Don’t Trust Your Gut: 3 Guidelines for Evidence-Based Recruiting

Chess Algorithms

Experience is generally good. Employers love job candidates with impressive track records. But when we on the hiring decision-making side start gaining experience in recruiting, there is a dark side that we need to be aware of if we still want to be effective.

The problem with growing experience in recruiting positions is that you start to gain confidence in your judgement. And that gut feeling about job candidates clouds the decision-making of even the best of us.

The essence of evidence-based recruiting is that you build your recruiting practice on the best available scientific evidence. What is scientific evidence? It is not expert opinions, TED Talks or blog posts. Why not? Because they are opinions without rigorous methodology backing them up. Sure, there is often wisdom in the words of HR influencers, but in order to be effective, basic evidence-based guidelines should be in place.

In the core of evidence-based recruiting should be a hiring algorithm. Algorithm is simply a formula that calculates the score of each of your job candidates. Algorithmic decision-making is simple – you hire the candidate with the highest score. But an algorithm won’t work without variables. It is the recruiter’s responsibility to build the formula – decide what kind of data to gather from the candidates and which factors matter the most. But where to start?

Screening methods – the fairest of them all

I/O psychologists have been studying selection methods with meta-analytic methods for around a 100 years, and there is a clear consensus that General Cognitive Ability (GCA) – also known as General Mental Ability (GMA) or Intelligence Quotient (IQ) – is the most versatile and powerful of the methods commonly in use. Considering how simple-to-use and cheap methods there are available, it is a mystery why these tests are not more widely adopted in practice.

Especially as a screening method, GCA measure is powerful for a couple of reasons. First, for most jobs, the job requirements aren’t set in stone. Especially in startups or companies working in dynamic markets, the contents of employees’ jobs tends to change a lot. GCA is a measure that indicates how well the candidate would be able to learn new things. Second, and related, when the job requirements are complex or new, higher information processing capacity, which is what GCA essentially measures, helps candidates perform better.

Research suggests, that the best predictive validity is achieved when GCA is coupled with other methods that preferably are “MECE” – mutually exclusive and collectively exhaustive. This means that the other methods used should be strong as well, but they should measure different constructs that GCA tests measure. Famous companies such as Google measure GCA together with other variables – namely, “Googleyness” – that they have internally found predictive for future performance. Some evidence-based factors found in I/O psychology are conscientiousness and integrity, and most companies would actually get better results with these methods than with using classic unstructured job interviews as a go-to method. But I bet that…

You are going to interview anyway, so here is how to do it right

One common mistake that many recruiters make is not structuring their job interviews.

How do you expect to compare the candidates if you ask each of them different questions? And how do you expect to hire actual talent if you let human error come in between? If you use the so called “free talk” method (the losing method) to interview candidates, you are bound to simply get along better with some candidates than with others. If the recruiter was changed, the result would most likely be different too, and this is not a good indicator of the reliableness of the interview.

Structuring interviews takes some work, but it’s principles are fairly simple. Essentially, structured interview is an employment interview where

  1. the same questions are asked of each candidate in the same order
  2. free talk is minimised
  3. the evaluation criteria for each question are determined beforehand

The two best types of questions are behavioral and situational. Behavioral questions ask about candidates’ past performance in order to predict how the candidate is likely to perform in the future. Situational questions present hypothetical situations and ask how the candidate would proceed in a given situation.

The outcome of designing the structured interview should be an “interview booklet”. This guide provides a set of predetermined questions (based on variables you have deemed to be necessary for success in the job), room for note-taking and a guide for evaluation. It should be written in a way that anyone even without recruiting experience would be able to run the interview.

If you want to be really professional, have interviewers write down the answers of each candidate, and let someone else evaluate the answers. This obviously takes time, and you need to make the call whether the added value is worth it.

Decision time? Enter Excel

So. You have built your hiring algorithm (hopefully based on GCA and other reliable variables) and collected data to measure those variables using tests and structured interviews. Now it is time to be humble, and let your new best friend Excel make the decision for you.

When you let an algorithm decide for you, you are going to get an improvement of about 50% in predicting work performance. And the interesting fact is that even the most experienced recruiters with years of experience fail more often than algorithms.

Let’s go one step further than that. Even when there is a group of experts, and when they have more data available than your excel table (the algorithmic decision-maker), their decisions are worse. Why is this and what can you do to improve?

A likely reason, as mentioned, is that these bad choices arise from various psychological biases. We as humans are overly influenced by first impressions, personalities and our own values, among other things. Because hiring decisions are essentially prediction problems – ”which candidate would perform the best in the job?” – we should use statistical algorithms which are tools originally built for prediction problems.

This does not mean that experts are unimportant. They are a great source of insight in building the algorithm in the first place. But it does mean that HR professionals need to be humble and understand their limitations. Hiring managers need to be aware and continuously measure the success factors for each job in their company, but they need to restrain themselves when the decision-time comes.

Evidence-based decision-making is the first step towards next-generation recruiting. Most of the algorithmic methods discussed here are going to be adopted to various HR tech applications in the future, but by knowing the basics, you can already start making better decisions while waiting for Big Data and AI to become mainstream in the industry.

Further reading:

Danieli, O., Hillis, A., & Luca, M. (2016). How to Hire with Algorithms. Harvard Business Review,

Kuncel, N. R., Klieger, D. M., Connelly, B. S., & Ones, D. S. (2013). Mechanical versus clinical data combination in selection and admissions decisions: A meta-analysis. Journal of Applied Psychology, 98(6), 1060.

Levashina, J., Hartwell, C. J., Morgeson, F. P., & Campion, M. A. (2014). The structured employment interview: Narrative and quantitative review of the research literature. Personnel Psychology, 67(1), 241-293.

Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in personnel psychology: Practical and theoretical implications of 85 years of research findings. Psychological bulletin, 124(2), 262.

Schmidt, F. L. (2002). The role of general cognitive ability and job performance: Why there cannot be a debate. Human performance, 15(1-2), 187-210.

If you want to share this article the reference to Max Korpinen and The HR Tech Weekly® is obligatory.