The accent in the interview

Two recordings of the same answer, read by the same actor in two accents, were played to a hundred people asked to rate the speaker's competence. The scores differed by a wide margin, and almost nobody reported noticing the accent as a factor.

The design matters. Because the words were identical, nothing about the content can explain the gap. Because the actor was the same person, nothing about the voice can either. What is left is the listener's association between a way of speaking and a set of expectations about education and reliability, applied within a second or two and then rationalised as a judgement about the answer.

This is one of the better-documented findings in applied linguistics, and it has an uncomfortable practical implication. Interview training usually teaches candidates to change the thing that is being judged: to flatten their vowels, slow down, adopt the accent of the organisation. It works, which is the problem. Advice that improves one candidate's chances while leaving the mechanism in place passes the cost to whoever cannot or will not follow it.

The alternative is structural. Panels that read anonymised written answers before meeting anyone, that score against a published rubric, and that record scores before discussion show much smaller accent effects. None of these changes asks a listener to be less biased, which is fortunate, because asking has never worked.

The finding is sometimes reported as proof that interviews are worthless. That overstates it. Interviews measure something real; they simply measure several other things at the same time, and a good process separates them instead of hoping that the listener will.