top of page

Unfairness of Background Checks

Many believe that a background check is necessary to create and maintain a safe workplace. The underlying assumption is that it can identify people who would pose a risk to the safety of workplace. I have a question though. How many people that wouldn't have harmed the company, if hired, are denied employment because of a background check?


I use Bayes' Theorem to answer this question. First, a few assumptions. We need a "prior," which is the rate of "harmful" individuals in the labor pool. That is, what's the percentage of people in the labor pool that would end up doing something "harmful" to the company? By "harmful," I mean what background check companies tell us that their product helps employers identify, i.e. risks of violence, theft, or harassment. There's no way to give a proper estimate. Those who conduct interviews for a living can probably provide a reasonable estimate. Since I am not one, I'll just say 1 out of 20 or 5 percent. So the prior is 5%.


Now, about the accuracy of background checks. It's important to note that the risk for "false positives" (a "harmful" person passing the check) isn't necessarily correlated with the risk for "false negatives" (a "harmless" person failing the check). I make assumptions for each.


First, I assume that a given background check correctly predicts that a person is "harmless" in 90% of cases. That seems to me quite high, but let's go with it. In other words, in 10% of the cases, the background check (incorrectly) passes a harmful person as "harmless" because he has no criminal record. Another way to think about this is what's the likelihood that a person without a criminal record will later do something "harmful," e.g. by violence, theft, or harassment. The "sensitivity" of the background check is estimated to be 90%.


On the other hand, I assume that the background check gets it wrong 33% of the time when it fails the person. In 67% of the time it says the person is "harmful," the background check is right. In other words, in 1 out of 3 cases when the check labels a harmless person as "harmful" because he has a record, the person wouldn't have harmed the company if hired. (This is quite low. In my experience, more than 9 out of 10 people with a criminal record turn out to be good employees, but I use the lower estimate just to be conservative.) The "specificity" of the background check is 67%.


With these assumptions, this is what we get.

If the person passed the background check:

  • True Negative: The probability that he is "harmless" = 99.22%

  • False Negative: The probability that he is "harmful" = 0.78%


So far, this would seem to support the fairness of a background check. The employer is 99% sure that he hired a harmless guy. Good, right? Well, not so fast. We have to look at the people who failed the background check too.

If the person failed the background check:

  • True Positive: The probability that he is "harmful" = 12.55%

  • False Positive: The probability that he is "harmless" = 87.45%


In other words, there is 87% chance that the employer denied employment to an individual who does not pose a risk of harm to the company. The employer can be sure that the decision was fair to the denied applicant in only 13% of the cases.


In short, employers likely care only about the "true negative." What I care about is the "false positive." You can mess with the slides below to change the model's assumptions and see how they affect the (un)fairness value (i.e. true negative and false positive). For example, if you agree with me that 9 out of 10 people with a criminal record actually makes good employees, you'd change the specificity to 10%.



You might notice quickly that the most important factor is the percentage of "harmful" people in the labor pool. Sadly, I don't think we'll ever get an accurate estimate for this number, but my experience tells me less than 1 out of 10 people looking for work will ever do anything "harmful" in the way that background check companies would have employers fear, i.e. by violence, theft, or harassment. The background check is unfair, then, to the other 9.

Recent Posts

See All

True Cost of Background Checks

All of Us or None v. Hamrick In May 2021, the appellate court in California ruled in All of Us or None v. Hamrick that trial courts in...

Comentários


bottom of page