MARCH 10 — Since its publication, “Racism in Recruitment: A Study on Racial Bias for Entry Level Jobs in Malaysia” by the Centre for Governance and Political Studies (Cent-GPS) has been criticised heavily on Twitterjaya.

Without doubt, if submitted to any peer-reviewed journal, it will be rejected.

Readers may find the paper here.

I will focus on two issues left generally untouched by others.

Advertisement

Firstly, the conceptual nuances of labour market discrimination that the paper conveniently ignored.

Secondly, the social ramifications of this methodologically flawed paper in the quest to making Malaysia a less racist country.

Is discrimination that simple?

Advertisement

An important caveat to note — no one with a sound mind will dispute that the labour market in Malaysia is exceptionally racist.

Evidence and theories in labour economics suggest that there are two types of discrimination — taste-based and statistical.

Taste-based discrimination covers the kind of racism typically that we think happen.

On the other hand, an employer may be discriminating between applicants, or employees, categorically because they don’t have enough information about the applicants at hand.

In this case, at least the racism ingrained in him/her does not play into the hiring equation.

Let’s proceed with a thought experiment.

Suppose someone applies for a job and does not get it. A candidate, of a different ethnicity or gender, with the same CV applies for the job and gets the offer instead. Can we conclude that the employer is racist?

Strictly speaking, no. The bar in identifying taste-based racism is profoundly high. Broadly, two things may have caused this outcome.

One, the employer really has an unethical disdain for people of the applicant’s ethnicity. In this case, the employer is racist and discriminates by taste.

Two, the employer realises that the CV is a bad gauge of productivity after the candidate is hired.

However, these productivity determinants cannot be observed throughout the hiring process, leaving the employer to resort to past performances (or hearsay) of employees, grouped by observables such as ethnicity, gender, school, location etc.

Now, imagine the employer magically has all the information about a candidate, beyond what can be captured in an interview, much less a CV.

If there is still this systematic preference against one ethnicity, then yes, the employer falls into category one.

In category two, the employer turns out to no longer discriminate anyone by ethnicity, gender etc because he/she can gauge the candidates’ productivity accurately.

Clearly, solving taste-based discrimination requires more than reforms in labour markets. Making the market more efficient, making flows of information more seamless, will only help if the discrimination that is going on is statistical.

How do we figure which one is the case?

Firstly, we need a randomised controlled experiment, which was exactly what the papers cited did. Both Bertrand and Mullainathan (2004) and Khalid and Lee (2016) did these. The scale of control on the size, characteristics and selection of CVs and applications were immense.

There were in excess of 3,000 – 4,000 CVs, all with varying names, while education, experience, skills and other characteristics were carefully balanced.

The number of firms considered also exceeded what Cent-GPS did.

Respectively, they submitted the CVs to an excess of 1,300 and 689 firms.

In comparison, from the methodology chapter itself, Cent-GPS appeared to have seven unbalanced CVs and a way lower number of firms considered.

Secondly, we need some help from the “tech” side of research. I must stress that simply taking differences of average responses is INSUFFICIENT to claim that discrimination occurs.

We must account for ALL differences in characteristics and information useful in the hiring process on both the firm’s and employee’s sides, when calculating the difference in call back rates (or hiring rate for that matter).

Cent-GPS made no attempt at accounting for these issues, much less pointing out that having a sample size of seven CVs, compared to 3,000 CVs, won’t provide them any space to do this correction.

Cent-GPS can claim that responses were different for the seven CVs sent but cannot claim that employers discriminated the candidates the way Khalid and Lee (2016) did.

Now, we talk about the social impact. If anything, this paper has really set back the fight against racism.

Noticed talks about vaccines causing autism? That unfounded claim came from a terribly designed experiment by Andrew Wakefield.

While his medical career was definitively over, the impact of his unethical paper remains in full swing today. Cent-GPS nearly did that with publishing and sensationalising this paper.

In labour economics, the literature and research on job market discrimination is truly voluminous. When published, Khalid and Lee (2016) and Bertrand and Mullainathan (2004) gained celebrity-like status in public policy.

They managed to influence convention and thinking in policymaking. Civil societies can build movements and campaigns based on their research.

Robust research underpins impactful and credible civil and policy movements.

By being dishonest with the results, we undermine researchers who relied on sound methods and movements. Now, when people talk about labour market discrimination, Cent-GPS comes to mind because of two things.

One, they claimed to have found evidence of racial discrimination in labour markets. The way this so-called study was sensationalised really points towards recent coincidental political developments.

Two, further into the future, we may find resistance against buying into research output on discrimination in markets, even if robust methods are used. That’s why there is a highly scrutinising, and painful, process of peer review.

Now, when people think of anti-discrimination policies, the response is likely going to be that “well, the results were not credible in that one paper we know of, why should we trust the next ones?”

Researchers, no matter what, must adhere to the immense bar of peer review and methodological robustness before making any claims. The reach of our output goes far beyond the academia, or the world of think tanks, with very real and long-lasting social ramifications. 

One bad apple really does spoil the bunch.

* All views, thoughts and opinions expressed belong solely to the author, and not necessarily to the author's employer or organisation.