What the interview is actually measuring

"Strong analytical skills" is not a competency anyone can measure in 45 minutes. "Writes correct SQL against an unfamiliar schema", "designs and reads an A/B test", "chooses a metric for a new feature" and "disagrees with a stakeholder without losing them" are: each one is a thing an interviewer can watch you do or ask you to evidence. Turning the job description's adjectives into six to eight observable competencies like these is the first move, and the build-along does it on Ren's job description.

The mapping in the figure is Ren's constructed loop, but the shape is common: the technical rounds each read one or two competencies deeply, and the behavioural conversation reads several broadly, through stories. That asymmetry decides where prep time goes. A candidate who spends three weeks on SQL puzzles and an evening on stories has prepared one round for a loop where stories are heard in several.

Some job descriptions are a wish list copied from three other roles. Two moves help. First, weight what appears in the "what you will do" section over the requirements list: duties describe the work, requirements often describe an imagined candidate. Second, ask the recruiter in the screen what the loop looks like and what each round focuses on. That is an ordinary, expected question, and the answer is the most reliable map you will get. Session 6 has more on the questions candidates ask.

Schmidt and Hunter (1998, Psychological Bulletin) summarised 85 years of research on 19 selection procedures. Their table put work sample tests at .54, general mental ability and structured interviews both at .51, job knowledge tests at .48 and unstructured interviews at .38, as correlations with later rated job performance. For two decades those were the numbers in every textbook and hiring guide.

Sackett, Zhang, Berry and Lievens (2022, Journal of Applied Psychology) re-examined how those estimates had been corrected for range restriction, the statistical adjustment for the fact that a study only sees people who were hired. They argued the corrections were often too large, especially in concurrent studies run on current employees, and recommended no range-restriction correction for those. Most validities fell by .10 to .20. Structured interviews came out at .42 and ranked first; job knowledge .40; work samples .33; cognitive ability .31; unstructured interviews .19. They also showed where the old .51 came from: an observed correlation of .28, raised to .37 by one correction and to .51 by the range-restriction step.

Oh, Le and Roth (2023, JAP) reanalysed Sackett's own data and argued that it substantially underestimates how severe range restriction is, and that the no-correction recommendation should be reconsidered. Ones and Viswesvaran (2023) disagreed that cognitive ability's validity is much lower than previously reported. Sackett and colleagues' reply (2023) agreed correction is right when the information about the applicant pool exists, and argued that in meta-analysis it usually does not. The abstracts of the first and third were read; the second in full.

This page does not resolve that argument, and neither should you. What a candidate can take from it is the part neither camp disputes: a structured interview predicts performance better than an unstructured one, in both sets of numbers. The size of the cognitive ability estimate is where they fight, and that is a question for the people designing tests, not for you preparing answers.

Questions based on a job analysis, the same for each candidate. The questions come from the competencies, so the job description predicts them. Each answer rated, on anchored scales. An answer is scored against descriptions of weak, adequate and strong evidence for that one competency, so wandering into a different competency earns nothing on this question. Notes taken. What you say is written down and compared later; specific actions and results survive into the notes, adjectives do not. Limited prompting and follow-up. The interviewer may not rescue a vague answer with "and what did you do then?", so the structure of the answer has to be yours.

Not every loop is fully structured, and the people interviewing you on any given day vary. Preparing for the structured version costs nothing in an unstructured one: clear evidence for the competency asked about is a good answer in both rooms.

The claim is usually credited to a study of interviewers' decision time. Frieder, Van Iddekinge and Raymark (2016) studied 166 interviewers and 691 applicants at a university career centre. Their abstract reports that more experienced interviewers and more rapport building went with quicker decisions, and that asking every candidate the same questions went with longer decision times. A reported breakdown of the same study puts only about 5 percent of decisions in the first minute and about 30 percent within five minutes, meaning most came later. Those percentages were seen in a secondary summary, not in the paper itself, so treat them as reported.

The "thirty seconds" version usually points at Ambady and Rosenthal (1993), who found that strangers' ratings of 13 college teachers from silent clips under 30 seconds predicted end-of-semester student evaluations. That is a study of teachers on silent video, not of hiring decisions. First impressions matter in interviews; the specific minute counts do not come from where they are said to.