial) live session, judged on a rubric rather than vibes, which is the heart of a structured interview.
How We Think About It
Our own vetting leans on realistic, low-anxiety evaluation, real problems in a normal setting, scored against a rubric, precisely because the research says a stressed whiteboard measures the wrong thing. It's part of why our five-stage process predicts retention: we're measuring engineering, not stage fright. See available engineers.
Frequently Asked Questions
Are whiteboard interviews accurate?
Less than people think. An NC State study found being watched cut coding performance by more than half, so a whiteboard partly measures stress tolerance rather than engineering ability.
Are take-home assignments better?
They measure real code quality, structure, and testing better than a stressed whiteboard, but they ask for hours of unpaid work and can be gamed. Keep them short, scoped, and respectful of the candidate's time.
What's the best technical interview format?
A humane mix: a short scoped take-home or a collaborative live session, judged on a rubric rather than gut feel. The goal is to measure how someone builds, not how they perform under a stranger's gaze.
Does the interview format affect hiring?
Yes, both ways. It changes what you measure, and candidates judge you on it when deciding whether to accept an offer.
The Bottom Line
A stressed whiteboard interview can cut a candidate's measured performance in half, which means it's partly testing the wrong thing. Use formats that measure how people actually build, keep them humane and scored on a rubric, and you'll stop rejecting strong engineers for the crime of being nervous.
Roberto Espinoza is CEO of Ruzora, which helps US startups hire pre-vetted senior LATAM engineers in 72 hours. See available engineers.
