How much discrimination in hiring?

A recent front-page headline in The Guardian newspaper proclaimed that ‘Minorities in UK face ‘shocking’ jobs bias’.

According to the report’s author Haroon Siddique, research had found that:

“…applicants from minority ethnic backgrounds had to send 80% more applications to get a positive response from an employer than a white person of British origin.”

It was further reported that:

“…discrimination against black Britons and those of South Asian origin – particularly Pakistanis – [was] unchanged over almost 50 years.”

These are alarming and demoralising claims, certainly. But behind them lies a great deal of uncertainty and nuance, and a slightly less negative picture than what one might infer from the headlines. The source for the article was a new study carried out by Dr Valentina Di Stasio and Professor Anthony Heath for a project conducted by the Centre for Social Investigation at Nuffield College, as part of a much wider country-comparative project funded by the European Commission.

The methodology used was to send applications to advertised jobs using fictitious CVs. Some were in the name of people with typically ethnic minority names while some were in the name of those with typically ethnic majority names. Ratios were then constructed of the share of successful majority applicants to successful minority applicants – so-called call-back ratios. A ratio of 1 means equality of treatment while a figure greater than 1 denotes discriminatory treatment in favour of whites and at the expense of the minority group in question. Such experiments are known as ‘correspondence tests’.

In fact, reading Nuffield’s report reveals that it claims ethnic minority people would need to send 60 per cent more applications in order to receive as many call-backs as the majority, not 80 per cent. The Guardian is out by 20 percentage points. (Perhaps the 80 per cent is referring to all non-white minorities and excluding the white ones, but I couldn’t see this figure in the source document.)

Little in the Nuffield report is provided to substantiate the claim of no change since the 1970s, particularly in the case of ethnic Pakistanis. All we have is:

“To put these findings in perspective, in a meta-analysis summarizing results from several field experiments conducted in Britain, we found that there has been virtually no change over time in the level of discrimination faced by Pakistanis: the studies from the 1970s and from the 21st century show almost identical patterns of discrimination despite being almost forty years apart.”

No more detail is given and there is no reference for this claim. No data from studies from the 1970s on call-back rates are given. The only meta-analysis referenced covers studies dating from 1990 to 2015.

I got in touch with Professor Heath and he directed me to his book Social Progress in Britain, which was published last year and is the basis for the CSI’s claim of no change. Inside is an analysis of all the correspondence tests carried out in this country from 1969 to 2016/17.

This analysis forms a central plank of the book – it is based on the ‘five giants’ identified by the Beveridge Report of 1942: want, ignorance, squalor, sickness, and idleness. It looks to appraise how much progress has been made in combating them. To this list, Heath proposes two more – unfairness and discord. The analysis of correspondence tests constitutes his key test of how fair we are regarding race and ethnicity, and how much progress has been made.

Heath’s conclusion is:

“Field experiments of racial discrimination strongly suggest that the non-discrimination principle is violated to much the same degree in 2017 as it was at the time of the first comparable field experiment nearly fifty years earlier.”

There are some objections that can be made to this bold claim. The first is that the evidence can be read to show a decrease in the discrimination rates for Pakistanis. From the 1969 study conducted by Roger Jowell and Patricia Prescott-Clarke, I was able to compute a call-back ratio for Pakistanis of 2.2. Heath’s latest study gives a figure of 1.7. However, the 1969 study was focused on ‘white-collar’ jobs. Heath’s most recent experiment gives a figure of 1.4 for ‘high-skilled’ occupations. That would be a decline of 36 per cent, assuming we are comparing like with like, and if true then today we would be much closer to parity of treatment than before.

Let us take a look more closely at Heath’s assembled evidence. The graph below is reproduced from his book. A closer inspection would suggest that while call-back rates for white minorities have remained constant, they appear to have worsened for black minorities. At the same time, one could easily interpret the trend for Asians either as one of decline as well as just random fluctuations without any clear direction – stability.

Is it really credible that discrimination should have increased for the West Indian/black Caribbean group? The experimental data show a rise from near-enough parity in 1969 to a call-back ratio of about 1.5 in the early 1970s, to just under 2 in 2008/9.

Figure 1. Correspondence test results – reprinted from Social Progress in Britain by Anthony Heath (2018) published by OUP

The chief problem however lies in the methodology. There is a distinction between internal and external validity of research design. The former refers to the extent to which the experiment or research design is free from errors and differences are accounted for only by the variable the researcher is interested in; in this case ethnicity. The latter is referring to the extent to which we can generalise the findings to the wider population. That is to say, can the experimental finding of a call-back ratio of 1.6 be said to represent what goes on in the real world?

The internal validity of the experiment is high since all other things between the fictional applicants are the same. However, I would argue the external validity is low. This is because all the different studies have slightly different sampling methodologies and are often referring to different sections of the labour market, cities or regions.

In order to assess the level of discrimination, one would need a representative sample of vacancies. This would be achieved by random sampling. Correspondence tests however work on the basis of sampling only those jobs that the researcher becomes aware of. The Nuffield researchers made use of an un-named website while the researchers in 1969 used local and national newspapers. Can it really be said with any surety that the different methods will be capturing the labour market in the same light?

Dr Didier Ruedin, an expert on correspondence tests, explained to me via email that researchers make use of systematic sampling in these types of studies. This is a form of random sampling whereby you sample every fourth, fifth, tenth, or whatever. However, this is only random sampling if you are choosing from a pool which has no selection effects. As Heath himself concedes in his book, all these studies target different sections of the labour market. It seems perfectly reasonable then to reconcile a decline (or stability) in discrimination for Pakistanis with a rise for West Indian/black Caribbeans, by saying that we are just observing measurement error. At the very least, we cannot discount the possibility that the results are confounded in this way. There is too much uncertainty.

Thus, I would argue that all these correspondence tests do is prove the existence of discrimination on grounds of race, ethnicity, and religion, but they do little to measure its extent, in terms of the risks that minorities face. Comparisons across time and place tell us very little.

What other evidence might we consider? The British Social Attitudes Survey has asked over numerous years, white people if they would object to having a black boss. This can be seen as a proxy for the extent to which whites are willing to discriminate against minorities. The most recent data come from 2006 and show that 3 per cent would mind a lot while 6 per cent would mind a little. These figures are down from 8.5 and 11.7 per cent respectively from 1983. (Without being explicit, Heath acknowledges this evidence in his book although it has no bearing on his conclusion of no change.)

Based on these data, I estimate that in 2019, just over 1 per cent of the white population would mind a lot having a black boss.

Figure 2. Share of whites who would mind a lot having a black or West Indian boss (with projection) – British Social Attitudes survey

However, such evidence might be confounded by a social desirability bias – perhaps, people are only less willing to admit to such feelings. Without overstating the matter, can we readily assume such sentiments translate into practice?

Another avenue would be to look at unemployment statistics. Data from the Labour Force Survey show that differences between minority groups and the white British majority are in decline. For instance, the difference in unemployment rates between Pakistanis/Bangladeshis and the white British was 9 percentage points in 2004, falling to 6 in 2017. For black people the same figures fell from 9 to 5. For Indians there has not been much change, although there was less difference to begin with – from 3 percentage points in 2004 to 2 in 2017.

However, these figures only evidence outcomes, not explanations. While regression analysis might reveal these gaps to be attributable to other factors like age and region, it will never be able to arrive at a causal estimate of discrimination since you cannot account for every possible alternative explanation. Such evidence is at best, only ever suggestive but not conclusive.

We might also consult ethnic minority people themselves via opinion polls. Survation polling on behalf of the think tank British Future found that 33.3 per cent of non-white ethnic minorities think there is less discrimination when applying for jobs than there was 50 years ago. 26.3 per cent think there was more, 28.5 per cent think it is about the same with 11.9 per cent admitting they did not know. So roughly evenly split.

Older cohorts are more likely to believe it has decreased as seen in the graph below. While one could argue that they are the best witnesses since they have been alive the longest and experienced Britain as it became more ethnically varied, the truth is these are just measures of perception. An elderly minority person might have a good feel for what applying for jobs was like in 1968, as well as some upsetting personal experiences, but will have less first-hand experience of today’s market. The converse will apply to the youngest – they might be more likely to think things are worse today, but they have neither direct experience of the past, nor an accurate and objective measurement with which to compare.

Figure 3. Non-white minority perception of racial discrimination when applying for jobs compared with 50 years ago – Survation/British Future

So where does this leave us? I don’t agree with Professor Heath that racial/ethnic discrimination is the same as it was some 50 years ago, because I don’t think he has come up with accurate measurements that can be compared across years. While older minority cohorts are likely to say things are better, they’ve no objective measurement either. And while white people are more likely to profess to being more accepting of minorities these days, we can’t assume that translates into practice.

In a blog for the British Academy, to accompany the research, Professor Heath calls for a “new initiative” to tackle discrimination in hiring. Bearing in mind that employers are increasingly using name-blind applications and similar ‘nudge’ approaches, it is not immediately evident what he has in mind. Quotas though are not the answer since there is no way of knowing just how many minority people an employer should hire since we can’t adequately measure preferences and availability of workers with any real precision. Just because we know discrimination is going on, doesn’t mean we know what the expected levels of minority representation should be. Trying to force the issue may make matters even worse.

While my suspicion is that things will have got better, I can’t evidence it and so I think I will just have to reserve judgement. My investigations began with a startling headline in a newspaper. I then found out that the newspaper article was potentially misleading. I further found out that the document that was the basis for it had not substantiated its claims and was insufficiently referenced. Finally, the evidence that was the basis for this document was both more nuanced and also did not hold up to scrutiny.

I checked other data sources but couldn’t advance any further on whether or not discrimination in employment had improved or not. The truth is we know it goes on, but we’ve no good way of measuring by how much. I wish academics and journalists would do more to convey this uncertainty since needlessly damning headlines will have real impacts. It makes it ever so slightly harder to convince those on the margins that there is a place for them in British society if you have newspapers crying blue murder without good reason.

 

 

 

 

 

 

Richard Norrie Written by:

Be First to Comment

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.