Hiring, and the filters · 4.6

Background checks, and their error rate

Background checks, and their error rate. What is actually the case, and how it compares with what is repeated.

A practical software reference for this part of the discussion is Monitask's overview of employee monitoring system.

A background check is a database query, databases contain errors, and the person who bears the cost of an error is the applicant.

Where the errors come from

Matching on name and date of birth rather than on a unique identifier. Records that were expunged or sealed and were not updated in a commercial copy of the data. Charges without conviction reported as though they were convictions. Records belonging to a different person with the same name.

Commercial screening firms buy court records in bulk and resell them, and the update path from a court correcting a record to a vendor correcting their copy is slow and in some cases absent.

Who fixes it

The applicant. American consumer reporting law gives a right to dispute and requires notice before an adverse decision, which is a real protection and operates on a timescale longer than most hiring processes.

By the time a dispute is resolved the job is filled. The remedy exists and the position is not restored, which is the general shape of remedies that operate after a decision.

Ban the box, and what happened

Policies removing the criminal history question from initial applications spread across American jurisdictions, intending to let applicants with records reach the interview stage.

Field experiments found an unintended effect. With the question removed, callback rates for young Black men without records fell relative to white applicants, apparently because some employers substituted inference for information.

The finding does not establish that the policy was wrong overall; it establishes that removing an information channel changes what employers use instead, and that the substitute can be worse.

Credit checks

Used by many employers for roles with no financial responsibility. The evidence that credit history predicts job performance or theft is weak, and several states have restricted the practice for that reason.

The strongest objection is circularity: financial difficulty is frequently caused by unemployment, and using it to screen for employment closes the loop.

Proportionality

Whether a record is relevant depends on the job, on how long ago it was, and on what it was. Guidance in several jurisdictions asks employers to consider those three rather than applying a blanket rule, and blanket rules remain common because they require no judgement.

An individualised assessment takes time and produces a decision somebody has to defend, which is precisely why it is a better process.

What an employer should check about the checks

Which vendor, what sources, how matching is done, how often data is refreshed, and what the dispute process looks like from the applicant's side.

Most employers buying screening cannot answer any of these, and the questions are answerable, because the vendor has documented answers and is rarely asked.

What an applicant can do

Obtain your own report before applying, from the same kind of source an employer would use. Errors found in advance can be disputed on your timetable rather than during a hiring process.

This is unequally available advice, since it costs money and time, and it is the only advice that works.

The wider point

Every screening step is an information channel with an error rate, and error rates are rarely stated. A check that is wrong two per cent of the time, applied to thousands of applicants, produces a predictable number of people wrongly excluded, none of whom appears in any statistic the employer sees.

Which is the recurring shape on this site: the cost falls outside the place where it would be measured.

Social media screening

Widely practised, rarely governed by policy, and it surfaces protected characteristics the employer is not permitted to consider and now cannot un-know.

Where it is done at all, having somebody outside the decision perform it and pass on only job-relevant findings is the arrangement that limits the damage.

How long a record should count

Evidence on reoffending finds that after a sufficient interval without further offences, the probability of a new offence approaches that of the general population. That interval is a matter of years rather than decades.

Few screening policies encode anything like it, and a blanket rule covering all records of all ages is treating a decaying signal as a permanent one.

The order of the steps

Running a check before an offer wastes money on candidates who will not be hired and exposes information the decision should not have had. After a conditional offer is both cheaper and cleaner, and it is what the guidance in most jurisdictions recommends.

For broader context, consult FTC background-check guidance.