Lecture 9: Recruitment and Discrimination

To choose from a set of candidates, what formal tests can be used?

• Performing job-related tasks

• Tests of physical ability

• Psychological tests of cognitive ability, honesty and personality

• Medical exams

• Drug tests

• Consulting public records → e.g. credit scores, criminal records, social media

When is testing most likely to be useful?

• Test is cheap to administer → costs inc. out-of-pocket costs + disutility of intrusive tests to workers

• Test is accurate → accuracy inc. type-I error (hiring someone unqualified) and type-II error (overlooking qualified candidate)

• Test is valid → actually measures what it’s supposed to measure

• Applicants have high variance in abilities → if applicants similar in terms of likely job performance, testing may be waste of money

• Workers likely to stay for long time → careful selection of workers is investment that pays off only if workers remain w/ firm

• Firing costs are high → when difficult or costly to dismiss bad workers, should invest more in testing candidates

• Alternative to testing NOT practical or effective

How effective are the most widely-used employment screening tests?

• Difficult to know → testing procedures often proprietary - companies who market tests keep them secret

• Testing companies → only have incentive to report positive results

Name direct evidence on the effectiveness of employee testing procedures

Hoffman, Mitchell, Lisa B. Kahn, and Danielle Li. "Discretion in hiring." The Quarterly Journal of Economics 133.2 (2018): 765-800

What were the facts of “Discretion in hiring”?

• Obtained employee performance info from 15 client firms of company who provides online job testing services

• Client firms hiring low-skill service workers → high and costly turnover (key outcome in study)

• HR managers responsible for hiring had access to candidate test results and encouraged to factor them in → BUT could override them w/ own discretion

• Core question: When managers override test recommendations, is employee turnover better or worse?

• Managers saw test results in form of 3 categories → low, middle or high scoring

• Manager considered to have ‘overridden’ test result → if hired from low/middle groups when someone from higher group available

• Authors controlled for differences in applicant pools, hiring times and locations

What were the results of “Discretion in hiring”?

• Managers who frequently overrode test results hired workers w/ higher turnover

• Workers from top-scoring group who were passed over by managers but then hired at later date → lower turnover than workers for whom they were originally passed over

• So managers often overrule test recommendations → biased or mistaken NOT only because they have superior private info

What was the main lesson of “Discretion in hiring”?

• Testing provided significant value over and above previous system relying only on manager discretion

• Note: Result only applied to turnover for low-skilled workers → different results may apply for other testing methods, other productivity measures or other types of workers

How can monitoring and incentivising workers be used as an alternative to testing?

i.e. Use strong ex post incentives and performance pay

• Strong ex post incentives → screening for intrinsic motivation becomes less important + employer won’t care as much about worker ability as worker bears cost of low output (e.g. 100% piece rate)

How can a probationary period be used as an alternative to testing?

• Worker performs job and only retained if performs adequately → probationary periods make sense when:

  • Bad workers can’t do much damage AND

  • Relevant tests are impractical or expensive

How can labour market intermediaries be used as an alternative to testing?

e.g. temporary help agencies, head-hunters, or certifying agencies to pre-screen applicants

• Hiring workers this way is becoming more and more common

How effective are labour market intermediaries?

Stanton and Thomas (2016) → study how small intermediary organisations arose on online labour market oDesk

• These “outsourcing agencies” typically founded by experienced oDesk workers who brought on additional inexperienced workers as affiliates

• Workers affiliated w/ outsourcing agency had much higher job-finding probabilities and wages at beginning of their careers → compared to similar workers w/o agency

How can inducing self-selection by workers be used as an alternative to testing?

• Structuring compensation to attract workers the firm desires → e.g. keeping overall level of compensation fixed but changing structure of pay to induce worker selection

• Advantage (compared to testing) → could have NO direct cost to employer

• Disadvantage → inducing self-selection can only reveal info workers know about themselves whereas tests can sometimes reveal strengths and weaknesses a worker didn’t even know

Describe an example of self-selection by workers

Firm hiring workers for 2 periods

• All workers produce q=6 output in each period w/ NO effort costs

• Firms must pay training costs of c=5 in period 1 only

• Workers have outside option value of v=3

• 2 types of workers:

  • Type-A workers → stay w/ firm for period 2 (given paid at least 3 units)

  • Type-B workers → leave at end of period 1

• Only workers know their own type

In the self-selection example, can the firm design a compensation package that encourages only the Type-As to apply?

• Assume workers paid w₁ and w₂ in periods 1 and 2 → each workers’s Present Value of Utility (PVU):

  • Type-A PVU if accept job = w₁ + w₂

  • Type-A PVU if reject job = 2v

  • Type-B PVU if accept job = w₁

  • Type-B PVU if reject job = v

• Thus workers will accept job offers under following conditions:

  • Type-As accept if w₁ + w₂ >= 2v

  • Type-Bs accept if w₁ >= v

• Type-As care about total amount they earn over 2 periods → BUT Type-Bs care about what they earn in period 1 only

• Thus, firm’s present value of profits (PVπ) → see pic

Under a flat wage profile, what will be a firm’s present value of profits?

i.e. Offers same wage in each period → assume firm offer lowest w that workers would accept (w=3)

• See pic

• Both worker types hired → BUT firm loses money

• Costs of training Type-B who leaves → outweighs returns to training Type-A who stays

Under a rising wage profile, what will be a firm’s present value of profits?

Instead of offering (w₁,w₂) = (3,3) → firm now offers (w₁,w₂) = (2,4)

• Type-A still accepts as w₁ + w₂ = 6 → still as good as outside option of 2v=6

• Type-B now reject offer as w₁=2 less than outside option of v=3

• W/ only Type A being hired → profits now be (see pic) → firm now makes profits

• Thus, firm uses rising wage profile to induce worker self-selection in which only workers who planned on staying apply to job → achieved this w/o changing overall generosity of pay

Why is it complicated to avoid bias when hiring employees?

• In practice, defining and identifying the “best” candidate can be murky exercise → poor info and unconscious biases can affect judgement when evaluators are trying their best to be objective

• Situations arise where most profitable thing to do is NOT necessarily the “right” thing to do

How common is it for employers to take race or gender into account when deciding whether to hire someone?

• To convincingly detect discrimination in hiring → should ideally compare employers’ responses to candidates identical in all other respects except their race/gender

• This is the basis for a widely used tool for measuring discrimination in recruiting → resume audit study

• Resume audit study = researchers randomly submit fictitious resumes that are identical except for prohibited characteristic to large sample of job ads and keep track of no. of callbacks received by various resumes

Name direct evidence of detecting discrimination in hiring

Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review, 94(4), 991-1013

What were the facts of “Are Emily and Greg more employable than Lakisha and Jamal”?

• Conducted resume audit study on racial discrimination in hiring in US

• Submitted fictitious resumes to white-collar job ads (sales, admin support, clerical services, customer services) in Boston and Chicago

• To manipulate applicants’ perceived race → resumes randomly assigned African-American names (e.g. Lakisha, Jamal) or white-sounding names (e.g. Emily, Greg)

What were the results of “Are Emily and Greg more employable than Lakisha and Jamal”?

• White names received 50% more callbacks for interviews than identical resumes w/ black name

• Authors showed this did NOT result from employers inferring applicant’s social class from their name

• Racial gap in callbacks did NOT vary across occupation, industry and employer size

Explain how conscious discrimination can occur due to tastes or preferences of employer/recruiter

• Preferences for interacting w/ members of one’s own ethnic group or distaste for employing members of another group have been openly expressed in no. of countries and periods of time

  • Explain large effects detected by Bertrand and Mullainathan (2004) during period when no longer socially acceptable to express such preferences

Explain how conscious discrimination can occur due to tastes or preferences of firm’s customers/employees

• e.g. Unbiased recruiter in India may be reluctant to hire low-caste worker for sales job → if they expect customers to be averse to interacting w/ low-caste salespeople

• Rational employer may be reluctant to hire openly gay worker → if employer expects this to cause serious disruption among existing workers

Explain how conscious discrimination can occur due to unbiased employer beliefs about the relative productivities of two groups

• e.g. Common for Chinese electronics producers to actively prefer women as assembly workers → as women on avg have better fine motor skills (Baker and Cornelson, 2016)

• Here, publicly observable characteristic (e.g. race, age, gender) serves as easy but imprecise measure of skill employer values → type of behaviour known as statistical discrimination

• i.e. When employer hires individual from one group over another group based on statistically unbiased belief that avg productivities of two groups are different for job in question

How can we tell whether employers are engaging in statistical discrimination as opposed to taste-based aversion to a particular group?

• If discrimination statistical → should diminish when more accurate info about individual workers’ productivities/qualifications made available

• Bertrand and Mullainathan (2004) tested for statistical discrimination by creating 2 types of resumes for each race → higher-quality resumes had more labour market experience, fewer gaps between jobs, more certifications and more likely to have email address

• If employers inferring such higher qualifications when seeing white-sounding name on resume → providing this extra info would reduce callback gap

• BUT employers’ tendency to favour white-sounding names was higher among high-quality resumes → suggests statistical discrimination NOT main motive for callback gap

Explain how conscious discrimination can occur due to biased employer beliefs about the productivities of two groups

• In statistical sense, beliefs biased when employer’s estimate of avg ability in group differs from group’s true avg ability

• Historically, common wisdom about types of jobs women are capable of performing → quite different from what we know today → suggest earlier beliefs was inaccurate

What is a way of measuring unconscious discrimination?

Implicit Association Test (IAT) - computer-based test measuring strength of person’s unconscious association between 2 broad concepts (e.g. “good” or “bad” on one hand, Republican or Democrat on other)

• Test based on fact that ppl can categorise objects or concepts more quickly when concepts feel “naturally” associated to them

• Tests routinely reveal more natural for both men and women to associate men w/ careers and women w/ families than other way around

What evidence of unconscious discrimination did Dan-Olof Rooth (2010) find?

• Swedish recruiters exhibiting unconscious biases and stereotypes toward Arab-Muslim men according to IAT → substantially less likely to invite those men to interviews

• Recruiters exhibiting strongest unconscious biases were often ones unaware of them or unwilling to openly report bias to survey researcher → suggests even researchers who don’t wish to take race or gender into account often do so unconsciously

Why is discrimination based on employer/recruiter tastes, biased beliefs or unconscious factors likely to harm the group being discriminated against and reduce profits?

• These criteria involve selecting workers for reasons other than qualifications and likely job performance → recruiters who use them will hire less-qualified workforce

• Firms engaging in these practices will underperform and be forced out of competitive markets

• As these forms of discrimination reduce profits → unsurprising many high-performing firms try hard to limit effects of these factors on recruiting, evaluation, retention and pay practices

Why is discrimination based on customer or coworker tastes likely to harm the group being discriminated against yet raise profits?

• e.g. 1950-60s - owners of Southern textile mills tried to hire black workers → risked major disruption, morale problems and even sabotage from exclusively white workforce

• Thus, business owners face a trade-off between doing “the right thing” and maximising their own profits

Why does statistical discrimination hurt the group being discriminated against but has more complex effects on profits?

• Short-run → can raise profits by raising avg suitability of workers hired for a job

• Longer-run → can be harmful to both firms and society as a whole by reducing worker incentives to improve their qualifications and skills

• If workers know they will be judged based on group membership rather than based on personal skill levels → workers’ incentives to work hard to improve personal skills reduced

• Reasoning applies to both workers favoured by discriminatory policies and those disfavoured by it:

  • Disfavoured gets discouraged (and angry)

  • Favoured gets lazy (and entitled)

Why does discrimination of all forms reduce productivity through ‘psychological’ mechanisms?

• Stereotype threat effects → when simply being perceived by someone else via lens of negative stereotype reduces performance

• Pygmalion effects → positive effect of leader’s high expectations on group’s performance

• Golem effects → negative effect of low expectations

Thus, evidence exists that low expectations of group’s performance can be self-fulfilling prophecy → discrimination can change worker behaviour in way that fulfils employers’ expectations of individual or group

In Steele and Aronson’s (1995) lab experiment, how did group-based expectations affect employee performance?

• Simply indicating rate before cognitive test reduced African-Americans’ performance

• Effect much weaker when subjects convinced test not being used to measure their abilities → suggests pressure and anxiety from fear of being judged (or fulfilling a negative stereotype) was cause of reduced performance

In Glover et al.’s (2016) field study, how did group-based expectations affect employee performance?

Used IAT to measure managers’ bias against African-origin workers

• When working w/ biased manager → minority cashiers were absent more often, spent less time at work, scanned items more slowly and took more time between customers

• BUT when assigned unbiased manager → minority cashiers performed significantly better than majority cashiers

• Thus, discrimination can cause poor employee performance → NOT just reflect employer’s expectation of poor performance

In Lavy and Sand’s (2015) study, how did group-based expectations affect employee performance?

Estimated effect of Israeli primary school teachers’ gender biases on student academic achievements later in life

• Measured teacher bias by comparing avg marking of boys and girls in classroom exam to respective means in blind national exam marked anonymously

• Being assigned to more gender biased teacher in primary school affected students’ subject choices and course performance in high school w/ potential implications for college majors and careers

• Thus, encountering discrimination at one point in time caused reduced achievement at later date

How can bias in hiring be reduced by changing the decision environment?

Recruiters sifting through large no. of resumes under time pressure → susceptible to mental short cuts (e.g. stereotypes, emotional “gut” reactions)

• Psychologists formalised this idea as difference between System 1 and System 2 cognitive functioning

• System 1 → our intuitive system - fast, automatic, effortless, implicit and emotional

• System 2 → our reasoning system - slower, conscious, effortful, explicit and logical

• As System 1 decisions are more susceptible to unconscious bias, simple often-overlooked way to reduce bias in hiring → change decision environment to give recruiters access to time and quiet needed to make System 2 decisions

What evidence supports reduced bias in hiring by changing the decision environment?

• Large body of psychological research examines decision quality of judges → judges make lower quality decisions under conditions that are:

  • Tiring - e.g. working long hrs

  • Stressful - e.g. heavy, backlogged or diverse caseloads

How can bias in hiring be reduced by monitoring the recruiters?

• e.g. Firms may periodically assess recruiters’ performance by using objective measures of workers they hire

• These assessments can help companies train effective recruiters and if necessary, reassign less effective recruiters to other tasks

How does Parsons et al.’s (2011) study of MLB referees show monitoring decision-makers can reduce bias?

Referees slightly favour pitchers of their own race by expanding the strike zone

• Referee biases only present in ballparks w/o computerised QuesTrec cameras → which provide more objective and accurate measures of each pitch’s location

• In parks w/o cameras, biases strongest in games w/ low attendance and for pitches not pivotal for an at-bat → both situations where referee’s decisions are being scrutinised less closely

• Thus, closer monitoring of decision-makers can improve quality of decisions by eliminating conscious or unconscious racial bias

How can bias in hiring be reduced by using blind recruiting?

Involves concealing worker’s gender, race or age from recruiters at certain stages of recruitment process

• Goldin and Rouse’s (2000) study of auditions for major US symphony orchestras → blind recruiting success story

• From 1970-80s, most major orchestras introduced screens in recruiting process that concealed identity of musician during auditions

• Study used fixed effects regressions to control for each musician’s own quality → i.e. comparing same musician’s success w/ and w/o screen

• Found screen increases probability woman advanced out of prelim round and greatly increases likelihood female contestant will be winner in final round

What evidence is there for mixed results from blind recruiting?

Behagel et al. (2015) → large study of French companies found anonymous applications reduced hiring of minority candidates

• Result potentially due to companies actually trying to favour minorities when applicants’ minority status was not concealed

How can blind recruiting policies backfire?

Restricting employers’ access to certain pieces of info may induce employers to rely more on other worker characteristics that signal the missing info

• e.g. Doleac and Hansen’s (2016) study of “ban the box” policies adopted by some US states → policies prevent employers from conducting background checks until late in job application process

• Policies actually had opposite of intended effect → reduced employment rate of young low-skilled black men by 5.1% and young low-skilled Hispanic men by 2.9%

• “Blinding” some attributes (like criminal history) may incentivise employers to rely on other attributes (like race) to fill in hidden info